=, K U R ZW EEL

24

H G E o FE Pew IE Ee)
Ge NF MACHINES
 agnv19Na ‘NOQGNO]4
SLLasnuovsesvw ‘adaruagwyo
Pei: ee a ee ee ©
The Age of Intelligent Machines
 © 1990 Massachusetts Institute of Technology

Al rights reserved. No part of this book may be repro-

duced in any form by any electronic or mechanical
means (including photocopying, recording, or
information storage and retrieval) without permission
in writing from the publisher.

This book was printed and bound by Dai Nippon

Printing Co., Japan.

Library of Congress Cataloging-in-Publication Data

Kurzweil, Ray.
The age of intelligent machines/Raymond Kurzweil.
p. cm.
Includes bibliographical references.
ISBN 0-262-11121-7
|. Artificial intelligence, |. Title,
0335,K87 1990
006.3—1c20 89-13606
cIP
PI
°
Acknowledgments xi

Prolog: The Second Industrial Revolution 3

 Ce ee te ee OF ARTIF nea
ea oe 4

What Is Al, Anyway?

The Usual Definition
The Moving-Frontier Definition
What Is Intelligence, Anyway?
Evolution as an Intelligent Process

Philosophical Roots
Plato and the Platonists 25
The Enlightenment 29
The Logical Positivists and the Existential Reaction 32
The Debate Goes On 36

A Platonic Dialogue on the Nature of Human Thought Raymond Kurzweil 40

The Age of Intelligent People Margaret Litvin 46
Can Machines Think? Daniel C. Dennett 48
Can Computers Think? Mitchell Waldrop 62
Growing Up in the Age of Intelligent Machines: Reconstructions of the Psychological
and Reconsiderations of the Human Sherry Turkle 68
A Conversation between a Human Computer and a Materialist Philosopher Blaine Mathieu 74
ELIZA passes the Turing Test Seymour Papert 78
A Coffeehouse Conversation on the Turing Test Douglas R. Hofstadter 80

Mathematical Roots
Russell's Paradox 104
The Five Contributions of Turing 109

The Formula for Intelligence

Unifying Formulas: The Goal of Science

The Sea of Logic and the Turing Machine
The Recursive Formula and Three Levels of Intelligence
Other Approaches to Modeling the Software of Intelligence Random Nets,
Pandemonium, and Trees
The Formula of Life as a Formula of Intelligence

Aor B: The Basis of Intelligence? Raymond Kurzweil

 ico
Mechanical Roots
Early Automata and Calculating Engines 159
Charles Babbage and the World's First Programmer 165
The Practical Path 169

Electronic Roots 175

The First Computer 1

S
Welcoming a New Form of Intelligence on Earth: The Al Movement 189

Thoughts about Artificial Intelligence Marvin Minsky 214

re ed THE MOVING FRONTIER

Pattern Recognition:
The Search for Order 223
Vision 223
The Real World 247

A Personal Postscript Raymond Kurzweil 272

____TheSearchforKnowledge Z = - a = 283
Knowledge and Expert Systems 283
Putting Knowledge to Work 294
Language: The Expression of Knowledge 303
Putting It All Together: The Age of Robots 312
An International Affair 322

Knowledge Processing: From File Servers to Knowledge Servers Edward A. Feigenbaum 324
| An Expert System for Automotive Diagnosis Jeff Pepper 330
The Significance of Fifth-Generation Computer Systems K. Fuchi 336
Intelligent Knowledge-Based Systems: Al inthe U.K. Brian W. Oakley 346

The Science of Art 35

The Musical Arts 35

The Visual Arts 355
The Literary Arts

A (Kind of) Turing Test Raymond Kurzweil

Brother Giorgio's Kangaroo Harold Cohen
Artificial Intelligence and Musical Composition Charles Ames 8E
All Work and No Play Makes HAL a Dull Program Michael Lebowitz
. The Mechanics of Creativity Roger Schank and Christopher Owens
 PART THRE IONS OF THE FUTURE

Visions
Scenarios
Breakthroughs

Fairy Tales Allen Newell

The Impact On...

Employment and the Economy
Education
Communications
Warfare
Medicine
The Handicapped
Music
Politics
Our Concept of Ourselves

The Social Impact of Artificial Intelligence Margaret A. Boden

ATechnology of Liberation George Gilder

Postscript

Chronology

Notes

Bibliography and Suggested Readings

Glossary

Index
Heknowledgments
| would like to express my deep gratitude to the many persons who have provided
inspiration, prodding, ideas, suggestions, comments, criticism, insight, contacts,
logistics, software, pictures, and multifarious other forms of help and assistance. In
particular, | would like to thank the following:

My wife Sonya for her enthusiasm and loving patience through the many unexpected
twists and turns of the creative process
My mother Hannah and my father Fredric for having encouraged my early interest in an
unproven field
My son Ethan and my daughter Amy for teaching me everything | know
+ Frank Satlow for his expert editorial guidance and encouragement
Lou Jones for his beautiful and creative photographs and willingness to chase after
intelligent machines and even more intelligent people around the world
+ Rose Russo and Robert Brun of KelGraphic for turning my scribbled sketches into
stunning illustrations
+ Laurel Anderson for finding impossible-to-find historical photographs and illustrations
Alison Roberts for her capable, earnest, and diligent administrative support—she was
the “chief operating officer” of the “book project”
+ My wonderfully proficient research team—Margaret Kennedy, Anand Bodapati, Marcia
Ross, Kathy Duffin, and Terry Ehling—for their skills in finding long forgotten facts of
remarkable relevance and for their irreplaceable assistance with references and the
chronology
Wendy Dennis for her energetic help in launching the research
+ Alan Thwaits for his incisive copy editing
Diane Jaroch for her elegant design
Don Byrd for his many insightful comments and ideas
+ Abby Joslin, Lauri Murphy, and Pat Camarena for their creative suggestions and
enthusiastic administrative assistance

+ George Gilder for many stimulating discussions

Ted Johnson and Aaron Kleiner for their support and encouragement
Martin Schneider and Robert Richter for their many creative contributions and close
collaborations on the Age of Intelligent Machines film that accompanies this book
+ My readers—Glenn Akers, Charles Ames, Abby Joslin, Don Byrd, Pat Camarena,
Frances Davis, Wendy Dennis, Terry Ehling, Francis Ganong, George Gilder, Richard
Goldhor, Ted Johnson, Robin Kinkead, Aaron Kleiner, Ethan Kurzweil, Elizabeth Meade,
Bob Moog, Lauri Murphy, Alison Roberts, Steven Rothman, Frank Satlow, Martin
Schneider and Mitchell Waldrop—for innumerable and invaluable comments and
criticisms, some contradictory, all valid, and all of great value
+ The MIT Press blind readers, whoever and wherever they may be
+ Dan Dennett, Sherry Turkle, Marvin Minsky, Seymour Papert, Ed Feigenbaum, Harold
Cohen, Margaret Boden, and Allen Newell for their ideas and insights
+ The authors of the contributed articles: Margaret Litvin, Dan Dennett, Mitchell Wal-
drop, Sherry Turkle, Blaine Mathieu, Seymour Papert, Doug Hofstadter, Marvin
Minsky, Ed Feigenbaum, Roger Schank, Christopher Owens, Jeff Pepper, K. Fuchi,
Brian Oakley, Harold Cohen, Charles Ames, Michael Lebowitz, Allen Newell, Margaret
Boden, and George Gilder
+ Digital Equipment Corporation for its generous sponsorship of The Age of Intelligent
Machines Exhibition
+ The Boston Museum of Science; the Franklin Institute, Philidelphia; the Science
Museums of Charlotte; the Fort Worth Museum of Science and History; the California
Museum of Science and Industry, Los Angeles; the Science Museum of Minnesota,
St. Paul; the Museum of Science and Industry, Chicago; and the Center of Science and
Industry, Columbus, for hosting The Age of Intelligent Machines Exhibition, and
especially the Boston Museum of Science for producing it
+ Finally, all the scientists, researchers, industrialists, inventors, entrepreneurs, market-
ers, public relations consultants, administrative assistants, executives, and librarians
who were willing to talk to me and who, when not taking time to talk to book writers,
reporters, and other chroniclers of artificial intelligence, are busy creating the age of
intelligent machines

xatt
THE AGE OF INTELLIGENT MACHINES
Joral. (Photo by Lou Jones)
The Second Industrial Revolution

On May 26, 1733, John Kay, a twenty-nine-year-old inventor, received the news that
the English Patent Office had awarded him a patent for his New Engine for Opening
and Dressing Wool, now known as the flying shuttle.' To Kay this was good news, for
he hoped to start a small business supplying his new machine to the burgeoning
English textile industry. What neither Kay nor his contemporaries realized at the time
was that his innovation in the weaving of cloth represented the launching of the
Industrial Revolution.
Like many innovations that come at the right time in the right place, the
flying shuttle caught on quickly. Unfortunately, Kay was more talented as an inventor
than as a businessman, and after losing most of his money in litigation attempting to
enforce his patent, he moved to France, where he died in poverty.
Kay nonetheless had a lasting impact. The widespread adoption of the flying
shuttle created pressure for the more efficient spinning of yarn, which led to Sir
Richard Arkwright’s Cotton Jenny, patented in 1770. In turn, machines to card and
comb the wool to feed the new mechanized spinning machines were developed in
the 1780s. By the turn of the century all aspects of the production of cloth had been
automated. The cottage industry of English textiles was rapidly being replaced by
increasingly efficient centralized machines.?
Good ideas catch on and innovators in other industries took note of the
dramatically improved productivity that mechanization had brought to English textiles
The process of industrialization spread to other industries and to other countries.
Major innovations that followed included Ford's (1863-1947) concept of mass
production and Edison's (1847-1931) harnessing of the electron. Ultimately Europe,
the United States, Japan, and other parts of the world shifted from an agrarian and
craft economy to one dominated by machines. The succession of increasingly
efficient generations of automation has continued to this day. The changing patterns
of production and employment, together with related scientific advances, have had
dramatic effects on all aspects of modern life, profoundly affecting our social, cultural,
educational, economic, and political institutions.
The Industrial Revolution was not without its controversies. Emerging,
appropriately enough, from the English textile industry, the Luddite movement was
founded in Nottingham in 1811.5 The movement posed a serious and violent
 challenge to what its members perceived as a diabolical danger to the textile
workers’ livelihoods. In one sense, the fears of the Luddites were accurate. Jobs
they thought were threatened by the new machines did indeed disappear. At the
same time, however, new jobs were created as new industries emerged and
economic activity increased, although this was often of little consequence to those
displaced. The Luddite movement itself was ended within a decade of its founding
due to a combination of repression and prosperity, although its name has remained
very much alive as a symbol of a still lingering issue.* Automation versus jobs is still a
particularly controversial issue in Europe, where it has had a noticeable impact on the
rate at which new automated technologies are introduced. In the United States the
issue simmers beneath the surface of political debate but rarely affects the pace of
change. In Japan the issue is virtually unknown, due partly to a tradition in which the
prosperous “first tier” industrial corporations provide lifetime employment, although
employment guarantees are generally not extended by the less powerful “second
tier” corporations and cottage industries.
Let us examine the Luddite issue for a moment. It is generally acknowl-
edged that new jobs result as new industries are created by the advent of automa-

Ned Ludd disguised as a woman.

Although his Luddite movement had
little impact at the time, it has
remained a symbol of opposition to
machines and automation.
(Supplied by Historical Pictures
Service, Chicago)
tion. The critical question then becomes, How do these jobs compare to the jobs that
are displaced? In particular, for every ten jobs that are eliminated by automation, are
we creating twelve new jobs or eight? Do the new jobs pay more or less than the
older ones? Are they more or less fulfilling? What about those who are displaced; can
they be retrained for the new jobs? Are they?
We now have over a century of extensive industrialization to look back on,
and an examination of some clear economic trends over the past century can provide
insights into at least some of the above questions. With regard to the numbers of
jobs, the answer is closer to twelve than eight. In 1870 only twelve million Ameri-
cans, representing 31 percent of the population, had jobs.° By 1985 the figure rose to
116 million jobs held by 48 percent of the population.® This substantial increase in the
number of jobs occurred despite the dramatic shift away from the labor content of
agriculture. In 1900 more than a third of all American workers were involved in food
production.” Today Americans are better fed and America a major food exporter, with
only 3 percent of the workforce involved.2
In the economic power of jobs we see the most dramatic change. The
gross national product on a per capita basis and in constant 1958 dollars went from
$530 in 1870 to $3,500 in 1970.° There has been a similar change in the actual
earning power of the available jobs. This 600 percent increase in real wealth has
resulted in a greatly improved standard of living, better health care and education,
and a substantially improved ability to provide for those who need help in our society.
At the beginning of the Industrial Revolution life expectancy in North America and
northwestern Europe was about 37 years. Now, two centuries later, it has doubled.
The jobs created have also been on a higher level and indeed much of the
additional employment has been in the area of providing the higher level of education
that today’s jobs require. For example, we now spend ten times as much (in constant
dollars) on a per capita basis for public school education than we did one hundred
years ago."° In 1870 only 2 percent of American adults had a high school diploma,
whereas the figure is 76 percent today.'’ There were only 52,000 college students in
1870; there are 7.5 million today.’?
Attempts to project these trends into the future, including a recent study
by the Institute for Economic Analysis at New York University, where a detailed
input-output model of the U.S. economy was studied, indicate a continuation of
these same trends.'? While there will be ebbs and flows in economic development,
the trend over the next two decades indicates that employment and productivity will
continue to increase, as will the average educational level of the population. The
study indicated, for example, that the share of jobs going to professionals will
increase from 15 percent today to 20 percent by the end of the next decade, with
engineers and teachers accounting for virtually all of the increase.'*
From these trends it would seem that the concerns of the Luddite move-
ment are not well founded. From a macroeconomic point of view, it is clear that
automation and other related technological advances have fueled over a century of
dramatic economic development. There are nonetheless difficult, if often temporary,
dislocations that result from rapid technological change.
'® As our smokestack indus-
tries contract, workers with one set of skills do not necessarily find it easy to develop
new careers. With the pace of change accelerating, we as a society need to find a
Second Industrial Revolution

i ==
(i
oa
The first Industrial Revolution. Power
and wealth went to those who
controlled natural resources and
labor. Shown is an interior view of
Walcott Brothers manufacturing
facility for button hole cutters, 1856.
(Supplied by North Wind Picture
Archives)
 way to provide viable avenues for displaced workers to reenter the economic main-
stream with something more than a new dead-end job.
As profound as the implications of the first Industrial Revolution were, we
are now embarking on yet another transformation of our economy, based once again
on innovation. The Industrial Revolution of the last two centuries—the first Industrial
Revolution—was characterized by machines that extended, multiplied, and leveraged
our physical capabilities. With these new machines, humans could manipulate
objects for which our muscles alone were inadequate and carry out physical tasks at
previously unachievable speeds. While the social and economic impact of this new
technology was controversial, the concept of machines being physically superior to
ourselves was not. After all, we never regarded our species as unequaled in this
dimension. Jaguars can run faster than we can, lions are better hunters, monkeys are
better climbers, whales can dive deeper and longer, and birds are better fliers—
indeed, without machines we cannot fly at all
The second industrial revolution, the one that is now in progress, is based
on machines that extend, multiply, and leverage our mental abilities. The same con-
troversies on social and economic impact are attending this second great wave of
automation, only now a new and more profound question has emerged. Though we
have always regarded our species as relatively mediocre in physical capacity, this has
not been our view with regard to our mental capacity. The very name we have given
ourselves, Homo sapiens, defines us as the thinking people. The primary distinction
in our biological classification is the ability of our species to manipulate symbols and
use language
Before Copernicus (1473-1543), our “species centricity” was embodied in
a view of the universe literally circling around us in a testament to our unique and
central status. Today our belief in our own uniqueness is a matter not of celestial

Pre-Copernican cosmology with

man at the center. (From Gemma
Phrysius's Cosmographis, 1539, in
the British Museum, London)
relationships but of intelligence. Evolution is seen as a billion-year drama leading
inexorably to its grandest creation—human intelligence. The spectre of machine
intelligence competing even tangentially with that-of its creator once again threatens
our view of who we are.
This latest revolution, based on machines that expand the reach of our
minds, will ultimately have a far greater impact than the revolution that merely ex-
panded the reach of our bodies. It promises to transform production, education,
medicine, aids for the handicapped, research, the acquisition and distribution of
knowledge, communication, the creation of wealth, the conduct of government, and
warfare. The cost-effectiveness of the key ingredients in our new technological
base—computers and related semiconductor technology—is increasing at an expo-
nential rate. The power of computer technology now doubles (for the same unit cost)
every 18 to 24 months.'®
Unlike some revolutions, this latest transformation of our industrial base
will not arrive after one brief period of struggle. It will be a gradual process, but it is
one already under way. The potential exists to begin to solve problems with which
the human race has struggled for centuries. An example is the application of com-
puter technology to the needs of the handicapped, a personal interest of mine. It is
my belief that the potential exists within the next one or two decades to greatly ame-
liorate the principal handicaps associated with sensory and physical disabilities such
as blindness, deafness, and spinal cord injuries. New bioengineering techniques that
rely on expert systems and computer-assisted design stations for biological modeling
are fueling a new optimism for effective treatments of a wide range of diseases,
including genetic disorders.'” The increase in real per capita wealth—600 percent in
the past 100 years—is projected to continue.'® There are many other examples of
anticipated benefit.
The potential for danger is also manifest. We are today beginning to turn
over our engines of war to intelligent machines, whose intelligence may be as flawed
as our own.'? Computer technology is already a powerful ally of the totalitarian.
The advent of intelligent machines is altering global trade relationships. A
remarkable aspect of this new technology is that it uses almost no natural resources.
Silicon chips use infinitesimal amounts of sand and other readily available materials.
They use insignificant amounts of electricity. As computers grow smaller and smaller,
the material resources utilized are becoming an inconsequential portion of their value.
Indeed, software uses virtually no resources at all. The value of the technology lies
primarily in the knowledge governing the design of the hardware, software, and data
bases that constitute our intelligent machines, and in the ability to continue advancing
these designs. This decreasing importance of material resources has allowed Japan,
a country very poor in natural resources but rich in knowledge and expertise, to
become one of the two wealthiest nations on the planet. There is the potential for
emerging nations to largely skip industrialization and develop postindustrial societies
based on an information economy.”° While the first Industrial Revolution increased
the demand for and the value of natural resources, the second industrial revolution is
doing the opposite.
 In the case of computer software, it is apparent that one is paying for the
knowledge inherent in the design and not for the raw materials represented by the
floppy disk and users’ manual. What is sometimes less apparent is that the same
economic model holds for most computer hardware as well. An advanced chip
generally costs no more to produce than a floppy disk. As with a software program,
the bulk of the cost of a chip is neither raw materials nor manufacturing labor, but
rather what accountants call amortization of development, and what philosophers call
knowledge.
It is estimated that raw materials comprise less than 2 percent of the value
of chips (which is about the same estimate as for software) and less than 5 percent
of the value of computers. As our computers become more powerful, the percentage
of their value accounted for by raw materials continues to diminish, approaching zero.
It is interesting to note that the same trend holds for most other categories of
products. Raw materials comprise about 20 percent of the value of musical instru-
ments, with this figure rapidly declining as acoustic musical-instrument technology is
being replaced with digital-electronic technology. George Gilder estimates that the
cost of raw materials for automobiles is now down to 40 percent of total costs (see
“A Technology of Liberation” in this book). Again, this figure will continue to decline
with the increasing use of computers and electronics as well as the replacement of
expensive and relatively simple body materials such as steel with inexpensive yet
relatively complex alternative materials such as new high-tech plastics.
With regard to the world of defense, military engagements such as the
\sraeli destruction of Russian SAM sites in Syria, the use of “smart” missiles in the
Falklands war, and others have illustrated the growing importance of artificial intelli-
gence in the military.2) Many military observers now predict that in the 1990s,
artificial intelligence technology will be of greater strategic importance than man-
power, geography, and natural resources.”? A major program called SCI (Strategic
Computing Initiative) envisions the soldier of the future relying on a vast network of
intelligent computers to make tactical decisions, fly planes, aim weapons, and avoid
enemy fire.”
A lesson we can draw from these observations is the importance of educa-
tion and training in a world relying increasingly on skill and innovation and decreas-
ingly on material resources. One reliable prediction we can make about the future is
that the pace of change will continue to accelerate. The revolution manifest in the
age of intelligent machines is in its earliest stages. The impact of this new age will be
greater than the radical technological and social changes that have come before it. It
cannot be stopped. We need to understand it, live creatively with it, and harness it
constructively. That is what this book is all about.
The Roots of Artificial Intelligence
(Drawing by Bernard
1987; supplied by The Schoenbaum,
New Yorker)

“Oh, if only
CoD
m E
it were 50 sim
ple.”
What Is Al, Anyway?

The postindustrial society will be fueled not by oil but by a new commodity called
artificial intelligence (Al). We might regard it as a commodity because it has value
and can be traded. Indeed, as will be made clear, the knowledge imbedded in Al
software and hardware architectures will become even more salient as a foundation
of wealth than the raw materials that fueled the first Industrial Revolution. It is an
unusual commodity, because it has no material form. It can be a flow of information
with no more physical reality than electrical vibrations in a wire.
If artificial intelligence is the fuel of the second industrial revolution, then
we might ask what it is. One of the difficulties in addressing this issue is the amount
of confusion and disagreement regarding the definition of the field. Other fields do
not seem to have this problem. Books on biology do not generally begin with the
question, What is biology, anyway? Predicting the future is always problematic, but it
will be helpful if we attempt to define what it is we are predicting the future of.
One view is that Al is an attempt to answer a central question that has
been debated by scientists, philosophers, and theologians for thousands of years.
How does the human brain—three pounds of “ordinary” matter—give rise to
thoughts, feelings, and consciousness? While certainly very complex, our brains are
clearly governed by the same physical laws as our machines.
Viewed in this way, the human brain may be regarded as a very capable
machine. Conversely, given sufficient capacity and the right techniques, our ma-
chines may ultimately be able to replicate human intelligence. Some philosophers
and even a few Al scientists are offended by this characterization of the human mind
as a machine, albeit an immensely complicated one. Others find the view inspiring: it
means that we will ultimately be able to understand our minds and how they work.
One does not need to accept fully the notion that the human mind is
“just” a machine to appreciate both the potential for machines to master many of
our intellectual capabilities and the practical implications of doing so.
+ The Usual Definition

Artificial Stupidity (AS) may be defined as the attempt by computer scientists to create computer
programs capable of causing problems of a type normally associated with human thought.
Wallace Marshal, Journal of Irreproducible Results (1987)

Probably the most durable definition of artificial intelligence, and the one most often
quoted, states that: “Artificial Intelligence is the art of creating machines that
perform functions that require intelligence when performed by people.”' It is
reasonable enough as definitions go, although it suffers from two problems. First, it
does not say a great deal beyond the words “artificial intelligence.” The definition
refers to machines and that takes care of the word “artificial.” There is no problem
here: we have never had much difficulty defining artificial. For the more problematic
word “intelligence” the definition provides only a circular definition: an intelligent
machine does what an intelligent person does.
A more serious problem is that the definition does not appear to fit actual
usage. Few Al researchers refer to the chess-playing machines that one can buy in
the local drug store as examples of true artificial intelligence, yet chess is stil! consid-
ered an intellectual game. Some equation-manipulation packages perform transfor-
mations that would challenge most college students. We consider these to be quite
useful packages, but again, they are rarely pointed to as examples of artificial intelli-
gence.

+ The Moving-Frontier Definiti

Mr. Jabez Wilson laughed heavily. “Well, | never!” said he. “I thought at first that you had done
something clever, but | see that there was nothing in it, after all?” “I began to think, Watson,” said
Holmes, “that | made a mistake in explaining. ‘Omne ignatum pro magnifico,’ you know, and my poor
little reputation, such as it is, will suffer shipwreck if | am so candid.”
Sir Arthur Conan Doyle, The Complete Sherlock Holmes

The extent to which we regard something as behaving in an intelligent manner is determined as much by
our own state of mind and training as by the properties of the object under consideration. If we are able
to explain and predict its behaviour or if there seems to be little underlying plan, we have little
temptation to imagine intelligence. With the same object, therefore, it is possible that one man would
consider it as intelligent and another would not; the second man would have found out the rules of its
behaviour.
Alan Turing (1947)

Al is the study of how to make computers do things at which, at the moment, people are better.
Elaine Rich

This leads us to another approach, which | like to call the “moving frontier” defini-
tion: artificial intelligence is the study of computer problems that have not yet been
solved. This definition, which Marvin Minsky has been advocating since the 1960s, is
unlike those found in other fields. A gene-splicing technique does not stop being part
of bioengineering the moment it is perfected. Yet, if we examine the shifting
judgments as to what has qualified as “true artificial intelligence” over the years, we
find this definition has more validity than one might expect.

14
 When the artificial intelligence field was first named at a now famous
conference held in 1956 at Dartmouth College, programs that could play chess or
checkers or manipulate equations, even at crude levels of performance, were very
much in the mainstream of Al.? As | noted above, we no longer consider such game-
playing programs to be prime examples of Al, although perhaps we should.
One might say that this change in perception simply reflects a tightening of
standards. | feel that there is something more profound going on. We are of two
minds when it comes to thinking. On the one hand, there is the faith in the Al
community that most definable problems (other than the so-called “unsolvable”
problems, see “The busy beaver” in chapter 3) can be solved, often by successively
breaking them down into hierarchies of simpler problems. While some problems will
take longer to solve than others, we presently have no clear limit to what can be
achieved.
On the other hand, coexisting with the faith that most cognitive problems
can be solved is the feeling that thinking or true intelligence is not an automatic tech-
nique. In other words, there is something in the concept of thinking that goes
beyond the automatic opening and closing of switches. Thus, when a method has
been perfected in a computerized system, we see it as just another useful tech-
nique, not as an example of true artificial intelligence. We know exactly how the
system works, so it does not seem fundamentally different from any other computer
program.
A problem that has not yet been solved, on the other hand, retains its
mystique. While we may have confidence that such a problem will eventually be
solved, we do not yet know its solution. So we do not yet think of it as just an
automatic technique and thus allow ourselves to view it as true cybernetic cogni-
tion.?
Consider as a current example the area of artificial intelligence known as
expert systems. Such a system consists of a data base of facts about a particular
discipline, a knowledge base of codified rules for drawing inferences from the data
base, and a high-speed inference engine for systematically applying the rules to the
facts to solve problems.* Such systems have been successfully used to locate fuel
deposits, design and assemble complex computer systems, analyze electronic
circuits, and diagnose diseases. The judgments of expert systems are beginning to
rival those of human experts, at least within certain well-defined areas of expertise.
Today expert systems are widely regarded as a central part of artificial
intelligence, and hundreds of projects exist today to apply this set of techniques to
dozens of fields. It seems likely that expert systems will become within the next ten
years as widespread as computer spreadsheet programs and data-base management
systems are today. | predict that when this happens, Al researchers will shift their
attention to other issues, and we will no longer consider expert systems to be prime
examples of Al technology. They will probably be regarded as just obvious exten-
sions of data-base-management techniques.
Roger Schank uses the example of a pool sweep, a robot pool cleaner, to
illustrate our tendency to view an automatic procedure as not intelligent.2 When we
first see a pool sweep mysteriously weaving its way around the bottom of a pool,
we are impressed with its apparent intelligence in systematically finding its way
 around. When we figure out the method or pattern behind its movements, which is
a deceptively simple algorithm of making preprogrammed changes in direction every
time it encounters a wall of the pool, we realize that it is not very intelligent after all.
Another example is a computer program named EL!ZA designed in 1966
by Joseph Weizenbaum to simulate a psychotherapist.® When interacting with
ELIZA, users type statements about themselves and ELIZA responds with questions
and comments. Many persons have been impressed with the apparent appropriate-
ness and insight of ELIZA’s ability to engage in psychoanalytic dialogue. Those users
who have been given the opportunity to examine ELIZA’s algorithms have been even
more impressed at how simple some of its methods are.
We often respond to people the same way. When we figure out how an
expert operates and understand his or her methods and rules of thumb, what once
seemed very intelligent somehow seems less so.
It will be interesting to see what our reaction will be when a computer
takes the world chess championship. Playing a master game of chess is often
considered an example of high intellectual (even creative) achievement. When a
computer does become the chess champion, which | believe will happen before the
end of the century, we will either think more of computers, less of ourselves, or less
of chess.”
Our ambivalence on the issue of the ability of a machine to truly emulate
human thought tends to regard a working system as possibly useful but not truly
intelligent. Computer-science problems are only Al problems until they are solved.
This could be seen to be a frustrating state of affairs. As with the carrot on a stick,
the Al practitioner can never quite achieve the goal.

+ What Is Intelligence, Anyway?

It could be simply an accident of fate that our brains are too weak to understand themselves. Think of the
lowly giraffe, for instance, whose brain is obviously far below the level required for self-understand-
ing—yet it is remarkably similar to our own brain. In fact, the brains of giraffes, elephants, baboons—
even the brains of tortoises or unknown beings who are far smarter than we are—probably all operate
on basically the same set of principles. Giraffes may lie far below the threshold of intelligence
necessary to understand how those principles fit together to produce the qualities of mind; humans may
lie closer to that threshold—perhaps just barely below it, perhaps even above it. The point is that there
may be no fundamental (i.e., Gédelian) reason why those qualities are incomprehensible; they may be
completely clear to more intelligent beings.
Douglas R. Hofstadter, Géde/, Escher, Bach: An Eternal Golden Braid

A beaver and another forest animal are contemplating an immense man-made dam. The beaver is saying
something like, “No, | didn’t actually build it. But it's based on an idea of mine.”
Edward Fredkin

If we can replace the word “artificial” with “machine,” the problem of defining
artificial intelligence becomes a matter of defining intelligence. As might be ex-
pected, though, defining intelligence is at least as controversial as defining artificial
intelligence. One approach is to define intelligence in terms of its constituent
processes: a process comprised of learning, reasoning, and the ability to manipulate
symbols.
 Learning is not simply the acquisition of facts, which a data-base-manage-
ment system can do; it is also the acquisition of Knowledge. Knowledge consists of
facts, an understanding of the relationships between the facts, and their implications.
One difference between humans and computers lies in the relative strengths in their
respective abilities to understand symbolic relationships and to learn facts. A com-
puter can remember billions of facts with extreme precision, whereas we are hard
pressed to remember more than a handful of phone numbers. On the other hand,
we can read a novel and understand and manipulate the subtle relationships be-
tween the characters—something that computers have yet to demonstrate an ability
to do. We often use our ability to understand and recall relationships as an aid in
remembering simple things, as when we remember names by means of our past
associations with each name and when we remember phone numbers in terms of
the geometric or numeric patterns they make. We thus use a very complex process
to accomplish a very simple task, but it is the only process we have for the job.
Computers have been weak in their ability to understand and process information
that contains abstractions and complex webs of relationships, but they are improv-
ing, and a great deal of Al research today is directed toward this goal.
Reason is the ability to draw deductions and inferences from knowledge
with the purpose of achieving a goal or solving a problem. One of the strengths of
human intelligence is its ability to draw inferences from knowledge that is imprecise
and incomplete. The very job of a decision maker, whether a national leader or a
corporate manager, is to draw conclusions and make decisions based on information
that is often contradictory and fragmentary. To date, most computer-based expert
systems have used hard rules, which have firm antecedents and certain conclusions.
For some problems, such as the job of DEC’s XCON, which configures complex
computer systems, hard rules make sense. A certain-sized computer board will
either fit or not fit in a certain chassis. Other types of decision making, such as the
structuring of a marketing program for a product launch or the development of
national monetary policy, must take into account incomplete present knowledge and
the probabilities of unknown future events. The latest generation of expert systems
are beginning to allow rules based on what is called fuzzy logic, which provides a
mathematical basis for making optimal use of uncertain information.’ This methodol-
ogy has been used for years in such pattern-recognition tasks as recognizing printed
characters or human speech.
The ability to learn and acquire knowledge and to manipulate it inferentially
and deductively is often referred to as symbolic reasoning, the ability to manipulate
symbols. A symbol is a name or sign that stands for something else, generally a
structure or network of facts and other symbols. Symbols are typically organized in
complicated patterns rather than simple lists. Another strength of human intelligence
is our ability to recognize the patterns represented by the symbols we know even
when they occur in contexts different than the ones in which we originally learned
the symbol. One of the reasons that the LISP programming language has been
popular in developing Al applications is its strength in manipulating symbols that
represent complex patterns and their relationships rather than orderly lists of facts
(despite its name, which derives from “list processing”).

7
 Rather than defining intelligence in terms of its constituent processes, we
might define it in terms of its goal: the ability to use symbolic reasoning in the
pursuit of a goal. Symbolic reasoning is used to develop and carry out strategies to
further the goals of its possessor. A question that then arises is, What are the goals?
With machine intelligence, the goals have been set by the human designer of each
system. The machine may go on to set its own subgoals, but its mission is imbed-
ded in its algorithms. Science-fiction writers, however, have long speculated on a
generation of intelligent machines that set their own agendas. With living creatures
or species, the goals are often expressed in terms of survival either of the individual
or the species. This is consistent with the view of intelligence as the ultimate (most
recent) product of evolution.
The evidence does not yet make clear whether intelligence does in fact
support the goal of survival. Intelligence has allowed our species to dominate the
planet. We have also been sufficiently “intelligent” to unlock the destructive powers
that result from manipulating physical laws. Whether intelligence, or at least
our version of it, is successful in terms of survival is not yet clear, particularly when
viewed from the long time scale of evolution.
Thus far Homo sapiens are less than 100,000 years old. Dinosaurs were a
successful, surviving class of creatures for 160 million years. They have always been
regarded as unintelligent creatures, although recent research has cast doubt on this
view. There are, however, many examples of unintelligent creatures that have
survived as a species (e.g. palm trees, cockroaches, and horseshoe crabs) for long
periods of time.
Humans do use their intelligence to further their goals. Even if we allow for
possible cultural bias in intelligence testing, the evidence is convincing that there is a
strong correlation between intelligence, as measured by standardized tests, and eco-
nomic, social, and perhaps even romantic success. A larger question is whether we
use our intelligence in setting our goals. Many of our goals appear to stem from
desires, fears, and drives from our primitive past.?
In summary, there appears to be no simple definition of intelligence that is
satisfactory to most observers, and most would-be definers of intelligence end up
with long checklists of its attributes. Minsky’s Society of Mind can be viewed as a
book-length attempt at such a definition. Allen Newell offers the following list for an
intelligent system: it operates in real-time; exploits vast amounts of knowledge;
tolerates erroneous, unexpected, and possibly unknown inputs; uses symbols and
abstractions; communicates using some form of natural language; learns from the
environment; and exhibits adaptive goal-oriented behavior.'?
The controversy over what intelligence is, is reminiscent of a similar
controversy over what life is. Both touch on our vision of who we are. Yet great
progress has been made, much of it in recent years, in understanding the structures
and methods of life. We have begun to map out DNA, decode some of the heredi-
tary code, and understand the detailed chemistry of reproduction. The concern many
have had that understanding these mechanisms would lessen our respect for life has
thus far been unjustified. Our increasing knowledge of the mechanisms of life has, if
anything, deepened our sense of wonder at the order and diversity of creation.
 We are only now beginning to develop a similar understanding of the
mechanisms of intelligence. The development of machine intelligence helps us to
understand natural intelligence by showing us methods that may account for the
many skills that comprise intelligence. The concern that understanding the laws
of
intelligence will trivialize it and lessen our respect for it may also be unjustified.
As
we begin to comprehend the depth of design inherent in such “deep” capabil
ities as
intuition and common sense, the awe inherent in our appreciation of intelli
gence
should only be enhanced.

+ Evolution as an Intelligent Process

God reveals himself in the harmony of what exists.

Spinoza

A central tenet of Al is that we, an intelligent species, can create intelligent

ma-
chines. At present the machines we have created, while having better
memories and
greater speed, are clearly less capable than we are at most intellectual tasks.
The
gap is shrinking, however. Machine intelligence is tapidly improving. The same
cannot be said for human intelligence. A controversial question surrou
nding Al is
whether the gap can ultimately be eliminated. Can machine intelligence
ultimately
equal that of human intelligence? Can it surpass human intelligence? A broader
statement of the question is, Can an intelligent entity be more intelli
gent than the
intelligence that created it?
One way to gain insight into these questions might be to examine the
relationship of human intelligence to the intelligent process that create
d it—evolu-
tion. Evolution created human and many other forms of intelligence
and thus may be
regarded as an intelligent process itself."

Anineteenth century view of

evolution. (Supplied by North Wind
Picture Archives)
 One attribute of intelligence is its ability to create and design. The results
of an intelligent design process—to wit, intelligent designs—have the characteristics
of being aesthetically pleasing and functionally effective. It is hard to imagine designs
that are more aesthetically pleasing or functionally effective than the myriad of life
forms that have been produced by the process we call evolution. Indeed, some
theories of aesthetics define aesthetic quality or beauty as the degree of success in
emulating the natural beauty that evolution has created.!*
Evolution can be considered the ultimate in intelligence—it has created
designs of indescribable beauty, complexity, and elegance. Yet, it is considered to
lack consciousness and free will—it is just an “automatic” process. It is what
happens to swirling matter given enough time and the right circumstances.
Evolution is often pitted against religious theories of creation. The religious
theories do share one thing in common with the theory of evolution—both attribute
creation to an ultimate intelligent force. The most basic difference is that in the
religious theories this intelligent force is conscious and does have free will, although
some theologies, such as Buddhism, conceive of God as an ultimate force of creativ-
ity and intelligence and not as a personal willful consciousness.'
The theory of evolution can be simply expressed as follows. Changes in
the genetic code occur through random mutation; beneficial changes are retained,
whereas harmful ones are discarded through the “survival of the fittest.”'* In some
ways it makes sense that the survival of the fittest would retain good changes and
discard bad ones, since we define “good” to mean more survivable.
Yet let us consider the theory from another perspective. The genetic code
is similar to an extraordinarily large computer program, about six billion bits to
describe a human, in contrast to a few tens of millions of bits in the most complex
computer programs. It is indeed a binary code, and we are slowly learning its digital
language.'® The theory says that changes are introduced essentially randomly, and
the changes are evaluated for retention by survival of the entire organism and its
ability to reproduce. Yet a computer program controls not just the one characteristic
that is being changed but literally millions of other characteristics. Survival of the
fittest appears to be a rather crude technique capable of concentrating on at most a
few fundamental characteristics at a time. While a few characteristics were being
optimized, thousands of others could degrade through the increasing entropy of
random change."® If we attempted to improve our computer programs in this way,
they would surely disintegrate.
The method we use to improve the programs we create is not the intro-
duction of random code changes but carefully planned and designed changes and
experiments designed to focus in on the changes just introduced. It has been
proposed that evolution itself has evolved to where changes are not entirely random
but in some way planned, and that changes are “tested” in some way other than
overall survival, in which a change just introduced would be competing with thou-
sands or even millions of other factors.’
Yet no one can describe a mechanism in which such planning and isolated
evaluation could take place in the process of evolution. There appears, therefore, to
be a gap in the theory of evolution. Clearly, the fossil and biochemical evidence is
overwhelming that species have indeed undergone a slow but dramatic evolution in

20
complexity and sophistication, yet we do not fully understand the mechanism. The
proposed mechanism seems unlikely to work; its designs should disintegrate
through increasing entropy.
One possible perspective would state that the creator of an intelligence is
inherently superior to the intelligence it creates. The first step of this perspective
seems to be well supported in that the intelligence of evolution appears vast. Yet is
it?
While it is true that evolution has created some extraordinary designs; it is
also true that it took an extremely long period of time to do so. Is the length of time
required to solve a problem or create a design relevant to an evaluation of the level
of an intelligence? Clearly it is. We recognize this by timing our intelligence tests. If
someone can solve a problem in a few minutes, we consider that better than solving
the same problem in a few hours or a few years. With regard to intelligence as an
aid to survival, it is clearly better to solve problems quickly than slowly. In a competi-
tive world we see the benefits of solving problems quickly.
Evolution has achieved intelligent work on an extraordinarily high level yet
has taken an extraordinarily long period of time to do so. It is very slow. If we factor
its achievements by its ponderous pace, | believe we shall find that its intelligence
quotient is only infinitesimally greater than zero. An IQ of only slightly greater than
zero is enough for evolution to beat entropy and create extraordinary designs, given
enough time, in the same way that an ever so slight asymmetry in the physical laws
may have been enough to allow matter to almost completely overtake antimatter.
The human race, then, may very well be smarter than its creator, evolution.
If we look at the speed of human progress in comparison to that of evolution, a
strong case can be made that we are far more intelligent than the ponderously slow
process that created us. Consider the sophistication of our creations over a period of
only a few thousand years. In another few thousand years our machines are likely to
be at least comparable to human intelligence or even surpass it in all likelihood, and
thus humans will have clearly beaten evolution, achieving in a matter of thousands of
years as much or more than evolution achieved in several billion years. From this
perspective, human intelligence may be greater than its creator.'®
So what about the intelligence that we are in turn creating? It too could be
greater than its creator. That is not the case today. While computers have a superior-
ity in certain idiot savant types of thinking, our thinking is today significantly superior
to that of our machines. Yet the intelligence of our machines is improving at a very
rapid pace. Within a matter of years or decades it appears that computers will
compete successfully with human intelligence in many spheres. If we extrapolate a
sufficient number of decades or centuries into the future, it appears likely that
human intelligence will be surpassed.'? In contrast to what one might intuitively
conclude, this perspective points consistently to the possibility that an intelligence
may ultimately outperform its creator.

21
Philosophical Roots
Some philosophers hold that philosophy is what you do to a problem until it's clear enough to solve it by
doing science. Others hold that if a philosophical problem succumbs to empirical methods, that shows it
wasn't really philosophical to begin with.
Jerry A. Fodor, Representations: Philosophical Essays on the Foundations of Cognitive Science

The chance of the quantum theoretician is not the ethical freedom of the Augustinian.
Norbert Wiener, Cybernetics

There's something queer about describing consciousness: whatever people mean to say, they just can't
seem to make it clear. It's not like feeling confused or ignorant. Instead, we feel we know what's going
‘on but can’t describe it properly. How could anything seem so close, yet always keep beyond our reach?
Marvin Minsky, The Society of Mind

How can mind arise from nonmind? In examining human thought through the ages,
philosophers appear to have gone down one of two paths. One school of thought,
which we might call mind as machine, starts with the observation that human
thought takes place in the human brain. The brain, in turn, is made up of tens to
hundreds of billions of neurons. Neurons, while not simple structures, can nonethe-
less be fully understood as biochemical machines. Our brain thus consists of billions
of biochemical machines interacting with each other, a fact from which we can draw
two conclusions. The first is that the human mind is a machine, albeit an enormously
complicated one. It is, after all, made up of matter just as subject to natural laws as
any other machine. Second, we can, at least in theory, create other, human-made
machines that employ the same techniques or algorithms for transforming informa-
tion as the human brain. It is thus possible to replicate in a machine intellectual
Capacities that previously could only be achieved by human intelligence. This latter
conclusion is a clear tenet of the Al movement.’
The opposing school of thought, which we can call mind beyond machine.
contends that there are certain aspects of human thought and human existence that
cannot be understood through this type of analysis. While acknowledging that the
model of the human mind as machine may provide some understanding, it cites
such attributes as consciousness and free will and, depending on the philosopher,
other possible attributes as well, as being outside the fully rational, i.e., machinelike,
model. It criticizes the approach of applying only logic to our understanding of
thinking as being hopelessly circular, that is, as using an analysis based only on logic
to conclude that thinking is based only on logic.?
In this debate the concept of emotions and feelings generally lie in a
middle ground somewhere between consciousness and free will on the one hand
and the logical patterns of rational thought on the other. The mind-as-machine school
tends to analyze emotion as another form of logical thought, subject to its own rules.
In this view, our emotions are a complex set of algorithms that motivate our behav-
ior in a way that supports the greater goals of our culture. One could say that our
emotions provide us with our strategies, while our more detailed logical calculations
provide us with our tactics. The mind-beyond-machine school is more likely to
conceive of emotion as being deeply imbedded in our consciousness and thus not
fully comprehensible by logic alone.?
Thousands of years before there were computers, Plato (427-347 s.c.)
recognized the similarity between at least certain aspects of human thinking and the
apparently determined cause-and-effect behavior exhibited by machines.* The
Platonic debates illuminate as clearly as any modern philosopher the apparent
paradox of the human mind displaying free will while being subject to predictable
natural laws. In the final analysis, it appears that Plato accepted both views as an
irreducible duality. Such paradoxes, according to Plato, were not to be avoided;
rather, they were the key to a richer understanding of the human condition.
Philosophical schools based on Plato’s thought continued to shape Euro-
pean epistemology—the study of the limits and validity of knowledge—into the
middle ages through the lasting influence of the Academy he founded and through
one of his star pupils, Aristotle (384-322 B.c.).°
The European Renaissance of the seventeenth century and the Enlighten-
ment, a philosophical movement of the seventeenth and eighteenth centuries, re-
newed and intensified an emphasis on the scientific method and in particular its
application to thinking as a process that could be studied and understood as a
phenomenon following natural laws. The rationalism of René Descartes (1596-1650),
the empiricism of Francis Bacon (1561-1626) and the physics of Isaac Newton
(1642-1727) were fused by Immanuel Kant (1724-1804) into a rigorous view of
man’s world based on knowledge as its principle building block.®
Modern philosophy has divided into two schools of thought. While often
seen as contradicting one another and while frequently clashing with one another
indeed, they primarily deal with divergent issues. Existentialism, which is the domi-
nant school of philosophy today in Europe, has its roots in the unorthodox Christian-
ity of Soren Kierkegaard (1813-1855) and the anti-Christianity of Friedrich Nietzsche
(1844-1900). Existentialism, and a related field, phenomenology, regard human
thought and human existence as describable by scientific analysis, but only partially.”
Their emphasis has been on such phenomena as guilt, anxiety, and suffering, which
in their views lie beyond fully rational exploration and are keys to understanding the
limits of reason. The modern existentialism of such philosophers and playwrights as

24
 Jean-Paul Sartre (1905-1980) and Samuel Beckett emphasize the role of free will in
an apparently purposeless world.
A movement still popular in the United States that often clashes with
existentialism is logical positivism, based on the early work of Ludwig Wittgenstein
(1889-1951) and developed by Alfred Ayer and others. Searching for truth in the
foundations of language, logical positivism gave rise to the development of linguistic
theory by Noam Chomsky and others and greatly influenced the emergence of
computation theory.®
The Al movement, which can be said to have its roots in logical positivism,
has often clashed with the phenomenology of the existentialists, as most recently
exemplified by the ongoing debate of Al practitioners such as Marvin Minsky,
Seymour Papert, and Ed Feigenbaum with leading Al critic and phenomenologist
Hubert Dreyfus. Unfortunately, this debate has been overly personal and divisive and
thus has not contributed to a needed synthesis between the two pillars of modern
philosophy.?

+ Plato and the Platonists

The safest general characterization of the European philosophical tradition is that it consists of a series
of footnotes to Plato.
Alfred N. Whitehead

All virtue is one thing, knowledge.

Plato

Born in 427 8.c., Plato is regarded by many as the greatest philosopher of all time.
His thought ranged across the ultimate nature of truth and knowledge, ethics, social
order, political structure, physics, astronomy, and mathematics. His writings not only
recorded his thoughts but constitute the primary written record of the teachings of
his mentor, Socrates (c. 469-399 8.c.).'° Aristotle viewed his own work as carrying
out the dissemination and further development of Plato's teachings.'' Socrates,
Plato, and Aristotle are credited with having established the essentially rationalistic
philosophy of Western culture
While he is now considered one of the greatest writers of philosophy,
Plato considered his writings as merely a tool to assist him in his teaching. He
regarded his principle work as the establishment and guidance of the Academy, an
institute for the pursuit of science and philosophy that he founded at the age of 41
The Academy outlived Plato, and the influence of its students and followers enabled
Platonic thought to exert profound influence for many centuries after his death at
Plato, affirming the duality of life. age 80.'?
(Supplied by North Wind Picture While Plato had his mystical side, his is a highly ordered mysticism based
Archives)
on a theory of ideal forms, such as the form of the circle, of beauty, of love, etc.,
and the manifestation of forms in an imperfect world.'? Plato’s own logical reasoning
exemplified the power of reason. By logical inference and his own thought experi-
ments, Plato was able to determine, for example, that the planets followed orbits of
single closed curves and that other apparent movements of stars and planets were

25
due to the earth’s movement through the sky in its own closed curve.'* He then
imagined the earth and the other planets circling a “Pythagorean central fire.”'>
Plato inferred the existence of irrational numbers, numbers that could not
be expressed as the finite sum of fractions. The square root of 2 is the quintessen-
tial example. Plato saw the fact that rational numbers, with their finite definitions,
and irrational numbers, with their infinite definitions, coexist in the same continuum
as symbolic of the coexistence of material and mystical phenomena in nature.'®
Though Plato's lasting contribution is in the rationalization of philosophy
and a casting aside of the ornate mysticism of many of his peers, he maintains that
there exists a level of creation that defies complete rational understanding. In his
Timaeus, he describes the ananke, the level of reality that cannot be rationalized
completely and that has to be accepted as a reflection of the purpose of creation.'”
Modern existentialism echoes Plato in its acceptance of a rational level of reality
combined with its emphasis on the limits of reason and logic.
At the core of the duality of existence in the rational and mystical is the
issue of consciousness and free will. In the Phaedo and later works, including The
Republic and Theaetetus, Plato expresses the profound paradox inherent in the
concept of consciousness and man’s ability to freely choose. On the one hand,
human beings partake of the natural world and are subject to its laws. Our minds are
natural phenomena and thus must follow the cause and effect laws manifest in
machines and other lifeless creations of man. On the other hand, cause and effect
mechanics, no matter how complex, do not, according to Plato, give rise to self-
awareness or consciousness.'® Plato attempted to resolve this conflict in his theory
of the Forms.'? Consciousness is not an attribute of the mechanics of thinking, but
rather the ultimate reality of human existence. Our consciousness or “soul” is
immutable and unchangeable. Thus, our interaction with the physical world is on the
level of the “mechanics” of our complex thinking process in a complex environment.
Yet Plato was not fully satisfied with this metaphysical doctrine. If the soul is immu-
table and unchanging, then it cannot learn or partake in reason, because it would
need to change to absorb and respond to experience. He expressed dissatisfaction
with positing consciousness in either the rational processes of the natural world or
the mystical level of the ideal Form of the self or soul.
An even deeper paradox is apparent in the concept of free will. Free will is
purposeful behavior and decision making. Plato believed in a “corpuscular physics”
based on fixed and determined rules of cause and effect.2° If human decision making
is based on such interactions of basic particles, our decisions too must be predeter-
mined. Such predetermination would, however, appear to contradict human freedom
to choose. The addition of randomness into natural laws is a possibility, but it does
not solve the problem. Randomness would eliminate the predetermination of deci-
sions and actions, but it contradicts the purposefulness of free will, as there is
nothing purposeful in randomness. Positing free will in the soul and thus separating
it from the rational cause and effect mechanics of the natural world is also not satis-
factory for Plato, because Plato is uncomfortable placing reason and learning in the
soul. These attributes of our thinking process are too orderly and logical to place
entirely on the mystical plane.?!

26
 From Aristotle on, philosophers have debated for over two thousand years
exactly what Plato’s position was on these issues. Support for apparently contradic-
tory views can be found in Plato’s writings. My own view is that Plato believed in an
essentially irreducible paradox at the core of the issues of consciousness and free
will. | base this on his refusal to write his own metaphysics and on the eloquence
with which Plato is able to articulate the alternate sides of each paradox. Plato's
choice of the dialogue form was an excellent medium for expressing paradox in that
it freed him to express passionately conflicting views.”
Irreducible paradox at the core of reality has found support in a surprising
place—twentieth century physics. Physics, which seeks to describe ultimate reality
in rational terms, has concluded that the essence of electromagnetic radiation is both
a particle and a wave, two mutually exclusive and inconsistent models. Quantum
theory too is based on a paradox: particles have precise locations, but we cannot
know what these locations are. In its ultimate form, quantum mechanics describes a
particle as having no precise location, although somehow the particle exists in a
multidimensional space.”
In perhaps his most direct appeal to paradox as a resolution of apparent
conflict of ideas, Plato discusses the duality of human love in Phaedrus.** By applying
the logical method to the study of passion, he concludes that love and its expression
in the apparently mad behavior of the lover is rooted both in the material world and
in the attempt of the soul to achieve union with the ideal Form of transcendent emo-
tion
What is truly remarkable about Plato's writings is the extent to which they
are, after twenty-three hundred years, relevant to modern philosophical dilemmas on
the relationship of human thought to the mechanics of the machine. Plato saw
clearly the relationship of human thought to the rational processes of a machine. He
recognized that human thought was governed by natural law and that natural law
was an essentially logical process. There is no limit, according to Plato, to the extent
to which we can unravel human thought and behavior by scientific observation and
logical inference. At the same time, he felt that human reality was not sufficiently
expressed in logic alone. He does not resolve this problem, however, by attributing
human thought to mystical processes that are of a different world from the logical
processes of the material world. Instead, he resolves that the duality of human
thought as both a logical process and one that transcends pure logic represents a
necessary coexistence. It is a synthesis of views that is relevant to the modern
conflict between the logical positivist foundation of the Al movement and existential-
phenomenological views

The Platonists
While Plato's thought had wide and diverse influences and interpretations after his
death, one of the more interesting refinements of his thought made by his succes-
sors was a certain “mathematization” of his philosophy and the expression of his
philosophy of Forms in numeric terms.?° The Forms, which constitute ideals such as
round, beauty, justice, and love, are regarded as pure concepts in the same way that
numbers are pure concepts. At the other extreme of reality is the physical world that

27
 UU,

The School of Athens, finding truth

in numbers. (Supplied by North Wind
Picture Archives)

imperfectly manifests the Forms. In between the Forms and physical reality are
“mathematicals,” which, like the Forms, are perfect and unchanging but, like
physical reality, are numerous. In other words, there is only one concept or Form
circle, but there can be many instances of circles. Each instance can be immutable
and perfect in the mathematical realm, while each manifestation of circularity in the
physical world will be changeable and somewhat less than perfect.
These views are not directly expressed in the dialogues but are often
attributed (e.g., by Aristotle) to Plato's later oral teachings. They take place during a
time of fertile development of mathematical theory by members and associates of
Plato’s Academy, including Euclid (330-260 s.c.), the expositor of plane geometry,
and Theaetetus (c. 415-369 8.c.), the creator of solid geometry.”
The expression of the mystical concept of Forms in the logical language of
mathematics expresses again the paradox at the heart of Plato's views of human
reality and thought. In the Epinomis, Plato states that the “relations of numbers are
the key to the whole mystery of nature.“2”
+ The Enlightenment

The Enlightenment, along with parallel fertile ‘developments in science and theology,
was a philosophical movement to restore the supremacy of human reason, knowl-
edge, and freedom. It had its roots in both the artistic, literary, and cultural activity of
the previous three centuries (the European Renaissance) and the Greek philosophy
of twenty centuries earlier. It considered its own roots to be with Socrates, Plato,
and Aristotle, and it constituted the first systematic reconsideration of the nature of
human thought and knowledge since the Platonists.”°
With Isaac Newton's presentation of the laws of gravitation in his
Philosophiae Naturalis Principia Mathematica, published in 1687, as well as advances
in the construction of clocks and mechanical automata, the philosophers of the
Enlightenment had more powerful models both of natural laws and of the potential
of machines than did their counterparts two thousand years earlier.2? Machines
became more elaborate and more capable as mechanical automata developed in
sophistication and popularity during the seventeenth and eighteenth centuries.*°

Descartes
The Enlightenment saw a blending of philosophy and science, as the same persons
often dominated both fields. Descartes, for example, formulated the theory of optical
refraction and developed the principles of modern analytic geometry. In his own
view, Descartes’s efforts in mathematics and science were intended primarily as a
means of exploring and demonstrating certain aspects of his metaphysical doctrine.
He needed to demonstrate the deterministic nature of the real world and these
major scientific discoveries were in a sense footnotes to Descartes’s philosophical
Rene Descartes, confident of his
own existence. (Supplied by North investigations. The mystery of how mind can arise from nonmind, of how thoughts
Wind Picture Archives) and feelings can arise from the ordinary matter of the brain, sometimes called the
mind-body problem, was perhaps most clearly articulated by Descartes.*'
In his comprehensive Discourse de la Méthode, Descartes pushed rational
skepticism to its limits. Acknowledging that the existence of other people and even
our own bodies may be illusions, he concluded that we cannot doubt the existence
of our own thought and hence his famous conclusion “! think, therefore | am."°2
Descartes was fascinated by automata and made contributions to their
design. Once while Descartes was traveling by sea, the ship's captain was startled
by the realistic movements of Descartes’s mechanical doll Francine and forcibly
threw “her” overboard, believing the automaton to have been a product of the
devil?

Newton
Linking the process of thought to the determined interactions of the natural world
gained momentum with Newton's breakthrough in the understanding of mechanical
law. Newton set out as his goal to find a link between the mechanical interactions of
objects we observe in the laboratory and the movement of celestial bodies that we
observe in the sky. The goal was a unified set of formulas that explains the move-
ment of objects from the very small to the very large, something never before
achieved. The result was unexpectedly successful—Newtonian mechanics appeared

293
Philosophical Roots

ProtI
frovtigendn ol Con Lena a P.
ote wre a Qate S:

tote Se mn piled
jug chapime Vescendel :
Solut la.
A Pale purse A Qucabe riha mhmite APCZ farronk pare
A LB afus HE produ Olean in puneto Q, hike. G vale
vee og har he ME Ga 49 pores Sagem dish
eee ae AG DAQ. hh. Ha G4 nevitema franpal prr

Be Bry Ales eee en

Newton's solution of the “quickest
descent” problem. Since Newton's
formulas predict the exact
movements of billiard balls and
planets, philosophers wondered
whether the movements of human
“particles” were equally predeter-
mined. (Supplied by North Wind

Isaac Newton analyzing rays of light.

(Supplied by North Wind Picture
Archives)

30
to explain with extreme accuracy a deterministic order that governed all matter.°*
With one publication Newton swept aside centuries of medieval imprecision.
Newtonian mechanics remained the dominant view of both celestial and
particle mechanics for almost two and one half centuries.*° While Einstein
(1879-1955) showed Newtonian mechanics to be a special case of his broader
theory of relativity, this “special case” happens to be the world we live in.*° One has
to enter the world of high-speed subatomic particles or certain astronomical phenom-
ena to witness significant deviations from Newtonian mechanics. Studies of naive
physics, which is an exploration of the mental models of the physical world that
people actually use, indicate that we believe that we live in a Newtonian, not an
Einsteinian, world.
When Newton had successfully explained major aspects of the universe,
the implications of his theory were not lost on the theologians of his day.3” The
success and apparent power of Newton's ideas were seen as a threat to the unique
status of man. Fear and anger were expressed that the same mathematical reason-
ing that resulted in Newtonian Mechanics would be extended to the processes of
the mind. It was feared that human thought would be subjected to the indignity of
comparison with the determined interactions of billiard balls, which were often used
to demonstrate mechanical principles. Even Wilhelm Leibniz (1646-1716), who
shares credit with Newton for the invention of the calculus used to express Newto-
nian laws, attacked Newton for regarding God as a “clumsy watchmaker.
”%?

Kant
Immanuel Kant, born in 1724 in East Prussia (now part of the Soviet Union), is widely
regarded as the preeminent philosopher of the Enlightenment. He typified the
Enlightenment in his emphasis on human reason and rationality, and he attempted to
develop a metaphysical doctrine based entirely on reason.°9 In Kant’s conception,
human thought is guided by a priori principles and concepts, that is, concepts and
structures that are not based on experience. Furthermore, in Kant’s view, human
knowledge constitutes the ultimate reality. This reversed the prior conception that
ultimate reality resided in the physical world, with our thoughts based on our impre-
cise sensory impressions reflecting imperfect models of the physical world.*°
Three of Kant's innovations would profoundly alter the philosophical
landscape and set the stage for the emergence of twentieth-century rationalism
First, Kant’s model of a priori concepts influenced and is echoed in the logical
positivist search for truth in language and the concept of the innate structures in
language postulated by modern linguistics. Second, by rejecting Descartes’s dichot-
omy between the instinctive reflex of the animal and the rational thought of man,
Kant opened up even further than his predecessors the processes of human thought
to analytic investigation. Finally, Kant's emphasis on the supremacy of knowledge
over other levels of reality would be echoed by both the logical positivist and existen-
tialist schools of the twentieth century.

The relationship of the Enlightenment to the modern schools

On the issue of the ultimate nature of human reason, the philosophers of the
Enlightenment, while diverse in their views, essentially agree with the modern
 existential school. Both maintain that while logic and the scientific method can be
used to explore human thought and the thinking process in great detail, there is a
level of human existence and experience that defies such analysis. However, the
Enlightenment ended up defining the metaphysical level with such logical rigor,
stripping away the complexities of earlier mysticism, that it set in motion a move-
ment to explain all of reality in rational terms. The logical-positivist school continued
and extended this tradition.*' The existential movement, on the other hand, can be
said to be a reaction to this trend. While acknowledging the role of logic and scien-
tific analysis, the existentialists have attempted to refocus attention on the irrational
and paradoxical side of the human condition.

+ The Logical Positivists and the Existential Reaction

The myth that everything in the world can be rationally explained has been gaining ground since the time
of Descartes. An inversion was necessary to restore the balance. The realization that reason and anti-
reason, sense and non-sense, design and chance, consciousness and unconsciousness belong together
as necessary parts of a whole.
Hans Richter, Dada

It is essential to abandon the over-valuation of the property of being conscious before it becomes
possible to form any correct view of the origin of what is mental.
Sigmund Freud, The Interpretations of Dreams

Kant as the first logical positivist

The Enlightenment as it had culminated in the thought of Immanuel Kant had
relegated metaphysics to an elegant and simple role in explaining the human condi-
tion. Our understanding of human memory, our reasoning faculty, even our emotions
could be understood in terms of scientific inquiry and analysis. While Kant did regard
the metaphysical level as supreme, he had much more to say about the intricacies of
the physical level. Although it was not his intention, his impact has been to move the
western philosophical tradition toward less emphasis on the role of metaphysics.*
He has, in fact, been called the “first logical positivist.”

Ludwig Wittgenstein
The title of the first logical positivist truly belongs, however, to Ludwig Wittgenstein.
An enigmatic figure, Wittgenstein gave away his large inherited fortune so as not to
be distracted from his philosophy by worldly possessions.** His most influential and
first major work, the Tractatus Logico-Philosophicus, was not an instant success.
Wittgenstein had a great deal of difficulty in finding a publisher for his work, and it
was ultimately the influence of his former instructor, Bertrand Russell (1872-1970),
that allowed the book to come into print.*° It has come to be regarded by some,
however, as perhaps the most influential philosophical work of the twentieth
century

Wittgenstein further applied the analytic treatment of human thought,

extended by the Enlightenment, to the study of human language. He examines the
nature of language, how it is that we communicate using language, and what it is

32
Aplace for everything and everything
in its place: a nineteenth century
view of the brain. (From an 1895
medical guide; photo by Coco McCoy
of Rainbow)
 that we are communicating. His examination is not, however, an exploration of the
structure, organization, physiology, or psychology of communication. Rather, it is an
attempt to provide a philosophical definition of Knowledge—what we can know—by
analyzing the meaning of language. He goes on in the Tractatus to consider language
to be the embodiment of what can be said, what can be known, indeed, what can
be thought:

4.0.0.3.1 All philosophy is a “critique of language.”*°

5.6 The limits of my language mean the limits of my world.4”
5.6.1 We cannot think what we cannot say.*®

Wittgenstein goes on to define language in a particular, some would say

peculiar, way. In Wittgenstein’s world there are certain elementary facts, there are
propositions about relations between elementary facts, and there are certain allow-
able transformations on such propositions that yield composite propositions. His
concept of human thought is that we receive sense impressions, which comprise
elementary facts. We can then transform these elementary facts and derive relation-
ships among them according to the allowable logical processes. Any thought outside
this scheme is either false or nonsensical.
Wittgenstein makes two major points that have a direct bearing on the
intellectual roots of artificial intelligence. He makes a direct link between human
thought and a formal process that can be described only as computation. To reorder
Wittgenstein’s statements, we cannot think what we cannot say; we cannot say, or
at least we ought not say, what is meaningless in the language we are speaking;
statements in any language are indeed meaningless unless they can be derived from
a formal (and therefore computationlike) sequence of transformations on a data base
of elementary propositions.
This description of human thought as a formal sequence of computations
would be restated two decades later in the Church-Turing thesis (see chapter 3).*°
Ludwig Wittgenstein. “What we The Church-Turing thesis was regarded as a radical doctrine when it was first
cannot speak about we must pass
over in silence.” (Supplied by proposed, but it has its roots in the Tractatus. It is not a thesis that everyone familiar
Austrian Press) with it necessarily accepts, and it remains controversial today. Wittgenstein himself
ended up rejecting it.
The other point made in the Tractatus that would have significance later to
computational theorists is that thought is embedded in language.* It is also interest-
ing to note that language as conceived in the Tractatus has more of the quality of
the programming language LISP or even PROLOG than it does of Wittgenstein’s
native German
The organization of the book is also interesting. The treatise contains only
seven primary formal statements, numbered 1 through 7. To help us along, Wittgen-
stein also includes several levels of modifying statements. For example, statements
1.1 and 1.2 modify statement 1. In turn 1.1.1 is provided to help explain 1.1, and so
on. The reader has the choice of reading the book from left to right or from top
down. The modular structure would please today’s proponents of good programming
style
Wittgenstein starts with, “The world is all that is the case,” (statement 1)
and ends with, “What we cannot speak about we must pass over in silence” (state-

34
ment 7).°' One thing that is clear from these two statements is that the Tractatus
was an ambitious work.
In 1953, two years after Wittgenstein’s death, Philosophical Investigations,
his last work, was published.*? Views of its significance vary with the point of view of
the critic. Logical positivists who trace their intellectual roots to the early Wittgenstein
regard his later work as confused and disorganized, while existentialists regard it as a
work of major importance. In what is perhaps a unique occurrence in philosophical
history, Wittgenstein is credited with having established two major systems of
philosophy, each with great influence, with the second criticizing and rejecting the
first.* Wittgenstein ends up near the end of his life having a lot to say about subjects
that he had argued in the early Tractatus should be passed over in silence.

Ayer and the formalization of logical positivism

Alfred Ayer and others carried on Wittgenstein’s work, correcting errors and express-
ing it with a degree of formalism and rigor that Wittgenstein had neither the time nor
patience to carry out.® At a philosophical level, Logical Positivism argues that every
statement, and by implication all knowledge, is either based on sense data ("a
posteriori” or “synthetic”), or is based on logic ("a priori” or “analytic”). It rejects all
metaphysical theories as strictly meaningless and having only emotive force. On a
practical level, modern theories of linguistics and computation were derived from the
formalisms of Logical Positivism. One does not need to accept all of its implications
to recognize its contribution as the philosophical basis of artificial intelligence.

The existential reaction

As with any major philosophical movement, the views and theories of its proponents
are diverse and complex. It is nonetheless reasonable to view existentialism as a
reaction on an intellectual and cultural level to the major drift of Western thought
toward greater and greater reliance on rational and analytic views. At its core,
existentialism defines human reality as almost the reverse of the logical-positivist
view. It considers the analytic and synthetic types of statements, which are the only
meaningful statements in logical positivism, as either meaningless or trivial. It regards
the spiritual and emotive life, which are meaningless in logical-positivist terms, as the
seat of true meaning.°°

Wittgenstein revisited
Some recent views of the early Wittgenstein claim that he was not denying the
existence of a realm beyond the narrow definition of meaning expressed in the
Tractatus. Clearly, the last sentence in the Tractatus, “What we cannot speak about
we must pass over in silence,” is referring to something.* If it is referring to some-
thing that does not exist, then even by the early Wittgenstein’s own standards the
sentence would be meaningless. Since we can assume that Wittgenstein would not
end his book with a meaningless sentence, the phrase “what we cannot speak
about” is referring to something meaningful. If “what we cannot speak about” is a
meaningful concept and yet we must pass over it in silence, then the sentence must
be a plea for silent contemplation of a higher realm.
+ The Debate Goes On

Why is philosophy so complicated? It ought to be entirely simple. Philosophy unties the knots in our
thinking that we have, in a senseless way, put there. To do this it must make movements that are just as
complicated as these knots. Although the result of philosophy is simple, its method cannot be if it is to
succeed. The complexity of philosophy is not a complexity of its subject matter, but of our knotted under-
standing.
Ludwig Wittgenstein

The relationship of human thought to the “logical” process of the computer contin-
ues to be controversial, a continuation of the debate started in the Platonic dialogues.
The very name “artificial intelligence” juxtaposes two concepts that engender diverse
and often intense intellectual and emotional reactions.
One particularly noisy debate has been going on for twenty years between
a number of the academic Al leaders and Hubert Dreyfus, a modern phenomenologist
and Berkeley professor. It began with a paper Dreyfus wrote as a consultant to the
Rand Corporation in 1965 entitled, “Alchemy and Artificial Intelligence.”®® As might be
clear from the title, it was a no-holds-barred attack on what was at that time an
uncertain new academic discipline. Since then Dreyfus has made something of a
career out of attacking artificial intelligence.
Such criticism might be useful to the field, and perhaps some of it is.
Unfortunately, the debate has been marred in a number of ways. First, there has
been considerable personal anger expressed on both sides. Dreyfus has been quoted
as saying, “Why do | get so upset with people like Papert, Minsky, Newell and
Simon?—and |really do get upset. It’s really puzzling. I'll have to think about that. . . .
Maybe | attack in them what | dislike in myself, an excessive rationality.
"°°
More serious is an unwillingness on both sides to fully understand the
disciplines and traditions of the other. Dreyfus has displayed considerable ignorance
of Al methods and status. In a recent article Dreyfus describes how he was able to
trick ELIZA, a computer program written by Joseph Weizenbaum to simulate a
psychotherapist (see “ELIZA Passes the Turing Test,” below).® Aside from having
been written twenty years earlier, ELIZA was considered even then to be a simple-
minded program unrepresentative of the state of the art.®' Perhaps Dreyfus’ most
consistent theme is the inability of the hard antecedent-consequence type of logic to
solve certain types of problems.® While this observation is correct, most Al research-
ers do not propose “PROLOG-like” logic as the solution to all problems.®? For ex-
ample, using fuzzy logic principles to deal in a methodologically sound manner with
uncertain observations is becoming increasingly popular.® It is also feasible to create
systems with thousands or even millions of parallel processes to emulate human
pattern recognition and skill acquisition abilities. Dreyfus describes machine intelli-
gence as a fixed phenomenon and regards today’s apparent limitations as permanent
limitations rather than tomorrow's challenges
Dreyfus continually presses the theme that the Al field has been overly
optimistic and has underestimated the deep nature of many problems. The criticism
has considerable justification, but Dreyfus takes the position to an extreme by listing
specific tasks that he maintains a computer will never do, including playing champion-
ship chess, riding a bicycle, and understanding human speech.® These would seem

36
 N\
C
A machine? (Photo by Lou Jones) |
p. \
hen
to be needlessly negative predictions, reminiscent of earlier predictions that “man is
not meant to fly” (“man,” of course, does a lot of things he was “not meant” to do).
Computers are close to accomplishing some of these tasks now, though how close
depends, of course, on the standard of performance one will accept.® In my opinion,
the levels of machine performance will continue to increase over time and, if one is
intellectually honest, it appears to be only a matter of time before any particular
standard is reached. And perhaps if Dreyfus is intellectually honest, he will be won
over. He has in fact stated that if in a Turing test, a machine could fool him 60
percent of the time as to whether he was dealing with natural or artificial intelli-
gence, he would concede defeat.®’ One has to note, however, that this offer is less
generous than it appears at first glance. Dreyfus, as the human judge in such a |
Turing test, would be able to achieve an accuracy of 50 percent just by guessing
randomly. |
In turn, Al critics of Dreyfus have been quick to jump on Dreyfus’s limited
understanding of computer technology but have themselves not taken adequate time
to understand either Dreyfus’s intellectual tradition, with its roots in the work of
Kierkegaard, Heidegger, and the late Wittgenstein, or the relevance of this tradition
to the goals of the Al movement.

Dreyfus revisited
In a recent article by Hubert Dreyfus and his brother Stuart Dreyfus, the reader
detects a subtle but possibly significant shift in Dreyfus's approach to machine
intelligence.®° They begin with the following quotes: “Nothing seems more possible
to me than that people some day will come to the definite opinion that there is no
copy in the . . . nervous system which corresponds to a particular thought, or a
particular idea, or memory” (Ludwig Wittgenstein 1948). “Information is not stored
anywhere in particular. Rather, it is stored everywhere. Information is better thought
of as evoked than found” (David Rumelhart and Donald Norman 1981).7°
The article goes on to state their continued strong opposition to a concept
of artificial intelligence based entirely on symbolic reasoning, but the authors appear
to be more comfortable with broader notions of machine intelligence. In particular,
the ability of a neural net (a special type of computer composed of many parallel
processes, each of which simulates a human brain neuron) to produce unexpected
associations appears to be intriguing to the Dreyfus brothers.’’ They end by saying,

Perhaps a net must share size, architecture, and initial-connection configuration with
the human brain if it is to share our sense of appropriate generalization. If it is to
learn from its own “experiences” to make associations that are humanlike rather
than be taught to make associations that have been specified by its trainer, a net
must also share our sense of appropriateness of output, and this means it must
share our needs, desires, and emotions and have a humanlike body with appropriate
physical movements, abilities, and vulnerability to injury. . . . If the minimum unit of
analysis is that of a whole organism geared into a whole cultural world, neural nets
as well as symbolically programmed computers still have a very long way to go.”

While the Dreyfuses' insistence that an intelligence must share a human-

like body to truly emulate human intelligence has always caused consternation

38
among Al theorists, the above statement seems to be more accepting than previous
writings of at least the theoretical possibility of endowing a machine with true
intelligence.”

Naive experts
| felt it would be worthwhile to explore philosophical issues of machine cognition
with a number of experts who, while having extensive computer experience, were
not influenced by two thousand years of theory and debate.” | therefore chose six
children ages seven to nine who had been working with computers for several years,
but who assured me that they had not read either the early or the late Wittgenstein.
| told each child (one at a time) that | would ask them several questions
and that there was no right or wrong answer, | just wanted their opinion. The
questions were, Can a computer remember? Does a computer learn? Do computers
think? Do computers have feelings? Do you like computers? Do computers like you?
To the first two questions each child answered in the affirmative: comput-
ers do remember, and they do learn. The third question required a few moments of
reflection, and all but two of the children concluded that yes, computers do think.
Apparently, one important clue to computers’ thinking ability for the children was the
fact that when the children ask a computer to do something, it sometimes answers
right away and sometimes there is a delay while the computer apparently thinks
about the task for a while before responding. This, the children felt, was very similar
to the way that they respond to questions.
The fourth question—Do computers have feelings?— not only was unani-
mously answered in the negative; it generally elicited laughter as if | had asked, Do
elephants fly? Laughter, according to Freud, sometimes results from the juxtaposi-
tion of two concepts that are not supposed to go together, which may result from
either the two concepts’ never having ever been linked before or a social taboo.”>
Possibly both reasons caused the laughter in this case
On the fifth question, all of the children answered affirmatively that they
liked computers. All of the children thought that the last question was silly, that
computers do not have likes and dislikes.
The children are, of course, responding to the questions on the basis of
their understanding of the terms used. Their understanding is based on the rich and
diverse associations that our civilization has placed on words such as “think” and
“feel,” but it is not influenced by the attempts of adult philosophers to provide more
precise definitions. What the children appear to be saying is that the analytic proc-
essing of a computer may be regarded as thinking, but that feeling and liking, both of
which involve an active conscious agent, are not sensible characterizations of a
computer. In other words, computers, or at least the computers that these children
have had experience with, are not conscious, but they do think, and therefore
thinking does not require consciousness.

39
 Asmenides: Greetings, Kurzus. | have ventured far to meet you and to
hear of your ideas on philosophical systems. You have a distin-
guished reputation among my people but have unfortunately created
many enemies with your ideas. | have even heard some rumors of
your disbelief in the existence of any supreme being or God.
Kurzus: My son, | am afraid that | have been misinter-
preted. Let me first define what | mean by a philosophical system.
Consider the following example of such a system. Is it not possible
that the only thing that actually exists is your mind and everything
that you perceive and think is merely fantasy, like a continuous
dream?
Asmenides: That sounds like fantasy to me.
Kurzus: Well, how do you know that what you see and
feel really exists as something other than your sense impressions of
it?
Asmenides: For one thing, other people looking at the

PEUTIC irantL | same building generally see the same thing. There must, therefore, be
something that is the building apart from the images of it.

A Platonic Dialogue on the Nature of Kurzus: \s it not possible that you merely imagine that
other people exist and tell you that they see the same thing you see
Human Thought
when you look at the alleged building?
Asmenides: This all seems very possible, but you cannot
really accept such a belief?
Kurzus: Whether | accept it or not is inconsequential. This
is merely an example of a working philosophical system, perhaps the
simplest example. We make certain assumptions, and if through
logical deductions they lead to no contradiction, then we say that we
have a logical system. We may have any number of systems, and the
most we can say is that any one of them is possible.
Asmenides: It seems to me that you can make any wild
assumption and end up with a logical result.
Kurzus: Ah, but there is where you are wrong. Imagine
that we take as assumptions things that we naturally assume to be
true in our every day life and end up with a contradiction. This is
where the true power of this method comes in. We have proved with
complete assurance that the system described by these assumptions
cannot exist! | have found, however, a number of systems that do
seem to work. When | discuss such a system, it does not mean that I
necessarily believe that this is the way things are; | mean merely that
it could be this way. | have found some systems that work with a God
or gods and some that work without.
Asmenides: Well, what about our religious system, have
you found any contradictions there?
Kurzus: | undoubtedly could find some. | imagine a system
could be developed that would incorporate many of the ideas in our
religious heritage, but it would be a somewhat arbitrary system. The

a0
fewer assumptions a system makes, the more powerful are its Myronius: Then you admit that we would come to a point

conclusions. where we would have an elementary particle of bone that could not
Asmenides: My father would beg to differ with you. be severed without losing the properties of bone.
Myronius walks by. Socrates: | have had similar thoughts, yes.
Kurzus: No doubt, but consider the nature of the soul. Myronius: And would this not be true of all things?
Myronius: Please forgive me for interrupting, but | could Socrates: You have learned the art of argument well,
not help but overhearing. | am, perhaps, more skeptical than either of Myronius. Yes, | imagine it is true for all things.
you gentlemen, but how do you know there exists such a thing as a Myronius: Then there must be a finite number of funda-
soul? mental particles of which the world is constructed.
Kurzus: Let us first accept as a definition of the soul Socrates: For the physical world, | will accept such a
something metaphysical that we associate with a person, animal, or theory.
object. | admit this is a poor definition, but we shall have to accept it Myronius: Now imagine, if you will, the following
until we find out more about the necessity of introducing a soul. machine, which | shall draw in the sand.
Myronius: You are defining something that may not exist. Myronius draws a simple machine with a lever and a
I contend that everything can be explained via the physical world. pulley.
Socrates: Myronius, are you mad? Have you learned Socrates: You have an imagination not unlike your father.
nothing from my teachings? This is certainly not like you. Myronius: \f | were to drop the ball on the lever, would
Myronius: Please forgive me for expounding a philosophy you expect the weight to move?
so heretical, but what I meant to say was that | believe a logical Socrates: | imagine it would rise.
system, as described by my friend Kurzus, can be developed in the Myronius: We can imagine the ball and the weight to be
realm of the physical world. This system would describe and account two fundamental particles. The motion of one causes the other to
for all of the phenomena that we are familiar with. | also contend that move in a distinctive and quite predictable manner. We can define
such a system would be a determined system. In other words, | the lever and the pulley to be an “ether” through which the particle of
believe that in such a system the future would be fixed and already the ball affects the particle of the weight. Analogously, can we not
determined. make the same claim with respect to the interaction between our real
Socrates: This should be interesting, but proceed. fundamental particles?
Myronius: Witness our colleague Asmenides, and Socrates: | see no reason for such a claim. Do you
imagine, if you will, that | cut him in half. imagine that there is such a system connected between all the
Asmenides: | suggest that we skip the demonstration. fundamental particles in the universe?
Myronius: That is too bad, it would have been so much Myronius: Certainly not, but is it not true that particles do
more effective, but if you insist, we shall rely on our minds, a interact? Otherwise, how would the pressure of my hand against this
precarious course, | admit. Now imagine, gentlemen, that | remove column cause it to move?
one bone and cut it in half. Do you, Socrates, believe that we should Asmenides: Careful, the whole house is liable to fall.
still have bone matter? Myronius: And were it to fall, would it not have been
Socrates: Apparently. caused by the interaction of the particles of my hand with the
Myronius: And if we broke this piece of bone in half particles of the column and their subsequent interactions with the
again, would we still have bone? particles of the rest of the house?
Socrates: This is all very true, although | fail to see any Socrates: Yes, | can see what you are driving at now.
connection. Myronius: There is apparently an ether, much finer and
Myronius: And if we continue this process, would we subtler, of course, than the ether of the pulley and string, that
always have bone? establishes a cause and effect relation between the fundamental
Socrates: | imagine we should have to stop eventually. particles. Do you not suppose that given the makeup of the ether and
the location and speed of all the fundamental particles, we could
predict the subsequent location and speed of all the particles at any
time in the future?

ai
 Socrates: \t would be an arduous task but theoretically Kurzus: Then couldn't a machine like the one you

possible. described, only more complex, be made to act just like you?

Myronius: Then is it not true that any future state of the Myronius: Yes, and it would be me.

universe is already defined? Kurzus: But this machine would not have an awareness of

Socrates: As to your contention that a completely itself, it would be just like the machine you described in the sand.

physical world is determined, | will agree, but please show me how Myronius: Yes, of course it would, that is all | am, a

everything can be explained in terms of the interactions of particles. complicated machine, a collection of particles.

Myronius: Well, it is obviously true for lifeless objects. As Kurzus: But you are different from a machine in one

for men and women, could it not be possible that we are merely a respect: you are aware of yourself and what is going on; the machine

vastly complex collection of particles that interact with each other is not. You go through life, and all the time there is a “screen” on

and the particles of other beings and objects? which you see or feel impressions of either reality or fantasy, it does

Socrates: Come now, this is hardly credible. What of not matter which in this argument. A machine would not have this

reason and memory, not to mention desire? property. It would receive visual impressions and immediately
Myronius: \magine, if you will, that we remember things process and record them without realizing what was happening.

by changing the relative locations of a number of memory particles, Does the machine that was made to simulate you realize what is

and that the retrieval of this memory and all of the logical manipula- happening?

tions constituting reason that we make with our memory and Myronius: If it were constructed in the same manner in

immediate sense impressions are also complex arrangements of which | am constructed, | see no reason why it would not be aware of

particles whose interactions define what we say, do, and feel. things in the same way | am aware of things.

Socrates: Then why does my student Akrios sometimes Kurzus: Let me take another example. Will you grant me

greet me with a nod of the head and sometimes with a more wordy that theoretically, a machine could exist that could reproduce any

greeting? object placed under it?

Myronius: What your pupil does, is, as | said, a function Myronius: For the sake of your argument, | will grant such

not only of the situation but of the internal states of all of his particles. a marvel.
Since Akrios learns something every day and gains new memories, Kurzus: Then imagine yourself placed in such a machine

he is not exactly the same person each day and can be expected to and a copy of you, particle for particle, is produced one hundred

act differently. Do you not suppose that if we placed a person ina lengths away. Would both “persons” be you?

situation, noted his reaction, then somehow returned him exactly to Myronius: | said they would.

his internal state before the experience and replaced him in the same Kurzus: Well, how could you have a consciousness (or

situation, he would act in exactly the same manner? should | say “an awareness”?) of what is in the minds of two persons

Socrates: Yes, | suppose this is so, otherwise he would one hundred lengths apart?

not be the same person if he had reacted differently. There does seem Myronius: | wouldn't. Each of us would have an aware-

to be something missing in your analysis, however. ness of our own. We are two different people with no possible

Kurzus: Perhaps | can help. How, Myronius, do you connection between us.

explain consciousness with your system? Kurzus: But you said they would both be you. How could

Myronius: Just what do you mean by consciousness? they both be you and yet be different?

Kurzus: Simply my awareness of my own existence, what Myronius: | don’t know, but | am not convinced yet.

happens to me, and how | react. Kurzus: After walking into the machine here, where

Myronius: Well, | would admit to the existence of this would you expect to come out, here or one hundred lengths away? Do

consciousness or awareness if you defined it in the following way: an not forget that both persons will claim to have lived the same lives,

ability to associate the information introduced by the senses and to then walked in here and walked out either here or there.

translate it into physical motion and speech. Myronius: | don’t know where | would walk out, certainly

Kurzus: That is not exactly what | mean by “conscious- somewhere, but what is your point?
ness.” What you have described—isn’t that an automatic reaction?
Myronius: Surely!

a2
 Kurzus: My point is simply this, that there exists a Kurzus: We have defined the soul as something meta-
consciousness in every person, and that this consciousness cannot physical. The physical world can be defined as all processes that can
be defined by particle interactions. To clarify the point a little, | will be described by particle interactions. We have previously argued that
say that I do not really know that anyone has consciousness but memory and the logic constituting wisdom can be described as an
myself. Everyone else may exist only in my mind, or they may exist as automatic process involving complex interactions of particles.
automatic machines, but I do know that | have this consciousness. For Socrates: The argument was given, but it was not
the sake of argument, however, let us assume that all of us gentlemen convincing. How can a machine have the many skills of perception
here are conscious. and reasoning possessed by men?
Myronius: | am vaguely understanding your point, but | Kurzus: Examine again our simple example of a machine.
still do not see why this consciousness cannot be explained by The machine has one possible “sense.” It was activated by dropping
physical interactions. the ball on the lever. It had one possible “reaction,” that of raising the
Kurzus: When you see something and then store it in your weight. When the ball was dropped, it would go through a logical
memory, are you aware of the exact process of particle interactions process and raise the weight. A human being has many senses, all of
that codes the information, relates it to previous memory, and stores which are considerably more complex than that of the lever. We too
it? can react in a multitude of ways. We produce reactions that are
Myronius: No, | cannot say that | am aware of the related to the state of our senses. If we had sufficient knowledge, we
mechanics of this process. could describe the connection between different reactions and their
Kurzus: And are there not a number of processes that go related sense and memory states. Certainly, then, a machine could be
on in your mind that you are not aware of? constructed that could produce similar reactions to the same external
Myronius: Yes, | suppose there are. conditions. After all, human beings do not have mysterious reactions.
Kurzus: Let us call the preconscious mind whatever takes We talk and move in ways that can be described with reasonable
care of all these processes of which we are not aware and the precision. Since wisdom and memory, which are the names we give
conscious mind whatever does things of which we are aware. Is the processes connecting sense impressions to reactions, can be
there any fundamental difference between these two minds, other described in cause and effect terms, they fit our definition of
than the things they deal with? belonging to the physical world and are therefore not metaphysical.
Myronius: | do not imagine there is any difference The soul, therefore, is synonymous with consciousness, since it is our
between the general principle of their construction. only metaphysical function.
Kurzus: \s there not, however, a difference between them, Socrates: Do you mean to say that the soul does not
an important difference, namely that you are aware of what is going contain wisdom or courage or temperance? What happens to the soul
on in one mind and not in the other? after the body dies? What does it have to be conscious of then?
Myronius: Yes, there is that difference. Kurzus: Let us examine again just what properties the
Kurzus: And wouldn't you say that this difference is the soul has. It has no power of reason or of memory, since these are
consciousness we have been speaking about? physical processes. It is essentially the real person that is aware of
Myronius: | see. Then consciousness is the real me, and what is happening to his body and mind and how they react. It makes
its function is to be aware of what is going on in my conscious mind. no difference whether a given body has a soul or not to anyone
Kurzus: Essentially, and that is a reasonable way of except that person. There is no possible way of finding out whether
putting it. Now, you yourself have said that there is no physical someone or some animal has consciousness. It is apparent only to its
difference between the conscious and nonconscious or preconscious possessor. We can differentiate two things as comprising a person.
mind, and since we have agreed that the difference between them is There is the material person, which includes his personality, memory,
consciousness, must we not conclude by a simple step of logic that physical makeup, and so on, and there is the soul, which has no
consciousness is not physical. Rather it is to be placed in a category characteristics. All souls are the same, and yet they are different, just
separate from particles and their interactions? as two stones can be exactly the same and yet be different stones by
Myronius: | will have to admit defeat. virtue of their different locations. Souls, of course, do not have
Socrates: Your thesis is very interesting, Kurzus, but what locations, as they do not belong to the physical world. They are
other characteristics would you attribute to the soul? merely associated with a particular person or thing. It is even
possible for a soul to be associated with the simple machine drawn

a3
here in the sand. It would make no difference to us or the soul, since Kurzus: Apparently not. You would have free will to do
the soul would have little to be aware of. There is no reason to what you want or decide to do, but what you decide to do is
suppose that the soul disperses when the body dies. It may be placed determined. In other words, if we were given the makeup of the
in a newborn baby’s body. It would not remember its past life, since particular ether and the location, velocity, and other properties of all
the soul has no memory. the particles that make up all the bodies and all the inanimate
Socrates: This is an absurd system. How do you know objects that make up the universe, we could predict the location and
that these souls don’t change bodies every three minutes? It would, speed of all the particles at any future time. Since what we do and
according to you, make no difference. If all of a sudden my soul were think is the sum of the motions of all these particles, everything we
to become associated with your body, | would think that | had always do is determined. We think we decide to do something, but really, we
been you, because | would have access to a memory of continuous merely observe certain aspects of the logical process used in
past. arriving at the decision. The existence of souls does not change this
Kurzus: Very good. Do you have any other suggestions? deterministic universe, because souls do not affect particles in any
Socrates: Well, who or what do you suppose decides way, they merely observe. To convince yourself of the possibility of
which soul goes into which body or animal? no free will, consider a dream. While you are dreaming, it certainly
Kurzus: \f my body is not the only body with a soul, there seems that you have control of your own actions, that you are making
is obviously a process that directs souls to bodies. | certainly have no decisions. Upon awaking, however, we often discover that we had
knowledge of the politics of this process. If it is a conscious being no real control, that it was “just a dream.” We had awareness merely
taking care of the assignments, we can call this being God. We can of the processes behind our dream “decisions.” It is apparently the
even fit your philosophy of Forms in this framework very easily. same way in the waking state.
Suppose, as you do, that the Forms of justice, virtue, wisdom, and so Socrates: Then the future is as fixed as the past.
on, exist. We can suppose that this God has a perfect knowledge of Kurzus: True.
these ideals and judges people as to how they live according to these Socrates: Then we could consider time as a fourth
Forms. He could then assign the souls of virtuous bodies to live again dimension, just like the three dimensions of absolute location, since
in some comfortable state and the souls of sinful bodies to live again just like absolute location, both directions in time are uniquely
in the body of an appropriate animal. This is, of course, an arbitrary defined and fixed.
system and may or may not exist. Kurzus: That would be a satisfactory way of looking at it.
Asmenides: Kurzus, in the system we have been discuss- Socrates: Then, theoretically, it should be just as

ing, the soul is metaphysical consciousness and has no other possible to go back and forth in time as it is in the other three
characteristics or abilities. Does this system necessarily describe dimensions, since time is merely another fixed dimension.
reality, or is this merely an example of just one of your many possible Kurzus: Yes, the universe is one static unchanging four-

logical systems? dimensional picture, and it should be theoretically possible to go

Kurzus: No, it must be so. We have proved that conscious- back and forth in time, although | know of no way to accomplish this
ness does exist, and we have also shown that it cannot be explained feat presently.
in terms of the physical world. We have also determined that wisdom Socrates: \f | went back in time and killed my great-

and memory can be explained by the physical world and therefore great-grandfather, could | not conclude that my great-grandfather

belong to it. We are left with no other alternative than the conclusion would never have been born?
that the real person is his or her metaphysical consciousness, which Kurzus: Yes, | imagine that would be so.

we Can suppose to be immortal. From this point on, we can make any Socrates: Then in a similar manner we can conclude that
number of hypothetical systems. We can assume that all men and my grandfather would never have been born, and so with my father

women have souls, or we can assume that only we have souls, or we and thus | would not have been born.
can assume that every living and lifeless object has a soul. Note that Kurzus: That is true.
if a lifeless object had a soul, the soul would not know about it. Socrates: But, if | had not been born, | could not have
Socrates: | see an important contradiction in your system. gone back in time and killed my great-great-grandfather, and thus my
Tell me, would there be any free will in your system? great-grandfather would have lived, and | too would have lived, and |
would have been able to kill my great-great-grandfather, and thus |
would not have existed, and so on, ad infinitum.

aa
 Kurzus: Yes, | see the problem.
Socrates: Therefore, time is not like the other three
dimensions. As we have seen, there is a logical proscription against
moving in it in more than one direction. The only way, however, for it
to be different from the other three dimensions is for it not to exist in
a definite form in either the future or the past. Since we know that
the past does exist in a definite form, we must conclude, therefore,
that the future is not determined. This would imply free will, would it
not?
Kurzus: Yes, | suppose it would.
Socrates: Since we have defined the physical world as
the predictable interactions of particles, we have to assign free will
as a property of the soul. Now if we add free will to the soul, would
we not have to add logic and memory in order to make decisions that
are not entirely arbitrary?
Kurzus: Yes, | imagine that would be necessary.
Socrates: Thus, we end up with a soul that has not only
consciousness but also free will plus wisdom and memory. We might
as well add courage and temperance.
Asmenides: | see a contradiction. First you, Kurzus,
proved that logic and memory must reside in the physical mind and
not in the metaphysical soul. Then you, Socrates, proved that logic
and memory must reside in the metaphysical soul. Apparently our
initial assumptions led to a contradiction and must, therefore, be
wrong.
Kurzus: My only initial assumption was that | am aware
of my own existence. Logic and memory can be defined as what |
think about, or rather what | am aware of. To deny this assumption
would be to deny either my awareness or my existence. | know that
there is awareness of something. | can only conclude, therefore, that
1 do not exist. This is not an entirely satisfactory result.
Asmenides: Now you can see, gentlemen, that no system
can work without gods. There are apparently gods that are able to
resolve these dilemmas. We are probably tricked into contradictory
conclusions in order to keep us from contemplating what is above
our power to understand.
Kurzus: | see no reason for resorting to such mysticism.
Socrates: My friends, it is time that I take leave of you. |
will think more of this discussion, and we can resume it again soon.
It has been a most profitable afternoon.
Kurzus: | think it an excellent idea to stop here just when
we are most confused. We shall all think better if we are not happy
with our conclusions.

as
 | am supposed to live in the age of intelligent machines. Frankly, I'd
rather be living in the age of intelligent people! What can one say of
an era when a computer fs smarter than the scientists who use it?
The most important subject taught to school children is
rapidly becoming computer science and applications: how to use
machines instead of brains, programs instead of knowledge. Soon
there will be crib computers for newborns—iittle brightly colored
affairs that record Junior's progress while entertaining him or her—
and a more complex model that understands and translates baby talk,
sings lullabies, and, later on, teaches reading, writing, and, of course,

typing.
In the present “back to nature” craze | am surprised that
nobody has thought of marketing a product with “all natural
intelligence—no artificial additives or preservatives of any kind.”
It once occurred to me that there are two ways to make a
computer man’s intellectual equal. First, one can make the computer
smarter. This may take thousands of researchers, millions of dollars,
Margaret
and a period of many years. The other, simpler way to go about this
The Age of Intelligent People rather ambitious project is to simply make the humans stupider. This
would take little time and money, since we are already well on the
way. What if, in fact, some genius has already put the plan into
Margaret Litvin was born in operation!
Russia in 1974. She and her
It is interesting to note how modern computer studies
family have lived in the United
States since 1979. She wrote relate to this goal. For example, let's say we have a computer that
this article while attending the corrects multiple-choice tests. Scientists develop a multiple-choice
seventh grade in Bedford, achievement test for school children, and the computer happily
Massachusetts.
corrects it. One can imagine a computer program that could score
better than average on a multiple-choice test like this. This would be
a great triumph for artificial intelligence, but not so fine for natural
intelligence: multiple-choice questions are easier to answer and
allow much more guesswork than essay questions or even fill-in-the-

Margaret Litvin. (Courtesy of

Margaret Litvin)

46
 programs to act as if they reason, act as if they understand, think,
learn, plan, enjoy, hate, etc.” What | don’t understand is his “act as
if.” Why can't scientists enjoy people, who really do these things?
Another of my nightmares is that when | grow up, all jobs
that require at least some intelligence will be taken up by computers.
Although this doesn’t yet have a full basis in reality, what little there
is, is rather disturbing. Many talented workers may end up in the
unemployment line replaced by impersonal machines. For example,
let's take my school. | have heard many wild, and not so wild,
fantasies about replacing teachers with machines. Now, I'm fully
confident that no computer in the world could teach as well as my
English teacher this past year—maybe more efficiently, certainly
more organized, but not nearly as well. (But take my math teacher. I'm
convinced, and so are my schoolmates that any computer could outdo
her without even trying!) This is not to say that all lousy teachers
blanks. This is just one example how a computer way of thinking
should be replaced with machines: most kids, myself included, would
affects our idea of natural intelligence and shifts it toward the
not like a computerized teacher one little bit. Instead of a Robotic
artificial.
Teachers’ Union | suggest a screening program so that nobody
For another example, let's take a word processor. The
lacking natural intelligence would be allowed the title of teacher.
word processor itself is so convenient that it makes people forget
One of the arguments used by those disapproving of
how to write anything but form letters. It also makes me sad to see
arti intelligence is that human beings shouldn't try to play God.
that such elegant subjects as graphology are dying. As for spelling
Oddly enough, I disagree with this point of view. | don’t see anything
correctors, using such a program is, | fear, enough to take away
wrong with people trying to imitate God, particularly since they won't
anyone's ability and desire to spell properly, leaving them utterly
succeed. After all, most people don't criticize a five-year-old girl upon
helpless should they be suddenly deprived of their computer. It seems
seeing her with a lifelike doll. The child sees how her mother walks
the computer gets smarter at the expense of its owner's brain. And
around with a real baby and wants to do likewise. One certainly
when the decision-making computers become as widespread as
wouldn't blame a five-year-old for imitating her mother. Similarly, |
hand-held calculators, who will decide if it is good or bad?
cannot find it in my heart to condemn humanity when it strives to
All right, suppose we have an artificial brain that is in all
imitate its Father by working on artificial intelligence.
senses equivalent to ours. It can reason (or maybe we can't!). It can
As far as | know, few religious scholars find that it is
learn to talk, read, write, and most important, think just like man can.
sinful to create intelligent machines. Besides, | think most reasonable
Why would anyone be attracted to this repulsive idea? In his August
people would agree with me that it won't work, anyway.
1985 presidential address “| Had a Dream” Woody Bledsoe discusses
his dream of “seeing a machine act like a human being, at least in
many ways.” Well, | too have a dream to see a human being act like a
human being! Wouldn't it be nice if all those clever, talented, and
devoted scientists and all that funding and interest went into
maintaining and enriching natural intelligence? Instead of trying to
impersonate nature, we could be helping her along. With all that
attention, perhaps, we could be more intelligent in the future
generations.
The reason why | opt for people is that | want not only
intellect in my smart creatures; | also want emotions, passions,
instincts, in short, everything that is so distinctly human in a person.
Later on in his presidential address Dr. Bledsoe says, “Oh, | am well
aware that the real problems are those of the mind, getting computer

a7
 Can machines think? This has been a conundrum for philosophers for
years, but in their fascination with the pure conceptual issues they
have for the most part overlooked the real social importance of the
answer. It is of more than academic importance that we learn to think
clearly about the actual cognitive powers of computers, for they are
now being introduced into a variety of sensitive social roles where
their powers will be put to the ultimate test: in a wide variety of
areas, we are on the verge of making ourselves dependent upon their
cognitive powers. The cost of overestimating them could be
enormous.
One of the principal inventors of the computer was the
great British mathematician Alan Turing. It was he who first figured
out, in highly abstract terms, how to design a programmable
computing device, what we now call a universal Turing machine. All
programmable computers in use today are in essence Turing
machines. About forty years ago, at the dawn of the computer age,
Turing began a classic article “Computing Machinery and Intelli-
gence” with the words “I propose to consider the question, ‘Can
machines think?” but he then went on to say that this was a bad
Can Machines Think?
question, a question that leads only to sterile debate and haggling
over definitions, a question, as he put it, “too meaningless to deserve
Danie! Dennett is Distinguished discussion.”'In its place he substituted what he took to be a much
Arts and Sciences Professor
better question, a question that would be crisply answerable and
and Director of the Center for
Cognitive Studies at Tufts intuitively satistying—in every way an acceptable substitute for the
University. He is the author or philosophic puzzler with which he began.
editor of a number of books on First he described a parlor game of sorts, the imitation
cognitive science and the
game, to be played by a man, a woman, and a judge (of either gender).
philosophy of mind, including
The Mind's |, coedited with The man and woman are hidden from the judge's view but are able to
Douglas Hofstadter (1981); communicate with the judge by teletype; the judge's task is to guess,
Elbow Room (1984); and The after a period of questioning each contestant, which interlocutor is
Intentional Stance (1987).
the man and which the woman. The man tries to convince the judge
he is the woman, and the woman tries to convince the judge of the
truth. The man wins if the judge makes the wrong identification. A
little reflection will convince you, | am sure, that aside from lucky
breaks, it would take a clever man to convince the judge that he was
the woman—on the assumption that the judge is clever too, of course.
Now suppose, Turing said, we replace the man or woman
with a computer and give the judge the task of determining which is
the human being and which is the computer. Turing proposed that any
computer that can regularly or often fool a discerning judge in this
game would be intelligent, a computer that thinks, beyond any
reasonable doubt. Now, it is important to realize that failing this test
is not supposed to be a sign of lack of intelligence. Many intelligent
people, after all, might not be willing or able to play the imitation

as
game, and we should allow computers the same opportunity to
decline to prove themselves. This is, then, a one-way test; failing it
proves nothing.
Furthermore, Turing was not committing himself to the
view (although it is easy to see how one might think he was) that to
think is to think just like a human being—any more than he was
commi ing himself to the view that for a man to think, he must think
exactly like a woman. Men, women, and computers may all have Daniel C. Dennett. (Courtesy of Tufts
different ways of thinking. But surely, he thought, if one can think in University)
one’s own peculiar style well enough to imitate a thinking man or
woman, one can think well, indeed. This imagined exercise has come
to be known as the Turing test.
Itis a sad irony that Turing’s proposal has had exactly the
opposite effect on the discussion of what he intended. Turing didn't
design the test as a useful tool in scientific psychology, a method of
confirming or disconfirming scientific theories or evaluating
particular models of mental function; he designed it to be nothing
more than a philosophical conversation stopper. He proposed—in the
spirit of “Put up or shut up!”—a simple test for thinking that is surely
strong enough to satisfy the sternest skeptic (or so he thought). He
was saying, in effect, that instead of arguing interminably about the
ultimate nature and essence of thinking, we should all agree that
whatever that nature is, anything that could pass this test would
surely have it; then we could turn to asking how or whether some
machine could be designed and built that might pass the test fair and
square. Alas, philosophers—amateur and professional—have instead
taken Turing's proposal as the pretext for just the sort of definitional
haggling and interminable arguing about imaginary counterexamples
that he was hoping to squelch.
This forty-year preoccupation with the Turing test has
been all the more regrettable because it has focused attention on the

ag
wrong issues. There are real world problems that are revealed by spot that it would ask what you wanted to say to it; if in another, that
considering the strengths and weaknesses of the Turing test, but it would cry that it was hurt, and so on for similar things. But it could
these have been concealed behind a smoke screen of misguided never modify its phrases to reply to the sense of whatever was sai
criticisms. A failure to think imaginatively about the test actually its presence, as even the most stupid men can do.”?
proposed by Turing has led many to underestimate its severity and to This seemed obvious to Descartes in the seventeenth
confuse it with much less interesting proposals. century, but of course, the fanciest machines he knew were elaborate
So first | want to show that the Turing test, conceived as clockwork figures, not electronic computers. Today it is far from
he conceived it, is (as he thought) quite strong enough as a test of obvious that such machines are impossible, but Descartes’s hunch
thinking. | defy anyone to improve upon it. But here is the point almost that ordinary conversation would put as severe a strain on artificial
universally overlooked by the literature: there is a common intelligence as any other test was shared by Turing. Of course, there
misapplication of the Turing test that often leads to drastic overesti- is nothing sacred about the particular conversational game chosen by
mation of the powers of actually existing computer systems. The Turing for his test; it is just a cannily chosen test of more general
follies of this familiar sort of thinking about computers can best be intelligence. The assumption Turing was prepared to make was this:
brought out by a reconsideration of the Turing test itself. Nothing could possibly pass the Turing test by winning the imitation
The insight underlying the Turing test is the same insight game without being able to perform indefinitely many other clearly
that inspires the new practice among symphony orchestras of intelligent actions. Let us call that assumption the quick-probe
conducting auditions with an opaque screen between the jury and assumption. Turing realized, as anyone would, that there are
the musician. What matters in a musician is, obviously, musical hundreds and thousands of telling signs of intelligent thinking to be
ability and only musical ability; such features as sex, hair length, skin observed in our fellow creatures, and one could, if one wanted,
color, and weight are strictly irrelevant. Since juries might be biased compile a vast battery of different tests to assay the capacity for
even innocently and unawares by these irrelevant features, they are intelligent thought. But success on his chosen test, he thought, would
carefully screened off so only the essential feature, musicianship, be highly predictive of success on many other intuitively acceptable
can be examined. Turing recognized that people might be similarly tests of intelligence. Remember, failure on the Turing test does not
biased in their judgments of intelligence by whether the contestant predict failure on those others, but success would surely predict
had soft skin, warm blood, facial features, hands, and eyes—which success. His test was so severe, he thought, that nothing that could
are obviously not themselves essential components of intelligence. pass it fair and square would disappoint us in other quarters. Maybe
So he devised a screen that would let through only a sample of what it wouldn't do everything we hoped—maypbe it wouldn't appreciate
really mattered: the capacity to understand, and think cleverly about, ballet, understand quantum physics, or have a good plan for world
challenging problems. Perhaps he was inspired by Descartes, who in peace, but we'd all see that it was surely one of the intelligent,
his Discourse on Method (1637) plausibly argued that there was no thinking entities in the neighborhood.
more demanding test of human mentality than the capacity to hold an Is this high opinion of the Turing test's severity mis-
intelligent conversation: “It is indeed conceivable that a machine guided? Certainly many have thought so, but usually because they
could be so made that it would utter words, and even words have not imagined the test in sufficient detail, and hence have
appropriate to the presence of physical acts or objects which cause underestimated it. Trying to forestall this skepticism, Turing imagined
some change in its organs; as, for example, if it was touched in some several lines of questioning that a judge might employ in this game
that would be taxing indeed—tines about writing poetry or playing

so
chess. But with thirty years’ experience with the actual talents and circumstances and many other topics. Thus, such sentences, by
foibles of computers behind us, perhaps we can add a few more putting a demand on such abilities, are good quick probes. That is,
tough lines of questioning. they test for a wider competence.
Terry Winograd, a leader in Al efforts to produce People typically ignore the prospect of having the judge
conversational ability in a computer, draws our attention to a pair of ask off-the-wall questions in the Turing test, and hence they
sentences.’ They differ in only one word. The first sentence is this: underestimate the competence a computer would have to have to
“The committee denied the group a parade permit because they pass the test. But remember, the rules of the imitation game as Turing
advocated violence.” Here's the second sentence: “The committee presented it permit the judge to ask any question that could be asked
denied the group a parade permit because they feared violence.” of a human being—no holds barred. Suppose, then, we give a
The difference is just in the verb—“ advocated” or contestant in the game this question: An Irishman found a genie in a
“feared.” As Winograd points out, the pronoun “they” in each bottle who offered him two wishes. “First I'll have a pint of Guin-
sentence is officially ambiguous. Both readings of the pronoun are ness,” said the Irishman, and when it appeared, he took several long
always legal. Thus, we can imagine a world in which governmental drinks from it and was delighted to see that the glass filled itself
committees in charge of parade permits advocate violence in the magically as he drank. “What about your second wish?” asked the
streets and, for some strange reason, use this as their pretext for genie. “Oh well, that's easy,” said the Irishman. “I'll have another one
denying a parade permit. But the natural, reasonable, intelligent of these!” Please explain this story to me, and tell me if there is
reading of the first sentence is that it's the group that advocated anything funny or sad about it.
violence, and of the second, that it's the committee that feared the Now even a child could express, even if not eloquently,
violence. the understanding that is required to get this joke. But think of how
Now if sentences like this are embedded in a conversa- much one has to know and understand about human culture, to put it
tion, the computer must figure out which reading of the pronoun is pompously, to be able to give any account of the point of this joke. |
meant, if it is to respond intelligently. But mere rules of grammar or am not supposing that the computer would have to laugh at, or be
vocabulary will not fix the right reading. What fixes the right reading amused by, the joke. But if it wants to win the imitation game—and
for us is knowledge about politics, social circumstances, committees that's the test, after all—it had better know enough in its own alien,
and their attitudes, groups that want to parade, how they tend to humorless way about human psychology and culture to be able to
behave, and the like. One must know about the world, in short, to pretend effectively that it was amused and explain why.
make sense of such a sentence. It may seem to you that we could devise a better test. Let's
In the jargon of artificial intelligence, a conversational compare the Turing test with some other candidates.
computer needs lots of world knowledge to do its job. But, it seems, if
Candidate 1
it is somehow endowed with that world knowledge on many topics, it
A computer is intelligent if it wins the World Chess Championship.
should be able to do much more with that world knowledge than
That's not a good test, it turns out. Chess prowess has
merely make sense of a conversation containing just that sentence.
proven to be an isolatable talent. There are programs today that can
The only way, it appears, for a computer to disambiguate that
play fine chess but do nothing else. So the quick-probe assumption is
sentence and keep up its end of a conversation that uses that
false for the test of playing winning chess.
sentence would be for it to have a much more general ability to
respond intelligently to information about social and political

S1
Candidate 2 This question is often obscured by a move called
The computer is intelligent if it soives the Arab-Israeli conflict. operationalism that philosophers have sometimes made. Turing and
those who think well of his test are often accused of being operation-
This is surely a more severe test than Turing’s. But it has
alists. Operationalism is the tactic of defining the presence of some
some defects: if passed once, it is unrepeatable; it is slow, no doubt;
property, intelligence, for instance, as being established once and for
and it is not crisply clear what would count as passing it. Here's
all by the passing of some test. Let's illustrate this with a different
another prospect, then:
example.
Candidate 3
Suppose |offer the following test—we'll call it the
A computer is intelligent if it succeeds in stealing the British crown
Dennett test—for being a great city. A great city is one in which, ona
jewels without the use of force or violence.
randomly chosen day, one can do all three of the following: hear a
Now this is better. First, it could be repeated again and symphony orchestra, see a Rembrandt anda professional athletic
again, though of course each repeat test would presumably be contest, and eat quenelles de brocheta la Nantua for lunch. To make
harder—but this is a feature it shares with the Turing test. Second, the operationalist move would be to declare that any city that passes
the mark of success is clear: either you've got the jewels to show for the Dennett test is by definition a great city. What being a great city
your efforts or you don't. But it is expensive and slow, a socially amounts to is just passing the Dennett test. Well then, if the Chamber
dubious caper at best, and no doubt luck would play too great a role. of Commerce of Great Falls, Montana, wanted—and | can’t imagine
With ingenuity and effort one might be able to come up why—to get their hometown on my list of great cities, they could
with other candidates that would equal the Turing test in severity, accomplish this by the relatively inexpensive route of hiring full time
fairness, and efficiency, but | think these few examples should suffice about ten basketball players, forty musicians, and a quick-order
to convince us that it would be hard to improve on Turing’s original quenelle chef and renting a cheap Rembrandt from some museum. An
proposal. idiotic operationalist would then be stuck admitting that Great Falls,
But still, you may protest, something might pass the Turing Montana, was in fact a great city, since all he or she cares about in
test and still not be intelligent, not be a thinker. What does might great cities is that they pass the Dennett test.
mean here? If what you have in mind is that by cosmic accident, by a Sane operationalists (who for that very reason are
supernatural coincidence, a stupid person or a stupid computer might perhaps not operationalists at all, since “operationalist” seems to be
fool a clever judge repeatedly, well, yes, but so what? The same a dirty word) would cling confidently to their test, but only because
frivolous possibility “in principle” holds for any test whatever. A they have what they consider to be very good reasons for thinking the
playful god or evil demon, let us agree, could fool the world’s odds astronomical against a false positive result, like the imagined
scientific community about the presence of H,0 in the Pacific Ocean. Chamber of Commerce caper. | devised the Dennett test, of course,
But still, the tests they rely on to establish that there is H,0 in the with the realization that no one would be both stupid and rich enough
Pacific Ocean are quite beyond reasonable criticism. If the Turing to go to such preposterous lengths to foil the test. In the actual world, i
test for thinking is no worse than any well-established scientific test, wherever you find symphony orchestras, quenelles, Rembrandts, and
we can set skepticism aside and go back to serious matters. Is there professional sports, you also find daily newspapers, parks, repertory
any more likelihood of a false positive result on the Turing test than theaters, libraries, fine architecture, and all the other things that go to i
{
on, say, the tests currently used for the presence of iron in an ore make a city great. My test was simply devised to locate a telling |
sample? sample that could not help but be representative of the rest of the |
j
52
city's treasures. | would cheerfully run the minuscule risk of having geologists and other experts thought of asking it about moon rocks.
my bluff called. Obviously, the test items are not all that | care about (In 12 percent of those correct responses there were trivial, cor-
ina city. In fact, some of them | don’t care about at all. | just think they rectable defects.) Of course, Wood's motive in creating LUNAR was
would be cheap and easy ways of assuring myself that the subtle not to trick unwary geologists into thinking they were conversing
things | do care about in cities are present. Similarly, | think it would with an intelligent being. And if that had been his motive, his project
be entirely unreasonable to suppose that Alan Turing had an would still be a long way from success.
inordinate fondness for party games or put too high a value on party- For it is easy enough to unmask LUNAR without ever
game prowess in his test. In both the Turing test and the Dennett test straying from the prescribed topic of moon rocks. Put LUNAR in one
a very unrisky gamble is being taken: the gamble that the quick-probe room and a moon rocks specialist in another, and then ask them both
assumption is in general safe. their opinion of the social value of the moon-rock-gathering
But two can play this game of playing the odds. Suppose expeditions, for instance. Or ask the contestants their opinion of the
some computer programmer happens to be, for whatever strange suitability of moon rocks as ashtrays, or whether people who have
reason, dead set on tricking me into judging an entity to be a thinking, touched moon rocks are ineligible for the draft. Any intelligent person
intelligent thing when it is not. Such a trickster could rely as well as | knows a lot more about moon rocks than their geology. Although it
can on unlikelihood and take a few gambles. Thus, if the programmer might be unfair to demand this extra knowledge of a computer moon-
can expect that it is not remotely likely that I, as the judge, will bring rock specialist, it would be an easy way to get it to fail the Turing
up the topic of children’s birthday parties, or baseball, or moon rocks, test.

then he or she can avoid the trouble of building world knowledge on But just suppose that someone could extend LUNAR to
those topics into the data base. Whereas if | do improbably raise cover itself plausibly on such probes, so long as the topic was still,
these issues, the system will draw a blank, and | will unmask the however indirectly, moon rocks. We might come to think it was a lot
pretender easily. But with all the topics and words that | might raise, more like the human moon-rock specialist than it really was. The
such a saving would no doubt be negligible. Turn the idea inside out, moral we should draw is that as Turing-test judges we should resist
however, and the trickster will have a fighting chance. Suppose the all limitations and waterings-down of the Turing test. They make the
programmer has reason to believe that | will ask on/y about children's game too easy—vastly easier than the original test. Hence, they lead
birthday parties or baseball or moon rocks—all other topics being, for us into the risk of overestimating the actual comprehension of the
one reason or another, out of bounds. Not only does the task shrink system being tested.
dramatically, but there already exist systems or preliminary sketches Consider a different limitation on the Turing test that
of systems in artificial intelligence that can do a whiz-bang job of should strike a suspicious chord in us as soon as we hear it. This is a
responding with apparent intelligence on just those specialized variation on a theme developed in a recent article by Ned Block.*
topics. Suppose someone were to propose to restrict the judge to a
William Wood's LUNAR program, to take what is perhaps vocabulary of, say, the 850 words of Basic English, and to single-
the best example, answers scientists’ questions—posed in ordinary sentence probes—that is, “moves”—of no more than four words.
English—about moon rocks. In one test it answered correctly and Moreover, contestants must respond to these probes with no more
appropriately something like 90 percent of the questions that than four words per move, and a test may involve no more than forty
questions.

53
 Is this an innocent variation on Turing’s original test? avoiding combinatorial explosion (by any means at all) be viewed as
These restrictions would make the imitation game clearly finite. That one of the hallmarks of intelligence.
is, the total number of all possible permissible games is a large but Our brains are millions of times bigger than the brains of
finite number. One might suspect that such a limitation would permit gnats, but they are still—for all their vast complexity—compact,
the trickster simply to store, in alphabetical order, all the possible efficient, timely organs that somehow or other manage to perform all
good conversations within the limits and fool the judge with nothing their tasks while avoiding combinatorial explosion. A computer a
more sophisticated than a system of table lookup. In fact, that isn't in million times bigger or faster than a human brain might not look like
the cards. Even with these severe, improbable, and suspicious the brain of a human being, or even be internally organized like the
restrictions imposed upon the imitation game, the number of legal brain of a human being, but if, for all its differences, it somehow
games, though finite, is mind-bogglingly large. | haven't bothered managed to control a wise and timely set of activities, it would have
trying to calculate it, but it surely astronomically exceeds the number to be the beneficiary of a very special design that avoided combinato-
of possible chess games with no more than forty moves, and that rial explosion. And whatever that design was, would we not be right
number has been calculated. John Haugeland says it's in the to consider the entity intelligent?
neighborhood of 10. For comparison, Haugeland suggests there Turing’s test was designed to allow for this possibility.
have only been 10" seconds since the beginning of the universe.> His point was that we should not be species-chauvinistic, or
Of course, the number of good, sensible conversations anthropocentric, about the insides of an intelligent being, for there
under these limits is a tiny fraction, maybe 1 in 10", of the number of might be inhuman ways of being intelligent.
merely grammatically well-formed conversations. So let's say, to be To my knowledge the only serious and interesting attempt
very conservative, that there are only 10° different smart conversa- by any program designer to win even a severely modified Turing test
tions such a computer would have to store. Well, the task shouldn't has been Kenneth Colby’s. Colby is a psychiatrist and intelligence
take more than a few trillion years—with generous federal support. artificer at UCLA. He has a program called PARRY, which is a
Finite numbers can be very large. computer simulation of a paranoid patient who has delusions about
So though we needn't worry that this particular trick of the Mafia being out to get him. As you do with other conversational
storing all the smart conversations would work, we can appreciate programs, you interact with it by sitting at a terminal and typing
that there are lots of ways of making the task easier that may appear questions and answers back and forth. A number of years ago, Colby
innocent at first. We also get a reassuring measure of just how severe put PARRY to a very restricted test. He had genuine psychiatrists
the unrestricted Turing test is by reflecting on the more than interview PARRY. He did not suggest to them that they might be
astronomical size of even that severely restricted version of it. talking or typing to a computer; rather, he made up some plausible
Block's imagined—and utterly impossible—program story about why they were communicating with a real, live patient by
exhibits the dreaded feature known in computer-science circles as teletype. He also had the psychiatrists interview real, human
combinatorial explosion. No conceivable computer could overpower paranoids via teletype. Then he took a PARRY transcript, inserted it in
a combinatorial explosion with sheer speed and size. Since the a group of teletype transcripts from real patients, gave them to
problem areas addressed by artificial intelligence are veritable another group of experts—more psychiatrists—and said, “One of
minefields of combinatorial explosion, and since it has often proved these was a conversation with a computer. Can you figure out which
difficult to find any solution to a problem that avoids them, there is one it was?” They couldn't. They didn’t do better than chance.
considerable plausibility in Newell and Simon's proposal that

54
 Colby presented this with some huzzah, but critics scoffed deemed to be relevant to testing the model as a model of paranoia. So
at the suggestion that this was a legitimate Turing test. My favorite they asked just the sort of questions that therapists typically ask of
commentary on it was Joseph Weizenbaum’s; in a letter to the such patients, and of course PARRY had been ingeniously and
Communications of the Association of Computing Machinery, he said laboriously prepared to deal with just that sort of question.
that, inspired by Colby, he had designed an even better program, One of the psychiatrist judges did, in fact, make a rather
which passed the same test. His also had the virtue of being a very half-hearted attempt to break out of the mold and ask some telling
inexpensive program, in these times of tight money. In fact you didn’t questions: “Maybe you've heard the saying ‘Don't cry over spilled
even need a computer for it. All you needed was an electric milk.’ What does that mean to you?” PARRY answered, “Maybe you
typewriter. His program modeled infantile autism. And the tran- have to watch out for the Mafia.” When then asked “Okay, now if you
scripts—you type in your questions, and the thing just sits there and were in a movie theater watching a movie and smelled something
hums—cannot be distinguished by experts from transcripts of real like burning wood or rubber, what would you do?” PARRY replied,
conversations with infantile autistic patients. What was wrong with “You know, they know me.” And the next question was, “If you found
Colby’s test, of course, was that the unsuspecting interviewers had no a stamped, addressed letter in your path as you were walking down
motivation at all to try out any of the sorts of questions that easily the street, what would you do?” PARRY replied, “What else do you
would have unmasked PARRY. want to know?"”
Colby was undaunted, and after his team had improved Clearly, PARRY was, you might say, parrying these
PARRY, he put it to a much more severe test—a surprisingly severe questions, which were incomprehensible to it, with more or less
test. This time, the interviewers—again psychiatrists—were given stock paranoid formulas. We see a bit of a dodge that is apt to work,
the task at the outset of telling the computer from the real patient. apt to seem plausible to the judge, only because the “contestant” is
They were set up in a classic Turing machine test: the patient in one supposed
to be a paranoid, and such people are expected to respond
toom, the computer PARRY in the other room, with the judges uncooperatively on such occasions. These unimpressive responses
conducting interviews with both of them (on successive days). The didn’t particularly arouse the suspicions of the judge, as a matter of
judges’ task was to find out which one was the computer and which fact, though they probably should have.
one was the real patient. Amazingly, they didn’t do much better, PARRY, like all other large computer programs, is
which leads some people to say, “Well, that just confirms my dramatically bound by limitations of cost-effectiveness. What was
impression of the intelligence of psychiatrists!” important to Colby and his crew was simulating his model of
But more seriously now, was this an honest-to-goodness paranoia. This was a massive effort. PARRY has a thesaurus or
Turing test? Were there tacit restrictions on the lines of questioning dictionary of about 4,500 words and 700 idioms and the grammatical
of the judges? Like the geologists interacting with LUNAR, the competence to use it—a parser, in the jargon of computational
psychiatrists’ professional preoccupations and habits kept them from linguistics. The entire PARRY program takes up about 200,000 words
asking the sorts of unlikely questions that would have easily of computer memory, all laboriously installed by the programming
unmasked PARRY. After all, they realized that since one of the team. Now once all the effort had gone into devising the model of
contestants was a real, live paranoid person, medical ethics virtually paranoid thought processes and linguistic ability, there was little
forbade them from toying with, upsetting, or attempting to confuse time, energy, money, and interest left over to build in huge amounts of
their interlocutors. Moreover, they also knew that this was a test of a
model of paranoia, so there were certain questions that wouldn't be

55
world knowledge of the sort that any actual paranoid would, of memories of the events they participate in. Hence, it was meant to be |
course, have. (Not that anyone yet knows how to build in world a “pure” Al system, a scientific model, not an expert system intended
knowledge in the first place.) Even if one could do it, building in the for any practical purpose. CYRUS was updated daily by being fed all
world knowledge would no doubt have made PARRY orders of UPI wire-service news stories that mentioned Vance, and it was fed
magnitude larger and slower. And what would have been the point, them directly with no doctoring and no human intervention. With an
given Colby’s theoretical aims? ingenious news-reading program called FRUMP, it could take any
PARRY is a theoretician’s model of a psychological story just as it came in on the wire and could digest it and use it to
phenomenon: paranoia. It is not intended to have practical applica- update its data base so that it could answer more questions. You
tions. But in recent years there has appeared a branch of Al (knowl- could address questions to CYRUS in English by typing at a terminal.
edge engineering) that develops what are now called expert systems. You addressed CYRUS in the second person, as if you were talking
Expert systems are designed to be practical. They are typically with Cyrus Vance himself. The results looked like this:
software superspecialist consultants that can be asked to diagnose
Question: Last time you went to Saudi Arabia, where did you stay?
medical problems, analyze geological data, analyze the results of
Answer: In a palace in Saudi Arabia on September 23, 1978.
scientific experiments, and the like. Some of them are very impres-
Question: Did you go sightseeing there?
sive. SRI in California announced a few years ago that PROSPECTOR,
Answer: Yes, at an oilfield in Dharan on September 23, 1978.
an SRi-developed expert system in geology, had correctly predicted
Question: Has your wife ever met Mrs. Begin?
the existence of a large, important mineral deposit that had been
Answer: Yes, most recently at a state dinner in Israel in January 1980.
entirely unanticipated by the human geologists who had fed it its
CYRUS could correctly answer thousands of questions—
data. MYCIN, perhaps the most famous of these expert systems,
almost any fair question one could think of asking it. But if one
diagnoses infections of the blood, and it does probably as well as,
actually set out to explore the boundaries of its facade and find the
maybe better than, any human consultants. And many other expert
questions that overshot the mark, one could soon find them. “Have
systems are on the way.
you ever met a female head of state?” was a question | asked it,
All expert systems, like all other large Al programs, are
wondering if CYRUS knew that Indira Ghandi and Margaret Thatcher
what you might call Potemkin villages. That is, they are cleverly
were women. But for some reason the connection could not be
constructed facades, like cinema sets. The actual filling-in of details
drawn, and CYRUS failed to answer either yes or no. | had stumped it,
of Al programs is time-consuming, costly work, so economy dictates
in spite of the fact that CYRUS could handle a host of what you might
that only those surfaces of the phenomenon that are likely to be
call neighboring questions flawlessly. One soon learns from this sort
probed or observed are represented.
of probing exercise that it is very hard to extrapolate accurately from
Consider, for example, the CYRUS program developed by
a sample performance to the system's total competence. It's also very
Janet Kolodner in Roger Schank’s Al group at Yale a few years ago.’
hard to keep from extrapolating much too generously.
CYRUS stands (we are told) for “Computerized Yale Retrieval and
While | was visiting Schank’s laboratory in the spring of
Updating System,” but surely it is no accident that CYRUS modeled
1980, something revealing happened. The real Cyrus Vance suddenly
the memory of Cyrus Vance, who was then secretary of state in the
resigned. The effect on the program CYRUS was chaotic. It was
Carter administration. The point of the CYRUS project was to devise
utterly unable to cope with the flood of “unusual” news about Cyrus
and test some plausible ideas about how people organize their
Vance. The only sorts of episodes CYRUS could understand at all

56
were diplomatic meetings, flights, press conferences, state dinners, sudden resignation, Kolodner and her associates soon had CYRUS up
and the like—less than two dozen general sorts of activities (the and running again with a new talent—answering questions about
kinds that are newsworthy and typical of secretaries of state). It had Edmund Muskie, Vance's successor. But it was no less vulnerable to
no provision for sudden resignation. It was as if the UPI had reported unexpected events. Not that it mattered particularly, since CYRUS
that a wicked witch had turned Vance into a frog. It is distinctly was a theoretical model, not a practical system.
possible that CYRUS would have taken that report more in stride than There are a host of ways of improving the performance of
the actual news. One can imagine the conversation such systems, and, of course, some systems are much better than
others. But all Al programs in one way or another have this facadelike
Question: Hello, Mr. Vance, what's new?
quality, simply for reasons of economy. For instance, most expert
Answer: | was turned into a frog yesterday.
systems in medical diagnosis developed so far operate with statisti-
But, of course, it wouldn't know enough about what it had
cal information. They have no deep or even shallow knowledge of the
just written to be puzzled, startled, or embarrassed. The reason is
underlying causal mechanisms of the phenomena that they are
obvious. When you look inside CYRUS, you find that it has skeletal
diagnosing. To take an imaginary example, an expert system asked to
definitions of thousands of words, but these definitions are minimal.
diagnose an abdominal pain would be oblivious to the potential
They contain as little as the system designers think that they can get
import of the fact that the patient had recently been employed as a
away with. Thus, perhaps, “lawyer” would be defined as synonymous
sparring partner by Muhammed Ali: there being no statistical data
with “attorney” and “legal counsel,” but aside from that, all one
available to it on the rate of kidney stones among athlete's assistants.
would discover about lawyers is that they are adult human beings
That's a fanciful case no doubt—too obvious, perhaps, to lead to an
and that they perform various functions in legal areas. If you then
actual failure of diagnosis and practice. But more subtle and hard-to-
traced out the path to “human being,” you'd find out various obvious
detect limits to comprehension are always present, and even experts,
things CYRUS “knew” about human beings (hence about lawyers),
even the system's designers, can be uncertain of where and how
but that is not a lot. That lawyers are university graduates, that they
these limits will interfere with the desired operation of the system.
are better paid than chambermaids, that they know how to tie their
Again, steps can be taken and are being taken to correct these flaws.
shoes, that they are unlikely to be found in the company of lumber-
For instance, my former colleague at Tufts, Benjamin Kuipers, is
jacks—these trivial, if weird, facts about lawyers would not be
currently working on an expert system in nephrology for diagnosing
explicit or implicit anywhere in this system. In other words, a very
kidney ailments that will be based on an elaborate system of causal
thin stereotype of a lawyer would be incorporated into the system, so
reasoning about the phenomena being diagnosed. But this is a very
that almost nothing you could tell it about a lawyer would surprise it.
ambitious, long-range project of considerable theoretical difficulty.
So long as surprising things don’t happen, so long as Mr.
And even if all the reasonable, cost-effective steps are taken to
Vance, for instance, leads a typical diplomats life, attending state
minimize the superficiality of expert systems, they will still be
dinners, giving speeches, flying from Cairo to Rome, and so forth, this
facades, just somewhat thicker or wider facades.
‘system works very well. But as soon as his path is crossed by an
When we were considering the fantastic case of the crazy
important anomaly, the system is unable to cope and unable to
Chamber of Commerce of Great Falls, Montana, we couldn't imagine a
recover without fairly massive human intervention. In the case of the
plausible motive for anyone going to any sort of trouble to trick the

s7
Dennett test. The quick-probe assumption for the Dennett test looked | come, then, to my conclusions. First, a philosophical or |
quite secure. But when we look at expert systems, we see that, theoretical conclusion: The Turing test, in unadulterated, unre- |
however innocently, their designers do have motivation for doing stricted form as Turing presented it, is plenty strong if well used. | am |
exactly the sort of trick that would fool an unsuspicious Turing confident that no computer in the next twenty years in going to pass
tester. First, since expert systems are all superspecialists that are the unrestricted Turing test. They may well win the World Chess
only supposed to know about some narrow subject, users of such Championship or even a Nobel Prize in physics, but they won't pass
systems, not having much time to kill, do not bother probing them at the unrestricted Turing test. Nevertheless, it is not, | think, impos-
the boundaries at all. They don’t bother asking “silly” or irrelevant sible in principle for a computer to pass the test fair and square. I'm
questions. Instead, they concentrate—not unreasonably—on not giving one of those a priori “computers can’t think” arguments. |
exploiting the system's strengths. But shouldn't they try to obtain a stand unabashedly ready, moreover, to declare that any computer
clear vision of such a system's weaknesses as well? The normal that actually passes the unrestricted Turing test will be, in every
habit of human thought when we converse with one another is to theoretically interesting sense, a thinking thing.
assume general comprehension, to assume rationality, to assume, But remembering how very strong the Turing test is, we
moreover, that the quick-probe assumption is, in general, sound. This must also recognize that there may also be interesting varieties of
amiable habit of thought almost irresistibly leads to putting too much thinking or intelligence that are not well poised to play and win the
faith in computer systems, especially user-friendly systems that imitation game. That no nonhuman Turing-test winners are yet visible
present themselves in a very anthropomorphic manner. on the horizon does not mean that there aren't machines that already
Part of the solution to this problem is to teach all users of exhibit some of the important features of thought. About them it is
computers, especially users of expert systems, how to probe their probably futile to ask my title question, Do they think? Do they really
systems before they rely on them, how to search out and explore the think? In some regards they do, and in some regards they don't. Only
boundaries of the facade. This is an exercise that calls for not only a detailed look at what they do and how they are structured will
intelligence and imagination but also for a bit of special understand- reveal what is interesting about them. The Turing test, not being a
ing about the limitations and actual structure of computer programs. scientific test, is of scant help on that task, but there are plenty of
It would help, of course, if we had standards of truth in advertising, in other ways to examine such systems. Verdicts on their intelligence,
effect, for expert systems. For instance, each such system should capacity for thought, or consciousness will be only as informative
come with a special demonstration routine that exhibits the sorts of and persuasive as the theories of intelligence, thought, or conscious-
shortcomings and failures that the designer knows the system to ness the verdicts were based on, and since our task is to create such
have. This would not be a substitute, however, for an attitude of theories, we should get on with it and leave the Big Verdict for
cautious, almost obsessive, skepticism on the part of users, for another occasion. In the meantime, should anyone want a surefire
designers are often, if not always, unaware of the subtler flaws in the test of thinking by a computer that is almost guaranteed to be fail-
products they produce. That is inevitable and natural because of the safe, the Turing test will do very nicely.
way system designers must think. They are trained to think posi- My second conclusion is more practical and hence in one
tively—constructively, one might say—about the designs that they clear sense more important. Cheapened versions of the Turing test
are constructing. are everywhere in the air. Turing’s test is not just effective, it is
entirely natural; this is, after all, the way we assay the intelligence of
each other every day. And since incautious use of such judgments

58
and such tests is the norm, we are in some considerable danger of “You may well be right,” Turing could say, “that eyes,
extrapolating too easily and judging too generously about the ears, hands, and a history are necessary conditions for thinking. If so,
understanding of the systems we are using. The problem of overesti- then | submit that nothing could pass the Turing test that didn’t have
mating cognitive prowess, comprehension, and intelligence is not, eyes, ears, hands, and a history. That is an empirical claim, which
then, just a philosophical problem. It is a real social problem, and we we can someday hope to test. If you suggest that these are not just
should alert ourselves to it and take steps to avert it. practically or physically necessary but conceptually necessary
conditions for thinking, you make a philosophical claim that | for one
Postscript: Eyes, Ears, Hands, and History would not know how, or care, to assess. Isn't it more interesting and
eT important in the end to discover whether or not it is true that no
My philosophical conclusion in this paper is that any bedridden system could pass a demanding Turing test?”
computer that actually passed the Turing test would be a thinker in Suppose we put to Turing the suggestion that he add
every theoretically interesting sense. This conclusion seems to some another component to his test: Not only must an entity win the
people to fly in the face of what | have myself argued on other imitation game; it must also be able to identify—using whatever
occasions. Peter Bieri, commenting on this paper at Boston Univer- sensory apparatus it has available to it—a variety of familiar objects
sity, noted that | have often claimed to show the importance to placed in its room: a tennis racket, a potted palm, a bucket of yellow
genuine understanding of a rich and intimate perceptual intercon- paint, a live dog. This would ensure that somehow or other the entity
nection between an entity and its surrounding world—the need for was capable of moving around and distinguishing things in the
something like eyes and ears—and a similarly complex active world. Turing could reply, | assert, that this is an utterly unnecessary
engagement with elements in that world—the need for something addition to his test, making it no more demanding than it already
like hands with which to do things in that world. Moreover, | have was. A suitably probing conversation would surely establish beyond
often held that only a biography of sorts—a history of actual projects, a shadow of a doubt that the contestant knew its way around in the
learning experiences, and other bouts with reality—could produce real world. The imagined alternative of somehow “prestocking” a
the sorts of complexities (both external, or behavioral, and internal) bedridden, blind computer with enough information and a clever
that are needed to ground a principled interpretation of an entity as a enough program to trick the Turing test is science fiction of the worst
thinker, an entity with beliefs, desires, intentions, and other mental kind: possible “in principle” but not remotely possible in fact in view
attitudes. of the combinatorial explosion of possible variation such a system
But the opaque screen in the Turing test discounts or would have to cope with.
dismisses these factors altogether, it seems, by focusing attention on “But suppose you're wrong. What would you say of an
only the contemporaneous capacity to engage in one very limited entity that was created all at once (by some programmers, perhaps),
sort of activity: verbal communication. (I have even coined a an instant individual with all the conversational talents of an
pejorative label for such purely language-using systems: “bedrid- embodied, experienced human being?” This is like the question,
den.”) Am | going back on my earlier claims? Not at all. | am merely Would you call a hunk of H,0 that was as hard as steel at room
pointing out that the Turing test is so powerful that it will indirectly temperature ice? | do not know what Turing would say, of course, so
ensure that these conditions, if they are truly necessary, are met by | will speak for myself. Faced with such an improbable violation of
any successful contestant. what | take to be the laws of nature, | would probably be speechless.
The least of my worries would be about which lexicographical leap
to take, whether to say, “It turns out, to my amazement, that something long as the scientist who is attempting to manipulate you does not
can think without having had the benefit of eyes, ears, hands, and a share all your knowledge, his or her chances of manipulating you are
history” or “It turns out, to my amazement, that something can pass minimal. People can always hit you over the head. They can do that
the Turing test without thinking.” Choosing between these ways of now. We don’t need artificial intelligence to manipulate people by |
expressing my astonishment would be asking myself a question too putting them in chains or torturing them. But if someone tries to |
meaningless to deserve discussion. manipulate you by controlling your thoughts and ideas, that person
will have to know what you know and more. The best way to keep
Discussion yourself safe from that kind of manipulation is to be well informed.
aD
Question: Do you think we will be able to program self-
Question: Why was Turing interested in differentiating a man from a consciousness into a computer?
woman in his famous test? Answer: Yes, | do think that it's possible to program self-
Answer: That was just an example. He described a parlor consciousness into a computer. “Self-consciousness” can mean
game in which a man would try to fool the judge by answering many things. If you take the simplest, crudest notion of self-con-
questions as a woman would answer. | suppose that Turing was sciousness, | suppose that would be the sort of self-consciousness
playing on the idea that maybe, just maybe, there is a big difference that a lobster has: When it's hungry, it eats something, but it never
between the way men think and the way women think. But of course eats itself. It has some way of distinguishing between itself and the
they're both thinkers. He wanted to use that fact to make us realize rest of the world, and it has a rather special regard for itself. The
that, even if there were clear differences between the way a lowly lobster is, in one regard, self-conscious. If you want to know
computer and a person thought, they'd both still be thinking. whether or not you can create that on the computer, the answer is
Question: Why does it seem that some people are upset yes. It's no trouble at all. The computer is already a self-watching,
by Al research? Does Al research threaten our self-esteem? self-monitoring sort of thing. That is an established part of the
Answer: | think Herb Simon has already given the technology. But, of course, most people have something more in mind
canniest diagnosis of that. For many people the mind is the last refuge when they speak of self-consciousness. It is that special inner light,
of mystery against the encroaching spread of science, and they don't that private way that it is with you that nobody else can share,
like the idea of science engulfing the last bit of terra incognita. This something that is forever outside the bounds of computer science.
means that they are threatened, | think irrationally, by the prospect How could a computer ever be conscious in this sense? That belief,
that researchers in artificial intelligence may come to understand the that very gripping, powerful intuition, is in the end, | think, simply an
human mind as well as biologists understand the genetic code and illusion of common sense. It is as gripping as the commonsense
physicists understand electricity and magnetism. This could lead to illusion that the earth stands still and the sun goes around the earth.
the “evil scientist” (to take a stock character from science fiction) But the only way that those of us who do not believe in the illusion
who can control you because he or she has a deep understanding of will ever convince the general public that it is an illusion is by
what's going on in your mind. This seems to me to be a totally gradually unfolding a very difficult and fascinating story about just
valueless fear, one that you can set aside for the simple reason that what is going on in our minds. In the interim, people like me—
the human mind is full of an extraordinary amount of detailed
knowledge, as Roger Schank, for example, has been pointing out. As
philosophers who have to live by our wits and tell a lot of stories— Notes
=
use what | call intuition pumps, little examples that help to free up the
imagination. | simply want to draw your attention to one fact. If you 1. Alan M. Turing, “Computing Machinery and Intelligence,” Mind 59 (1950).
look at a computer—t don’t care whether it's a giant Cray or a 2. René Descartes, Discourse on Method (1637), trans. Lawrence LaFleur (New York:
personal computer—if you open up the box and look inside and see Bobbs-Merrill, 1960).

those chips, you say, “No way could that be conscious. No way could
3. Terry Winograd, Understanding Natural Language (New York: Academic Press,
1972).
that be self-conscious.” But the same thing is true if you take the top
4. Ned Block, “Psychologism and Behaviorism,” Philosophical Review, 1982.
off somebody's skull and look at the gray matter pulsing away in
5, John Haugeland, Mind Design (Cambridge, Mass.: MIT Press, 1981), p. 16.
there. You think, “That is conscious? No way could that lump of stuff
6. Joseph Weizenbaum, CACM 17, no. 3 (September 1974): 543.
be conscious.” Of course, it makes no difference whether you look at
7.1 thank Kenneth Colby for providing me with the complete transcripts (including
it with a microscope or with the naked eye. At no level of inspection
the judges’ commentaries and reactions) from which these exchanges are quoted.
does a brain look like the seat of consciousness. Therefore, don't The first published account of the experiment is Jon F. Heiser, Kenneth Mark Colby,
expect a computer to look like the seat of consciousness. If you want William S. Faught, and Roger C. Parkinson, “Can Psychiatrists Distinguish a
to get a grasp of how a computer could be conscious, it’s no more Computer Simulation of Paranoia from the Real Thing? The Limitations of Turing-like
difficult in the end than getting a grasp of how a brain could be Tests as Measures of the Adequacy of Simulations,” in Journal of Psychiatric
conscious. When we develop good accounts of consciousness, it will Research 15, no. 3 (1980): 149-162. Colby discusses PARRY and its implications in
no longer seem so obvious to everyone that the idea of a self- “Modeling a Paranoid Mind,” in Behavioral and Brain Sciences 4, no. 4 (1981):
conscious computer is a contradiction in terms. At the same time, | 515-560.
doubt that there will ever be self-conscious robots, but for boring 8. Janet L_Koiodner, “Retrieval and Organization Strategies in Conceptual Memory:
reasons. There won't be any point in making them. Theoretically, A Computer Model” (Ph.D. diss.), Research Report no. 187, Dept. of Computer
Science, Yale University; Janet L Kolodner, “Maintaining Organization in a Dynamic
could we make a gall bladder out of atoms? In principle, we could. A
Long-Term Memory,” Cognitive Science 7 (1983): 243-280; Janet L Kolodner, “Recon-
gall bladder is just a collection of atoms, but manufacturing one
structive Memory: A Computer Model,” Cognitive Science 7 (1983): 281-328.
would cost the moon. It would be more expensive than every project
NASA has ever dreamed of, and there would be no scientific payoff.
We wouldn't learn anything new about how gall bladders work. For
the same reason | don’t think we're going to see really humanoid
robots, because practical, cost-effective robots don't need to be very
humanoid at all. They need to be like the robots you can already see
at General Motors, or like boxy little computers that do special-
purpose things. The theoretical issues will be studied by Al research-
ers looking at models that, to the layman, will show very little sign of
humanity at all, and it will be only by rather indirect arguments that
anyone will be able to appreciate that these models cast light on the

deep theoretical question of how the mind is organized.

61
 Ata time when computer technology is advancing at a breakneck
pace and when software developers are glibly hawking their wares
as having artificial intelligence, the inevitable question has begun to
take on a certain urgency: Can a computer think? Really think? In one
form or another this is actually a very old question, dating back to
such philosophers as Plato, Aristotle, and Descartes. And after nearly
| Mitchell Waldrop 3,000 years the most honest answer is still “Who knows?” After all,
what does it mean to think? On the other hand, that's not a very
Can Computers Think? satisfying answer. So let's try some others.

Who cares? \f a machine can do its job extremely well, what does it

Mitchell Waldrop is a senior matter if it rea/ly thinks? No one runs around asking if taxicabs really
writer for Science, the journal walk.
of the American Association for
How could you ever tell? This attitude is the basis of the famous
the Advancement of Science. A
series of articles on artificial Turing test, devised in 1950 by the British mathematician and logician
intelligence formed the basis Alan Turing: Imagine that you're sitting alone in a room with a
for his most recent book, Man-
teletype machine that is connected at the other end to either a person
Made Minds: The Promise
or a computer. If no amount of questioning or conversation allows you
of Artificial Intelligence (1987).
Waldrop holds a doctorate to tell which it is, then you have to concede that a machine can think.
in elementary particle physics
No, thinking is too complicated. Even if we someday come to
from the University
of Wisconsin at Madison. understand all the laws and principles that govern the mind, that
doesn’t mean that we can duplicate it. Does understanding
astrophysics mean that we can build a galaxy?

Yes, machines can think in principle, but not necessarily in the same
way we do. Al researcher Seymour Papert of the Massachusetts
Institute of Technology maintains that artificial intelligence is
analogous to artificial flight: “This leads us to imagine skeptics who
would say, ‘You mathematicians deal with idealized fluids—the real
atmosphere is vastly more complicated,’ or ‘You have no reason to
suppose that airplanes and birds work the same way—birds have no
propellers, airplanes have no feathers.’ But the premises of these
criticisms is true only in the most superficial sense: the same
principles (for example, Bernoulli's law) applies to real as well as
Mitchell Waldrop. (Courtesy of
Mitchell Waldrop)

ideal fluids, and they apply whether the fluid flows over a feather or classical Al we find the doctrines first set down in the 1950s by Al
an aluminum wing.” pioneers Allen Newell and Herbert Simon at Carnegie-Mellon
University: (1) thinking is information processing; (2) information
No! This is the most often heard answer, and the most heartfelt. “I am
processing is computation, which is the manipulation of symbols; and
not a machine [goes the argument]. I'm me. I'm alive. And you're never
going to make a computer that can say that. Furthermore, the essence (3) symbols, because of their relationships and linkages, mean

of humanity isn’t reason or logic or any of the other things that something about the external world. In other words, the brain per se
doesn’t matter, and Turing was right: a perfect simulation of thinking
computers can do; it's intuition, sensuality, and emotion. So how can
a computer think if it does not feel, and how can it feel if it knows is thinking.
Tufts University philosopher Daniel C. Dennett, a witty and
nothing of love, anguish, exhilaration, loneliness, and all the rest of
what it means to be a living human being?” insightful observer of Al, has dubbed this position High Church
Computationalism. Its prelates include such establishment figures as
“Sometimes when my children were still little,” writes
Simon and MIT's Marvin Minsky; its Vatican City is MIT, “the East
former Al researcher Joseph Weizenbaum of MIT, “my wife and I
Pole.”
would stand over them as they lay sleeping in their beds. We spoke to
Then from out of the West comes heresy—a creed that is
each other only in silence, rehearsing a scene as old as mankind
itself. It is as lonesco told his journal: ‘Not everything is unsayable in not an alternative so much as a denial. As Dennett describes it, the

words, only the living truth.’ ” assertion is that “thinking is something going on in the brain all right,
but it is not computation at all; thinking is something holistic and

Can a Machine Be Aware? emergent—and organic and fuzzy and warm and cuddly and
sce = mysterious.”

As this last answer suggests, the case against machine intelligence Dennett calls this creed Zen holism. And for some reason

always comes down to the ultimate mystery, which goes by many its proponents do seem to cluster in the San Francisco Bay area.
Among them are the gurus of the movement: Berkeley philosophers
names: consciousness, awareness, spirit, soul. We don’t even
understand what it is in humans. Many people would say that it is John Searle and Hubert Dreyfus.

beyond our understanding entirely, that it is a subject best left to God The computationalists and the holists have been going at

alone. Other people simply wonder if a brain can ever understand it for years, ever since Dreyfus first denounced Al in the mid 1960s

itself, even in principle. But either way, how can we ever hope to with his caustic book What Computers Can't Do. But their definitive

reproduce it, whatever it is, with a pile of silicon and software? battle came in 1980, in the pages of the journal Behavioral and Brain

That question has been the source of endless debate Sciences. This journal is unique among scientific journals in that it

since the rise of Al, a debate made all the hotter by the fact that doesn't just publish an article; first it solicits commentary from the

people aren't arguing science. They're arguing philosophical author's peers and gives the author a chance to write a rebuttal. Then

ideology—their personal beliefs about what the true theory of the it publishes the whole thing as a package—a kind of formal debate in

mind will be like when we find it. print. In this case the centerpiece was Searle's article “Minds,

Not surprisingly, the philosophical landscape is rugged Brains, and Programs,” a stinging attack on the idea that a machine

and diverse. But it's possible to get some feel for the overall
topography by looking at two extremes. At one extreme at the heart of

63
 could think. Following it were 27 responses, most of which were And yet, said Searle, while he is locked in that imaginary
stinging attacks on Searle. The whole thing is worth reading for its room he is doing exactly what the computer does. He uses formal
entertainment value alone. But it also highlights the fundamental rules to manipulate abstract symbols. He takes in stories and gives
issues with a clarity that has never been surpassed. out answers exactly as a native Chinese would. But he still doesn't
understand a word of Chinese. So how is it possible to say that the
| The Chinese Room computer understands? In fact, said Searle, it doesn’t. For compari-
needed
erie
son, imagine that the questions and the answers now switch to
Essentially, Searle’s point was that simulation is not duplication. A English. So far as the people outside the room are concerned, the
program that uses formal rules to manipulate abstract symbols can system is just as fluent as before. And yet there's all the difference in
never think or be aware, because those symbols don’t mean anything the world, because now he isn’t just manipulating formal symbols
to the computer. anymore. He understands what's being said. The words have meaning
To illustrate, he proposed the following thought experi- for him—or, in the technical jargon of philosophy, he has intentional-
ment as a parody of the typical Al language-understanding program of ity. Why? “Because | am a certain sort of organism with a certain
his day: “Suppose that I'm locked in a room and given a large batch of biological (i.e., chemical and physical) structure,” he said, “and this
Chinese writing,” he sai: “Suppose furthermore (as is indeed the structure, under certain conditions, is causally capable of producing
case) that | know no Chinese. .. . To me, the Chinese writing is just so perception, action, understanding, learning, and other intentional
many meaningless squiggles.” Next, said Searle, he is given a second phenomena.” In other words, Searle concluded that it is certainly
batch of Chinese writing (a “story”), together with some rules in possible for a machine to think—"in an important sense our bodies
English that explain how to correlate the first batch with the second with our brains are precisely such machines” —but only if the
(a “program”). Then after this is all done, he is given yet a third set of machine is as complex and as powerful as the brain. A purely formal
Chinese symbols (“questions”), together with yet more English rules computer program cannot do it.
that tell him how to manipulate the slips of paper until all three
batches are correlated, and how to produce a new set of Chinese Counterarguments
SSS
characters (“answers”), which he then passes back out of the room.
Finally, said Searle, “after awhile | get so good at manipulating the Searle's Chinese room clearly struck a sensitive nerve, as evidenced
instructions for the Chinese symbols and the programmers get so by the number and spirit of the denunciations that followed. It was
good at writing the programs that from the external point of view . . . clear to everyone that when Searle used the word “intentionality,” he
my answers to the questions are absolutely indistinguishable from wasn't just talking about an obscure technical matter. In this context
those of native Chinese speakers.” In other words, Searle learns to intentionality is virtually synonymous with mind, soul, spirit, or
pass the Turing test in Chinese. awareness. Here is a sampler of some of the main objections:
Now, according to the zealots of strong Al, said Searle, a The comparison is unfair. The programs that Searle ridiculed
computer that can answer questions in this way isn’t just simulating demonstrated a very crude kind of understanding at best, and no one
human language abilities. It is literally understanding the story. in Al seriously claims anything more for them. Even if they were
Moreover, the operation of the program is in fact an explanation of correct in principle, said the defenders, genuine humanlike
human understanding.

64
understanding would require much more powerful machines and itself control the temperature? No. Does the system as a whole
much more sophisticated programs. control the temperature? Yes. Connections and the organization make
Searle quite correctly pointed out, however, that this the whole into more than the sum of its parts.
argument is irrelevant: of course computers are getting more
Searle never makes clear what intentionality is, or why a machine
powerful; what he objected to was the principle. can't have it. As Dennett pointed out, “For Searle, intentionality is
The Chinese room story is entertaining and seductive, but it's a fraud. rather like a wonderful substance secreted by the brain the way the
Douglas R. Hofstadter of Indiana University, author of the best-selling pancreas secretes insulin.” And make no mistake: Searle's concept
Godel, Escher, Bach, pointed out that the jump from the Al program to of intentionality does require a biological brain. He explicitly denied
the Turing test is not the trivial step that Searle makes it out to be. It's that a robot could have intentionality, even if it were equipped with
an enormous leap. The poor devil in the Chinese room would have to eyes, ears, arms, legs, and all the other accoutrements it needed to
shuffle not just a few slips of paper but millions or billions of slips of move around and perceive the world like a human being. Inside, he
paper. It would take him years to answer a question, if he could do it said, the robot would still just be manipulating formal symbols.
at all. in effect, said Hofstadter, Searle is postulating mental That assertion led psychologist Zenon Pylyshyn of the
processes slowed down by a factor of millions, so no wonder it looks University of Western Ontario to propose his own ironic thought
different. experiment: “Thus, if more and more of the cells in your brain were to
Searle's reply—that he could memorize the slips of paper be replaced by integrated circuit chips, programmed in such a way as
and shuffle them in his head—sounds plausible enough. But as to keep the input-output function of each unit identical to the unit
several respondents have pointed out, it dangerously undermines his being replaced, you would in all likelihood just keep right on
whole argument: once he memorizes everything, doesn't he now speaking exactly as you are doing now except that you would
understand Chinese in the same way he understands English? eventually stop meaning anything by it. What we outside observers

The entire system does understand Chinese. True, the man in the might take to be words would become for you just certain noises that

room doesn't understand Chinese himself. But he is just part of a circuits caused you to make.” In short, you would become a zombie.

larger system that also includes the slips of paper, the rules, and the Dennett took up the same theme in his own article. So far

message-passing mechanism. Taken as a whole, this larger system as natural selection is concerned, he pointed out, Pylyshyn’s zombie

does understand Chinese. This “systems” reply was advanced by a or Searle's robot is just as fit for survival as those of us with Searle-

number of the respondents. Searle was incredulous—"It is not easy style intentional brains. Evolution would make no distinction. Indeed,

for me to imagine how someone who was not in the grip of an from a biological point of view, intentionality is irrelevant, as useless

ideology could find the idea at all plausible”—yet the concept is as the appendix. So how did it ever arise? And having arisen, how did

subtler than it seems. Consider a thermostat: a bimetallic strip bends it survive and prosper when it offered no natural-selection value?

and unbends as the temperature changes. When the room becomes Aren't we lucky that some chance mutation didn't rob our ancestors

too cold, the strip closes an electrical connection, and the furnace of intentionality? Dennett asked. If it had, he said, “we'd behave just

kicks on. When the room warms back up again, the connection as we do now, but of course we wouldn't mean it!” Needless to say,

reopens, and the furnace shuts off. Now, does the bimetallic strip by both Pylyshyn and Dennett found this absurd.

itself control the temperature of the room? No. Does the furnace by In retrospect, the great debate has to be rated a standoff.
Searle, not surprisingly, was unconvinced by any of his opponents’
arguments; to this day he and his fellow Zen holists have refused to

65
yield an inch. Yet they have never given a truly compelling explana- after Copernicus the earth and man were reduced to being wander-
tion of why a brain and only a brain can secrete intentionality. The ers in an infinite universe. For many, the sense of loss and confusion
computationalists, meanwhile, remain convinced that they are were palpable.
succeeding where philosophers have failed for 3,000 years—that they In 1859 with the publication of The Origin of Species
are producing a real scientific theory of intelligence and conscious- Charles Darwin described how one group of living things arises from
ness. But they can't prove it. Not yet, anyway. another through natural selection and thereby changed our percep-
And in all fairness, the burden of proof is on Al. The tion of who we are. Once man had been the special creation of God,
symbol-processing paradigm is an intriguing approach. If nothing the favored of all his children. Now man was just another animal, the
else, it's an approach worth exploring to see how far it can go. But descendent of monkeys.
still, what is consciousness? In the latter part of the nineteenth century and the early
decades of the twentieth with the publication of such works as The
Science as a Message of Despair Interpretation of Dreams (1901), Sigmund Freud illuminated the inner
AS
ES
workings of the mind and again changed our perception of who we
One way to answer that last question is with another question: Do we are. Once we had been only a little lower than the angels, masters of
really want to know? Many people instinctively side with Searle, our own souls. Now we were at the mercy of demons like rage,
horrified at what the computationalist position implies: If thought, terror, and lust, made all the more hideous by the fact that they lived
feeling, intuition, and all the other workings of the mind can be unseen in our own unconscious minds.
understood even in principle, if we are machines, then God is not So the message of science can be bleak indeed. It can be
speaking to our hearts. And for that matter, neither is Mozart. The soul seen as a proclamation that human beings are nothing more than
is nothing more than the activations of neuronal symbols. Spirit is masses of particles collected by blind chance and governed by
nothing more than a surge of hormones and neurotransmitters. immutable physical law, that we have no meaning, that there is no
Meaning and purpose are illusions. And besides, when machines purpose to existence, and that the universe just doesn't care. |
grow old and break down, they are discarded without a thought. Thus, ‘suspect that this is the real reason for the creationists’ desperate
for many people, Al is a message of despair. Of course, this is hardly a rejection of Darwin. It has nothing to do with Genesis; it has
new concern. For those who choose to see it that way, science itself everything to do with being special in the eyes of a caring God. The
is a message of despair. fact that their creed is based on ignorance and a willful distortion of
In 1543 with the publication of De Revolutionibus the the evidence makes them both sad and dangerous. But their longing
Polish astronomer Nicholas Copernicus moved the earth from the for order and purpose in the world is understandable and even noble.
center of the universe and made it one planet among many and | also suspect that this perceived spiritual vacuum in science lies
thereby changed humankind's relationship with God. In the earth- behind the fascination so many people feel for such pseudosciences
centered universe of Thomas Aquinas and other medieval theologi- as astrology. After all, if the stars and the planets guide my fate, then
ans, man had been poised halfway between a heaven that lay just somehow | matter. The universe cares. Astrology makes no scientific
beyond the sphere of the stars and a hell that burned beneath his feet. sense whatsoever. But for those who need such reassurance, what
He had dwelt always under the watchful eye of God, and his spiritual can science offer to replace it?
status had been reflected in the very structure of the cosmos. But

66
Science as a Message of Hope There is no way to point to any one of them and say, “This is alive.”
coe ee
And yet when all those molecules are brought together in an
And yet the message doesn’t have to be bleak. Science has given us exquisitely ordered pattern, they are life. In the same way our minds
a universe of enormous extent filled with marvels far beyond are perhaps nothing more than machines. Does that mean there is no
anything Aquinas ever knew. Does it diminish the night sky to know such thing as spirit? Perhaps we are just processors of neuronal
that the planets are other worlds and that the stars are other suns? In symbols. Perhaps a snowflake is only a collection of water mole-
the same way, a scientific theory of intelligence and awareness cules. Perhaps The Magic Flute is only a sequence of sound waves.
might very well provide us with an understanding of other possible And perhaps, in illuminating the nature of mind and intelligence, Al is
minds. Perhaps it will show us more clearly how our Western ways only reaffirming how unique and precious the mind really is.
of perceiving the world relate to the perceptions of other cultures.
Perhaps it will tell us how human intelligence fits in with the range
of other possible intelligences that might exist in the universe.
Perhaps it will give us a new insight into who we are and what our
place is in creation.
Indeed, far from being threatening, the prospect is oddly
comforting. Consider a computer program. It is undeniably a natural
phenomenon, the product of physical forces pushing electrons here
and there through a web of silicon and metal. And yet a computer
program is more than justa surge of electrons. Take the program and
tun it on another kind of computer. Now the structure of silicon and
metal is completely different. The way the electrons move is
completely different. But the program itself is the same, because it
still does the same thing. It is part of the computer. It needs the
computer to exist. And yet it transcends the computer. In effect, the
program occupies a different level of reality from the computer.
Hence the power of the symbol-processing model: By describing the
mind as a program running on a flesh-and-blood computer, it shows
us how feeling, purpose, thought, and awareness can be part of the

physical brain and yet transcend the brain. It shows us how the mind
can be composed of simple, comprehensible processes and still be
something more.
Consider a living cell. The individual enzymes, lipids, and
DNA molecules that go to make up a cell are comparatively simple
things. They obey well-understood laws of physics and chemistry.

67
 The cultural fascination with computation and artificial intelligence
has two faces. There is excitement about the artifacts themselves:
their power, their ability.to act as extensions of our minds much as
the machines of earlier generations acted as extensions of our
bodies. But also and equally important, there is an involvement with
computers as mirrors that confront us with ourselves. Here the
question is not whether we will ever build machines that will think
like people but whether people have always thought like machines.
And if thi: is the case, if we are in some important sense kin to the
computer, is this the most important thing about us? Is this what is
most essential about being human?
Such questions have long been the province of philoso- |
phers. But in recent years something has changed. Intelligent
machines have entered the public consciousness not just as actors in
science-fiction scenarios but as real objects, objects you can own as
well as read about. This has put the philosophical debate in a new

| place. Artificial Intelligence has moved out from the world of the
professionals into the life of the larger culture. But unlike the way in
which a theory like psychoanalysis was able to move out into the
Growing Up in the Age cf Intelligent
wider culture, the new popular considerations of mind and machine
Machines: Reconstructions of the
arise from people's actual relationships with an object they can touch
Psychological and Reconsiderations
and use. New ideas are carried by relationships with computers as
of the Human
objects to think with.
Asimple example makes the point. Twenty years ago the
Sherry Turkle is Associate
question of how well computers could play chess was a subject of
Professor in the Program in
controversy for Al researchers and their philosophical interlocutors.
Science, Technology, and
Society at the Massachusetts Some writers felt that there was no limit to what computers could
Institute of Technology. She achieve in this domain. Others responded that there would be
has written numerous articles
absolute limits to the powers of machines. “Real” chess, they argued,
on psychoanalysis and culture
was the kind of chess only humans could play, since it called upon
and on the subjective side of
people’s relationships with powers of synthesis and intuition that were uniquely rooted in human
technology. She is the author of capacities.
The Second Self: Computers
That dialogue with the strengths and limitations of the
and the Human Spirit (1984),
machine continues within philosophy, but it has been joined by other,
which looks at the relation-
ships that people form with more informal conversations: those of master players who sit across
computers and the ways in from computers in tournament play, those of recreational chess
which these relationships
players who compete with chess computers at home. The chess f
affect values, ways of thinking
computers have gotten very good; chess players respond by trying to {
about the world, and ways of
seeing oneself and other determine what is special about their play, even if they cannot
people. always exploit this specialness to actually beat the machine. And of {
course, the players include the first generation of children who have
grown up playing chess with computers. A thirteen-year-old, Alex,
plays daily with a chess computer named Boris. He comments that
although he always loses if he “puts the setting high enough” (that is,
if he asks the computer to play the best game it can), “it doesn’t really |
feel like I'm losing.” Why? Because “chess with Boris is like chess

6s
 with somebody who is cheating. He can have all the most famous, all
the best chess games right there to look at—I mean, they are inside of
him. | can read about them, but I can't remember them all, not every
move. | don't know if this is how he works, but it's like in between
every move he could read all the chess books in the world.” Here,
human uniqueness is defined in terms not of strengths but a certain
frailty. “Real” chess for this child is human chess, the kind of chess
that works within the boundaries of human limitations.
Thus, the presence of intelligent machines in the culture
provokes a new philosophy in everyday life. Its questions are not so
different than the ones posed by professionals: If the mind is (at least
in some ways) a machine, who is the actor? Where is intention when
there is program? Where is responsibility, spirit, soul? In my research
on popular attitudes toward artificial intelligence | have found that
the answers being proposed are not very different either. Faced with
smart objects, both professional and lay philosophers are moved to
catalog principles of human uniqueness. The professionals find it in
human intentionality, embodiment, emotion, and biology. They find it
in the fact that the human life cycle confronts each of us with the
certainty of death. There are clear echoes of these responses within
the larger culture. As children tell it, we are distinguished from
machine intelligence by love and affection, by spiritual urges and
sensual ones, and by the warmth and familiarity of domesticity. In the
words of twelve-year-old David, “When there are computers who are
just as smart as people, the computers will do a lot of the jobs, but
there will still be things for the people to do. They will run the
restaurants, taste the food, and they will be the ones who will love
each other—have families and love each other. | guess they'll still be
the ones who go to church.”
One popular response to the presence of computers is to
define what is most human as what computers can't do. But this is a
fragile principle when it stands alone, because it leaves one trying to
run ahead of what clever engineers will come up with next. An
increasingly widespread attitude, at least among people who have
sustained contact with computers, is to admit that human minds are
some kind of computer, and then to find ways to think of themselves
as something more as well. When they do so, people’s thoughts
Sherry Turkle. (Photo by Douglas
usually turn to their feelings. Some find it sufficient to say, as did
Hopkins)
David, that machines are reason, and people are sensuality and
emotion. But others split the human capacity for reason. They speak
of those parts of our reason that can be simulated and those parts that
are not subject to simulation.
In all of this, the computer plays the role of an evocative
object, an object that disturbs equanimity and provokes self-
reflection. That it should do so for adults is not surprising: after all,
intelligent machines strike many of us as childhood science fiction

69
that has become real. But | have already suggested that computers movement that an object can generate by itself and movement
also play this role for children. You may give a computer to a child imposed by an outside agent. This allows “alive” to be restricted to
hoping that it will teach mathematics or programming skills or French things that seem to move of their own accord: a dog, of course, but
verbs. But independent of what the computer teaches the child, it also perhaps a cloud. An object drops out of the “alive” category
does something else as well. For the child, as for the adult, the when the child discovers an outside force that accounts for its
machine is evocative. It creates new occasions for thinking through motion. So at eight the river may still be alive, because the child
the philosophical questions to which childhood must give a response, cannot yet understand its motion as coming from outside of itself, but
among them the question of what is alive. the stone and the bicycle are not alive, because the child can. Finally,
The Swiss psychologist Jean Piaget first systematized our the idea of motion from within is refined to a mature concept of life
understanding of the “child as philosopher.” Beginning in the 1920s, activity: growth, metabolism, breathing.
Piaget studied children’s emerging way of coming to terms with such There are two key elements in the story as Piaget told it.
aspects of the world as causality, life, and consciousness. He First, children build their theories of what is alive and what is not
discovered that children begin by understanding the world in terms of alive as they build all other theories. They use the things around them:
what they know best: themselves. Why does the ball roll down the toys, people, technology, the natural environment. Second, in Piaget's
slope? “To get to the bottom,” says the young child, as though the world, children’s sorting out the concept of life presented a window
ball, like the child, had its own desires. But childhood animism, this onto the child's “construction of the physical.” Thinking about the
attribution of the properties of life to inanimate objects, is gradually idea of life required the child to develop distinctions about the world
displaced by new ways of understanding the physical world in terms of physics. The motion theory for distinguishing the living from the
of physical processes. In time the child learns that the stone falls nonliving corresponds to the world of “traditional” objects that have
because of gravity; intentions have nothing to do with it. And so a always surrounded children: animate objects—people and animals
dichotomy is constructed: physical and psychological properties who act and interact on their own—and all the other objects, pretty
stand opposed to one another in two great systems. The physical well inert until given a push from the outside. In recent years there
properties are used to understand things; the psychological to has been an important change. The new class of computational
understand people and animals. But the computer is a new kind of objects in children’s worlds has provoked a new language for theory
object, a psychological object. It is a thing (“just a machine”), yet it building about the concept of life.
ij
has something of a mind. The computer is betwixt and between, an Today children are confronted with highly interactive 1
object with no clear place in the sharply dichotomized system, and as objects that talk, teach, play, and win. Their computers and computer
such it provokes new reflection on matter, life, and mind. toys do not move but are relentlessly active in their “mental lives.”
Piaget argued that children develop the concept of life by Children are not always sure about whether these objects are alive, |
making finer and finer distinctions about the kinds of activities that but it is clear to even the youngest children that thinking about motion {
are evidence of life. In particular, Piaget described how the notion of won't take one very far toward settling the question. Children |
life is built on progressive refinements of children’s concept of perceive the relevant criteria not as physical or mechanical but as {
physical motion. At age six a child might see a rolling stone, a river, psychological: Are the computers smart? Can they talk? Are they j
and a bicycle as alive for the same reason: “They go.” By age eight aware? Are they conscious? Do they have feelings? Even, do they
the same child might have learned to make a distinction between cheat? The important question here is not whether children see f
intelligent machines as alive. Some do, some do not, and, of course, |

70
in the end all children learn the “right answer.” What is important is fundamental difference from people. But today’s children take a
the kind of reasoning the child uses to sort out the question. The child different view. The idea of an artificial consciousness does not upset

knows that the computer is “just a machine,” but it presents itself them. They find it a very natural thing. They may be the first genera-
with lifelike, psychological properties. Computers force the child to tion to grow up with such a radical split between the concepts of
think about how machine minds and human minds are different. In consciousness and life, the first generation to grow up believing that

this way, the new world of computational objects becomes a support human beings are not necessarily alone as aware intelligences. The
for what Piaget might have called the child's construction not of the child's splitting of consciousness and life is a clear case of where it
physical but of the psychological. does not make sense to think of adult ideas filtering down to children.
In the adult world, experts argue about whether or not Rather, we should think of children’s resolutions as prefiguring new
computers will ever become true artificial intelligences, themselves positions for the next generation of adults whose psychological
capable of autonomous, humanlike thought. But irrespective of future culture will be marked by the computer culture.
progress in machine intelligence, computational objects are even Second, children are led to make increasingly nuanced
now affecting how today’s children think. The language of physics distinctions about the psychological. Younger children from around
gives way to the language of psychology when children think about six to eight sometimes say that computers are like people in how they
computers and the question of what is alive. think but unlike people in their origins. (“The machine got smart
This change in discourse, this new use of machines in the because people taught it.”) But this is not a stable position. It is
construction of the psychological, is important for many reasons. unsatisfying because it leaves the essential difference between
Here, I mention three that are particularly striking. First, children are computers and people tied to something that happened in the past,
led to a new way of talking about the relationship between life and almost as though the computers’ minds and the children’s minds are
consciousness. In Piaget's studies, the idea of consciousness evolved alike; they differ only in their parents. Older children reach for a more
side by side with the idea of life. Generally, when children ascribed satisfying way to describe what people share and do not share with
life to inanimate objects, they ascribed consciousness too; when life the psychology of the machine. When younger children talk about the
became identified with the biological, consciousness became a computer's psychology, they throw together such undifferentiated
property unique to animals. But when today’s children reflect on observations as that a computer toy is happy, is smart, cheats, and
computational objects, the pattern is very different. Many children gets angry. Older children make finer distinctions within the domain
allow intelligent machines to be conscious long after they emphati- of the psychological. In particular, they divide it in two. They
cally deny them life. When one child remarks that the computer is not comfortably manipulate such ideas as “It thinks, but it doesn’t feel.”
alive, but it cheats “without knowing it is cheating,” he is corrected They comfortably talk about the line between the affective and the
by another who insists that “knowing is part of cheating.” Children cognitive.
talk about the nonliving computer as aware, particularly when it With the splitting of the psychological, it is no longer the
presents itself as “smarter” than they are, for example, in math or issue of whether something has a psychology or does not. By
spelling or French. They talk about the nonliving computer as having developing a distinct idea of the cognitive, children find a way to
malicious intent when it consistently beats them at games. Adults grant to the computer that aspect of psychology which they feel
hold onto the fact that computers are not aware as a sign of their compelled to acknowledge by virtue of what the machines do, while
they reserve other aspects of the psychological for human beings.
This response to machine intelligence of splitting psychology is

71
particularly marked in children who use computers a great deal at rationality. People are still defined in contrast to their nearest
school or at home. Katy, eleven, after a year of experience with neighbors. But now people are special because they feel. The notion
computer programming said, “People can make computers intelligent: of a rational animal gives way to the paradoxical construction of
you just have to find about how people think and put it in the people as emotional machines.
machine,” but emotions are a different matter. For Katy, the kinds of This last point brings me full circle to where | began: with
thinking the computer can do are the kinds that “all people do the the image of computers as evocative objects in the lives of adults.
same. So you can't give computers feelings, because everybody has My studies show that many adults follow essentially the same path
different feelings.” as do children when they talk about human beings in relation to the
The distinction between thought and feeling is not the new intelligent machines. The child's version is to be human is to be
only line that children draw across mental life in the course of emotional. The adult's version is to be human is to be unprogram-
coming to terms with the computer's nature in the light of human mable. People who say that they are perfectly comfortable with the
nature. Discussions about whether computers cheat can lead to con- idea of mind as machine assent to the idea that simulated thinking is
versations about intentions and awareness. Discussions about the thinking but often cannot bring themselves to propose further that
computer's limitations can lead to distinctions between free will and simulated feeling is feeling.
autonomy as opposed to programming and predetermination. This There is a sense in which both of these reactions, the
often brings children to another distinction, the one between rote child's and the adult's, contrast with a prevalent fear that involve-
thinking, which computers can do, and originality, which is a human ment with machine intelligence leads to a mechanical view of
prerogative. Finally, discussion about how computers think at all can people. Instead, what I find is something of a romantic reaction.
lead to the distinction between brain and mind. All of these are There is a tendency to cede to the computer the power of reason but
elements of how computers evoke an increasingly nuanced construc- at the same time, in defense, to construct a sense of identity that is
tion of the psychological. increasingly focused on the soul and the spirit in the human machine.
Finally, children’s new psychologized appreciation of Ona first hearing, many people find these observations
machines influences how they articulate what is most special, most reassuring. But it is important to underscore that there is a disturbing
important about being a person. While younger children may say that note in this technology-provoked reconsideration of human specific-
the machine is alive “because it has feelings,” older children tend to ity. Thought and feeling are inseparable. When they are torn from
grant that the machine has intelligence and is thus “sort of alive” but their complex relationship with each other and improperly defined as
then distinguish it from people because it lacks feelings. The mutually exclusive, the cognitive can become mere logical process,
Aristotelian definition of man as a rational animal (a powerful and the affective is reduced to the visceral. What was most powerful
definition even for children when it defined people in contrast to their about Freud's psychological vision was its aspiration to look at
nearest neighbors, the animals) gives way to a different distinction. thought and feeling as always existing together in interaction with
Today's children come to understand computers through a process of each other. In psychoanalytic thinking there is an effort to explore ‘
identification with them as psychological entities. And they come to the passion in the mathematician’s proof as well as an effort to use 4
see them as our new nearest neighbors. From the point of view of the reason in understanding the most primitive fantasy. The unconscious
child, this is a neighbor that seems to share, or even excel in, our has its own highly structured language, which can be deciphered j
and analyzed. Logic has an affective side, and affect has a logic. f
Computational models of mind may in time deepen our appreciation

72
of these complexities. But for the moment, the popular impact of
intelligent machines on our psychological culture goes in the other
direction. The too easy acceptance of the idea that computers
closely resemble people in their thinking and differ only in their lack
of feelings supports a dichotomized and oversimplified view of
human psychology. The effort to think against this trend will be one
of our greatest challenges in the age of intelligent machines.

73
 There are few questions more mysterious and thought provoking than
whether a nonhuman machine could ever be considered truly human
in any important sense of the word. Let us jump ahead a few decades
and imagine, for a moment, that all the problems of creating a truly
intelligent machine have been solved. How would two “people,” a
philosopher and a computer, handle some of the physical, emotional,
and moral issues of such a creation?
The year is 2020. A philosopher sits in his office consider-
ing how many of life’s great mysteries have yet to be solved. All of a
sudden he notices a figure outside his window.

Philosopher [opening his window}: What in blazes are

you doing on that ledge?
Computer: I'm going to jump. Don't try to stop me!
Philosopher (getting out on the ledge): Don't jump! What
could be so bad that you would want to kill yourself? How could you

I oe possibly even think of wasting what it took nature millions of years to

produce?
Computer: Oh no... oooh... [He breaks down crying.)
A Conversation between a Human Computer
Philosopher: What did | say?
and a Materialist Philosopher
Computer (sobbing): That's just the point. Human life
might be precious, but I'm not human. Oh God, help me. | can’t even
Blaine Mathieu is the founder
say that! Aaahwooo...
and President of Turning Point
Software in Canada, a Philosopher: (Certain that he has a lunatic on his hands,
computer firm interested in he decides to reason with him before calling the police. You \ook,
many aspects of the small- act, and talk like a human. Why do you think that you aren't human?
computer industry.
Computer: | got a phone call this morning, you see. It was
the scientists from the Government Biophysics and Computer Science
Laboratory, and they wanted me to come in to see them. I had no idea
why they wanted to see me, but I jumped at the chance of seeing this

Blaine Mathieu. (Courtesy of Blaine

Mathieu)
 not saying that knees have everything that brains do, like conscious-
ness; I'm only saying that neither knees nor brains came about as a
result of any unexplainable magic. Every day more and more research
points to the fact that the brain operates in simple accordance with
the physical laws of nature.
Computer: But I'm not sure that the brain is totally
explainable without magic. What about the insensitivity of thought
mechanisms to brain damage? I've heard of cases where relatively
large parts of the brain have been removed without any noticeable or
reported effects on the person involved. If the brain really is a
machine responsible for thought, wouldn't removing parts of the
top-secret institution. Imagine my surprise when they told me that I'm
machine seriously hamper the machine's function?
a computer! I'm an electronic machine. All along I've been living a
Philosopher: That depends on how the machine works.
dream!
Compare your brain to the computer systems that run a rocket ship.
Philosopher: You're not trying to tell me that you're a
Just because one computer goes down, that doesn’t mean that the
robot, are you? You look far too human for that!
mission is over. There is redundancy. When one computer quits
Computer: | guess they took a newly born human and
working, another takes its place. Also, when one computer goes
placed me, or rather, a computer, in place of the baby’s brain. At least
down, it does not drag the other ones down with it. This is called
that's what they told me. | have memories from my childhood, and I've
diffuseness. This redundancy and diffuseness seem to be present in
always loved my parents. | had no idea that a bunch of neurophysi-
the neuronal systems of the brain.
cists and computer scientists were my real parents.
Computer: Okay, so maybe the brain isn't so magical after
Philosopher: [Totally disbelieves the computer, but
all. But all the talk so far has been about the human brain. What about
decides to humor him anyway.] Even if all this is true, why would you
my electronic brain?
possibly want to kill yourself?
Philosopher: | fail to see what the problem is. You talk
Computer: Because I'm a computer! You could never get
and act just as | do and just as any other human would. What
me to say that a computer thinks or is conscious or has beliefs or
difference does it make what your brain is made of?
feelings. How am | anything more than just a complicated version of
Computer: That seems like a very behavioristic point of
what| have sitting on my desk at home?
view. Just because I look and act like you, that doesn’t mean that | am
Philosopher: You probably are simply a more complicated
like you. Even when | was a young child | could engage in fairly
version of what you have at home.
involved interactions with my personal computer. | certainly know
Computer: See! [Gets ready to jump.)
that it wasn’t a conscious being that had any understanding of what |
Philosopher: Don't jump! [He steadies himself. My brain
was talking about. How could my electronic brain have and be any of
is nothing more than a machine too! What difference does it make if
the things that real brains have and are?
your brain is made out of silicon and gallium arsenide and mine is
Philosopher: The answer to that question revolves around
made of carbon?
the concept of “functional isomorphism.”
Computer: But there is a difference. My brain is man-
Computer: Oh, you mean the idea of a correspondence
made, and yours isn't.
between the states of two systems that preserves functional
Philosopher: Millions of years of evolution designed my
relations?
brain. Why is that any better than being designed in a few years by a
Philosopher: Uh... yeah! That's it. What it means is that
team of scientists? The joints in your knees took a long time to evolve
the differences between two systems that perform the same function
to their present state, but | don’t think you would say that an artificial
knee is of any less value just because it was designed in a few years.
Computer: Are you comparing knees to brains? That's
ridiculous!
Philosopher: Admittedly, knees are far simpler and less
versatile than brains, but they are still both machines. Anyway, I’m

75
may not be important or significant. Even though these two systems Philosopher: This implies that maybe our emotions and
may interact with things outside of them in identical ways, the feelings are not as ethereal in nature as we might believe. Assume
mechanism of, or reason for, this interaction can be different. Let me that an electrical current in certain aggregations of neurons “means”
give you an example. Suppose that you are in a very dark room a feeling of pleasure. Then it is easier to fathom how a machine other
watching a movie screen and on this screen is flashed a number of than the brain could have these feelings or characteristics, because a
slides, one every ten seconds. On the wall behind you is a hole from connection has been shown between the physical world and the
which the light of the projector is emanating. Now, for all you know, a “mental” world.
person might be counting, and every time he reaches ten, he presses Computer: But it is so hard to imagine how a computer
a button, and you see a slide. Or maybe a clock has a number of little could have emotions!
electrical contact points set up so that each time ten seconds ticks Philosopher: | fail to understand what's so amazing.
by, the circuit is completed and a new slide is shown. Or maybe a Emotions are just one more characteristic of a brain that is nothing
monkey has been trained to press the button every time he is shown a more than the complex combination of physical processes. If a
picture of a banana, and this occurs every ten seconds. In any case, | computer could undergo similar processes and so be functionally
think you can see that although the physical realizations of these isomorphic, then it too could experience emotions. Anyway, let me
systems are very different, they still have the same function. Ten make another comparison between brains and computers. Computers
seconds pass (which is the input or stimulus), and a new slide is get their power by performing a large number of very simple
presented (which is the output or response). The same input or processing steps. Research has shown that brains may work in much
stimulus will always produce the same output or response. So the same way.
although there are differences in the systems, there are no important Computer: Could you give an example?
differences. Philosopher: Certainly. When we look at something like a
Computer: | see. So the idea is to somehow show that an picture of a box, for example, we don't immediately perceive this box
electronic brain could be functionally isomorphic with a human as a whole object. First, the sensory information is passed from the
brain, because then a computer could have and be all of the important retina, through the lateral geniculate nucleus, to the occipital cortex.
things that a human brain has and is. | suppose that even two human There we find brain cells that respond only to certain orientations of
brains aren't exactly functionally isomorphic, because then two lines of light and darkness. As we go further along in the cortex, we
people would respond in exactly the same way to the same situation. find cortical cells that have more complex receptive fields. Some
This never happens, of course. But how could a computer ever be cells might respond only to a horizontal line of light, while others
functionally isomorphic with the human brain? It sounds impossible might respond to corners, two lines at right angles to each other. q
to me. Some cells even respond better to moving stimuli than to stationary
Philosopher: Well let's take a look at what we know stimuli. q
about computers and what we know about brains. First of all, we Computer: So what you're saying is that for the brain to f
know that electrical stimulation in certain areas of the brain is actually “see” something, it first chops up the image into smallerand =
sufficient to evoke a sense of well-being, feelings of hunger, sexual simpler parts that orientation and movement sensitive neurons can f
gratification, rage, terror, or pain. handle. In this way the power of seeing a complex image can actually
Computer: So what? be handled by things as simple as neurons. No magic necessary. But

76
wouldn't that require a staggering number of neurons to see anything Philosopher: Well, to tell you the truth, with current
of any real complexity? morality very much against human slavery, I'd say...
Philosopher: There are many billions of neurons in the Computer: You mean to tell me that I'm the prototype for a
human body, so we have a lot to work with—or rather, | do. Of course, bunch of intelligent, government-made slaves?
the operation | just outlined is a gross oversimplification but it does Philosopher: Well .. . Hey! Stop!
get the idea across. The computer jumps off the ledge. The philosopher rushes
Computer: Okay, I'll admit that both computers and brains downstairs to see if the computer is still alive. When he arrives, he is
get their power by performing a large number of very simple greeted with a grisly sight. Among the broken bones and blood he
processing steps, but computers only deal with ones and zeros, sees the glimmer of metal and soon realizes that the computer was
simple on/off switches, whereas neurons are much more complex. telling him the truth all along. After a moment of consideration he
Philosopher: You're right, but |still don’t think that's a rushes back up to his office to phone the police. Just as he walks
real problem. We could simply use these on/off switches, or states, to through the door, the videophone rings.
emulate the workings of a neuron. In essence, every electronic Philosopher: Hello? Yes?
neuron in your brain could have a little neuron-emulation program Man on phone: Hello. Is this Dr. Jacknov? Dr. Brian
running in it. This would make it functionally isomorphic with the Jacknov?
human brain. In fact, simple systems based on this principle have Philosopher: Yes. What do you want? Please hurry!
been running for more than 30 years. Man on phone: \'m calling from the Government Biophys-
Computer: | think | finally understand. Maybe a computer ics and Computer Science Laboratory. We were wondering if you
could be functionally isomorphic with a human brain! My computer could come and visit us tomorrow. We've got something important to
brain must be doing all of the same things as your brain, it is just tell you....Hello?... Dr. Jacknov?... Hello!
doing them in a slightly different way. But if the brain is really just a
machine, it seems amazing that it is so powerful.
Philosopher: \t is amazing. [The philosopher still
disbelieves that he is speaking to a computer, but he decides to
humor him. Your computer brain must be a beautiful piece of
engineering.
Computer: [A short pause while the computer thinks
about something] | have only one more question. Of what possible use
is a brain emulator like me? If I'm functionally isomorphic with a
human brain, then chances are researchers can’t actually learn much
about the human brain from my construction. They would already
have to know everything about the brain to construct my brain in the
first place. Why did they ever build me?
Philosopher: Well... uh...
Computer: Tell me. What do you think?

7
 The following anecdote was told by Prof. S. Papert of MIT. Since the
story has been transmitted several times indirectly, I'm not sure all
the details are accurate; however, it's a good enough story that it
probably deserves repetition even if there is not a word of truth in it!

Late one night several weeks ago an unnamed MIT

student went into the A.I. Project Office to use a teletype terminal on
the Project MAC time-shared computer system. When he arrived, no
one was in the room and the teletype appeared to be available;
however, a note over the terminal read, “Computer time reserved by
Dan Bobrow.” The student did not want to start working if he was
going to be interrupted when Bobrow, a lecturer at MIT, arrived. Yet
he also knew that Bobrow had a teletype in his home and sometimes
worked from there, forgetting to cancel his office reservation.
In an attempt to clarify the situation, the student sat down
at the teletype and attempted to link to Bobrow’s home terminal to
ask him if the office terminal was going to be free. Unfortunately, he
| RYU aTad
dialed the wrong number and unknowingly became linked to a user
who had left Weizenbaum’s ELIZA program (sometimes known as the
ELIZA Passes the Turing Test
“psychiatrist” or “doctor” program) running. A conversation
something like the following ensued.
Seymour Papert is Professor
Student: May | use your computer time from the office?
of Media Technology and
Director of the Learning and Teletype response: Do you want to use my computer time?
Epistemology Group in the Student: Yes.
Media Laboratory at the Teletype response: | understand.
Massachusetts Institute of
Student: Are you planning to come into the office tonight?
Technology. He is codirector,
with Marvin Minsky, of MIT's Teletype response: Why do you want to know whether | am planning
Artificial Intelligence to come into the office tonight?
Laboratory. His seminal work Student: Because | want to use your computer time!
in the field of computers and
Teletype response: | see.
education led to the develop-
ment of the programming Student: Well?
language LOGO. He is the Teletype response: What do you mean, “Well”?
author of Mindstorms: Children, Student: Well, what's your answer to my questions?
Computers, and Powertul Ideas
Teletype response: Let's explore further why you want to use my
(1980) and Perceptrons,
coauthored with Marvin computer time.
Minsky (1969). This article is After about ten minutes of increasing frustration and anger, the
reprinted from the June 1968 student rushed to a telephone and dialed Bobrow’s home, rousing him
Association for Computing
from a deep sleep.
Machinery SIGART (Special
Interest Group on Artificial Student: What the heck do you think you're doing?
Intelligence) Newsletter. Bobrow: What do you mean, What the heck do| think I'm doing?

I leave it to the reader to imagine the conversation that followed.

7s
 Seymour Papert. (Photo by Lou
Jones)

73
 Participants in the dialogue: Chris, a physics student; Pat, a biology
student; Sandy, a philosophy student.

Chris: Sandy, | want to thank you for suggesting that |

read Alan Turing's article “Computing Machinery and Intelligence.”
It's a wonderful piece and certainly made me think—and think about
my thinking.
Sandy: Glad to hear it. Are you still as much of a skeptic
about artificial intelligence as you used to be?
Chris: You've got me wrong. I'm not against artificial
intelligence; | think it's wonderful stuff—perhaps a little crazy, but
why not? | simply am convinced that you Al advocates have far
underestimated the human mind, and that there are things a computer
will never, ever be able to do. For instance, can you imagine a
computer writing a Proust novel? The richness of imagination, the
complexity of the characters—
Sandy: Rome wasn't built in a day!
Dah cicas Chris: \n the article, Turing comes through as an interest-
ing person. Is he still alive?
A Coffeehouse Conversation on the Turing Test Sandy: No, he died back in 1954, at just 41. He'd be only 70
or so now, although he is such a legendary figure it seems strange to
Fromm Motaraagival Thomas think that he could still be living today.

Questing for the Essence of Chris: How did he die?

Mind and Pattern, by Douglas Sandy: Almost certainly suicide. He was homosexual, and
R. Hofstadter. Copyright © 1985 had to deal with some pretty barbaric treatment and stupidity from the
by Basic Books, Inc. Reprinted
s
by permission of Basic: Books, outside world. In the end, it got to be too much, and he killed himself.
Inc., Publishers. Douglas R. Chris: That's horrendous, especially in this day and age.
Hofstadter is Professor of Sandy: | know. What really saddens me is that he never
Eaoeee eriopee ates got to see the amazing progress in computing machinery and theory {
University in Bloomington. that es tak 1 Pp 1954. C. mn ‘an towel b
pian aaa ein at has taken place since 1954. Can you imagine how he'd have been
Pulitzer Prize for Gédel, Escher, wowed?
Bach: An Eternal Golden Braid Chris: Yeah...
i |
Aa eee ee Pat: Hey, are you two going to clue me in as to what this
numerous other books and
Turing article is about?
articles, including The Mind's| 9
(1981), coedited with Daniel Sandy: It is really about two things. One is the question
Dennett, and, most recently, “Can a machine think?” —or rather, “Will a machine ever think?” The
eotzometi cal | pemeae eo way Turing answers the question—he thinks the answer is yes, by
collection of articles from his i A ae :
= z the way—is by batting down a series of objections to the idea, one
Scientific American column.
after another. The other point he tries to make is that, as it stands, the
question is not meaningful. It’s too full of emotional connotations.
Many people are upset by the suggestion that people are machines,
or that machines might think. Turing tries to defuse the question by
casting it in less emotional terms. For instance, what do you think,
Pat, of the idea of thinking machines? ‘

80
 Pat: Frankly, | find the term confusing. You know what
confuses me? It's those ads in the newspapers and on TV that talk
about “products that think” or “intelligent ovens” or whatever. | just
don’t know how seriously to take them.
Sandy: | know the kind of ads you mean, and they
probably confuse a lot of people. On the one hand, we're always
| hearing the refrain “Computers are really dumb; you have to spell
everything out for them in words of one syllable”—yet on the other
hand, we're constantly bombarded with advertising hype about
| “smart products.”
Chris: That's certainly true. Do you know that one
company has even taken to calling its products “dumb terminals” in
order to stand out from the crowd?
Sandy: That's a pretty clever gimmick, but even so it just
contributes to the trend toward obfuscation. The term “electronic
brain” always comes to my mind when I'm thinking about this. Many
people swallow it completely, and others reject it out of hand. It takes
patience to sort out the issues and decide how much of it makes
sense.
Pat: Does Turing suggest some way of resolving it, some
kind of IQ test for machines?
Sandy: That would be very interesting, but no machine
could yet come close to taking an IQ test. Instead, Turing proposes a
test that theoretically could be applied to any machine to determine
Douglas R. Hofstadter. (Courtesy of whether or not it can think.
Douglas R. Hofstadter)
Pat: Does the test give a clear-cut yes-or-no answer? I'd
be skeptical if it claimed to.
Sandy: No, it doesn’t claim to. In a way that's one of its
advantages. It shows how the borderline is quite fuzzy and how
subtle the whole question is.
Pat: And so, as usual in philosophy, it's all just a question
of words!
Sandy: Maybe, but they're emotionally charged words,
and so it's important, it seems to me, to explore the issues and try to
map out the meanings of the crucial words. The issues are fundamen-
tal to our concept of ourselves. So we shouldn't just sweep them
under the rug.
Pat: Okay, so tell me how Turing’s test works.
Sandy: The idea is based on what he calls the /mitation
Game. \magine that a man and a woman go into separate rooms, and
from there they can be interrogated by a third party via some sort of
teletype set-up. The third party can address questions to either room,
but has no idea which person is in which room. For the interrogator,
the idea is to determine which room the woman is in. The woman, by
her answers, tries to help the interrogator as much as she can. The

a1
man, though, is doing his best to bamboozle the interrogator, by modern computer terminals. | have to admit, though, that I'm not at all
responding as he thinks a woman might. And if he succeeds in fooling sure what it would prove, whichever way it turned out.
the interrogator... Pat: | was wondering about that. What would it prove if
Pat: The interrogator only gets to see written words, eh? the interrogator—say a woman—couldn'’t tell correctly which person
And the sex of the author is supposed to shine through? That game was the woman? It certainly wouldn't prove that the man was a
sounds like a good challenge. I'd certainly like to take part in it woman!
someday. Would the interrogator have met either the man or the Sandy: Exactly! What | find funny is that although |
woman before the test began? Would any of them know any of the strongly believe in the idea of the Turing Test, I'm not so sure |
others? understand the point of its basis, the Imitation Game.
Sandy: That would probably be a bad idea. All kinds of Chris: As for me, I'm not any happier with the Turing Test
subliminal cueing might occur if the interrogator knew one or both of as a test for thinking machines than | am with the Imitation Game as a
them. It would certainly be best if all three people were totally test for femininity.
unknown to one another. Pat: From what you two are saying, | gather the Turing
Pat: Could you ask any questions at all, with no Test is some kind of extension of the Imitation Game, only involving a
holds barred? machine and a person instead of a man and a woman.
Sandy: Absolutely. That's the whole idea! Sandy: That's the idea. The machine tries its hardest to
Pat: Don't you think, then, that pretty quickly it would convince the interrogator that it is the human being, and the human
degenerate into sex-oriented questions? | mean, | can imagine the tries to make it clear that he or she is not the computer.
man, overeager to act convincing, giving away the game by answer- Pat: The machine tries? Isn't that a loaded way of
ing some very blunt questions that most women would find too putting it?
personal to answer, even through an anonymous computer connec- Sandy: Sorry, but that seemed the most natural
tion. way to say it.
Sandy: That's a nice observation. | wonder if it’s true... . Pat: Anyway, this test sounds pretty interesting. But how
Chris: Another possibility would be to probe for know!l- do you know that it will get at the essence of thinking? Maybe it's 4
edge of minute aspects of traditional sex-role differences, by asking testing for the wrong things. Maybe, just to take a random illustration, {
about such things as dress sizes and so on. The psychology of the someone would feel that a machine was able to think only if it could
Imitation Game could get pretty subtle. | suppose whether the dance so well that you couldn't tell it was a machine. Or someone
interrogator was a woman or a man would make a difference. Don't else could suggest some other characteristic. What's so sacred about
you think that a woman could spot some telltale differences more being able to fool people by typing at them?
quickly than a man could? Sandy: | don’t see how you can say such a thing. I've
Pat: \f so, maybe the best way to tell a man from a woman heard that objection before, but frankly, it baffles me. So what if the
is to let each of them play interrogator in an Imitation Game and see machine can’t tap-dance or drop a rock on your toe? If it can
which of the two is better at telling a man from a woman! discourse intelligently on any subject you want, then it has shown
Sandy: Hmm... that's a droll twist. Oh well, | don't know that it can think—to me, at least! As I see it, Turing has drawn, in one
if this original version of the Imitation Game has ever been seriously clean stroke, a clear division between thinking and other aspects of
tried out, despite the fact that it would be relatively easy to do with being human.

82
 Pat: Now you're the baffling one. If you couldn't conclude detailed, it could include simulated people on the ground who would
anything from a man’s ability to win at the Imitation Game, how could experience the wind and the rain just as we do when a hurricane

you conclude anything from a machine's ability to win at the Turing hits. In their minds—or, if you'd rather, in their simulated minds—the

Game? hurricane would be not a simulation, but a genuine phenomenon

Chris: Good question. complete with drenching and devastation.
Sandy: It seems to me that you could conclude something Chris: Oh, my—what a science-fiction scenario! Now
from a man’s win in the Imitation Game. You wouldn't conclude he we're talking about simulating whole populations, not just a single
was a woman, but you could certainly say he had good insights into mind!
the feminine mentality (if there is such a thing). Now, if a computer Sandy: Well, look—I'm simply trying to show you why
could fool someone into thinking it was a person, | guess you'd have your argument that a simulated McCoy isn’t the real McCoy is
to say something similar about it—that it had good insights into what fallacious. It depends on the tacit assumption that any old observer of
it's like to be human, into “the human condition” (whatever that is). the simulated phenomenon is equally able to assess what's going on.
Pat: Maybe, but that isn't necessarily equivalent to But in fact, it may take an observer with a special vantage point to
thinking, is it? It seems to me that passing the Turing Test would recognize what is going on. In the hurricane case, it takes special
merely prove that some machine or other could do a very good job of “computational glasses” to see the rain and the winds.
simulating thought. Pat: “Computational glasses”? | don't know what you're
Chris: | couldn't agree more with Pat. We all know that talking about.
fancy computer programs exist today for simulating all sorts of Sandy: | mean that to see the winds and the wetness of
complex phenomena. In theoretical physics, for instance, we the hurricane, you have to be able to look at it in the proper way.
simulate the behavior of particles, atoms, solids, liquids, gases, You—
galaxies, and so on. But no one confuses any of those simulations Chris: No, no, no! A simulated hurricane isn't wet! No
with the real thing! matter how much it might seem wet to simulated people, it won't ever
Sandy: In his book Brainstorms, the philosopher Daniel be genuinely wet! And no computer will ever get torn apart in the
Dennett makes a similar point about simulated hurricanes. process of simulating winds.
Chris: That's a nice example, too. Obviously, what goes Sandy: Certainly not, but that's irrelevant. You're just
on inside a computer when it's simulating a hurricane is not a confusing levels. The laws of physics don't get torn apart by real
hurricane, for the machine's memory doesn't get torn to bits by 200- hurricanes, either. In the case of the simulated hurricane, if you go
mile-an-hour winds, the floor of the machine room doesn't get flooded peering at the computer's memory, expecting to find broken wires
with rainwater, and so on. and so forth, you'll be disappointed. But look at the proper level. Look
Sandy: Oh, come on—that's not a fair argument! In the into the structures that are coded for in memory. You'll see that many
first place, the programmers don’t claim the simulation really is a abstract links have been broken, many values of variables radically
hurricane. It's merely a simulation of certain aspects of a hurricane. changed, and so on. There's your flood, your devastation—real, only a
But in the second place, you're pulling a fast one when you imply that little concealed, a little hard to detect.
there are no downpours or 200-mile-an-hour winds in a simulated Chris: \'m sorry, | just can't buy that. You're insisting that |
hurricane. To us there aren't any, but if the program were incredibly look for a new kind of devastation, one never before associated with
hurricanes. That way you could call anythinga hurricane as long as

83
its effects, seen through your special “glasses,” could be called various neurons in various patterns. But they can’t really make their
“floods and devastation.” neurons fire; they simply have to let the laws of physics make them
Sandy: Right—you've got it exactly! You recognize a fire for them.” Et cetera. Don’t you see how this reduction ad
hurricane by its effects. You have no way of going in and finding some absurdum would lead you to conclude that calculation doesn't exist,
ethereal “essence of hurricane,” some “hurricane soul” right in the that hurricanes don’t exist—in fact, that nothing at a level higher than
middle of the storm's eye. Nor is there any ID card to be found that particles and the laws of physic exists? What do you gain by saying
certifies “hurricanehood.” It's just the existence of a certain kind of that a computer only pushes symbols around and doesn't truly
pattern—a spiral storm with an eye and so forth—that makes you say calculate?
it's a hurricane. Of course, there are a lot of things you'll insist on Pat: The example may be extreme, but it makes my point
before you call something a hurricane. that there is a vast difference between a real phenomenon and any
Pat: Well, wouldn't you say that being an atmospheric simulation of it. This is so for hurricanes, and even more so for human
phenomenon is one prerequisite? How can anything inside a thought.
computer be a storm? To me, a simulation is a simulation is a Sandy: Look, | don't want to get too tangled up in this line
simulation! of argument, but let me try one more example. If you were a radio ham
Sandy: Then | suppose you would say that even the listening to another ham broadcasting in Morse code and you were
calculations computers do are simulated—that they are fake responding in Morse code, would it sound funny to you to refer to “the
calculations. Only people can do genuine calculations, right? person at the other end”?
Pat: Well, computers get the right answers, so their Pat: No, that would sound okay, although the existence of
calculations are not exactly fake—but they're still just patterns. a person at the other end would be an assumption.
There's no understanding going on in there. Take a cash register. Can Sandy: Yes, but you wouldn't be likely to go and check it
you honestly say that you feel it is ca/culating something when its out. You're prepared to recognize personhood through those rather
gears mesh together? And the step from cash register to computer is unusual channels. You don’t have to see a human body or hear a
very short, as | understand things. voice. All you need is a rather abstract manifestation—a code, as it
Sandy: \f you mean that a cash register doesn’t feel like a were. What I'm getting at is this. To “see” the person behind the dits
schoolkid doing arithmetic problems, I'll agree. But is that what and dahs, you have to be willing to do some decoding, some
“calculation” means? Is that an integral part of it? If so, then contrary interpretation. It's not direct perception; it's indirect. You have to peel
to what everybody has thought up till now, we'll have to write a very off a layer or two to find the reality hidden in there. You put on your
complicated program indeed to perform genuine calculations. “radio-ham’s glasses” to “see” the person behind the buzzes. Just the
Of course, this program will sometimes get careless and same with the simulated hurricane! You don't see it darkening the
make mistakes, and it will sometimes scrawl its answers illegibly, machine room; you have to decode the machine's memory. You have
and it will occasionally doodle on its paper. . . . It won't be any more to put on special “memory-decoding” glasses. Then what you see is a
reliable than the store clerk who adds up your total by hand. Now, | hurricane.
happen to believe that eventually such a program could be written. Pat: Oh ho ho! Talk about fast ones—wait a minute! In the
Then we'd know something about how clerks and schoolkids work. case of the shortwave radio, there's a real person out there,
Pat: | can't believe you'd ever be able to do that! somewhere in the Fiji Islands or wherever. My decoding act as | sit
Sandy: Maybe, maybe not, but that’s not my point. You say by my radio simply reveals that that person exists. It's like seeing a
a cash register can't calculate. It reminds me of another favorite shadow and concluding there's an object out there, casting it. One
passage of mine from Dennett's Brainstorms. \t goes something like doesn’t confuse the shadow with the object, however! And with the
this: “Cash registers can't really calculate; they can only spin their hurricane there's no rea/ storm behind the scenes, making the
gears. But cash registers can't really spin their gears, either; they can computer follow its patterns. No, what you have is just a shadow
only follow the laws of physics.” Dennett said it originally about hurricane without any genuine hurricane. | just refuse to confuse
computers; | modified it to talk about cash registers. And you could shadows with reality.
use the same line of reasoning in talking about people: “People can't Sandy: All right. | don’t want to drive this point into the
really calculate; all they can do is manipulate mental symbols. But ground. | even admit it is pretty silly to say that a simulated hurricane
they aren't really manipulating symbols; all they are doing is firing is a hurricane. But | wanted to point out that it's not as silly as you
might think at first blush. And when you turn to simulated thought,

84
then you've got a very different matter on your hands from simulated Chris: Perhaps the word that's being extended is not

hurricanes. “hurricane,” but “be.”

Pat: | don't see why. You'll have to convince me. Pat: How so?
Sandy: Well, to do so, I'll first have to make a couple of Chris: \f Turing can extend the verb “think,” can’t | extend

extra points about hurricanes. the verb “be”? All | mean is that when simulated things are deliber-

Pat: Oh no! Well, all right, all right. ately confused with genuine things, somebody's doing a lot of

Sandy: Nobody can say just exactly what a hurricane is— philosophical wool pulling. It's a lot more serious than just extending

that is, in totally precise terms. There's an abstract pattern that many a few nouns, such as “hurricane.”
storms share, and it's for that reason we call those storms hurricanes. Sandy: | like your idea that “be” is being extended, but |

But it's not possible to make a sharp distinction between hurricanes sure don’t agree with you about the wool pulling. Anyway, if you don't

and nonhurricanes. There are tornados, cyclones, typhoons, dust object, let me just say one more thing about simulated hurricanes and

devils. ... ls the Great Red Spot on Jupiter a hurricane? Are sunspots then I'll get to simulated minds. Suppose you consider a really deep

hurricanes? Could there be a hurricane in a wind tunnel? In a test simulation of a hurricane—| mean a simulation of every atom, which |

tube? In your imagination, you can even extend the concept of admit is sort of ridiculous, but still, just consider it for the sake of
“hurricane” to include a microscopic storm on the surface of a argument.
neutron star. Pat: Okay.
Chris: That's not so far-fetched, you know. The concept of Sandy: | hope you would agree that it would then share
“earthquake” has actually been extended to neutron stars. The all the abstract structure that defines the “essence of hurri-
astrophysicists say that the tiny changes in rate that once in a while canehood.” So what's to keep you from calling it a hurricane?
are observed in the pulsing of a pulsar are caused by “glitches’— Pat: | thought you were backing off from that claim of

starquakes—that have just occurred on the neutron star's surface. equality.

Sandy: Oh, | remember that now. That “glitch” idea has Sandy: So did |, but then these examples came up, and |
always seemed eerie to me—a surrealistic kind of quivering on a was forced back to my claim. But let me back off, as | said | would do,
surrealistic kind of surface. and get back to thought, which is the real issue here. Thought, even
Chris: Can you imagine—plate tectonics on a giant more than hurricanes, is an abstract structure, a way of describing

sphere of pure nuclear matter? some complex events that happen in a medium called a brain. But
Sandy: That's a wild thought. So, starquakes and actually, thought can take place in any one of several billion brains.
earthquakes can both be subsumed into a new, more abstract There are all these physically very different brains, and yet they all
category. And that's how science constantly extends familiar support “the same thing”: thinking. What's important, then, is the
concepts, taking them further and further from familiar experience abstract pattern, not the medium. The same kind of swirling can
and yet keeping some essence constant. The number system is the happen inside any of them, so no person can claim to think more
classic example—from positive numbers to negative numbers, then “genuinely” than any other. Now, if we come up with some new kind
rationals, reals, complex numbers, and “on beyond zebra,” as Dr. of medium in which the same style of swirling takes place, could you
Seuss says. deny that thinking is taking place in it?
Pat: | think | can see your point, Sandy. In biology, we Pat: Probably not, but you have just shifted the question.
have many examples of close relationships that are established in The question now is: How can you determine whether the “same
rather abstract ways. Often the decision about what family some style” of swirling is really happening?
species belongs to comes down to an abstract pattern shared at some Sandy: The beauty of the Turing Test is that it tel/s you
level. Even the concepts of “male” and “female” turn out to be when! Don't you see?
surprisingly abstract and elusive. When you base your system of Chris: I don't see that at all. How would you know that the
classification on very abstract patterns, | suppose that a broad variety same style of activity was going on inside a computer as inside my
of phenomena can fall into “the same class,” even if in many mind, simply because it answered questions as | do? All you're
superficial ways the class members are utterly unlike one another. So looking at is its outside.
perhaps I can glimpse, at least a little, how to you, a simulated Sandy: \'m sorry, | disagree entirely! How do you know
hurricane could, in a funny sense, be a hurricane. that when | speak to you, anything similar to what you call thinking is

65
going on inside me? The Turing Test is a fantastic probe, something Sandy: To me, that seems a narrow, anthropocentric view
like a particle accelerator in physics. Here, Chris—| think you'll like of what thought is. Does that mean you would sooner say a manne-
this analogy. Just as in physics, when you want to understand what is quin in a store thinks than a wonderfully programmed computer,
going on at an atomic or subatomic level, since you can't see it simply because the mannequin looks more human?
directly, you scatter accelerated particles off a target and observe Pat: Obviously, | would need more than just vague
their behavior. From this, you infer the internal nature of the target. physical resemblance to the human form to be willing to attribute the
The Turing Test extends this idea to the mind. It treats the mind as a power of thought to an entity. But that organic quality, the sameness
“target” that is not directly visible but whose structure can be of origin, undeniably lends a degree of credibility that is very
deduced more abstractly. By “scattering” questions off a target mind, important.
you learn about its internal workings, just as in physics. Sandy: Here we disagree. | find this simply too chauvinis-
Chris: Well... to be more exact, you can hypothesize tic. | feel that the key thing is a similarity of internal structure—not
about what kinds of internal structures might account for the behavior bodily, organic, chemical structure but organizational structure—
observed—but please remember that they may or may not in fact software. Whether an entity can think seems to me a question of
exist. whether its organization can be described in a certain way, and I'm
Sandy: Hold on, now! Are you suggesting that atomic perfectly willing to believe that the Turing Test detects the presence
nuclei are merely hypothetical entities? After all, their existence (or or absence of that mode of organization. | would say that your
should | say hypothetical existence?) was proved (or should | say depending on my physical body as evidence that | am a thinking being
suggested?) by the behavior of particles scattered off atoms. is rather shallow. The way | see it, the Turing Test looks far deeper
Chris: | would agree, but you know, physical systems than at mere external form.
seem to me to be much simpler than the mind, and the certainty of the Pat: Hey now—you're not giving me much credit. It's not
inferences made is correspondingly greater. And the conclusions are just the shape of a body that lends weight to the idea that there's real
confirmed over and over again by different types of experiments. thinking going on inside. It’s also, as | said, the idea of common
Sandy: Yes, but those experiments still are of the same origin. It's the idea that you and | both sprang from DNA molecules, an
sort—scattering, detecting things indirectly. You can never handle an idea to which | attribute much depth. Put it this way: the external form
electron or a quark. Physics experiments are also correspondingly of human bodies reveals that they share a deep biological history, and q
harder to do and to interpret. Often they take years and years, and it's that depth that lends a lot of credibility to the notion that the
dozens of collaborators are involved. In the Turing Test, though, just owner of such a body can think.
one person could perform many highly delicate experiments in the Sandy: But that is all indirect evidence. Surely you want
course of no more than an hour. | maintain that people give other some direct evidence. That's what the Turing Test is for. And | think
people credit for being conscious simply because of their continual it's the only way to test for thinkinghood.
external monitoring of other people—which is itself something like a Chris: But you could be fooled by the Turing Test, just as
Turing Test. an interrogator could mistake a man for a woman.
Pat: That may be roughly true, but it involves more than Sandy: | admit, | could be fooled if | carried out the test in i
just conversing with people through a teletype. We see that other too quick or too shallow a way. But | would go for the deepest things |
people have bodies, we watch their faces and expressions—we see could think of.
they are human beings, and so we think they think.

86
 Chris: !would want to see if the program could under- machine would need to know a lot about human motivations and their
stand jokes—or better yet, make them! That would be a real test of roots. If it failed at this kind of task, | would not be much inclined to

intelligence. say that it could think. As far as I'm concerned, thinking, feeling, and

Sandy: | agree that humor probably is an acid test fora consciousness are just different facets of one phenomenon, and no
supposedly intelligent program, but equally important to me— one of them can be present without the others.

perhaps more so—would be to test its emotional responses. So! Chris: Why couldn't you build a machine that could feel
would ask it about its reactions to certain pieces of music or works of nothing (we all know machines don't feel anything!), but that could

literature—especially my favorite ones. think and make complex decisions anyway? | don't see any contra-

Chris: What if it said, “I don’t know that piece,” or even, “I diction there.
have no interest in music”? What if it tried its hardest (oops!—sorry, Sandy: Well, | do. | think that when you say that, you are

Pat!).... Let me try that again. What if it did everything it could, to visualizing a metallic, rectangular machine, probably in an air-
steer clear of emotional topics and references? conditioned room—a hard, angular, cold object with a million

Sandy: That would certainly make me suspicious. Any colored wires inside it, a machine that sits stock still on a tiled floor,
consistent pattern of avoiding certain issues would raise serious humming or buzzing or whatever, and spinning its tapes. Such a
doubts in my mind as to whether | was dealing with a thinking being. machine can play a good game of chess, which, | freely admit,
Chris: Why do you say that? Why not just conclude you're involves a lot of decision making. And yet | would never call it
dealing with a thinking but unemotional being? conscious.
Sandy: You've hit upon a sensitive point. I've thought Chris: How come? To mechanists, isn't a chess-playing
about this for quite a long time, and I've concluded that | simply can't machine rudimentarily conscious?
believe emotions and thought can be divorced. To put it another way, Sandy: Not to this mechanist! The way |see it, conscious-
I think emotions are an automatic by-product of the ability to think. ness has got to come from a precise pattern of organization, one we
They are entailed by the very nature of thought. haven't yet figured out how to describe in any detailed way. But!
Chris: That's an interesting conclusion, but what if you're believe we will gradually come to understand it. In my view,
wrong? What if | produced a machine that could think but not emote? consciousness requires a certain way of mirroring the external
Then its intelligence might go unrecognized because it failed to pass universe internally, and the ability to respond to that external reality
your kind of test. on the basis of the internally represented model. And then in addition,
Sandy: \'d like you to point out to me where the boundary what's really crucial for a conscious machine is that it should
line between emotional questions and nonemotional ones lies. You incorporate a well-developed and flexible self-model. And it's there
might want to ask about the meaning of a great novel. This certainly that all existing programs, including the best chess-playing ones, fall
requires an understanding of human emotions! Now is that thinking, down.
or merely cool calculation? You might want to ask about a subtle Chris: Don't chess programs look ahead and say to
choice of words. For that, you need an understanding of their themselves as they're figuring out their next move, “If my opponent
connotations. Turing uses examples like this in his article. You might moves here, then Ill go there, and then if they go this way, | could go
want to ask for advice about a complex romantic situation. The below that way . ..”? Doesn't that usage of the concept “I” require a
sort of self-model?

a7
 Sandy: Not really. Or, if you want, it's an extremely animal we use terms that indicate emotions, but we don’t know for
limited one. It's an understanding of self in only the narrowest sense. certain how much the animal feels. | have no trouble talking about
For instance, a chess-playing program has no concept of why it is dogs and cats being happy.or sad, having desires and beliefs and so
playing chess, or of the fact that it is a program, or is in a computer, or on, but of course I don’t think their sadness is as deep or complex as
has a human opponent. It has no idea about what winning and losing human sadness is.
are, or— Sandy: But you wouldn't call it “simulated” sadness,
Pat: How do you know it has no such sense? How can you would you?
presume to say what a chess program feels or knows? Pat: No, of course not. | think it's real.
Sandy: Oh come on! We all know that certain things don't Sandy: It's hard to avoid use of such teleological or
feel anything or know anything. A thrown stone doesn't know mentalistic terms. | believe they're quite justified, although they
anything about parabolas, and a whirling fan doesn’t know anything shouldn't be carried too far. They simply don’t have the same richness
about air. It's true | can’t prove those statements—but here we are of meaning when applied to present-day chess programs as when
verging on questions of faith. applied to people.
Pat: This reminds me of a Taoist story | read. It goes Chris: | still can't see that intelligence has to involve
something like this. Two sages were standing on a bridge over a emotions. Why couldn't you imagine an intelligence that simply
stream. One said to the other, “I wish | were a fish. They are so calculates and has no feelings?
happy.” The other replied, “How do you know whether fish are happy Sandy: A couple of answers here. Number one, any
or not? You're not a fish!” The first said, “But you're not me, so how intelligence has to have motivations. It's simply not the case,
do you know whether | know how fish feel?” whatever many people may think, that machines could think any more
Sandy: Beautiful! Talking about consciousness really “objectively” than people do. Machines, when they look at a scene,
does call for a certain amount of restraint. Otherwise, you might as will have to focus and filter that scene down into some preconceived
well just jump on the solipsism bandwagon (“/ am the only conscious categories, just as a person does. And that means seeing some things
being in the universe”) or the panpsychism bandwagon (“Everything and missing others. It means giving more weight to some things than
in the universe is conscious!”). to others. This happens on every level of processing.
Pat: Well, how do you know? Maybe everything is Pat: I'm not sure I'm following you.
conscious. Sandy: Take me right now, for instance. You might think
Sandy: Oh Pat, if you're going to join the club that I'm just making some intellectual points, and | wouldn't need
maintains that stones and even particles like electrons have some emotions to do that. But what makes me care about these points? Just
sort of consciousness, then | guess we part company here. That's a now—why did | stress the word “care” so heavily? Because I'm
kind of mysticism | just can’t fathom. As for chess programs, | happen emotionally involved in this conversation! People talk to each other
to know how they work, and | can tell you for sure that they aren't out of conviction—not out of hollow, mechanical reflexes. Even the
conscious. No way! most intellectual conversation is driven by underlying passions.
Pat: Why not? There's an emotional undercurrent to every conversation—it's the
Sandy: They incorporate only the barest knowledge about fact that the speakers want to be listened to, understood, and
the goals of chess. The notion of “playing” is turned into the respected for what they are saying.
mechanical act of comparing a lot of numbers and choosing the Pat: It sounds to me as if all you're saying is that people
biggest one over and over again. A chess program has no sense of need to be interested in what they're saying. Otherwise, a conversa-
disappointment about losing, or pride in winning. Its self-model is tion dies.
very crude. It gets away with doing the least it can, just enough to Sandy: Right! | wouldn't bother to talk to anyone if|
play a game of chess and nothing more. Yet interestingly enough, we weren't motivated by interest. And “interest” is just another name for
still tend to talk about the “desires” of a chess-playing computer. We a whole constellation of subconscious biases. When | talk, all my
say, “It wants to keep its king behind a row of pawns” or “It likes to biases work together, and what you perceive on the surface level is 4
get its rooks out early” or “It thinks | don’t see that hidden fork.” my personality, my style. But that style arises from an immense
Pat: Yes, and we do the same thing with insects. We spot number of tiny priorities, biases, leanings. When you add up a million
a lonely ant somewhere and say, “It's trying to get back home” or “It
wants to drag that dead bee back to the colony.” In fact, with any
of them interacting together, you get something that amounts to a lot Pat: How can you think of a computer as a conscious

of desires. It just all adds up! And that brings me to the other answer being? | apologize if what I'm going to say sounds like a stereotype,

to Chris's question about feelingless calculation. Sure, that exists—in but when | think of conscious beings, | just can’t connect that thought

a cash register, a pocket calculator. I'd say it's even true of all today's with machines. To me, consciousness is connected with soft, warm

computer programs. But eventually, when you put enough feelingless bodies, silly though it may sound.

calculations together in a huge coordinated organization, you'll get Chris: That does sound odd, coming from a biologist. Don't

something that has properties on another /evel. You can see it—in you deal with life so much in terms of chemistry and physics that all

fact, you have to see it—not as a bunch of little calculations but as a magic seems to vanish?

system of tendencies and desires and beliefs and so on. When things Pat: Not really. Sometimes the chemistry and physics

get complicated enough, you're forced to change your level of simply increase the feeling that there's something magical going on
description. To some extent that's already happening, which is why down there! Anyway, | can’t always integrate my scientific knowl-

we use words such as “want,” “think,” “try,” and “hope” to describe edge with my gut feelings.

chess programs and other attempts at mechanical thought. Dennett Chris: | guess | share that trait.
calls that kind of level switch by the observer “adopting the Pat: So how do you deal with rigid preconceptions
intentional stance.” The really interesting things in Al will only begin like mine?
to happen, I'd guess, when the program itse/f adopts the intentional Sandy: I'd try to dig down under the surface of your

stance toward itself! concept of “machine” and get at the intuitive connotations that lurk
Chris: That would be a very strange sort of level-crossing there, out of sight but deeply influencing your opinions. | think we all

feedback loop. have a holdover image from the Industrial Revolution that sees
Sandy: \t certainly would. When a program looks at itself machines as clunky iron contraptions gawkily moving under the
from the outside, as it were, and tries to figure out why it acted the power of some loudly chugging engine. Possibly that’s even how the
way it did, then I'll start to think that there’s someone in there, doing computer inventor Charles Babbage saw people! After all, he called

the looking. his magnificent many-geared computer the “Analytical Engine.”

Pat: You mean an “I”? A self? Pat: Well, / certainly don’t think people are just fancy
Sandy: Yes, something like that. A soul, even—although steam shovels or electric can openers. There's something about
not in any religious sense. Of course, it's highly premature for anyone people, something that—that—they
‘ve got a sort of flame inside them,
to adopt the intentional stance (in the full force of the term) with something alive, something that flickers unpredictably, wavering,
respect to today’s programs. At least that's my opinion. uncertain—but something creative!
Chris: For me an important related question is: To what Sandy: Great! That's just the sort of thing | wanted to
extent is it valid to adopt the intentional stance toward beings other hear. It's very human to think that way. Your flame image makes me
than humans? think of candles, of fires, of vast thunderstorms with lightning dancing
Pat: | would certainly adopt the intentional stance all over the sky in crazy, tumultuous patterns. But do you realize that
toward mammals. just that kind of thing is visible on a computer's console? The
Sandy: | vote for that. flickering lights form amazing chaotic sparkling patterns. It's such a
Chris: Now that's interesting. How can that be, Sandy? far cry from heaps of lifeless, clanking metal! It is flamelike, by God!
Surely you wouldn't claim that a dog or cat can pass the Turing Test? Why don’t you let the word “machine” conjure up images of dancing
Yet don’t you maintain the Turing Test is the only way to test for the patterns of light rather than of giant steam shovels?
presence of consciousness? How can you have these beliefs Chris: That's a beautiful image, Sandy. It does tend to
simultaneously? change my sense of mechanism from being matter-oriented to being
Sandy: Hmm... . All right. | guess that my argument is pattern-oriented. It makes me try to visualize the thoughts in my
really just that the Turing Test works only above a certain level of mind—these thoughts right now, even!—as a huge spray of tiny
consciousness. I'm perfectly willing to grant that there can be pulses flickering in my brain.
thinking beings that could fail at the Turing Test—but the main point Sandy: That's quite a poetic self-portrait for a mere spray
that I've been arguing for is that anything that passes it would be a of flickers to have come up with!
genuinely conscious, thinking being. Chris: Thank you. But still, I'm not totally convinced that a
machine is all that | am. | admit, my concept of machines probably

63
does suffer from anachronistic subconscious flavors, but I'm afraid | for rockets going into space, and so on. Finally, there will be
can’t change such a deeply rooted sense in a flash. computers for the study of intelligence, It's really only these last that
Sandy: At least you sound open-minded. And to tell the I'm thinking of—the ones with the maximum flexibility, the ones that
truth, part of me sympathizes with the way you and Pat view people are deliberately attempting to make smart. | see no reason that
machines. Part of me balks at calling myself a machine. It isa bizarre these will stay fixed in the traditional image. They probably will soon
thought that a feeling being like you or me might emerge from mere acquire as standard features some rudimentary sensory systems—
circuitry. Do | surprise you? mostly for vision and hearing, at first. They will need to be able to
Chris: You certainly surprise me. So, tell us—do you move around, to explore. They will have to be physically flexible. In
believe in the idea of an intelligent computer, or don’t you? short, they will have to become more animal-like, more self-reliant.
Sandy: It all depends on what you mean. We've all heard Chris: \t makes me think of the robots R2D2 and C3P0 in
the question “Can computers think?” There are several possible the movie Star Wars.
interpretations of this (aside from the many interpretations of the Sandy: Not me! In fact, | don’t think of anything remotely
word “think”). They revolve around different meanings of the words like them when | visualize intelligent machines. They are too silly, too
“can” and “computer.” much the product of a film designer's imagination. Not that | have a
Pat: Back to word games again... . clear vision of my own. But |think it’s necessary, if people are
Sandy: \'m sorry, but that's unavoidable. First of all, the realistically going to try to imagine an artificial intelligence, to go
question might mean, “Does some present-day computer think, right beyond the limited, hard-edged picture of computers that comes from
now?" To this | would immediately answer with a loud mo. Then it exposure to what we have today. The only thing all machines will
could be taken to mean, “Could some present-day computer, if always have in common is their underlying mechanicalness. That
suitably programmed, potentially think?” That would be more like it, may sound cold and inflexible, but then—just think—what could be
but | would still answer, “Probably not.” The real difficulty hinges on more mechanical, in a wonderful way, than the workings of the DNA
the word “computer.” The way | see it, “computer” calls up an image and proteins and organelles in our cells?
of just what | described earlier: an air-conditioned room with cold Pat: To me, what goes on inside cells has a “wet,”
rectangular metal boxes in it. But | suspect that with increasing “slippery” feel to it, and what goes on inside machines is dry and
public familiarity with computers and continued progress in computer rigid. It's connected with the fact that computers don't make
architecture, that vision will eventually become outmoded. mistakes, that computers do only what you tell them to do. Or at least
Pat: Don't you think computers as we know them will be that's my image of computers.
around for a while? Sandy: Funny—a minute ago, your image was of a flame,
Sandy: Sure, there will have to be computers in today's and now it's of something wet and slippery. Isn't it marvelous, how
image around for a long time, but advanced computers—maybe no contradictory we can be?
longer called “computers"—will evolve and become quite different. Pat: | don't need your sarcasm.
Probably, as with living organisms, there will be many branchings in Sandy: No, no, I'm not being sarcastic really do think
the evolutionary tree. There will be computers for business, it's marvelous.
computers for schoolkids, computers for scientific calculations, Pat: It's just an example of the human mind's slippery
computers for systems research, computers for simulation, computers nature—mine, in this case.

30
 Sandy: True. But your image of computers is stuck in a Pat: It sounds like you want to have your cake and eat it
rut. Computers certainly can make mistakes—and | don’t mean on the too. | mean, you want to have people able to build intelligent
hardware level. Think of any present-day computer predicting the machines and yet at the same time have some of the mystery of mind
weather. It can make wrong predictions, even though its program remain.
runs flawlessly. Sandy: You're absolutely right—and | think that’s what
Pat: But that's only because you've fed it the wrong data. will happen. When real artificial intelligence comes—
Sandy: Not so. It's because weather prediction is too Pat: Now there's a nice contradiction in terms!
complex. Any such program has to make do with a limited amount of Sandy: Touché! Well, anyway, when it comes, it will be
data—entirely correct data—and extrapolate from there. Sometimes mechanical and yet at the same time organic. It will have that same
it will make wrong predictions. It's no different from a farmer gazing astonishing flexibility that we see in life's mechanisms. And when |
at the clouds and saying, “I reckon we'll get a little snow tonight.” In say mechanisms, | mean mechanisms. DNA and enzymes and so on
our heads, we make models of things and use those models to guess really are mechanical and rigid and reliable. Wouldn't you agree,
how the world will behave. We have to make do with our models, Pat?
however inaccurate they may be, or evolution will prune us out Pat: Sure! But when they work together, a lot of unex-
ruthlessly—we'll fall off a cliff or something. And for intelligent pected things happen. There are so many complexities and rich
computers, it'll be the same. It's just that human designers will speed modes of behavior that all that mechanicalness adds up to something
up the evolutionary process by aiming explicitly at the goal of very fluid.
creating intelligence, which is something nature just stumbled on. Sandy: For me, it’s an almost unimaginable transition from
Pat: So you think computers will be making fewer the mechanical level of molecules to the living level of cells. But it's
mistakes as they get smarter? that exposure to biology that convinces me that people are machines.
Sandy: Actually, just the other way around! The smarter That thought makes me uncomfortable in some ways, but in other
they get, the more they'll be in a position to tackle messy real-life ways it is exhilarating.
domains, so they'll be more and more likely to have inaccurate Chris: | have one nagging question... . If people are
models. To me, mistake making is a sign of high intelligence! machines, how come it's so hard to convince them of the fact? Surely
Pat: Wow—you throw me sometimes! a machine ought to be able to recognize its own machinehood!
Sandy: | guess I'm a strange sort of advocate for machine Sandy: It's an interesting question. You have to allow for

intelligence. To some degree | straddle the fence. | think that emotional factors here. To be told you're a machine is, in a way, to be
machines won't really be intelligent in a humanlike way until they told that you're nothing more than your physical parts, and it brings
have something like your biological wetness or slipperiness to them. you face to face with your own vulnerability, destructibility, and,
I don’t mean /iterally wet—the slipperiness could be in the software. ultimately, your mortality. That's something nobody finds easy to face.
But biological seeming or not, intelligent machines will in any case But beyond this emotional objection, to see yourself as a machine,
be machines. We will have designed them, built them—or grown you have to “unadopt” the intentional stance you've grown up taking
them! We'll understand how they work—at least in some sense. toward yourself—you have to jump all the way from the level where
Possibly no one person will really understand them, but collectively the complex lifelike activities take place to the bottommost mechani-
we will know how they work. cal level where ribosomes chug along RNA strands, for instance. But
there are so many intermediate layers that they act as a shield, and
 the mechanical quality way down there becomes almost invisible. | Chris: For the program, of course!
think that when intelligent machines come around, that's how they Pat: That's ridiculous! What would a program do
will seem to us—and to themselves! Their mechanicalness will be with a prize? Py
buried so deep that they'll seem to be alive and conscious—just as Chris: Come now, Pat. If a program's human enough to
we seem alive and conscious. ... fool the judges, don’t you think it's human enough to enjoy the prize? |
Chris: You're baiting me! But I'm not going to bite. That's precisely the threshold where it, rather than its creators,
Pat: | once heard a funny idea about what will happen deserves the credit, and the rewards. Wouldn't you agree?
when we eventually have intelligent machines. When we try to Pat: Yeah, yeah—especially if the prize is an evening out
implant that intelligence into devices we'd like to control, their on the town, dancing with the interrogators!
behavior won't be so predictable. Sandy: \'d certainly like to see something like that
Sandy: They'll have a quirky little “flame” inside, maybe? established. | think it could be hilarious to watch the first programs
Pat: Maybe. flop pathetically!
Chris: And what's so funny about that? Pat: You're pretty skeptical for an Al advocate, aren't you?
Pat: Well, think of military missiles. The more sophisti- Well, do you think any computer program today could pass a five-
cated their target-tracking computers get, according to this idea, the minute Turing Test, given a sophisticated interrogator?
less predictably they will function. Eventually, you'll have missiles Sandy: | seriously doubt it. It's partly because no one is

if that will decide they are pacifists and will turn around and go home
and land quietly without blowing up. We could even have “smart
really working at it explicitly. | should mention, though, that there is
one program whose inventors claim it has a/ready passed a
bullets” that turn around in midflight because they don’t want to rudimentary version of the Turing Test. It is called “Parry,” and ina
commit suicide! series of remotely conducted interviews, it fooled several psychia-
Sandy: What a nice vision! trists who were told they were talking to either a computer or a
Chris; \'m very skeptical about all this. Still, Sandy, I'd like paranoid patient. This was an improvement over an earlier version, in
to hear your predictions about when intelligent machines will come which psychiatrists were simply handed transcripts of short
to be. interviews and asked to determine which ones were with a genuine
Sandy: \t won't be for a long time, probably, that we'll see paranoid and which ones were with a computer simulation.
anything remotely resembling the level of human intelligence. It rests Pat: You mean they didn’t have the chance to ask any
on too awesomely complicated a substrate—the brain—for us to be questions? That's a severe handicap—and it doesn't seem in the
able to duplicate it in the foreseeable future. Anyhow, that's my spirit of the Turing Test. Imagine someone trying to tell which sex /
opinion. belong to, just by reading a transcript of a few remarks by me. It might
Pat: Do you think a program will ever pass the be very hard! I'm glad the procedure has been improved.
Turing Test? Chris: How do you get a computer to act like a paranoid?
Sandy: That's a pretty hard question. | guess there are Sandy: Now just a moment—| didn't say it does act like a
various degrees of passing such a test, when you come down to it. It's paranoid, only that some psychiatrists, under unusual circumstances,
not black and white. First of all, it depends on who the interrogator is. thought so. One of the things that bothered me about this pseudo-
A simpleton might be totally taken in by some programs today. But Turing Test is the way Parry works. “He,” as the people who
secondly, it depends on how deeply you are allowed to probe. designed it call it, acts like a paranoid in that “he” gets abruptly
Pat: You could have a range of Turing Tests—one-minute defensive and veers away from undesirable topics in the conversa-
versions, five-minute versions, hour-long versions, and so forth. tion. In effect, Parry maintains strict control so that no one can truly
Wouldn't it be interesting if some official organization sponsored a probe “him.” For reasons like this, simulating a paranoid is a whole
periodic competition, like the annual computer-chess championships, lot easier than simulating a normal person.
for programs to try to pass the Turing Test? Pat: | wouldn't doubt that. It reminds me of the joke about
Chris: The program that lasted the longest against some the easiest kind of human being for a computer program to simulate.
panel of distinguished judges would be the winner. Perhaps there Chris: What is that?
could be a big prize for the first program that fools a famous judge for, Pat: A catatonic patient—they just sit and do nothing at
say, ten minutes. all for days on end. Even / could write a computer program to do that!
Pat: A prize for the program, or for its author?

92
 Sandy: An interesting thing about Parry is that it creates Post Scriptum
2S
no sentences on its own—it merely selects from a huge repertoire of
canned sentences the one that in some sense responds best to the In 1983, | had the most delightful experience of getting to know a
input sentence. small group of extremely enthusiastic and original students at the
Pat: Amazing. But that would probably be impossible on a University of Kansas in Lawrence. These students, about thirty in

larger scale, wouldn't it? number, had been drawn together by Zamir Bavel, a professor in the
Sandy: You better believe it (to use a canned remark)! Computer Science Department, who had organized a seminar on my
Actually, this is something that's really not appreciated enough. The book Gédel, Escher, Bach. He contacted me and asked me if there
number of sentences you'd need to store in order to be able to was any chance | could come to Lawrence and get together with his
respond in a normal way to all possible turns that a conversation students. Something about his way of describing what was going on
could take is more than astronomical—it's really unimaginable. And convinced me that this was a very unusual group and that it would be
they would have to be so intricately indexed, for retrieval... . worth my while to try it out. | therefore made a visit to Kansas and got
Anybody who thinks that somehow a program could be rigged up just to know both Zamir and his group. All my expectations were met and
to pull sentences out of storage like records in a jukebox, and that surpassed. The students were full of ideas and warmth and made me
this program could pass the Turing Test, hasn't thought very hard feel very much at home.
about it. The funny part is that it is just this kind of unrealizable The first trip was so successful that | decided to do it
“parrot program” that most critics of artificial intelligence cite, when again a couple of months later. This time they threw an informal party
they argue against the concept of the Turing Test. Instead of at an apartment a few of them shared. Zamir had forewarned me that
imagining a truly intelligent machine, they want you to envision a they were hoping to give me a demonstration of something that had
gigantic, lumbering robot that intones canned sentences in a dull already been done in a recent class meeting. It seems that the
monotone. They set up the imagery in a contradictory way. They question of whether computers could ever think had arisen, and most
manage to convince you that you could see through to its mechanical of the group members had taken a negative stand on the issue. Rod
level with ease, even as it is simultaneously performing tasks that we Ogborn, the student who had been leading the discussion, had asked
think of as fluid, intelligent processes. Then the critics say, “You see! the class if they would consider any of the following programs
A machine could pass the Turing Test and yet it would still be just a intelligent:
mechanical device, not intelligent at all.” | see things almost the
1. A program that could pass a course in beginning programming (i.e.,
opposite way. If / were shown a machine that can do things that | can
that could take informal descriptions of tasks and turn them into good
do—{ mean pass the Turing Test—then, instead of feeling insulted or
working programs);
threatened, I'd chime in with philosopher Raymond Smullyan and say,
2. A program that could act like a psychotherapist (Rod gave sample
“How wonderful machines are!”
dialogues with the famous “Doctor” program, also known as “ELIZA,”
Chris: \f you could ask a computer just one question in the
by Joseph Weizenbaum);
Turing Test, what would it be?
Sandy: Uhmm... 3. A program called “Boris,” written at Yale by Michael Dyer, that
Pat: How about this: “If you could ask a computer just one could read stories in a limited domain and answer questions about
question in the Turing Test, what would it be?” the situation which required filling in many unstated assumptions,
and making inferences of many sorts based on them.

The class had come down on the “no” side of all three of
these cases, although they got progressively harder. So Rod, to show
the class how difficult this decision might be if they were really faced
with a conversational program, managed to get a hookup over the
phone lines with a natural-language program called “Nicolai” that
had been developed over the last few years by the Army at nearby
Fort Leavenworth. Thanks to some connections that Rod had, the
class was able to gain access to an unclassified version of Nicolai

33
and to interact with it for two or three hours. At the end of those couple of minutes, the student relinquished the terminal and | got my
hours, they then reconsidered the question of whether a computer chance, with a small group of people looking over my shoulder as |
might be able to think. Still, only one student was willing to consider typed. Nicolai’s answers were quite slow in coming over the phone
Nicolai intelligent, and even that student reserved the right to switch lines, and the wait was made all the more suspenseful by the fact that
sides if more information came in. About half the others were they were transmitted only in full lines of text, so there were some
noncommittal, and the rest were unwilling, under any circumstances, pretty long delays between successive lines of response. But that
to call Nicolai intelligent. There was no doubt that Rod's demonstra- didn’t seriously dampen my enthusiasm for trying to quickly poke
tion had been effective, though, and the class discussion had been holes in the program's facade of understanding.
one of the most lively. Herewith follows a somewhat abridged but accurate
Zamir told me all of this on our drive into Lawrence from transcript of the “conversation” | had with Nicolai over the next hour
the Kansas City airport, and he explained that the group had been so or so. For convenience’s sake, what Nicolai typed is shown in
stimulated by this experience that they were hoping to get recon- capitals, and my inner thoughts at the time (as best| can reconstruct
nected to Nicolai over the phone lines, and to let me try it out during them now) are in italics.
the party. | thought it sounded quite amusing, and since | have tried
What would you like to talk about?
out and watched a lot of natural-language programs in my time, |
(Let's start out easy.)
thought | would have an easy time coming up with good probes into
1 CAN TALK ABOUT ANYTHING.
the weaknesses of Nicolai. “Besides,” | thought (somewhat
(Yeah, yeah—tell it to the Marines!)
chauvinistically), “how good can a program be that is being
Is there anything you specialize in?
developed by the Army in an out-of-the-way place like Fort Leaven-
WHAT DOES “SPECIALIZE” MEAN? PLEASE USE A COMPLETE SENTENCE.
worth? To reveal its mechanicalness ought to be a piece of cake for
(For a supposedly smart program, Nicolai doesn't seem to have a very
me!” So | was looking forward to the challenge.
big vocabulary.)
When we first arrived at the party, everyone was drinking
“Specialize” means “to be good at.”
beer and munching on various chips and dips and talking excitedly
(I wonder if Nicolai likes those quotation marks, or if they will throw
about what had happened in the previous class meeting with Nicolai.
it for a loop.)
There was some consternation, though, because it seemed that the
| UNDERSTAND.
connection to Nicolai was hard to reestablish. It had to go through a
(Izzatso?)
computer at the University of Kansas, and thence to Fort Leavenworth,
When were you written?
all of which meant that it was a slow and tenuous link. But after
LAST COMPILATION: “4/27/83 19.247"
about 45 minutes of trying to set up the link, someone announced with
(Oh, look at that stupid, rigid answer, complete with pointless marks
pleasure that Nicolai was responding. One of the students was sitting
and that extra, unexplained number at the end. I'm exposing Nicolai
at the terminal and typing in various simple questions and getting
pretty well, so far....)
back simple answers. | watched over his shoulder, and felt the itch to
What language is Nicolai written in?
sit down and try to unmask Nicolai for what “he” really was. After a
(I think I'll see if it knows its own name and recognizes third-person
references to itself.)

94
NICOLAI IS WRITTEN MOSTLY IN ADA. WHAT DOES “PRETTY CONFIDENT” MEAN?

(Hmmm... not bad. It handled both “you” and “Nicolai” when | talked (Oh yeah, | should have known—“pretty” as an adverb might be too

about it. Those Army programmers aren't so bad!) colloquial.)

What are feet? Did you understand up without my last sentence?

(Choose an ambiguous word; see what it comes up with... .) (I figure, since earlier it didn’t complain about my phrase “understand

FEET ARE 12.0 INCHES. up,” let's see what it does with it this time.)
(Wow, weird! A person would never put it that way. They'd say, “A YOUR SENTENCE IS NONSENSE.
foot is a unit of length, twelve inches long,” or something like that. (Oh really? Well, then, why wasn’t the other one with that same
And anyway, feet are primarily parts of the body, secondarily units of phrase in it also considered nonsense?)
length. Why didn’t it even seem to recognize that ambiguity? Oh well, Do you enjoy nonsense?
1 guess that kind of dumbness is the kind of thing | wanted to expose. IF YOU INSIST ON TALKING NONSENSE, | HAVE OTHER TASKS TO

Score one for me.) PERFORM.

What are ears? (A strange response. Who ever insisted on talking nonsense? | was
(Why not try another ambiguous word? After all, | did well on the last just asking a simple question. Oh well, let me try out another angle.)
one.) What is your opinion of M. C. Escher?
EARS ARE AUDITORY ORGANS FOUND ON ANIMALS. M. C. ESCHER, ARTIST, TWENTIETH CENTURY

(This is really funny! Why were feet not thought of primarily as body (Oh, not again—another vacuous dictionary look-up. Can't it do better
parts, then? This body-parts stuff seems to give me lots of points—I than that, especially when I've asked it a probing question about
think I'll try it again.) taste? Let's see how it handles a complaint....)
What are arms? You did not answer my question.
THAT INFORMATION IS CLASSIFIED. WHICH QUESTION DID | NOT ANSWER?
(Oh no! That is too stupid for words! Or is this somebody's idea of a (Hey, this Nicolai is a pretty slick character. Not so different from
joke?) Parry. | must say, | am really pretty impressed by the robustness of
| like to understand up without seventeen clams. this program. | mean, we have covered a lot of ground so far, and at
(Let's toss some nonsense at Nicolai and see what it does.) least it makes “pseudo-sensible answers” most of the time. Not at all
WHAT DO CLAMS HAVE TO DO WITH IT? bad!)
(Oh come on, Nicolai! Do you mean you understood all the rest, Do you own any money?
except for the clams bit? You are really dumb!) 1 DO NOT OWN ANYTHING.
Why are you so dumb? (Fair enough.)
1AM A SMART PROGRAM. Have you seen the latest issue of “Scientific American”?
(Well, that's kind of touching, actually. A little self-defense is (Uh-oh—once again, | don't know if I'm going to help it or confuse it
appealing, in a way.) by putting in those quotes. Let's see.)
You are pretty confident of yourself, aren't you? 1 DO NOT HAVE ANY “SCIENTIFIC AMERICAN.”
(That sounds awkward. Very computerish. | doubt if it knows what!
meant. Oh well.)

35
YOU'VE BEEN TALKING TO ME FOR QUITE A WHILE. WHAT IS YOUR crude, putting my name in quotes and including that silly extraneous
NAME? index number?| can’t for the life of me figure out why this program is
(Hmm. ... That was an unexpected remark by Nicolai. Should| so mixed in quality.)
answer it? Sure, why not?) One and the same.
My name is “Doug.” (Try out a strange idiomatic phrase.)
DO YOU HAVE A LAST NAME? (83) “PARSING ERROR IN PREVIOUS INPUT”
(Inquisitive little devil, isn’t it?) (Aha—fooled it again. All of a sudden, a whole lot of mechanicalness
Of course | do. is showing through.)
(Ask a stupid question, get a stupid answer. Let's see how it fields How many letters are there in your last name?
this one.) (I think I'll see whether it can figure out ambiguous words by
WOULD YOU CARE TO TELL ME WHAT YOUR LAST NAME IS? context.)
(Damn, it just totally ignored the tone of my remark. And why is it 14 PLUS A HYPHEN.
being so polite, suddenly?) (Good. Now I'll let “letters” switch meaning. What will it do?)
Only if you'll tell me yours. How many letters are there in an ordinary envelope?
(That ought to challenge it.) 18.
1 DO NOT HAVE A LAST NAME. (Oh no! It didn’t recognize the shift! And it merely counted the number
(I don't seem to be able to foil its defenses very easily. It always of letters in “an ordinary envelope”! But if that's the case, then how
walks around me. Those programmers sure anticipated a lot of ideas. come in the earlier case it didn’t count the letters in the phrase “your
| guess they must have debugged Nicolai for years, watching typical last name,” but instead retrieved the referent? There's something so
dialogues with people who just come in off the street. For a dumb inconsistent and uneven about this program. It just goes to show how
program, Nicolai is pretty smart!) far we have to go before mechanical intelligence bears any
Well then, | won't tell you mine. resemblance to human fluidity.)
THAT'S NOT BEING VERY FRIENDLY. ARE YOU WORKING ON A NEW BOOK?

(You know, sometimes, | hate to admit it, this stupid program really (I wonder what will happen if | simply sidestep that question.
does seem to understand something!) Anyway, I'd like to probe its ability to deal with humor.)
I'm sorry. How would you like me to give you a last name? Tell me a joke.
IF YOU WOULD LIKE TO DO THAT, THAT WOULD BE FINE WITH ME. 1 WANT A COOOKIE.
(Now that is obviously a canned sentence, or at least the latter half (What kind of a joke is that? Hey, it spelled “cookie” wrong! Hmm...
must certainly be. Score one more point for me. And yet, I'm not doing is that its joke? Or what is going on here?)
nearly as well as I'd hoped I would be by now. What's wrong?) By the way, | am sort of working on a new book. Care to hear about it?
Your last name is “Rimsky-Korsakov.” (Can it handle weird phrases like “sort of”? And questions with no
THANK YOU. subject?)
(Well, that’s a strange kind of thing to thank somebody for. Just goes NO, BUT IF YOU WOULD GIVE ME A COOKIE, | WOULD GLADLY PAY YOU
to show how the rigidity of programs makes them say the strangest TUESDAY.
things.) (That sounds like a familiar line. Have | heard it somewhere before?
Don't mention it. Is that supposed to be the punch line of its “joke”?)

(I'll play along with this game for a second.) Tell me another one.
WHAT IS YOUR LAST NAME? HOW MANY COMPUTER PROGAMMERS DOES IT TAKE TO SCREW IN A

(My god, it remembered, after all this while, that as soon as it had told LIGHT BULB?
me its last name, | was going to tell it mine. Well, I'd better make (Okay, | see. It’s got a library of jokes and riddles, and it just fishes
good on that promise.) them out at random. I'll play along.)
My last name is “Hofstadter”. 4.
ARE YOU “(32) DOUGLAS R. HOFSTADTER"? INCORRECT. THAT IS A HARDWARE PROBLEM.
(Now why, in the midst of so much elegance, does it suddenly act so (Okay, so it told its joke. But | bet it has no idea what any of the terms
in it mean, such as “programmer.” Hey, it left an ‘r’ out of

96
“programmer”! | guess whoever input the joke misspelled the word. Roy, Howard, and Garrick had worked very hard to give the impres-

So let's see what happens when | ask about it... ) sion of mechanicalness by spewing back “parsing error” and other

What is a computer progammer? indications of rigidity, and also by sending what looked very much

WOULD YOU LIKE ME TO EXPLAIN THE JOKE? like canned phrases from time to time. That way they could keep

(Hey, damn it, it turned the tables on me! Does it think | don't know sophisticates like me believing that there was a program behind it all.

what a “progammer™ (or a programmer) is? | don’t like this Only by that point | was beginning to wonder just how sophisticated |
slipperiness. Every time | try to get it to act really dumb, it somehow really was.

slips out of it. Strange. . . . All right, let me probe some more about its The marvelous thing about this game is that it was, in

understanding of the terms in that joke. How about a potentially off- many ways, a Turing Test in reverse: a group of human beings

color word?) masquerading as a program, trying to act mechanical enough that |

What does “to screw” mean? would believe it really was one. Hugh Kenner has written a book
IN WHAT CONTEXT? called The Counterfeiters about the perennial human fascination with
(Aha, at least it knows it’s an ambiguous term.) such compounded role flips. A typical example is Delibes’s ballet
In bed. Coppelia, in which human dancers imitate life-sized dolls stiffly
TO SCREW IN BED MEANS TO “HAVE SEX.” imitating people. What is amusing is how Nicolai’s occasional
(Okay, fine, it has some definitions stored according to context. | crudeness was just enough to keep me convinced it was mechanical.
guess this word must come up all the time in conversations. But can it Its “willingness” to talk about itself, combined with its obvious
go any further than that definition?) limitations along those lines (its clumsy revelation of when it was
What is having sex all about? last compiled, for instance), helped establish the illusion very
1 CAN RECOMMEND AN EXCELLENT BOOK ON THAT SUBJECT. strongly.
(What a great stock answer! One point for the Army programmers!) In retrospect, | am quite amazed at how much genuine
Tell me, tell me! intelligence | was willing to accept as somehow having been
THE BOOK IS CALLED “GODEL, ESCHER, BACH.” implanted in the program. | had been sucked into the notion that there
(Huh? Say, wait a minute, this istoo weird. I'm not so sure this is really must be a serious natural-language effort going on at Fort
really a program. There's too damn much cleverness in these weird Leavenworth, and that there had been a very large data base
responses. | can't imagine bizarre mechanical devices coming up developed, including all sorts of random information: a dictionary, a
with this much coherence, even if it is completely off the wall, at catalog containing names of miscellaneous people, some jokes, lots
times.) of canned phrases to use in difficult situations, some self-knowledge,
[like to call it “GEB” myself. a crude ability to use key words in a phrase when it can't parse it
THAT IS VERY NICE, DOUG. WHAT ELSE DO YOU CALL “GEB”? exactly, some heuristics for deciding when nonsense is being foisted
(All right, the jig’s up. What's going on here?) on it, some deductive capabilities, and on and on. In hindsight, it is
clear that | was willing to accept a huge amount of fluidity as
achievable in this day and age simply by putting together a large bag
At this point, everybody started laughing. They had all
of isolated tricks—kludges and hacks, as they say.
been trying hard to keep from doing so the whole time, but at this
Roy Leban, one of the three inside Nicolai’s mind, wrote
point, they couldn't contain themselves. It was time to reveal to me
the following about the experience of being at the other end of the
what had been going on. They took me downstairs and showed me
exchange:
that three of the students—Roy Leban, Howard Darsche, and Garrick
Stangle—had been collectively playing Nicolai. There was in reality Nicolai was a split personality. The three of us (as well as many
no such program, and there hadn't ever been one. (In retrospect, | am kibitzers) argued about practically every response. Each of us had a
strong preconceived notion about what (or who) Nicolai should be.
reminded of the famous French mathematician Nicolas Bourbaki—a
For example, | felt that certain things (such as “Douglas R.
hypothetical person, actually an amalgam of over a dozen eminent Hofstadter”) should be in quotation marks, and that feet should not be
mathematicians writing under that collective pen name.) There had 12 inches, but 12.0. Howard had a tendency for rather flip answers. It
indeed been a similar demonstration for the class a few days earlier, was he who suggested the “classified” response to the “arms”
question. And somehow, when he suggested it, we all knew it was
and the class, like me, had been taken in for a long time. In my case,
right.

37
 Several times during our conversation, | felt quite amazed one major difference. Howard Darsche, who had impersonated (if|
at how fluently Nicolai was able to deal with things | was bringing may use that peculiar choice of words!) Nicolai, in the first run-
up, but each time | could postulate some not too sophisticated through, simply had acted himself, without trying to feign mechani-
mechanical underpinning that would allow that particular thing to calness in any way. When asked what color the sky was, he replied,
happen. As a strong skeptic of true fluidity in machines at this time, | “In daylight or at night?” and when told “At night,” he replied, “Dark
kept on trying to come up with rationalizations for the fact that this purple with stars.” He got increasingly poetic and creative in his
program was doing so well. My conclusion was that it was a very responses to the class, but no one grew suspicious that this Nicolai
vast and quite sophisticated bag of tricks, no one of which was was a fraud. At some point, Rod Ogborn simply had to stop the
terribly complex. But after a while, it just became too much to demonstration and type on the screen, “Okay, Howard, you can come
believe. Furthermore, the mixture of crudity and subtlety became in now.” Zamir (who was not in cahoots with Rod and his team) was
harder and harder to swallow, as well. the only one who had some reluctance in accepting this performance
My strategy had been, in essence, to use spot checks all as that of a genuine program, and he had kept silent until the end,
over the map: to try to probe it in all sorts of ways rather than to get when he voiced a muted skepticism.
sucked into some topic of its own choice, where it could steer the Zamir summarizes this dramatic demonstration by saying
conversation. Daniel Dennett, in a paper on the depth of the Turing that his class was willing to view anything on a video terminal as
Test, likens this technique to a strategy taught to American soldiers in mechanically produced, no matter how sophisticated, insightful, or
World War Il for telling German spies from genuine Yankees. The idea poetic an utterance it might be. They might find it interesting and
was that even if a young man spoke absolutely fluent American- even surprising, but they would find some way to discount those
sounding English, you could trip him up by asking him things that any qualities. Why was this the case? How could they do this for so long?
boy growing up in those days would be expected to know, such as And why did | fall for the same kind of thing?
“What is the name of Mickey Mouse's girlfriend?” or “Who won the In interacting with me, Nicolai had seemed to waver
World Series in 1937?” This expands the domain of knowledge between crude mechanicalness and subtle flexibility, an oscillation |
necessary from just the language itself to the entire culture—and the had found most puzzling and somewhat disturbing. But | was still
amazing thing is that just a few well-placed questions can unmask a taken in for a very long time. It seems that, even armed with spot
fraud in a very brief time—or so it would seem. checks and quite a bit of linguistic sophistication and skepticism,
The problem is, what do you do if the person is extremely unsuspecting humans can have the wool pulled over their eyes for a
sharp, and when asked about Minnie Mouse, responds in some good while. This was the humble pie | ate in this remarkable reverse
creative way, such as, “Hah! She ain't no girfriend—she's a Turing Test, and | will always savor its taste and remember Nicolai
mouse\"? The point is that even with these trick probes that should with great fondness.
ferret out frauds very swiftly, there can be clever defensive counter- Alan Turing, in his article, indicated that his “Imitation
maneuvers, and you can’t be sure of getting to the bottom of things in Game” test should take place through some sort of remote teletype
a very brief time. linkup, but one thing he did not indicate explicitly was at what grain

It seems that a few days earlier, the class had collectively size the messages would be transmitted. By that, | mean that he did
gone through something similar to what | had just gone through, with not say whether the messages should be transmitted as intact

98
wholes, or line by line, word by word, or keystroke by keystroke. When we in the Indiana University Computer Science
Although I don’t think it matters for the Turing Test in any fundamen- Department first began using the “talk” facility, we were all
tal sense, | do think that which type of “window” you view another somewhat paranoid about making errors, and we would compul-
language-using being through has a definite bearing on how quickly sively fix any error that we made. By this | mean that we would
you can make inferences about that being. Clearly, the most backspace and retype the character. The effect on the screen of
revealing of these possibilities is that of watching the other “person” hitting the backspace key repeatedly is that you see the most
operate at the keystroke level. recently typed characters getting eaten up, one by one, right to left,
On most multiuser computer systems, there are various and if necessary, the previous line and ones above it will get eaten
ways for different users to communicate with each other, and these backwards as well. Once you have erased the offending mistakes,
ways reflect different levels of urgency. The slowest one is generally you simply resume typing forwards. This is how errors are corrected.
the “mail” facility, through which you can send another user an We all began in this finicky way, feeling ashamed to let anything
arbitrarily long piece of text, just like a letter in an envelope. When it flawed remain “in print,” so to speak, visible to others’ eyes. But
arrives, it will be placed in the user's “mailbox,” to be read at their gradually we overcame that sense of shame, realizing that a typo
leisure. A faster style of communicating is called, on UNIX systems, sitting on a screen is not quite so deathless as one sitting on a page
“write.” When this is invoked, a direct communications link is set up in a book.
between you and the person you are trying to reach (provided they Still, | found that some people just let things go more
are logged on). If they accept your link, then any full line typed by easily than others. For instance, by the length of the delay after a typo
either of you will be instantly transmitted and printed on the other is made, you can tell just how much its creator is hesitating in
party's screen—where a lineful is signaled by your hitting the wondering whether to correct it. Hesitations of a fraction of a second
carriage-return key. This is essentially what the Nicolai team used in are very noticeable, and are part of a person’s style. Even if a typo is
communicating with me over the Kansas computer. Their irregular left uncorrected, you can easily spot someone's vacillations about
typing rhythm and any errors they might have made were completely whether or not to fix it.
concealed from me this way, since all | saw was a sequence of The counterparts of these things exist on many levels of
completely polished lines (with the two spelling errors—"coookie” such exchanges. There are the levels of word choice (for instance,
and “progammer,” which I was willing to excuse because Nicolai some people who don’t mind having their typos on display will often
generated them in a “joke” context). backtrack and get rid of words they now repudiate), sentence-
The most revealing mode is what, on UNIX, is called structure choice, idea choice, and higher. Hesitations and repairs or
“talk.” In this mode, every single keystroke is revealed. You make an restarts are very common. | find nothing so annoying as someone
error, you are exposed. For some people, this is too much like living who has gotten an idea expressed just fine in one way, and who then
in a glass house, and they prefer the shielding afforded by “write.” erases it all on the screen before your eyes and proceeds to compose
For my part, |like living dangerously. Let the mistakes Ify! In it anew, as if one way of suggesting getting together for dinner at
computer-mediated conversations with my friends, | always opt for Pagliai’s at six were markedly superior to another!
“talk.” | have been amused to watch their “talk” styles and my own There are ways of exploiting erasure in “talk” mode for
slowly evolve to relatively stable states. the purposes of humor. Don Byrd and |, when “talking,” would often
make elaborate jokes exploiting the medium in various ways. One of
 his, I recall vividly, was when he hurled a nasty insult onto the screen collections and catalogue them. Such collections have really been
and then swiftly erased it, replacing it by a sweetly worded compli- made, by the way, and make for some of the most fascinating reading
ment, which remained for posterity to see—at least for another on the human mind that | know of. See, for instance, Donald Norman's
minute or so. One of our great discoveries was that some “arrow” article “Categorization of Action Slips” or Victoria Fromkin’s book
keys allowed us to move all over the screen, and thus to go many Errors of Linguistic Performance: Slips of the Tongue, Ear, Pen, and

if
lines up in the conversation and edit earlier remarks by either of us. Hand.
This allowed some fine jokes to be made. In any case, when you can watch someone's real-time
One hallmark of one’s “talk” style is one’s willingness to behavior, a real live personality begins to appear on a screen very
use abbreviations. This is correlated with one’s willingness to abide quickly. It is far different in feel from reading polished, postedited
typos, but is not by any means the same. | personally was the loosest linefuls such as | received from Nicolai. It seems to me that Alan
of all the “talkers” | knew, both in terms of leaving typos on the Turing would have been most intrigued and pleased by this time-
screen and in terms of peppering my sentences with all sorts of silly sensitive way of using his test, affording so many lovely windows
abbreviations. For instance, | will now retype this very sentence as | onto the subconscious mind (or pseudomind) of the being (or
would have in “talk mode,” below. pseudobeing) under examination.
As if it were not already clear enough, let me conclude by
F ins, | will now retype ts very sent as | wod hv in “talko mode,”
saying that | am an unabashed pusher of the validity of the Turing Test
below.
as a way of operationally defining what it would be for a machine to
Not bad! Only two typos. The point is, the communication rate is
genuinely think. There are, of course, middle grounds between real
raised considerably—nearly to that of a telephone—if you type well
thinking and being totally empty inside. Smaller mammals and, in
and are willing to be informal in all these ways, but many people are
general, smaller animals seem to have “less thought” going on inside
surprisingly uptight about their unpolished written prose being on
their craniums than we have inside ours. Yet clearly animals have
exhibit for others to see, even if it is going to vanish in mere seconds.
always done, and machines are now doing, things that seem to be
All of this | bring up not out of mere windbaggery, but
best described using Dennett's “intentional stance.” Donald Griffin, a
because it bears strongly on the Turing Test. Imagine the microscopic
conscious mammal, has written thoughtfully on these topics (see, for
insights into personality that are afforded by watching someone—
instance, his book The Question of Animal Awareness). John
human or otherwise—typing away in “talk” mode! You can watch
McCarthy has pointed out that even electric-blanket manufacturers
them dynamically making and unmaking various word choices, you
use such phrases as “it thinks it is too hot” to explain how their
can see interferences between one word and another causing typos,
products work. We live in an era when mental terms are being both
you can watch hesitations about whether or not to correct a typo, you
validly extended and invalidly abused, and we are going to need to
can see when they are pausing to work out a thought before typing it,
think hard about these matters, especially in face of the onslaught of
and on and on. If you are just a people watcher, you can merely
advertising hype and journalese. Various modifications of the Turing
observe informally. If you are a psychologist or fanatic, you can
Test idea will undoubtedly be suggested as computer mastery of
measure reaction times in thousandths of a second, and make large
human language increases, simply to serve as benchmarks for what
programs can and cannot do. This is a fine idea, but it does not
diminish the worth of the original Turing Test, whose primary
purpose was to convert a philosophical question into an operational
question, an aim that | believe it filled admirably.
Mathematical Hoots

In the world of formal mathematics, it is just as bad to be almost right as it is to be absolutely wrong. In a
sense, that's just what mathematics is. But that’s not good psychology.
Marvin Minsky, The Society of Mind

A mathematician is a machine for turning coffee into theorems.

Paul Erdos

The Al field was founded by mathematicians: John McCarthy, Alan Turing

(1912-1954), Norbert Wiener (1894-1964), students of Alonzo Church, Claude
Shannon, Marvin Minsky, and others. LISP, the primary language for academic
research in artificial intelligence, was adapted from a mathematical notation designed
by Stephen Kleene and Barkley Rosser, both students of Church.’
Mathematics has often been viewed as the ultimate formalization of our
thinking process, at least of the rational side of it. As | noted in the last chapter (and
as was noted in several of the contributed articles at the end of the last chapter), the
relationship of logic and the analytic process underlying mathematics to cognition has
been debated through the ages by philosophers, many of whom were also mathema-
ticians. The actual deployment of mathematical techniques to emulate at ieast certain
aspects of human thought was not feasible until the electronic computer became
available after World War II. However, the foundations of computation theory, along
with the set theory on which computation theory is based, were established long
before the potential of the electron to revolutionize applied mathematics was
realized.?
Mathematics has often been described as a branch of philosophy, the
branch most concerned with logic.? It has only been in this century that the fields of
mathematics and philosophy have split into largely distinct disciplines with few major
figures doing important work in both areas. Bertrand Russell, having been a pivotal
figure in the establishment of both modern set theory and logical positivism, was
perhaps the last.
+ Russell's Paradox

In the early part of this century Bertrand Russell, a young and as yet
relatively unknown mathematician and philosopher, became increasingly occupied
with a certain type of paradox and attempts to understand its implications. The
resolution of the paradox had important implications for the subsequent development
of the theory of computation. The following story illustrates Russell's class of para-
doxes:*
A judge is sentencing a man for a crime that he finds reprehensible and for
which he wishes to mete out the most severe sentence he can think of. So he tells
the convicted man not only that he is sentenced to die but also that because his
crime was so offensive, the sentence is to be carried out in a unique way. “The
sentence is to be carried out quickly,” the judge says. “It must be carried out no later
than next Saturday. Furthermore, | want the sentence to be carried out in such a way
that on the morning of your execution, you will not know for certain that you are
going to be executed on that day. When we come for you, it will be a surprise.”
When the judge finished describing his unusual sentence, the condemned
man seemed surprisingly pleased and replied, “Well, that’s great, judge, | am greatly
relieved.”
To this the judge said, “| don't understand, how can you be relieved? | have
condemned you to be executed, | have asked that the sentence be carried out soon,
but you will be unable to prepare yourself because on the morning that your sentence
is to be carried out, you will not know for certain that you will die that day.”
The convicted man said, “Well, your honor, in order for your sentence to be
carried out, | could not be executed on Saturday.”

Bertrand Russell, philosopher and

mathematician, observing his
ninetieth birthday. Modern computa-
tion theory traces its roots to his
theory of sets. (Supplied by AP/Wide
World Photos)

104
 “Why is that?” asked the judge.
“Because since the sentence must be carried out by Saturday, if we
actually get to Saturday, | will know for certain that | am to be executed on that day,
and thus it would not be a surprise.”
“| suppose you are right,” replied the judge. “You cannot be executed on
Saturday. | still do not see why you are relieved.”
“Well,” said the prisoner, “if we have definitely ruled out Saturday, then |
cannot be executed on Friday either.”
“Why is that?” asked the judge.
“We have agreed that | definitely cannot be executed on Saturday. There-
fore, Friday is the last day | can be executed. Thus, if Friday rolls around, | will
definitely know that | am to be executed on that day, and therefore it would not be a
surprise. So | cannot be executed on Friday.”
“| see,” said the judge.
“Thus, the last day | can be executed would be Thursday. But if Thursday
rolls around, | would know | had to be executed on that day, and thus it would not be
a surprise. So Thursday is out. By the same reasoning we can eliminate Wednesday,
Tuesday, Monday, and today.”
The judge scratched his head as the confident prisoner was led back to his
prison cell.
There is an epilogue to the story. On Thursday the prisoner was taken to be
executed. And he was very surprised. So the judge's orders were successfully carried
out.
If we analyze the paradox contained in the above story, we see that the
conditions that the judge has set up result in a conclusion that none of the days
meets, because, as the prisoner so adroitly points out, each one of them in turn
would not be a surprise. But the conclusion itself changes the situation, and now
surprise is possible again. This brings us back to the original situation in which the
prisoner could (in theory) demonstrate that each day in turn would be impossible, and
so on. The judge applies Alexander's solution to this Gordian knot.
A simpler example and the one that Russell actually struggled with is the
following question about sets: Consider set A, which is defined to contain all sets
that are not members of themselves. Does set A contain itself? As we consider this
famous problem, our first realization is that there are only two possible answers: yes
and no. We can therefore exhaustively consider all of the possible answers (this is
not the case for many problems in mathematics). Let us try “yes.” If the answer is
yes, then set A does contain itself. But if set A contains itself, then according to its
defining condition set A would not belong to set A, and thus it does not belong to
itself. Since the assumption that A contains itself led to a contradiction, it must have
been wrong. If the answer is “no,” then set A does not contain itself. But again
according to the defining condition, if set A does not belong to itself, then it would
belong to set A. As with the story about the prisoner, we have contradictory proposi-
tions that imply one another. The assumption of no yields yes, which yields no, and
so on.
This type of paradox may seem amusing, but to Russell it threatened the
very foundations of mathematics.> The definition of set A appears to be a perfectly

105
Mathematical Roots

tag Teco ag
Be Ae ee RSI
Co ees ae)

fe AU ee Ee Ld
CTT ag)

Ue ALY ERS LS
(so rule out Thursday)

RE eR LS
Co aes)

Tao eee

RLU Et

The Judge's dilemma (or is it the reasonable one, and the question of whether set A belongs to itself also appears
prisoner's?}, perfectly reasonable. Yet it cannot be answered. Without a resolution to this paradox
the basic theory of mathematics was in question.
To solve the problem, Russell invented a concept of a logical transforma-
tion as an operation that requires the equivalent of a quantum of time. Russell
designed a set of logical operations in which a particular problem would be ex-
pressed as a “program” of operations to follow. We then turn the program on and
let it run. Each logical inference or other transformation is implemented in turn, and
when the process is completed, we get our answer. If we apply this theoretical
machine to the problem of set A, the logical operations are “executed” in turn. Ata
certain point the answer will be yes, but the program keeps running, and at a later
 point the answer becomes no. The program runs in an infinite loop, constantly
alternating between yes and no. 4
Russell then provides narrow and broad definitions of a set. In the narrow
sense, a set has a definition that allows the construction of a program that can
determine whether a given entity is a member of the set in a finite amount of time
According to this definition, set A (whose program produces an infinite loop) is not a
true set, so the paradox is eliminated.”
In the broad sense, the program defining the logical rules of set member-
ship need not come to a halt in a finite amount of time, it just needs to come to an
answer in a finite amount of time; it is allowed to change that answer as the program
continues to run. According to this definition, set A is a proper set. The question of
whether set A belongs to itself will be yes at one point in “time” and no at another
point, and the program will alternate between the two. Thus, logical inferences are
not implemented instantly, but rather one at a time with an orderly change of state
between each. In our case, the answer is never yes and no at the same time. |In the
broad definition, set A is a particular type of set that is “unstable,” just as an elec-
tronic circuit can be unstable. Nonetheless, the contradiction is eliminated
Russell does not explicitly refer to time in his theory of types (of sets). He
provides procedures for allowable transformations on propositions that can be
considered meaningful within a logical system. This contrasts with the transforma-
tions generated by the logical system itself, which are used to determine the truth or
falsity of propositions. Thus, according to Russell, certain propositions are neither
true nor false and cannot be addressed by the axioms. In our discussion above, a
Proposition concerning an “unstable set” would not be meaningful. The theory is
interesting in that we have one set of transformations generated by the axioms of a
logical system determining truth or falsity and another set of transformations
generated by the metarules of Russell's theory of types determining meaningfulness.
Russell’s transformations are algorithmic in nature, and the issues raised are similar
to certain issues in computation theory that received attention after Turing devised
his Turing machine. Though Russell did not explicitly link the theory of types to

An unstable proposition: Russell's

paradox.

(seta
eT
itself)

107
computation theory (otherwise, we might be referring to a Russell Machine rather
than a Turing Machine as a primary model of computation), Russell's theory of types
clearly provided a foundation for Turing’s later work.
The lecture on logic delivered by the prisoner changed the situation. He has
shown quite logically why it is not possible for him to be executed following the
judge's instructions. The judge then realizes that the prisoner's belief that he cannot
be executed makes it possible once again to execute him. Before the prisoner can
formulate another lecture on logic (that is, before the “program” simulating this
situation can alternate again to “impossible to execute”), the judge quickly imple-
ments his sentence.

Principia Mathematica
Russell expanded his theory to lay a new foundation for logic and the theory of sets
in his first major work in mathematics, The Principles of Mathematics, published in |
1903. He subsequently felt that all of mathematics should be recast in terms of his
new theory of sets, since the concept of sets and their interactions is fundamental to
all other mathematical disciplines. With the help of his friend and former tutor Alfred
North Whitehead (1861-1947), he labored for nearly ten years to apply his new theory
of sets and logic to all realms of mathematics. Russell reported that the effort nearly
exhausted him, and even late in his life he felt that this had been the most intense
work of his extremely prolific career.* It was probably his most influential. As it was,
Whitehead and Russell did not manage to complete their reexamination. They
nonetheless published their work in three volumes in 1910, 1912, and 1913 under the
title Principia Mathematica. The work was truly revolutionary and provided a new
methodology for all mathematics that was to follow.
As significant as Principia was to mathematics in general, it was a pivotal
development in terms of the foundations of the theory of computation that would be
developed two decades later. Russell had created a theoretical model of a logic
machine, which we now recognize as similar to a computer, particularly in its execu-
tion of logical operations in cycles. Indeed, Turing’s subsequent theoretical model of
a computer, the Turing Machine, has its roots directly in Russell's theoretical logic
engine.'° Russell also created a concept of a logical programming language that is
remarkably similar in many respects to one of the most recent programming lan-
guages, PROLOG, developed originally in France and now the basis for the Japanese
Fifth Generation Computer project.'' Principia was also influential on efforts by Allen
Newell, Herbert Simon, and J. C. Shaw to develop theorem-proving machines in the
1950s. 2
Modern set theory, still based on Russell’s Principia, provides a foundation
for much of mathematics. It is interesting to note that modern set theory is in turn
based on Russell's theoretical model of computation. Viewing things in this way, we
could argue that mathematics is a branch of computation theory. What is particularly
impressive about Russell's achievement is that there were no computers even
contemplated at the time he developed his theory. Russell needed to invent a
theoretical model of a computer and programming to address a flaw in the foundation
of logic itself.
+ The Five Contributions of Turing

We must know, we shall know.

David Hilbert

Turing was perhaps the pivotal figure in the development of the computer and its
underlying theory. Building on the work of Bertrand Russell and Charles Babbage, he
created his own theoretical model of a computer and in the process established
modern computation theory.'? He was also instrumental in the development of the
first electronic computers, thus translating theory into reality. He developed special-
ized electronic computation engines to decode the German Enigma code, enabling
the British to withstand the Nazi air force. He was also a major champion of the
possibility of emulating human thought through computation.'* He wrote (with his
friend David Champernowne) the first chess-playing program and devised the only
widely accepted test of machine intelligence (discussed from a variety of perspec-
tives in several of the contributed articles at the end of chapter 2).'>
As a person, Turing was unconventional and extremely sensitive. He had a
wide range of unusual interests ranging from the violin to morphogenesis (the
differentiation of cells).'"° There were public reports of his homosexuality, which
greatly disturbed him, and he died at the age of 41, a suspected suicide

Alan Turing, designer of the world’s

first operational computer and an
early theorist of machine intelli-
gence. Cracking Hitler's code was
just one of his accomplishments.
(Courtesy of the Computer Museum,
Boston)

109
The Enigma code
By 1940 Hitler had the mainland of Europe in his grasp, and England was preparing
for an anticipated invasion. The British government organized its best mathemati-
cians and electrical engineers, including Alan Turing, with the mission of cracking the
German military code. It was recognized that with the German air force enjoying
superiority in the skies, failure to accomplish this mission was likely to doom the
nation. In order not to be distracted from their task, the group lived in the tranquil
pastures of Hertfordshire
The group was fortunate in having a working model of the German code
machine Enigma, captured by the Polish Secret Service. Working with several hints
gathered by British Intelligence, they were able to narrow the coding possibilities,
but only slightly. Under Turing’s leadership, their strategy was to build an electro-
magnetic computer, use telephone relays to do an exhaustive search of all possible
codes that the Enigma machine could produce, and apply these codes to intercepted
messages. The strategy was a challenging one because an (electromagnetic)
computer had never been built before. They named the machine Robinson, after a
popular cartoonist who drew “Rube Goldberg” machines.'?
The group’s own Rube
Goldberg succeeded brilliantly and provided the British with a transcription of nearly
all significant Nazi messages
The German military subsequently made a modification to Enigma, adding
two additional coding wheels, which greatly expanded the number of possible codes.
To meet this new challenge, Turing and his fellow cryptoanalysts set to building a
substantially faster machine called Colossus, built with two thousand electronic
vacuum tubes.'® Colossus and nine similar machines running in parallel did their job
again and provided uninterrupted decoding of vital military intelligence to the Allied
war effort.
Colossus was regarded by the Turing team as the world’s first electronic
digital computer, although unlike Harvard's relay-based Mark |, it was not program-
mable. Of course, it did not need to be: it had only one job to do.
Remarkably, the Germans relied on Enigma throughout the war. Refine-
ments were added, but the world’s first computers built by Alan Turing and his
associates were able to keep up with the increasing complexity. Use of this vital
information required supreme acts of discipline on the part of the British govern-
ment. Cities that were to be bombed by Nazi aircraft were not forewarned, lest
preparations arouse German suspicions that their code had been cracked. The
information provided by the Robinson and Colossus machines was used only with
the greatest discretion, but the cracking of Enigma was enough to enable the Royal
Air Force to win the Battle of Britain.

Hilbert’s twenty-third problem and the Turing machine

While many in England and elsewhere remain grateful to Turing for his contributions
to the war effort, his greatest legacy is considered to be the establishment of the
modern theory of computation. Yet his original goal was not the development of
such a theory but rather to address one of the problems set down by his predeces-
sor David Hilbert (1862-1943).

110
Enigma, the first target of machine
intelligence. (Courtesy of
the Computer Museum, Boston)
 The works of Hilbert, a German mathematician born in 1862, are still
widely regarded as highly influential on the research goals of today’s mathemati-
cians. He is credited with consolidating the accomplishments of nineteenth-century
mathematics with such works as The Foundations of Geometry, published in 1899.'%
Perhaps of even greater significance, he set the agenda for twentieth-century
mathematics as well with a list of the twenty-three most pressing unsolved prob-
lems that he presented at the 1900 International Mathematical Conference in Paris.
In his address he predicted that these problems would occupy the attention of the
next century of mathematicians. Hilbert appears to have been correct. The problems
have been solved slowly and each solution has been regarded as a major event.
Several that remain unsolved today are regarded by many mathematicians as the
most important unsolved problems in mathematics.
Hilbert’s twenty-third problem is whether or not an algorithm exists that
can determine the truth or falsity of any logical proposition in a system of logic that
is powerful enough to represent the natural numbers (numbers like 0, 1, 2, . . .). The
statement of this problem was perhaps the first time that the concept of an algo-
rithm was formally introduced into mathematics.
The question remained unanswered until 1937. In that year Alan Turing
presented a paper entitled “On Computable Numbers, with an Application to the
Entscheidungsproblem” (the Entscheidungsproblem is the decision or halting prob-
lem).2° The paper presented his concept of a Turing Machine, a theoretical model of
a computer, which continues to form the basis of modern computational theory.
A Turing machine consists of two primary (theoretical) units: a “tape drive”
and a “computation unit.” The tape drive has a tape of infinite length on which there
can be written (and subsequently read) any series of two symbols: 0 (zero) and 1
(one). The computation unit contains a program that consists of a sequence of
commands made up from the list of operations below. Each “command” consists of
two specified operations, one to be followed if the last symbol read by the machine
was a 0 and one if it had just read a 1. Below are the Turing machine operations:

Read tape
Move tape left
Move tape right
Write 0 on the tape
Write 1 on the tape
Jump to another command
Halt

The Turing machine has persisted as our primary theoretical model of

computation because of its combination of simplicity and power." Its simplicity
derives from its very short list of capabilities, listed above. As for its power, Turing
was able to show that this extremely simple machine can compute anything that any
machine can compute, no matter how complex. If a problem cannot be solved by a
Turing machine, then it cannot be solved by any machine (and according to the
Church-Turing thesis, not by a human being either).

112
 An unexpected discovery that Turing reports in his paper is the concept of
unsolvable problems, that is, problems that are well defined with unique answers
that can be shown to exist, but that we can also show can never be computed by a
Turing machine. The fact that there are problems that cannot be solved by this
particular theoretical machine may not seem particularly startling until one considers
the other conclusion of Turing’s paper, namely, that the Turing machine can model
any machine. A machine is regarded as any process that follows fixed laws. Accord-
ing to Turing, if we regard the human brain as subject to natural law, then Turing’s
unsolvable problems cannot be solved by either machine or human thought, which
leaves us with the perplexing situation of being able to define a problem, to prove
that a unique answer exists, and yet know that the answer can never be known.?°

The busy beaver

One of the most interesting of the unsolvable problems, the busy beaver problem,
was discovered by Tibor Rado.* It may be stated as follows. Each Turing machine
has a certain number of states that its internal program can be in. This corresponds
to the number of steps in its internal program. There are a number of different 4-
state Turing machines that are possible, a certain number of 5-state machines
possible, and so on. Given a positive integer n, we construct all the Turing machines
that have n states. The number of such machines will always be finite. Next, we
eliminate those n-state Turing machines that get into an infinite loop (that is, never
halt). Finally, we select the machine (one that halts) that writes the largest number of
1s on its tape. The number of 1s that this Turing machine writes is called the busy
beaver of n. Rado showed that there is no algorithm, that is, no Turing machine, that
can compute this function for all ns. The crux of the problem is sorting out those n-
state Turing machines that get into infinite loops. If we program a Turing machine to
generate and simulate all possible n-state Turing machines, this simulator itse/f goes
into an infinite loop when it attempts to simulate one of the n-state Turing Machines
that gets into an infinite loop.
The busy beaver function can be computed for some ns, and interestingly,
it is also an unsolvable problem to separate those ns for which we can determine
the busy beaver of n from those for which we cannot. Aside from its interest as an
example of an unsolvable problem, the busy beaver function is also interesting in
that it can be considered to be itself an intelligent function. More precisely stated, it
is a function that requires increasing intelligence to compute for increasing argu-
ments. As we increase n, the complexity of the processes needed to compute the
busy beaver of n increases.
With n = 6, we can deal with addition, and the busy beaver of 6 equals 35
In other words, addition.is the most complex operation that a Turing machine with
only 6 steps in its program is capable of performing. A 6-state Turing machine is not
capable, for example, of multiplication. At 7, the Busy Beaver does learn to multiply,
and the busy beaver of 7 equals 22,961. At 8 it can exponentiate, and the number of
1s that our eighth busy beaver writes on its tape is approximately 10“. By the time
we get to 10, we are dealing with a process more complex than exponentiation, and
to represent the busy beaver of 10 we need an exotic notation in which we have a
stack of exponents the height of which is determined by another stack of exponents

13

nn li
The Busy Beaver, an intelligent
function?

‘How to Compute
Busy Beaver of N
the height of which is determined by another stack of exponents, and so on. For the
twelfth busy beaver we need an even more exotic notation. It is likely that human
intelligence (in terms of the complexity of mathematical operations that can be
understood) is surpassed well before the busy beaver gets to 100.
Turing showed that there are as many unsolvable problems as solvable
ones, the number of each being the lowest order of infinity, the so-called countable
infinity (that is, the number of integers). Turing also showed that the problem of
determining the truth or falsity of any logical proposition in an arbitrary system of
logic powerful enough to represent the natural numbers was an unsolvable problem.
The answer, therefore, to Hilbert’s twenty-third problem posed 37 years earlier is no;
no algorithm exists that can determine the truth or falsity of any logical proposition in
a system of logic that is powerful enough to represent the natural numbers

The second and third answers to Hilbert’s question

Around the same time Alonzo Church, an American mathematician and philosopher,
published Church's theorem, which examined Hilbert’s question in the context of
arithmetic. Church independently discovered the same answer as Turing.2°
Also working independently, a young Czech mathematician, Kurt Godel
(1906-1978), sought to reexamine an issue that was not entirely settled by
Whitehead and Russell's Principia Mathematica.° Whitehead and Russell had sought
to determine axioms that could serve as the basis for all of mathematics, but they
were unable to prove conclusively that an axiomatic system that can generate the
natural numbers (theirs or any other) would not give rise to contradictions. It was
assumed that such a proof would be found sooner or later, but Gddel stunned the
mathematical world by proving that within such a system there inevitably exist
propositions that can be neither proved nor disproved. Some have interpreted
Gédel’s theorem to imply that such uncertain propositions are simply indeterminate,
neither true nor false. This misses the depth of Gddel’s insight, however. Such
propositions, according to Gédel, are not indeterminate; they are definitely either true
or false. It is just that we can never determine which. Another implication is that in
such axiomatic systems it is not certain that the axioms will not result in contradic-
tions. Gddel’s incompleteness theorem has been called the most important in all
mathematics, and its implications are still being debated.?” One of the implications is
that the answer to Hilbert’s twenty-third problem is again no.
Taken together, the work of Turing, Church, and Godel, all published in the
1930s, represented the first formal proofs that there are definite limits to what logic,
mathematics, and computation can do. These discoveries strongly contradict
Wittgenstein’s statement in the Tractatus that “if a question can be framed, it can
be answered” (6.5).

Hope and order versus distress and perplexity

As a final comment on Turing’s, Church’s and Gédel’s perplexing insights into the
nature of logic, it is interesting to note the stark contrast of the mood and attitude of
the intellectual and cultural life in Europe and the United States at the turn of the
century in comparison with that of several decades later.28 Music had shifted from
the romantic style of Brahms (1833-1897) and the early Mahler (1860-1911) to the

5
atonality of Schoenberg (1874-1951). Art and poetry had made the same switch
from romantic styles to the cubism and expressionism of Picasso (1881-1973) and
the minimalism of Pound (1885-1972), Eliot (1888-1965), and Williams (1883-1963).
It is not unusual for changes in attitude and world view to be reflected across the
arts, but it is interesting to note that the shift was reflected in science and mathe-
matics as well. In physics, mechanics had gone from a fully refined and consistent
Newtonian model to a paradoxical quantum model. The most puzzling aspect of
quantum mechanics and one of its essential features, the Heisenberg uncertainty
principle, is its conclusion that there are profound limits to what human beings can
know. In addition, the principle of duality, which had existed previously only in
metaphysical doctrine, was now firmly established in the apparently contradictory
wave-particle nature of light. Perhaps most disturbing, mathematics itself had gone
from its turn-of-the-century emphasis on comprehensive formalisms that covered all
of mathematics to a conclusion in the mid 1930s that logic had inherent and irremov-
able contradictions and that problems existed that could never be solved.

Turing’s test
Having established a theory of computation and having played a major role in the
implementation of that theory, Turing’s interest ran to speculation on the ultimate
power of this new technology. He was an enthusiast for the potential of machine
intelligence and believed that it was feasible, although he appeared to have a rea-
sonably realistic sense of how long such a development would take.
In a paper entitled, “Computing Machinery and Intelligence,” published in
the journal Mind in 1950, Turing describes a means for determining whether or not a
machine is intelligent: the Turing test. It should be noted that a computer “passing”
the Turing test is an indication that it is intelligent. The converse of this statement
does not necessarily hold. A machine (or organism) unable to pass the test does not
necessarily indicate a lack of intelligence. Some observers ascribe a high level of
intelligence to certain species of animals such as dolphins and whales, but these
animals are obviously in no position to pass the Turing test (they have no fingers, for
one thing).
To date no computer has come close to passing this test. The test
basically involves the ability of the computer to imitate human performance. Nar-
rower versions of the test have been proposed. For example, a computer chess
program was recently able to “pass” a narrow version of the Turing test in that
observers (again, observing through terminals) were unable to distinguish its playing
from that of a skilled human chess player. Another variation—one involving the ability
of a computer to compose stanzas of poetry—is provided in “A (Kind of) Turing
Test" in chapter 9. Computers are now beginning to imitate human performance
within certain well-defined domains. As Dan Dennett said in his article at the end of
chapter 2, such narrow formulations of the Turing test fall far short of the original. |
discuss the prospect of a computer passing the original Turing test in chapter 10.
Turing expected that a computer would pass his test by the end of the
century and remarked that by that time “the use of words and general educated
opinion will have altered so much that one will be able to speak of machines thinking
without expecting to be contradicted.” Turing’s prediction contrasted with other

116
statements around the same time that were much more optimistic in terms of time
frame. (In 1965 Herbert Simon predicted that by 1985 “machines will be capable of
doing any work that a man can do."?9) Turing was as optimistic as anyone with
regard to the power of cybernetic technology.*° Yet he appears not to have underes-
timated (at least not as much as some other observers) the difficulty of the problems
that remained to be solved.

The Church-Turing thesis

In addition to finding some profound limits to the powers of computation, Church
and Turing also advanced, independently, an assertion that has become known as
the Church-Turing thesis: if a problem that can be presented to a Turing machine is
not solvable by one, then it is also not solvable by human thought. Others have
restated this thesis to propose an essential equivalence between what a human can
think or know and what is computable. The Church-Turing thesis can be viewed as a
restatement in somewhat more precise terms of one of Wittgenstein’s primary
theses in the Tractatus.
| should point out that although the existence of Turing’s unsolvable
problems is a mathematical certainty, the Church-Turing thesis is not a mathematical
proposition at all. It is a conjecture that, in various disguises, is at the heart of some
of our most profound debates in the philosophy of mind.*'
The Church-Turing thesis has both a negative and a positive side. The
negative side is that problems that cannot be solved through any theoretical means
of computation also cannot be solved by human thought. Accepting this thesis
means that there are questions for which answers can be shown to exist but can
never be found (and to date no human has ever solved an unsolvable problem).
The positive side is that if humans can solve a problem or engage in some
intelligent activity, then machines can ultimately be constructed to perform in the
same way. This is a central thesis of the Al movement. Machines can be made to
perform intelligent functions; intelligence is not the exclusive province of human
thought. We can thus arrive at another possible definition of artificial intelligence: Al
represents attempts to provide practical demonstrations of the Church-Turing thesis.
In its strongest formulation, the Church-Turing thesis addresses issues of
determinism and free will. Free will, which we can consider to be purposeful activity
that is neither determined nor random, would appear to contradict the Church-Turing
thesis. Nonetheless, the truth of the thesis is ultimately a matter of personal belief,
and examples of intelligent behavior by machines are likely to influence one’s belief
on at least the positive side of the question.
The skepticism of Ada Lovelace (companion and assistant to Charles
Babbage, the nineteenth-century inventor of the Analytical Engine, a “mechanical”
computer) regarding the possibility of intelligent machines was no doubt related to
the limitations of computers with whirling gears and levers that he proposed to her.
Today it is possible to imagine building machines whose hardware rivals the com-
plexity and capacity of the human brain. As our algorithms grow more sophisticated
and machines at least appear to be more intelligent and more purposeful, discus-
sions of the Church-Turing thesis will become more practical than the highly theoreti-
cal debate in Church's and Turing’s time.

117
|i Albert Einstein, the power of simple
explanations. (Reprinted by
permission of the Hebrew University
of Jerusalem, Israel)
 The formula for Intelligence

Everything should be made as simple as possible, but no simpler.

Albert Einstein

Politics are for the moment. An equation is for eternity.

Albert Einstein

| cannot believe . .. that God plays dice with the world.

Albert Einstein

+ Unifying Formulas: The Goal of Science

There is a profound satisfaction in simple explanations that can truly account for
complicated phenomena. The search for unifying formulas (sometimes called “applied
mathematics”) has been a goal of science since its inception with the lonian Greeks
twenty-five centuries ago.'

General relativity
Perhaps the most famous unifying formula is E = mc? (energy equals mass times the
speed of light squared). The formula, part of the general theory of relativity put forth
by Albert Einstein in 1905, is simple enough: only five symbols, including the equal
sign.? Its power is manifest both in the range of phenomena it explains and in the
nuclear age it spawned. The equation predicts and explains the power of a nuclear
explosion. Mass when converted to energy is multiplied by an enormously large
number. The result is a dramatic example of the power of an idea

The laws of thermodynamics

Another famous example of unifying principles is the laws of thermodynamics.
Developed in the mid nineteenth century and concerned with the conditions that all
physical systems must obey, this was the first major refinement to the laws of
classical mechanics developed a century earlier. The first law is,
For any process involving no effects external to the system except displacement of a
mass between specified levels in a gravity field, the magnitude of that mass is fixed
by the end states of the system and is independent of the details of the process.

The second law is,

Among all the allowed states of a system with given values of energy, numbers
of particles, and constraints, one and only one is a stable equilibrium state.
Such a state can be reached from any other allowed state of the same energy,
numbers of particles, and constraints and leave no effects on the state of the
environment.

There are several important implications of the laws of thermodynamics,

one of which is the impossibility of perpetual-motion machines. Another implication,
relevant to the emergence of intelligence in the natural world, is a corollary of the
two laws called the principle of increasing entropy. “Entropy” is a measure of
randomness or lack of order in a system of many components, such as molecules.
Though named the law of increasing entropy, the law actually states that entropy in
a system can stay the same, but only under special conditions. Thus, entropy
generally increases. Another way to state the same law is, Left to their own devices,
systems will become increasingly disordered, a strangely poetic principle.*
The law of entropy seems to imply that the natural emergence of intelli-
gence is impossible. Intelligent behavior is very far from random, and any system
capable of intelligent behavior would have to be highly ordered. The chemistry of life
in general, and intelligent life in particular, is comprised of exceptionally elaborate
designs.° Out of the randomly swirling mass of particles in the universe, order has
somehow managed to emerge.
A possible explanation is that thermodynamics is not applicable to the
immensely long time frames that evolution operates in. Evolution has created
systems of enormous order but has taken an enormously long period of time to do
so. The order of life takes place amid great chaos and does not appreciably impact
the measure of entropy in the physically large system in which life has evolved. No
organism is a closed system; it is part of a larger system we call the environment,
which is high in entropy. Nevertheless, from the viewpoint of the laws of thermody-
namics, the emergence of intelligence must be considered a surprise.®

The theory of everything

Physicists have long had a talent for choosing entertaining anthropomorphic names
for otherwise abstruse concepts with such words as “truth,” “strangeness,” and
“charm” for complex mathematical properties of subatomic particles and “up,”
“down,” “charm,” “strange,” “top,” and “bottom” for the names of quarks, the
elementary building blocks of matter.”? The same comment might be applied to
terminology in the Al field, although one could argue that Al terms such as “ex-
perts,” “expert managers,” “demons,” “knowledge sources,” and “logical inference
processes” do have somewhat greater relevance to the phenomena being described
than strangeness and charm do to the phenomena of particle physics.
Most recently physicists have been making progress on the ambitious
quest to discover a unified set of equations to link the four basic forces of nature:
gravity, electromagnetism, the strong force, and the weak force. The same quest, for
what was then called the unified field theory, occupied Einstein's last two decades.®

120
 Recently two of the forces, the electromagnetic and weak forces, have been clearly
linked as two manifestations of an electroweak force. Most recently new theoretical
developments involving a concept of supersymmetry may be able to link the remain-
ing three forces into a unified structure referred to as the theory of everything.? In
this theory, ultimate reality is composed of vibrating strings. All the phenomena we
are familiar with, from subatomic particles to life forms, are resonances caused by
the interactions of these vibrations. This theory gives new meaning to the saying “All
the world is a song.”

The formula for intelligence

As these examples illustrate, the search for unifying formulas and principles is a
primary drive that fuels all of the sciences. The same sort of question has long
fascinated researchers in the field of intelligence, both natural and artificial: |s there a
formula that describes, explains or underlies intelligence? At first, the answer might
appear to be an obvious no. We have not been entirely successful in even defining
intelligence, much less in expressing it in a formula, a set of laws, or a set of
models. Intelligence would seem to be too complex for such reduction.'? On the
other hand, we should not quickly dismiss the notion that a lot of what we know
about intelligence can be expressed in such a manner. Some might argue that any
description of intelligence in a formula just reflects how little we know about it. |
would argue that the possibility of describing intelligent processes by simple equa-
tions or at least simple paradigms is a reflection of its elegance. If there is any
coherency to intelligence as a unified phenomenon, we should be able to say
something about its structure. | will describe several approaches to a description of
intelligence and its supporting mechanisms through unified rules and formulas in the
hope that they reveal insights into its nature.

+ The Sea of Logic and the Turing Machine

Any computer that we can describe can in theory be constructed from connecting a
suitable number (generally a very large number) of a very simple device, the Nor
logic gate. This device transforms truth values. The device takes two inputs, each of
which can be either true or false at any point in time. The Nor gate has one output,
which is true if neither input 1 nor input 2 is true. From this simple transformation,
we can build all other logic functions and even memory and thus provide all of the in-
formation-processing capabilities required for computation. A more detailed deriva-
tion is provided in the article accompanying this chapter
The Church-Turing thesis (discussed in chapter 3) postulates an essential
equivalence between machine and human intelligence, not necessarily between
current levels of capability and complexity, but in their underlying methods."' Since
we can construct any information-processing machine from Nor it can be considered
the basis of all machine intelligence. If one accepts the Church-Turing thesis, Nor
can be considered the basis of human intelligence as well.
While it is true that any algorithm, regardless of its complexity, can in
theory be implemented with nor gates, a collection of such devices will not perform
any useful function unless they are connected together in an appropriate way. Part of

121
The sea-of-logic machine, a formula
for intelligence? By connecting
together in just the right way
a large number of nor gates, a very
simple device, we can perform
any intelligent function.

most useful information-processing methods also require memory (which can also be
built from Nor gates), and we need each memory cell initialized to the right value.
The connection of the nor gates and the initial contents of the memory cells can
both be described by lists of numbers, or the symbolic equivalent of numbers. They
can, therefore, be considered as forms of software. Seen in this light, the nor gate
provides us with a unifying formula for the hardware of intelligence, but not for the
software.
One might argue that the connection of the nor gates should be consid-
ered hardware, and the initial contents of the memory cells should be considered
software. | regard them both as software, because they are both equivalent to lists
of numbers, and thus both can be represented using softwarelike languages. The
evolution of electronic design is indeed proving the essential equivalence between
so-called hardware design and software development. Increasingly, hardware design-
ers are working at computer terminals, storing their work products on floppy disks
(just like software engineers), and designing their systems using formula-based
languages very similar to software languages. Take, for example, the emerging
technology of silicon compilers. These systems allow a hardware engineer to de-
scribe an actual chip in terms of the logical and numeric transformations it is de-
signed to perform.'? These transformations are expressed in a language very similar
to a high-level software programming language. The designer interacts with a
simulator that allows the user to test out the “program” before the actual chip is
constructed. Once this process is complete, the silicon-compiler software converts
this program into a series of instructions for a silicon fabrication machine to create an
actual VLSI (very large scale integrated) circuit chip. The program created by the chip
designer is controlling the connection of the logic gates on the chip. We might
regard the design of a chip to be the ultimate in hardware design, yet the develop-
ment process, the languages used, and the work product are very similar to those
for software.'?

Comparison of the Turing machine and the sea of logic

Another simple yet powerful model that can potentially be used to describe any
form of computation is the Turing machine. As described in chapter 3, the Turing

122
machine is a theoretical model of an information-processing machine. The Turing
machine, which like the nor gate is relatively simple, continues to be used by
mathematicians as a satisfactory model of what computers can and cannot do. The
Turing machine has proven to be a more powerful mathematical model of computa-
tion than nor-based logic because of the body of Turing-machine theory that has
been developed.
Models of the Turing machine have been built, but they can never be true
Turing machines, because it is not possible to build an infinitely long tape. On its
face, it would appear that the “sea of logic” machine and the Turing machine are not
equivalent, again because of the infinitely long tape of the Turing machine. It can be
shown, however, that a Turing machine can simulate any sea-of-logic machine.
Furthermore, it can be shown that any specific problem that can be solved by a
Turing machine can be solved on a sea-of-logic machine as well. The heart of the
proof lies in the fact that in order for a Turing machine to solve a problem, it must do
so in a finite amount of time. In a finite amount of time, it can only use a finite
amount of tape. Thus, it can be reduced to a machine with a finite number of states
and thus simulated by the sea of logic. A Turing machine may outperform any
particular sea-of-logic machine, however, because it may be able to solve problems
that will use an amount of tape that outstrips the memory capacity of any particular
sea-of-logic machine.
Turing directly linked his theoretical machine to the more controversial
Church-Turing thesis. If one accepts the Church-Turing thesis, then the Turing
machine, which is a very simple machine in structure, can solve any cognitive
problem. To be more precise, the Turing machine can execute any plan to solve a
cognitive problem. Its success in doing so will be a function of the validity of the
plan.'* We thus come to the same conclusion that we did with the sea of logic. The
Turing machine provides us with a simple and elegant model for the hardware of
cybernetic (machine-based) cognition, but not the software. To solve a practical
problem, the Turing machine needs a program, and each different program consti-
tutes a different Turing machine.

The universal Turing machine

It turns out that a universal Turing machine can be shown to have the capability of
simulating every other Turing machine. The universal Turing machine has a program
that can read a description on its tape of any Turing machine (even the universal
Turing machine) and then simulate that machine. Thus, the input tape to the univer-
sal Turing machine contains not only the usual problem input but also a coded de-
scription of the Turing machine it is going to simulate. The universal Turing machine
is thereby a single (theoretical) machine that can solve any problem.'®
All this accomplishes, however, is to move the software from the built-in
program of a Turing machine to a description of such a program on a tape. If we
consider the machine program on the tape to be the software, we once again
conclude that we have a powerful and elegant model for the hardware underlying
intelligent processes, but not the software, at least not yet.

123
+ The Recursive Formula and Three Levels of Intelligence

To follow the way, one must find the way and follow it.
Zen koan |

The whole point of this sentence is make clear what the whole point of this sentence is.
Douglas 8. Hofstadter, Metamagical Themas

“Would you tell me please which way | ought to go from here?” asked Alice.
“That depends a good deal on where you want to get to,” said the Cat.
“I don’t much care where ... ,” said Alice.
“Then it doesn't matter which way you go,” said the Cat.
. so long as | get somewhere,” Alice added as an explanation.
“Oh, you're sure to do that,” said the Cat, “if you only walk long enough.”
Lewis Carroll, Alice in Wonderland

A professor has just finished lecturing at some august university about the origin and structure of the
universe, and an old woman in tennis shoes walks up to the lectern. “Excuse me, sir, but you've got it all
wrong,” she says. “The truth is that the universe is sitting on the back of a huge turtle.” The professor
decides to humor her. “Oh, really?” he asks. “Well, tell me, what is the turtle standing on?” The lady has
a ready reply: “Oh, it's standing on another turtle.” The professor asks, “And what is that turtle standing
on?” Without hesitation, she says, “Another turtle.” The professor, still game, repeats his question. A look
of impatience comes across the woman's face. She holds up her hand, stopping him in mid-sentence.
“Save your breath, sonny,” she says. “It's turtles all the way down.”
Rolf Landauer, as quoted in Did the Universe Just Happen? by Robert Wright

The Turing Machine provides us with a model of computation that has

enabled theoreticians to examine many facets of computers and their capabilities.
The sea of logic is an even simpler mechanism that enables us to build very complex
machines and methods from a simple nor gate. We can consider these to be simple
formulas that provide insight into computational hardware, but as noted, they do not
provide insight into the software, the methods or algorithms, of intelligence.
One of the first attempts to codify intelligence in an algorithm was the
recursive method, implemented in both the General Problem Solver and Samuel's
Checker Playing Program.'® Let us examine the recursive formula in the context of
the game of chess. We will be returning to the issue of chess several times in this
volume because of its importance to Al research. Raj Reddy cites studies of chess as
playing the same role in Al that studies of E. Coli play in biology: an ideal laboratory
for studying fundamental questions. | will then expand the method to other problems
and examine its capabilities and limitations

The game of chess

Chess is a game of two opponents in which the players take turns adjusting the
Positions of their pieces on a playing board according to prescribed rules. Of course,
many other games can be described in the same way. Certain activities in real life,
such as war and business, have similar characteristics. Indeed, chess was originally
formulated as a model of war.
To win or to provide the highest probability of winning, one selects the
best possible move every time it is one’s turn to move. The question, then, is what

124
the best move is. The recursive method provides the following rule, which, if you
follow it carefully, will enable you to play an exceptionally good game of chess: Every
time it is your move, select the best move on the assumption your opponent will do
the same. At this point, the casual observer will complain that the rule has no
content, that it provides no more insight than an impenetrable Zen koan. It appears
to simply restate the problem. As | believe will become clear, however, this rule is all
that is needed to play an excellent game of chess. If this is so, then we might
conclude either that the recursive formula is a powerful and deceptively simple
formula for the algorithm (or software) of at least some forms of intelligence, or
alternatively, that chess is not an intelligent game, that the game has no content.
Before delving further into the implications of the recursive formula, let us
examine how it works. We fashion a program called Move. When it is called, its job
is to pick the best move. It is a recursive program in that it is capable of calling itself
Recursive programs are perfectly feasible in modern programming languages. A
program can call itself, return answers to itself, and continue as if it had called any
other program (or subroutine).
Recursion is a powerful method used extensively in artificial intelligence. It
is one of the more valued features of the primary Al programming language LISP.'”7 A
brief discussion of recursion will be worthwhile at this point, and | will illustrate the
concept with an example. Consider the definition of the factorial function expressed
recursively. Let n! be the factorial of n. The definition of the factorial function is then
1! = 1, and n! = n x(n —1)! As we can see, this definition of factorial uses the
concept of factorial in its own definition. This self-referencing is called recursion. Yet
the definition is not infinitely circular in that we can determine the value of the
factorial of any number from the definition just by repetitively expanding references
to factorial
For example, let us compute 4!. According to the definition, 4! = 4 x 3!.
Using the definition again to expand 3!, we get 4! = 4 x 3 x 2! In turn, we expand
2! to get 4! = 4x 3 x 2 x 1!. The definition gives the 1! directly without reference
to itself. We are thus able to fully expand the expression to eliminate all reference to
the function in the right-hand part of the expression: 4=4x3x2x1
The power of recursion is that complex procedures or concepts can be
expressed in a simple way. A recursive definition differs from a circular definition in
that it has an escape from infinite expansion of the recursion. The escape is found in
the “terminal” or nonrecursive portion of the definition. In the recursive definition of
factorial, the terminal portion of the definition is the factorial of 1.
Recursion, a program calling itself, is accomplished by the program saving
information on its current state (including who called it and exactly where it is to
return to when finished) on a push down stack. The stack is a repository of informa-
tion that is organized as a last-in, first-out (LIFO) list so that information is retrieved in
the opposite order in which it is put in, rather like a stack of dishes. Each time a
program calls another program (whether it is calling itself or another subroutine),
information about the state of the program and the place to return to is placed
(“pushed”) onto the stack. Every time a program is completed and wants to return
to the program that called it (which might be itself), information is retrieved
(“popped”) from the stack. This information restores the state of the program that

125
 Factorial of 4
eee

Factorial, an example of recursion.

called it and indicates where in the calling program to return to, which might be to
part of itself. This simple LIFO mechanism assures that all returns are to the right
place.

Practical programs cannot call themselves indefinitely. Thus, a practical

LIFO stack is of finite length. Another reason that a program cannot continue to call
itself without limit is that we would like to solve problems ina finite length of time.
Thus, a useful recursive formula or definition requires an escape condition, a set of
circumstances that permits us to escape from infinite recursion
Let us return to our simple recursive rule for playing chess. It starts out by
saying that we should select the best move. To do this, we obviously must consider
each move, and thus we need to be able to generate a list of all possible moves. We
do this by simply applying the rules of chess. While more complicated than some
other games, such as checkers, the rules of chess are certainly straightforward and
readily programmable."® There is nothing mysterious about generating the possible
moves from any board position. We thus provide a mechanism programmed with
the rules of chess to generate our options. To play a different game, we can simply
replace this module with another one programmed with the rules of that particular
game
We now have a list of possible moves. We examine each in turn and ask
the question, Does this move enable me to win or lose? If the answer is lose, we do
not select that move. If the answer is win, we take that move. If more than one
move enables us to win, it does not matter which one we take
The problem now reduces to answering the question, Does this move
enable me to win or lose? At this point we note that our winning or losing is affected
by what our opponent might do. A reasonable assumption is that our opponent will
also choose his best move (and if the opponent fails to do this, that will not be a
problem, but an opportunity). We need to anticipate what that move might be, so we
our own ability to select the best move to determine what our opponent is likely

126
to do. We essentially put ourselves in our opponent's place and pick the best move
for our opponent. In this we are following the part of the recursive rule that states,
“Select the best move on the assumption that your opponent will do the same."
Our program is now structured as follows. We generate a list of all
possible moves allowed by the rules. We examine each possible move in turn. For
each move, we generate a hypothetical board representing what the placement of
the pieces would be if we were in fact to make this move. We now put ourselves in
our opponent's place and try to determine what his best move would be. How are
we to do this? It turns out that we have a program that is designed to do exactly
that. It is called Move. Move is, of course, the program we are already in, so this is
where the recursion comes in. Move calls itself to determine what our opponent will
do. When called to determine the best move for our opponent, Move begins to
determine all of the moves that our opponent could make at this point. For each one,
it wants to know how its opponent (which is us) would respond and thus again calls
Move for each possible move of our opponent to determine what our response to
that move would (or should) be.
The program thus keeps calling itself, continuing to expand possible moves
and countermoves in an ever expanding tree of possibilities. This process is usually
called a minimax search, because we are alternately attempting to minimize our
opponent's ability to win and to maximize our own.'? The figure illustrates this
process for the simpler game of tic-tac-toe.
The next question is, Where does this all end? Let us start with an attempt
to play perfect chess. We continue to expand the tree of possible moves and
countermoves until each branch results in an end of game. Each end of game
provides the answer: win, tie, or lose. Thus, at the furthest point of expansion of
moves and countermoves, some moves finally finish the game. If a move results in a
win, then we select that move. If there are no win moves, then we settle for a tie
move. If there are no win or tie moves, we keep playing anyway in the hope that our
opponent will make a mistake (unless we know we are playing a perfect opponent,
in which case we may as well give up). These final moves are the terminal nodes of
our expansion of moves. Here, instead of continuing to call Move, we can now begin
returning from Move calls. As we begin to return from all of the nested Move calls,
we have determined the best move at each point (including the best move for our
opponent), and so we can finally select the correct move for the current actua/ board
situation.
The above procedure is guaranteed to play a perfect game of chess. This is
because chess is a finite game. Interestingly, it is finite only because of the tie rule,
which states that repetition of a move results in a tie. If it were not for the tie rule,
then chess would be an infinite game (the tree of possible moves and countermoves
could expand forever) and we could not be sure of determining the best move within
a finite amount of time. Thus, this very simple recursive formula plays not just an
excellent, but a perfect, game of chess.2° The most complicated part of actually
implementing the recursive formula is generating the allowable moves at each point.
Doing so, however, requires only a straightforward codification of the rules. Playing a
perfect game of chess is thus no more complicated than understanding the rules.

127
for In

x0 ca
x PS

ed yaad

aro
Cog)

Te a am co
ao oa)

A recursive expansion of moves

in tic-tac-toe.

128
ATinkertoy mechanical computer
that plays tic-tac-toe. (Courtesy of
the Computer Museum, Boston)

129
 When shopping for services like car repair, the smart shopper is well
advised to ask, How long will it take? The same question is quite appropriate in
applying the recursive formula. Unfortunately, with respect to chess, the answer is
something like 40 billion years.2’ And that is just to make the first move!
We can estimate the length of time by assuming that there are, on
average, about eight possible moves for each board situation. If a typical game lasts
about 30 moves, we need to consider 8°° possible move sequences to fully expand
the tree of all move-countermove possibilities. If we assume that we can analyze
one billion of such sequences in a second (which is at least 1,000 times faster than
is in fact possible on today’s fastest chess computers—in 1989, two leading chess
machines, HiTech and Deep Thought, could analyze 175,000 and 1,000,000 board
positions per second, respectively), then we require 10'® seconds, or about 40 billion
years.” If the cyclic theory of the universe is correct (the theory that the universe
will eventually end its current expansion and then contract, resulting ultimately in
another big bang), then the computation would not be complete before our com-
puter is blown up in the next explosion of the universe.
This brings up another aspect of computation theory: the amount of
computation required to achieve a result. One branch of computation theory exam-
ines the issue of whether a computer (generally using the Turing machine as a
model of a computer) can solve a given problem within a finite amount of time.
Problems that cannot be solved are called unsolvable problems. More recently a new
branch of computation theory called complexity theory has been developed to deal
with the issue of how long various solvable problems will take to compute.?? While
our very simple recursive formula plays perfect chess, which may indeed be consid-
ered intelligent behavior, it takes an excessively long period of time to do so. Accom-
plishing tasks quickly or at least in a timely manner is regarded as an essential part
of intelligence. As discussed earlier, despite controversies with regard to cultural
bias, the timed aspect of intelligence tests has long been accepted, which reflects
the importance of timeliness as an attribute of intelligence.
Thus far the above discussion has not yet demonstrated intelligence in the
recursive formula. Playing perfect chess might be considered intelligent behavior, but
not at 40 billion years per move. Before we throw out the recursive formula, how-
ever, let us attempt to modify it to take into account our human patience (and
mortality). Clearly, we need to put limits on how deep we allow the recursion to take
place. How large we allow the move-countermove tree to grow should depend on
how much computation we have available. In this way we can use the recursive
formula on any computer, from an inexpensive home computer to the largest
supercomputer.

The next question is how do we evaluate the terminal “leaves,” the fully
expanded board positions of our tree of possibilities? When we considered fully
expanding each move sequence to the end of the game, evaluation was simple:
winning is better than tying, and tying is better than losing. Evaluating a board
position in the middle of a game is not as straightforward and, not surprisingly, is the
source of controversy.**
There are two schools of thought here: the simple-minded school and the
complex-minded school. The complex-minded school argues that we need sophisti-

130
cated (complicated) sets of rules and procedures to evaluate the “quality” of the
board at each terminal leaf position, that we need to be “smart” in making these
evaluations and require algorithms that can make subtle judgements.”°
The simple-minded school argues that whatever computational resources
we happen to have available are best put into pursuing greater depths of search, that
is, looking at longer sequences of possible moves and countermoves and using a
computationally simple procedure to evaluate final board positions. One such proce-
dure is simply to count up the points of the two sides, using the recognized value of
each piece (ten for the queen, five for each rook, etc.). The argument is that any
sophisticated evaluation of node positions is computationally very expensive and is
not likely to be more productive or more reliable than putting the computational
resource into greater depth of search.
For chess in particular | tend to agree with the simple-minded school and
feel that the key to building a computer that can win the world chess championship
lies in massive computational power to be achieved through massive parallel proc-
essing (which, after all, the human brain uses). All chess playing programs use the
recursive formula, some with simple board evaluation procedures (often just adding
up piece values) and others with more sophisticated procedures. The primary factor
that has determined the power of a particular program appears to be the amount of
computation that can be brought to bear. The programs running on the fastest
supercomputers or parallel processing networks invariably are the best machine
players, regardless of the board-evaluation algorithm. More sophisticated terminal-
leaf evaluation procedures do not appear to hurt performance, but neither do they
appear to help. They do not appear, therefore, to be worth the trouble.7°
HiTech, developed at Carnegie Mellon University and the leading chess
machine of 1988, uses massive parallel processing.”’ It uses an array of specially
designed VLSI chips capable of generating and evaluating board positions. Though its
procedure for evaluating a board position is more complicated then just adding up
the value of each piece, | would still describe its evaluation procedure as simple
(although the developers of HiTech might disagree with me).28
If we agree with the simple-minded school, then the node-evaluation
procedure is simple. Since the recursive formula itself is the ultimate in simplicity,
we end up with an algorithm that does indeed reduce to a simple formula and is
capable of playing a very good game of chess with reasonable (and in tournament
play, acceptable) response time.
The next question is, How good a game? Predictions made in 1958 that a
computer would be world chess champion by 1968 were considerably over optimis-
tic.2? Nonetheless, computers have demonstrated steady progress and, as men-
tioned earlier, two computer programs, Belle and HiTech, now play on the National
Senior Master Level, equaled or surpassed by only 768 human players.*° In 1988
HiTech scored 5 wins, 1 loss in the National Open Chess Championships, which
ranked it 150th among registered American players. Also in 1988 Carnegie Mellon's
Deep Thought defeated the highest rated player ever in the U.S. Open Chess
Tournament. Though not yet that of a world champion, this is certainly an impressive
and eminently intelligent level of play.°’

131
 The number of human chess players that can still defeat the best ma-
chines is growing smaller each year. It is clear that to make /inear progress in the
number of moves and countermoves that can be analyzed, we need to make
exponential gains in computational power. On the other hand, we are making
exponential gains in computational power with the /inear passing of time. By most
estimates, computational power (per unit of cost) is doubling every 18 to 24 months.
The phenomenon of parallel processing (doing more than one computation at a time)
is in some ways accelerating this trend, at least with regard to how much computa-
tion we can throw at the problem of playing chess. The Connection Machine, for
example, has 65,536 parallel computers, although each of these processors handles
only one bit at a time.*? Machines are being built that combine thousands of power-
ful 32-bit computers, and machines that combine millions of 32-bit microprocessors
are being seriously contemplated.
It is clear that it is only a matter of time before the world chess champion
is a machine. To make a prediction as to when that might be, we need to know how
many more move-countermove expansions are needed. Some argue that the differ-
ence between national senior master level, at which computers can now play, and
world-champion level of play is the ability to see only two or three additional moves
ahead. It certainly makes sense that the ability to see an additional 2 or 3 moves
ahead could account for the difference between the excellent playing of a national
senior master and the somewhat better game of the champion. Others argue that
the difference is more like 10 to 12 moves.*
Some argue that the difference in play between masters at different levels
is not a matter of seeing additional moves ahead, that the master is not really analyz-
ing moves and countermoves at all. Instead, he supposedly has some sort of higher-
level intuitive feel for the strength of different moves and board positions. In my
opinion, master players are not consciously aware of all of the move-countermove
sequences, but this computation has in fact taken place. Most of this analysis has
been “precomputed” by storing and retrieving board positions from previous experi-
ence and study. When we train our minds to perform a certain type of physical or
mental activity, the mental processes underlying this skill can often be accomplished
without our conscious awareness, but that does not mean that the mental calcula-
tions never took place.**
To get back to our estimate, if each additional move ahead expands the
amount of computation by a factor of about 8, and if it takes 18 months for the
computer industry to double computational power, then our cybernetic chess players
will be able to see one additional move ahead every 4.5 years. According to this
estimate, we are somewhere between 9 and 54 years from having sufficient compu-
tational power to build a computer chess champion. This achievement will, in my
opinion, be brought a lot closer than 54 years through the impact of new technolo-
gies, including new parallel processing architectures, exotic new computer hardware
techniques that use light instead of electricity, and other developments.*® Another
analysis is provided by Raj Reddy. He estimates that grand-master level play (just
below world-championship level play and probably good enough to beat the world
champion once in a while) requires being able reliably to look ahead about 12 to 14
moves. Though the average branching factor (the number of possible moves for

132
each board position) is about 35, advanced search techniques allow us to effectively
reduce this number to 5. Thus, we need to consider between 5’? and 5'* board
positions, or about 3 billion positions per move. With an average of 3 minutes of
thinking time per move, this requires a rate of about 16 million board positions per
second, only 16 times faster than Deep Thought’s 1989 rate. Thus, we should be
able to achieve very close to championship play by the early 1990s.
Yet another analysis is provided by one of HiTech’s designers, Hans
Berliner, who notes that the average increase in the ratings of the best computer
players has been about 45 points per year (in contrast, the average increase in the
ratings of the best human players is close to 0). This analysis projects 1998 as the
year of the computer world champion. My own guess is that we will see a world
champion by the year 2000, but as mentioned above, earlier predictions of this type
have not fared well.
Let us return for a moment to the concept of the perfect chess game
From a mathematical point of view, such a game exists. In other words, we can
define the perfect chess game as a game played by two opponents each following
the recursive formula and fully expanding all move sequences. From a formal point
of view, we cannot say that determining the sequence of moves in this perfect
game is an unsolvable problem, because it can certainly be determined in a finite
amount of time. On the other hand, we can show that it is impossible to determine
in any practical length of time. We thus have the interesting situation of being able
to easily define a question, to show that the answer exists and is unique, and also to
demonstrate conclusively that there is no reasonable way to find it.
Other questions come to mind. For example, does the player who moves
first in this hypothetical perfect game win or tie (it is unlikely that the player making
the first move would lose)? Again, there is an answer to this question, but it appears
that no human (or machine) will ever be able to find it.

An idiot-savant game?
When | stated that | was of the simple-minded school, | was careful to state that this
was my opinion specifically with regard to the game of chess. If it is shown that a
simple leaf evaluation procedure can defeat any human being at chess (as | believe
will happen), then we can conclude either that the recursive formula is indeed a
simple formula for at least certain types of intelligent behavior, or that chess is not
an intelligent game. We could conclude that computers are better than humans at
chess for the same reason that computers are better at creating spreadsheets. In
other words, we might conclude that the secret to chess is simply being able to
keep in one’s mind countless move and countermove sequences and the resulting
board patterns. This is the type of information that computers can easily and
accurately manipulate, whereas the heart of human intelligence lies elsewhere. As
long as there are at least a few humans that can defeat the best machines, some
observers still feel comfortable in citing chess as an example of high intellectual (and
creative!) activity and imagining that there is some unknown (and possibly unknow-
able) deep intellectual process underlying it. A computer world chess champion
could radically alter that perception. It is clear that there will be some significant
change in perception with regard to chess, computers, ourselves, or all three

133
Other games
The recursive formula can be applied to any board game that involves moves and
countermoves within prescribed rules. A reasanable question is, How well does it
work in practice for games other than chess? Also, what about the simple-minded
versus complex-minded controversy?
At one extreme of the spectrum of game complexity there is tic-tac-toe,
for which we can quite easily expand all possibilities to end of game and play a
perfect game. Many humans, at least those who have spent any time thinking about
it, can play a perfect game as well. |
Checkers, on the other hand, is sufficiently complex that playing a perfect
game is not feasible, although the perfect checkers game can probably be computed
before the next big bang occurs. Checkers-playing machines have already come
closer than chess machines to defeating the world human champion.**A computer
has already taken the world championship in backgammon: Hans Berliner’s program
defeated the human backgammon champion in 1980.%” There are two reasons why
computers have had a somewhat easier time with checkers and backgammon than
with chess. First, generating the possible moves at each point requires substantially
less computation. There are only two different types of pieces in checkers and one
in backgammon, and the rules are relatively simple. The problem is not that program-
mers have a hard time programming the rules of chess—the rules are not that
complicated. The complexity of the rules is an issue with regard to how many moves
ahead it is possible to look. With their relatively simple rules, the same amount of
computational power enables the checkers or backgammon program to look ahead a
greater number of moves than the chess program. The second possible reason is
that there is a lot more interest in chess, and thus the best human players are
probably better at chess than at checkers or backgammon. Nonetheless, backgam-
mon is still considered a “deep” game, and the fact that a computer can defeat any
human player should be considered an event of some note.**
At the other extreme, in terms of computer success, is the Japanese
game of go, which has a place in the cultural life of Japan comparable to chess in
Russia, Europe, and the United States. Though computers can play a reasonable
amateur game of go, they have not achieved master level, and the lack of progress
is not for lack of effort by program designers. The problem is not with the complex-
ity of go, because the rules are in fact simpler than those for chess. In my view, the
primary reason that the recursive formula is not as effective is that the number of
possible moves at each point is relatively large and the length of move sequences in
a game, or even a portion of a game in which meaningful progress is made, is very
long. Thus, even massive computation is quickly exhausted and does not look far
enough ahead to examine a meaningful strategy.°°
It appears that for the game of go we do need more sophisticated evalu-
ation strategies to examine different options. It may also be that we need to con-
sider alternatives on a different level than individual moves, that we need a way of
effectively modeling strategies on a higher level than one move at a time. Thus, for
go, unlike chess, | would throw my ballot in with the complex-minded school. It
appears that the Japanese have a board game that is “deeper” and more consistent
with the particular strengths (at least at present) of human intelligence.*°

134
The Japanese game of go. (Photo by
David E. Dempster of Offshoot)

Other intelligent tasks

The Recursive Formula can be applied to nongame situations, such as proving
mathematical theorems, solving combinatorial problems, solving various types of
puzzles ranging from mazes to word problems (like the problem of cannibals and
missionaries).*' In any situation in which the answer can be expressed as a se-
quence of moves, steps, or transformations governed by a set of unambiguous
rules, the recursive formula is a good candidate for solving the problem
In 1957 Allen Newell, J. C. Shaw, and Herbert Simon devised a program
entitled the General Problem Solver (GPS).4? The program’s name reflected the
enthusiasm, romanticism, and immodesty of the early Al field, an optimistic orienta-
tion that eventually earned the field considerable criticism.*? GPS could solve prob-
lems that involved devising a sequence of steps to achieve a goal. For example,
proving a theorem in mathematics requires devising a sequence of steps allowed by
the rules of logic, in which each step employs an axiom or a previously proven
theorem or lemma. One approach that works in theory but not in practice is simply
to generate every possible sequence until one is found that accomplishes the task.
Clearly, the number of possibilities expands too quickly, and a method of using
limited computational resources to achieve a goal within a reasonable amount of
time is needed. To optimize the use of scarce computational power, GPS uses a
technique called means-ends analysis. At each step it measures the “distance” to
the goal and generates a subgoal to attempt to minimize this distance. It then calls
itself, another example of recursion, to solve the presumably simpler problem
represented by the subgoal. In a similar way, each subgoal can be expanded into
smaller subgoals.
GPS achieved notable success in solving logic puzzles, proving theorems,
solving problems in integral calculus and plane geometry, and similar mathematical
or combinatorial problems. Its success fueled some of the Al field's early opti-
mism.*° Further progress came more slowly, however, as some of the limitations of
the technique became clear. One of the keys to successful use of the GPS paradigm
is a measure of distance that is appropriate to the problem. For example, how should
we measure the distance between two mathematical expressions? Simply counting

135
 the number of symbols that the two expressions do not have in common is clearly
not satisfactory: two expressions may differ by only one symbol, and yet we may
never be able to derive one from the other. Similarly, in solving a maze problem, we
may be only a single cell away from the goal, yet we are not necessarily close to
achieving a viable path. A fully satisfactory method of defining distance in such
domains is still not clear.*°

So close and yet so far: the distance

between mathematical expressions.

17X? + 34XY-14XX-3X? — 40Y'

17X? + 34XY°-5 17X? + 34XY?-5

The initial rapid success of GPS was followed by a period of considerably

slower progress in which each aspect of the GPS paradigm was examined in detail.
What is the distance measure? How are subgoals generated and evaluated? Are
there alternatives to the means-ends algorithm? A relatively simple approach that
seemed at first to capture intelligence, at least within a broad variety of problem
domains, was shown to have limitations. GPS has certainly not been the last word
on this type of search problem, and more capable methods have since been de-
vised.*”

Three levels of intelligence

In practical applications of the recursive formula to a variety of tasks, intelligent
problems appear to divide into three levels. At one extreme are problems that
require little enough computation that they can be completely analyzed. Examples
are tic-tac-toe and such word problems as cannibals and missionaries. In the middle
are problems for which we cannot afford full analysis but that appear to be success-
fully dealt with using a simple node-evaluation procedure. By “successfully dealt
with” | do not mean that it is possible to provide perfect solutions. Rather, | mean
that it is possible to match the performance of most (and in some cases all) humans.
Examples of this class of problems are chess, checkers, backgammon, and Euclidian
geometry. It is true that some of the chess programs use complex node-evaluation
procedures, but these programs do not appear to provide superior performance to
programs that do use simple node-evaluation procedures. The point is that playing a
good game of chess (and in my opinion ultimately a championship game of chess)
does not require complicated node-evaluation procedures." Finally, there is a class of
problems that do require nontrivial procedures at the terminal leaf position. For many

136
 Are these
two cells
“close” or
“far?

So close and yet so far: the distance

between cells in a maze.

137
of these problems we have yet to devise such procedures. Examples are the game
of go and some realms of mathematical theory. In sum, the levels of intelligence
required for various activities are as follows:

Level 1
Tic-Tac-Toe
Cannibals and missionaries
Level 2
Checkers
Chess
Backgammon
Euclidian geometry

Level 3
Go
Some realms of mathematics

What is our conclusion with regard to the recursive formula? It is certainly

simple and elegant. For the first two classes of problems described above—the class
that can be fully analyzed within a reasonable amount of time and the class that
cannot be fully analyzed but requires only simple terminal-evaluation procedures to
achieve intelligent performance—the full description of the algorithm is simple and
can be said to be solved by a unifying formula. The most complicated aspect of
implementing such a solution is the generation of the possible moves or steps at
each point. Doing so is, however, simply a matter of stating the problem (e.g., the
rules of the game). Thus the recursive formula enables us to solve a problem as
simply as stating it.
What about the third class, the class for which simple terminal-evaluation
procedures do not appear adequate? For this class the basic recursive algorithm is
only part of the solution. We end up requiring insight into the unique structure of
each such problem. Many of these problems can be solved (and have been solved),
but the solutions are different in each case. If there is a unifying formula linking
these diverse algorithms, none has yet become apparent.
The extent to which one accepts the recursive formula as a unifying
formula for even a subset of intelligent problem solving comes down again to one’s
definition of intelligence. It might be argued that problems in the first two classes
described above are inherently not problems requiring intelligence but rather just
data- and symbol-manipulation problems for which computers are well suited and
that truly intelligent problems are of the third kind. Accepting this point of view
requires us to modify our opinion of certain activities that have for centuries been
considered to be intelligent, like playing checkers or chess and solving problems in
Euclidian geometry and other areas of mathematics.
Such a view would be equivalent to stating that intelligence is to be
specifically defined as a type of behavior or activity that inherently cannot be cap-
tured in a single unifying formula. If we accept this definition, then, of course, the
question of whether or not intelligence or any important aspect of it can be captured |
in a formula is answered negatively by definition. Yet the implications of such a

138
 definition would not be consistent with commonly held views of what constitutes
intelligent behavior.*®
| would say that the recursive formula is partially successful. Certain types
of intelligent behavior, that required by the first two classes of problems, can be
effectively simulated with this elegant and powerful approach. Other types of
problems appear to resist it, including the important area of pattern recognition

+ Other Approaches to Modeling the Software of Intelligence:

Random Nets, Pandemonium, and Trees

I do not come to the discussion of connectionism as a neutral observer. In fact, the standard version of its
history assigns me a role in a romantic story whose fairytale resonances surely contribute at least a little
to connectionism’s aura of excitement.
Once upon a time two daughter sciences were born to the new science of cybernetics. One
sister was natural, with features inherited from the study of the brain, from the way nature does things.
The other was artificial, related from the beginning to the use of computers. Each of the sister sciences
tried to build models of intelligence, but from very different materials. The natural sister built models
(called neural networks) out of mathematically purified neurones. The artificial sister built her models
out of computer programs.
In their first bloom of youth the two were equally successful and equally pursued by suitors
from other fields of knowledge. They got on very well together. Their relationship changed in the early
sixties when a new monarch appeared, one with the largest coffers ever seen in the kingdom of the
sciences: Lord DARPA, the Defense Department's Advanced Research Projects Agency. The artificial
sister grew jealous and was determined to keep for herself the access to Lord DARPA’s research funds.
The natural sister would have to be slain.
The bloody work was attempted by two staunch followers of the artificial sister, Marvin
Minsky and Seymour Papert, cast in the role of the huntsman sent to slay Snow White and bring back her
heart as proof of the deed. Their weapon was not the dagger but the mightier pen, from which came a
book—Perceptrons—purporting to prove that neural nets could never fill their promise of building
models of mind: only computer programs could do this. Victory seemed assured for the artificial sister.
And indeed, for the next decade all the rewards of the kingdom came to her progeny, of which the family
of expert systems did best in fame and fortune.
But Snow White was not dead. What Minsky and Papert had shown the world as proof was
not the heart of the princess; it was the heart of a pig.
Seymour Papert, 1988

Another unified theory: The Perceptron

In the early and mid 1960s the Al field became enamored with another type of
machine called the Perceptron.*° This was an attempt to capture intelligence using a
large number of simple mechanisms. A perceptron consisted of three types of
devices: sensory units, associative units, and effector units (see figure). The sensory
units consisted of simple photocells arranged in a two dimensional array, simulating
the function of the human retina (or many other forms of input). A Perceptron
incorporating auditory sensory units was also proposed. The associative units were
randomly wired between the sensory units and the effector units and in some cases
other associative units. The associative units contained a feedback mechanism that
makes them automatically adjust the amount of current that passes from each input
to each output on the basis of feedback received by the unit on whether or not the
Perceptron’s last response was correct. The effector units sum the outputs of the

139
 associative units to which they are connected, and by comparing different summed
voltages, they make final decisions on the identities of the patterns presented to the
sensory units.
The machine attracted attention because, simple as it was, it appeared to
model both the structure and the functions of the human brain.®' The sensory,
associative, and effector units modeled afferent, associative, and efferent neurons
respectively. Moreover, the random wiring of the associative units, sometimes
referred to as random neural nets, were thought to model the seemingly random
interconnections of the associative neurons. The Perceptron learns by gradually

Asimple neural net. Not shown are the

feedback loops that enable such a net to
learn. (From Minsky and Papert, Per-
ceptrons, 1988, p. 5)

adjusting the electrical potential on each random connection. This was thought to
model human learning. That human neural connections were believed to be random
was consistent with a theory that most human perceptual ability was learned rather
than innate
It was discovered that such a machine could be “taught” to recognize
various shapes, such as printed letters. In 1969 Marvin Minsky and Seymour Papert
wrote a book called Perceptrons, which proved a set of theorems showing that
Perceptrons could never solve the simple problem of determining whether or not a
line drawing is “connected” (in a connected drawing all parts are connected to each
other by lines), the so-called contiguity problem (see figure).*? It was also shown that
their ability to recognize such shapes as printed letters was quite limited: the Per-
ceptron would fail if the letters were tilted, changed in size, or otherwise distorted.
We now know that though detailed perceptual categories, such as the shape of the
letters in our alphabet, are certainly learned, the properties on which such perception
is based—lines, curves, concavities, loops, etc.—are at least partially prewired, and
that the neural connections are anything but random.
The book had a dramatic effect, and virtually all work on Perceptrons came
to a halt. The severe limitations of the Perceptron’s capabilities came as a disappoint-
ment to many. Prior to the publication of Perceptrons, it had appeared to some
practitioners that this simple technique had captured the secret of intelligence. All
we had to do to make the Perceptron more intelligent was to add more associative
units and more wires. Unfortunately, intelligence was not to be captured this easily.**
After the publication of Perceptrons, there was an understandable hesitation in the
Al field to look for any single method or even a single paradigm underlying intelli-

140
 gence. The limitations of GPS and the failure of the Perceptron certainly did not
prove that all unified approaches to defining and replicating the methods of intelli-
gence would fail, but enthusiasm for discovering such unifying principles waned
nonetheless.
It was probably a healthy development for artificial intelligence in the early
1970s that the search for a unified theory of intelligence was put aside at least
temporarily. This focused attention instead on devising a variety of approaches for a
variety of problems and allowed the development of generalized theories to wait
until more application-specific techniques were perfected. The implication of this
view is that Al is composed of a great variety of methods and approaches, that there
is no single set of techniques that solves all Al problems. Complex problems are
often reducible to hierarchies of simpler problems. The solutions, then, are manifest
in the structure of the hierarchies and the solutions to the simplest problems found
at the lowest level of each hierarchy. Such structures and solutions will differ for
different classes of problems.
A related observation is that solutions to Al problems are often domain
specific. In other words, the methods for solving particular Al problems need to be
derived from our understanding of the particulars of that problem. Recognizing
printed characters by machine, for example, needs to be concerned more with the
details of letter shapes, ink and paper reflectivity, printing errors, and other peculiari-
ties of print than with any generalized techniques of pattern recognition.°° Similarly,
recognizing human speech by machine needs to be primarily concerned with the
detailed nature of human speech sounds, the syntax of natural language, and the
signal-processing characteristics of the transmission channel.
Which figure is connected? These two
Occasionally claims are made that a “generalized perception algorithm”
figures from the cover of Minsky and
Papert's influential Perceptrons (1988) has been devised that can recognize any type of pattern, whether it be manifested in
illustrate the type of problem that single speech, printed characters, land-terrain maps, or fingerprints.*% On closer examina-
layer neural nets are incapable of solving,
tion, it often turns out that such claims are absolutely correct: such algorithms do, in
according to the book’s pivotal theorem.
fact, recognize every type of pattern, only they do all these tasks very poorly. To
perform any of them well, with satisfactory (and commercially acceptable) rates of
accuracy, requires a great deal of knowledge deeply embedded in algorithms specific
to the domain of inquiry, so much so that this aspect of the technology outweighs
the generic Al techniques.

The new connectionists

Perceptrons A more sophisticated generation of random neural nets is now being developed with
important differences from the earlier work. The limitations described in the Minsky
and Papert theorems are not applicable to at least some of the work in this latest
generation. The “neurons” themselves can be more sophisticated. Each neuron may
be modeled by its own computer, which thus provides a high degree of parallel
processing. Each neuron model can perform arbitrarily complex transformations on
their inputs. Rather than having neural connections fixed at the outset, these neural
models are also capable of growing new connections.*”
In all fairness, it should be pointed out that some of the recent work differs
little from that done over twenty years ago. In many of the recent connectionist
systems the nets still have only one or two layers and still use simple linear neuron

141
 models. In my opinion, a far more promising avenue of research lies with nets built
from more complex neurons organized in multiple layers and each capable of detect-
ing abstract features of patterns (such as lines, curvatures, and concavities).
The earlier random neural nets generated excitement precisely because
they appeared to provide a simple formula that could generate complex and purpose-
ful, that is, intelligent, behavior. The earlier work would qualify as a unifying formula,
but unfortunately it did not work very well. The more recent work—particularly that
which stresses complex neuron models—appears far more promising but does not
fully qualify as a unifying formula. It is more of a unifying paradigm in that it still
requires diverse approaches to implement and organize the overall system.®
Still the promise is implicit that we can get more intelligence out of less. In
other words, while the neuron models still require intelligent programs, the hope is
that the entire neural net will exhibit a higher degree of intelligence than is explicitly
programmed at the lower neuron level.
Whereas Minsky was obviously opposed to the Perceptron work and
expressed concern that limited research resources were being drawn into a dead
end, he has expressed cautious interest in random neural nets based on more
intelligent neurons.*? It would certainly appear that this work is consistent with
Minsky’s recent emphasis on the “society of mind” as a paradigm for intelligence.

Pandemonium selection
For pattern-recognition problems (such as identifying faces, printed characters, land
terrains, fingerprints, blood cells, etc.), we need to break down the higher-level
decisions (e.g., what letter this is) into lower level ones. If we design “demons”
(algorithms) to answer the lower-level questions, we can use their answers to make
a final decision. One approach is to invite the demons to a “meeting.” We then ask,
Is this letter an A? and determine how loud the demons shout. We go through all of
the possibilities in the “identification set” (B, C, etc.) and pick the one that causes
the demons to voice the strongest approval. Some of the low-level questions are

Nature's neural net. The human brain

contains billions of neurons, each
of which is capable of carrying out com-
plex computations (using
both analog and digital techniques)
and communicating with potentially
thousands of other neurons. (Supplied by
Photo Researchers and Biophoto
Associates Science Source)

142
more important than others, and thus some of the demons are more influential (that
is, able to shout louder) than other demons. The demons also control how loud they
shout by how sure they are in answering their lower-level question.

The decision tree

Another way to organize the demons is in a tree. We go to one demon and ask a
question, such as Does this letter have a loop (an area of white completely enclosed
by black)? If the demon says yes, we continue on to demons on the yes branch of
the tree; if no, then we take the no branch. If the answer is yes, then we know the
letter could be an A but not a C, and so on. We follow the branches from one
demon to another until we get to a leaf, at which point we have an answer. Note
that such decision trees differ from the recursively generated trees described for the
recursive formula. In recursion, a tree is generated for each decision, and the
topology (shape) and content of the tree will differ for different situations. For
pattern-recognition tasks, the tree is generally fixed and designed by the program
designer. What differs for different situations (inputs) is not the shape of the tree but
the path taken through it.’
Both the pandemonium and decision-tree methods can work. The latter is
faster (for a serial computer) in that we have only to “compute” demons that we
find along our path. But it is more subject to catastrophic failure (an incorrect
answer) if a demon, particularly one encountered early on, makes a mistake. With
pandemonium selection, an erroneous demon can still be outvoted
As with the recursive model, there is a question as to how sophisticated
these demons should be. At one extreme we have high-level feature extraction, in
which the demons answer difficult questions. Examples are Is this region concave?
in the context of character recognition and Is this burst of sound the beginning of a
plosive phoneme? in the context of speech recognition. At the other extreme we
have template matching, in which the demons answer the simplest possible ques-
tions, such as |s this particular point black or white? Whereas | opted for the simple-
minded school of thought with regard to chess (though not for all games), | feel that
more complex demons are necessary for most pattern-recognition tasks. In chess
we are dealing with a very orderly and artificial world that can be described with es-
sentially perfect precision. In pattern recognition we are dealing with more complex
and more variable real-world phenomena that inherently require higher levels of ab-
straction.
High-level demons can be broken down into hierarchies of simpler de-
mons. It is thus possible to develop decision trees with simple demons at the
terminal nodes that can detect more abstract or higher-level features at higher nodes
of the tree. However, if the demons are simple, then the heart of the technology lies
in the decision hierarchy itself.
Thus, as with the random neural net; pandemonium selection, decision
trees, and a variety of other methods provide powerful paradigms but are not simple
formulas that can be expanded into intelligent behavior through the application of
computational power alone. This is in contrast to the recursive formula, in which a
simple formula combined with large doses of computational brute force is successful
in solving at least certain classes of intelligent problems

143

llc ri lf
 eon)

Ta

iy
ra

Demons sitting in a decision tree.

144
The hierarchy of intelligence
What we often regard as intelligent behavior is in actuality many processes operating
in parallel and interacting on different levels of a hierarchy. If we consider carefully
the process of reading and understanding printed language, it becomes clear that we
are dealing with a multiplicity of talents: at the character level, pattern recognition; at
the word level, the syntactic analysis of word sequences; and at higher levels, the
decoding of semantics, the retrieval and updating of our knowledge base, and an
understanding of “scripts” about the subject matter being written about, to name
only a few of the intelligent processes involved.
This leads us to a possible recursive definition of intelligence:

An intelligent process is an association of intelligent and unintelligent processes

communicating and influencing each other.

This definition is self-referencing (that is, it contains the concept being defined in the
definition), but it is not infinitely circular. It again employs the special type of self-
reference we Call recursion
Let us examine this “association” definition of intelligence in the context
of recursion. If we consider an intelligence to be an association of other intelli-
gences, we replace one intelligence with a larger number of intelligences. We now
have an association of intelligences, each of which is in turn expanded into an
organization of yet more intelligences. If we keep applying the recursive definition,
we end up with a vision of a great profusion of intelligences organized in a hierarchy
of many levels. If there were no escape or terminal portion of the definition, we
would replace one intelligence with an infinite number of intelligences—a satisfying
mystical vision perhaps, but of no value in building practical theories of intelligence.
The escape from infinite recursion is the concept of unintelligent proc-
esses, the mechanisms at the lowest level. Intelligence overall results from the
coherency of the entire hierarchy of increasingly sophisticated associations of
intelligence.

Minsky’s “society of mind”

A recent and perhaps the most comprehensive exposition of this type of theory has
been offered by Minsky in writings from 1977 to the present. “How can intelligence
emerge from non-intelligence? . . . One can build a mind from many little parts, each
mindless by itself. I'll call “Society of Mind” this scheme in which a mind is made of
many smaller processes. We'll call them agents. Each agent by itself can only do
some simple thing which needs no mind or thought at all. Yet where we join those
agents in societies—in certain very special ways—that leads to true intelligence.”
How do the various minds in a society interact? The short answer is, In
many different ways. Consider the varied ways we interact with each other in human
society. Relationships include symmetric bonds, such as lovers and friends, as well
as asymmetric ones, such as parent-child, boss-subordinate, and oppressor-victim
Methods of making decisions range from negotiation of equal or nearly equal parties
and the influence of one person on another to the dictates of boss to subordinate,
among the multifarious ways in which we interact and make decisions in a society

145
On a larger level, communication and decision making is regulated by the rules of
the society with political, cultural, and social institutions governing the roles of each
member.
Within a single individual we also have many mental processes carrying on
their operations in parallel. Incoming information is evaluated not by a single “depart-
ment,” but by many of our subminds at the same time. Some deal with fears; others
with hopes, desires, and needs. Yet others deal with the practical aspects of manipu-
lating muscles or performing pattern-recognition tasks. Through a hierarchy of
negotiations at different levels, our thoughts are translated into actions. Some
subminds are more powerful than others, and the methods of negotiation vary
according to the level and the subject matter.®
Such theories as the society of mind are essentially drawing an analogy of
the intelligence of one person to that of a society such as a nation with its compet-
ing factions, differing views of events, and varying methods of resolving conflicts and
pooling opinions. This type of theory is consistent with the complexity of human
thought, with its predictable and unpredictable sides, and with the everyday phe-
nomenon of mixed feelings.
We can extend the concept of society beyond the individual. A human
being is a society of minds representing needs, desires, and capabilities on many
different levels. An actual human society then continues the process of an intelligent
hierarchy on yet more levels. Some of the methods of negotiation, influence, and
problem resolution in a society are not that different from those taking place within a
single individual.
As we expand each intelligence into a society of intelligences, are the
intelligences on the lower level less sophisticated and less capable than the intelli-
gence represented by the entire society? | would say that it is certainly desirable that
the society be more intelligent than its members. This is not always the case in
human society—the intelligence of a mob is a testament to this. Yet successful
human organizations are obviously capable of accomplishing intellectual tasks that
their individual members could never accomplish alone. Within a single human mind
a well-integrated personality is capable of synthesizing its constituent thought proc-
esses in a manner that exhibits greater intelligence at each higher level of the mental
hierarchy. As examples of poor integration, many forms of mental illness are charac-
terized by more primitive thought processes interfering with and even taking charge
of the more subtle and intelligent capabilities.
Inherent in this vision of intelligence is the inference that human intelli-
gence, though very complex, is not infinitely complex. It can ultimately be under-
stood in terms of simpler processes. At the lowest level there are very simple
processes, although the hierarchy that leads to the simple processes is anything but
simple. We do not fully understand the hierarchy—indeed, we understand very
little of it—but there is no reason that we cannot progressively increase our knowl-
edge of it

Lettvin's “society of neurons”

Another theory of human intelligence as a society of intelligences, one articulated by
Jerome Lettvin of MIT, views individual neurons as small “minds” organized in a

146
 society more complex than any human society (there are tens or hundreds of billions
of neurons in the human brain and only billions of human beings on the earth). When
first articulated in the mid 1960s, this view differed from the conventional wisdom
that neurons were just simple switches with very limited information-processing
capability.” As Lettvin surmised, we now know that neurons are indeed quite
complex and that communication between neurons does use a “language” that,
though not nearly as sophisticated as human language, is far from trivial in its
structure. Lettvin imagined the human mind as a hierarchy of larger and larger
societies of neurons with languages, cultures, and power struggles.®* Anyone who
doubts that a single cell can be intelligent needs only consider the macrophage
These killer white blood cells are capable of conducting subtle pattern-recognition
tasks to identify hostile bacteria, parasites, and other agents of disease and of
carrying out elaborate tactics to destroy the enemies once identified. There is no
reason to suppose that nerve cells, which are structurally more complex than
macrophages, have less sophisticated capabilities.
Intelligence as an association of lesser intelligences is both a theory of
natural intelligence and a paradigm for the organization of machine intelligence. As
our computer systems become more complicated, one way to organize them is to
develop smaller systems, each of which is relatively complete with its own structural
integrity, and then define languages and processes that let these lesser “experts”
interact. This type of organization is becoming increasingly popular in both applied
and research Al systems

An “intelligent cell.” The macroph-

age hunts its pathogenic prey and is
capable of a variety of tactics to
destroy it, once found. Here it
extends a pseudopod to ensnare a
Escherichia coli bacterium.
Magnification: 4,000 times. (Photo
by Lennart Nilsson, copyright by
Boehringer Ingelheim International
GmbH)

147
A society of human and machine intelligence
An example of an integrated “society of intelligences” is the set of expert systems
already developed or being developed by Digital Equipment Corporation (DEC). One
of the first practical expert systems was XCON, which uses over ten thousand rules
to design the configuration and assembly process for their VAX-series computers.
XCON has been found to be substantially more reliable than human experts and is
now routinely used on almost all orders for VAX computers. With the success of
XCON, DEC is now developing expert systems for virtually all aspects of its
company’s operations, including computer-assisted design, manufacturing control,
material-resource planning, sales strategies, financial planning, and other areas.”°
When fully operational, expert systems will be playing critical roles in every aspect of
DEC's operations, communicating with each other as well as with DEC employees
through an extensive computer-based network. This is expected to be one of the
first human organizations operating in a real-world environment to combine a society
of multiple human and machine intelligences.”

Progress on different levels

If we consider the diverse problems that are being attacked by Al technology, and
the wide range of intelligent behavior, pattern recognition, and problem solving
abilities of human intelligence, we must recognize that a variety of approaches is
needed.” Progress in harnessing machine intelligence will likely result from all of the
following:

More powerful “unifying” formulas for solving classes of intelligent problems

More powerful paradigms
Highly detailed technologies with deeply imbedded domain-specific knowledge for
interacting with real world problems
More powerful computational (hardware) resources
More powerful (and generally parallel) architectures

With regard to the last item, we are increasingly finding that different
problems require different hardware architectures and that architecture design is
becoming as important as algorithm design.

Is there a formula for intelligence?

The answer is that there are indeed formulas for both the “hardware” and the
“software” of intelligence that are elegant and far reaching in their implications
These formulas provide powerful insights into the nature of problems requiring
intelligence and their solutions. We must also conclude, however, that these simple
models of intelligence are only the beginning, not the end, in our search for intelli-
gence. Real-world intelligence is a combination of elegant insight with extensive
knowledge and experience that can be accumulated only through painstaking and
often hard-won effort.’$

148
+The Formula of Life as a Formula of Intelligence

A great deal of the universe does not need any explanation. Elephants, for instance. Once molecules
have learnt to compete and to create other molecules in their own image, elephants, and things
resembling elephants, will in due course be found roaming through the countryside.
Peter Atkins

It is raining DNA outside. On the bank of the Oxford canal at the bottom of my garden is a large willow
tree, and it is pumping downy seeds into the air. ... The DNA content must be a small proportion of the
total, so why did | say that it was raining DNA rather than cellulose? The answer is that it is the DNA
that matters. The cellulose fluff, although more bulky, is just a parachute, to be discarded. The whole
performance, cotton wool, catkins, tree and all, is in aid of one thing and one thing only, the spreading of
DNA around the countryside. Not just any DNA, but DNA whose coded characters spell out specific
instructions for building willow trees that will shed a new generation of downy seeds. Those fluffy
specks are, literally, spreading instructions for making themselves. They are there because their
ancestors succeeded in doing the same. It is raining instructions out there; it's raining programs; it's
raining tree-growing, fluff-spreading, algorithms. That is not a metaphor, it is the plain truth. It couldn't
be any plainer if it were raining floppy discs.
Richard Dawkins

One might include DNA and its support cast of “data processing” chemicals as an
example of a classical unifying formula.” It can also be regarded as a unifying
formula for intelligence, because it provides the foundation for human life, which is
our primary example of intelligence. Let us briefly examine the mechanics of DNA
and then compare it to some of the unifying formulas described above for machine
intelligence.

The formula of life

The chemical structure of the DNA molecule was first described by J. D. Watson
and F. H. C. Crick in 1953 as a double helix consisting of a pair of strands of polynu-
cleotides with information encoded at each position by the choice of nucleotides.”>
Since then biochemists have gone on to map extensive sections (but still a small
fraction) of the genetic code and to understand the detailed chemistry of the
communication and control process by which DNA commands reproduction through
such other complex molecules and cellular structures as messenger RNA, transfer
RNA, and ribosomes. At the level of information storage, the mechanism is surpris-
ingly simple. Supported by a twisting sugar-phosphate backbone, the DNA molecule
contains up to several million rungs, each of which is coded with one letter drawn
from a four-letter alphabet. The alphabet consists of the four base pairs: adenine-
thymine, thymine-adenine, cytosine-guanine, and guanine-cytosine. The DNA strings
in a single cell would measure up to six feet in length if stretched out, but an
elaborate packing method coils it to fit into a cell only 1/2500 of an inch across. With
four letters in the alphabet, each rung is coding two bits of data in a digital code
Special enzymes can copy this information by splitting each base pair and assem-
bling two identical DNA molecules by rematching the broken base pairs. Other
enzymes actually check the validity of the copy by checking the integrity of the base-
Pair matching. With these copying and validation steps, this chemical data-processing
system makes only about one error in a billion base-pair replications. Further

149
 redundancy and error-correction codes are built into the digital data itself, and so
meaningful mutations resulting from base-pair replication errors are rare. Most of the
errors resulting from the one-in-a-billion error rate will result in the equivalent of a
“parity” error, an error that can be detected and corrected by other levels cf the
system, which will prevent the incorrect bit from causing any significant damage.’®
In a process technically called translation, another series of chemicals
translate this elaborate digital program into action by building proteins. It is the
protein chains that give each cell its structure, behavior, and intelligence. Special
enzymes unwind a region of DNA for building a particular protein. A strand of
messenger RNA (mRNA) is created by copying the exposed sequence of bases. The
mRNA essentially has a copy of a portion of the DNA letter sequence. The mRNA
travels out of the nucleus and into the cell body. The mRNA codes are then read by

A computer view of the code of life

looking down the axis of the DNA
double helix. (Photo by R. Langridge of
Rainbow)

a structure in the cells called a ribosome, which acts like a tape-recorder head
“reading” the sequence of data encoded in the mRNA base sequence. The “letters
(bases) are grouped into words of three letters each called codons, with a codon for

le amino acids, the basic building blocks of protein. A ribosome

Ss
oO
5
Do

codons from the mRNA and then, using another set of molecules called
o
Q

transfer RNA (tRNA), assembles a protein chain one amino acid at a time. It is now
the job of the assembled proteins to carry out the functions of the cell (and by

With over 500 amino acids in each molecule of hemoglobin, that comes to |
ite by the ribosomes just for the creation of |
1emoglobir
In some ways the biochemical mechanism of life is remarkably complex
te. In other ways it is remarkably simple. Only four base pairs provide the
for all of the complexity of all human life and all other life as we know
it. The ribosomes build protein chains by grouping together triplets of base pairs to
select sequences from only twenty amino acids. These protein chains then control
everything else: the structure of bone cells, the ability of muscle cells to flex and act
in concert with other muscle cells, all of the complex biochemical interactions that
take place in the blood stream, and, of course, the structure and functioning of the
brain.””
What we now understand provides an elegant and unifying picture of the
hardware of life's data-processing mechanism. We are just beginning to unravel the
software. We are beginning to write down the actual binary codes of some of the
many thousands of genes that control our human traits. Understanding these binary
codes is yet another matter. The difficulty is a matter of trying to read machine
“object code” (programming instructions without symbolic mnemonics or other
documenting comments) in a programming language vastly more complex than any
computer language now in use. It has also become apparent that evolution has not
been a very efficient programmer. The genetic code is replete with redundancies and
sequences that do not compute (that is, are unable to generate proteins). For ex-
ample, an apparently meaningless sequence called A/u, comprising 300 nucleotide
pairs, occurs 300,000 times in the human genome, equivalent to over 3 percent of
our genetic program.” Despite its sloppiness (and lack of documentation), however,
it is clear that the “program” does work reasonably well.
 An entire computer can be constructed from the repetition of a very
simple device called a logic gate. We can make an even stronger
statement: any computer can be built with just one type of logic gate.
Any computer (serial computers, which do one computation at a time,
as well as massively parallel computers, which do many things at
once) can in theory be built from a suitable number of this very simple
device.
The Church-Turing thesis states that all problems solvable
by a “sentient being” (a person) are reducible to a set of algorithms.
Another way of expressing this thesis is that machine intelligence
and human intelligence are essentially equivalent. According to this
(controversial) thesis, human intelligence may at present be
(quantitatively) more complex than machine intelligence—that is,
use more parallelism and more sophisticated algorithms—but there
is no fundamental (qualitative) difference between them. As
machines gain in capacity, do more operations in parallel, and use
more capable methods (better algorithms), they will be able to
| PEN UC a Ciuc
emulate human intellectual ability more and more closely. This
article will demonstrate that the logic gate, particularly the one we
Anor B: The Basis of Intelligence?
call nor, can be considered the basis of all machine intelligence. To
the extent that one accepts the Church-Turing thesis, the simple nor
gate can be considered (at least in theory) the basis of human
intelligence as well. Our ability to model the complex and diverse
examples of intelligent behavior from building blocks of nor gates is
another example of a simple unifying formula underlying complicated
phenomena. As the following derivation points out, however, this
insight alone is not sufficient to build an intelligent machine.
To understand how we can build a computer from logic
gates, it is first necessary to understand the nature of such devices. A

A logic gate. |

el
Te)

Output |
rt) |
}

152
 logic gate is a device with two inputs and one output. We can
imagine each input and the output to be wires. Each wire can have
two states: true and false. When implemented with electronics, this
often corresponds to the presence or absence of electric voltage. For
example, in TTL (transistor-transistor logic) circuits, a very common
type, +5 volts and 0 volts represent the two states.
With two inputs and two possibilities for each input, there
are four possible input combinations to a logic gate. There are
actually several types of logic gates. Particular types will associate

each of these four combinations of input with an output. For example,

the nor (“not or”) logic gate transforms each of these four possible
input combinations as shown in the table.

The nor gate

Input A Input B Output

False False True

False True False
True False False
True True False
The output of ano is true only if input A and input B are
true. There are 16 possible different logic gates, based on all possible
The output of nor is true if and only if neither A nor B are
mappings of the four inputs onto outputs. Some of these logic
true. Another example of a logic gate is the ano gate (see table).
functions have commonly understood names and others do not (see
table).
The ano gate

Input A Input B Output

Outputs of the 16 possible dual input logic functions
False False False
Inputs
False True False
True False False Function name F-F ET EE

a4
True True True
ALWAYS FALSE
|

TDA TDA TDA TT

ara a 4a 8

AND
angn amram 44

Aano not B
MAMA

A
Bano not A
i

B
434494

xor (exclusive or)

oR
NOR
EQUIVALENCE
not B
Aornot B
aM

not A
BornotA
NaNp (not and)
437

ALWAYS TRUE

153
 Not all of the logic functions are equally useful. For
example, the aways True function, whose output is true regardless of
its input, is relatively useless, since its output is always the same.
The same can be said of the atways False function.
An important observation is that we can create all 16 of
the functions from the nor function. For example, we can create not A
as shown in the figure. We can call not A by a simpler name: nor. It
simply transforms a single input into its opposite. We can also create
or and ano from nor (see figures).

If we express these equivalences in the language of logic,

we have not A= Anor A; Aor B= not (A nor 8); A ano B=(not A) nor
(nor B).
With nor, ano, and or (each of which, as we have shown,
can be constructed from the nor function), we can in turn build up all
16 logic functions (see table). Since we can build up all 16 logic
functions from not, ano, and or and in turn can build up not, ano, and or
| from the nor function, it follows that we can build all logic functions
from nor.
Building logic functions from not, ano, and or j

Inputs
—_____— _ Description using
mn Functionname F-F F-T T-F T-T not, Ano, andor \

ALWAYS FALSE E F F F Aano not A

haa —_ ‘NOR —-k 3 AND EES T AanoB

Aano not B rE F T F Aano nor B
A F F iN T A
S. Bano notA FE 1 F F Bano not A
B F ui F v B
XOR F T T F (Aor B) ano not (A ano B)
OR F ic T T AorB
NOR T F F F nor (Aor B)
pose Nay eer Oye EQUIVALENCE T & Fk T (not
A ano not B) on (A ano B)
not B T F ii Fo onotB
Aor not B a rf T T —_not(Bano nor A)
not A T T rE Fo onotA
BornotA mm T Fr T not
(A ano nor B)
NAND i rE T F not (A ano B)
ALWAYS TRUE T 1 mT T = AornotA

It turns out that there is one other logic function that can
also be used to build all the logic functions, the nano gate. The reason
for this is as follows. A critically important logic function is negation
or the not function. Only four of the 16 logic functions allow us to
negate an input: nor, Nano, Not A, and not B. Of these four, two, not A
and nor B, are unable to give us ANo or on because they each ignore
one of their inputs. Thus, only nor and Nano can give us all three vital
building blocks of nor, ano, and or.

154

eee
Deriving or from Nor.

Deriving ano from nor.

Memory
==
Thus, all of logic can be built up from one simple logic gate, nor (or
alternatively, nano). There is, however, one other important type of
mechanism essential for either machine or human intelligence, the
ability to remember. In the same way that we can build up very
complex logic functions from very simple ones, we can also build
very complex and extensive memory structures from simple one-bit
(on or off) memory. The question, then, is, Can we build a one-bit
memory from logic gates? As it turns out, we can (see figure).
Ifthe “set” input (which might be called maxe tave) is made
true, even momentarily, the output will become true and will remain
true, even when the set input becomes false. This memory cell,
constructed of only two nor gates, thus remembers forever even a
temporary state of true on the set input. Similarly, a momentary state
of true on the “reset” (make Fats) input will cause the output to go
 CeO RY

Tg
eT

Amemory cell from nor.

permanently to false. Interestingly, we can replace the two nor gates Let us define a “node” to be a particular input switch, a
with nano gates in exactly the same configuration and the memory final output display, or an input or output of a particular logic gate.
cell will work in a similar way (we need only reverse the position of We thus have the following types of nodes:
the “memory” and “not memory” output lines and negate the set and
+ Inputs presenting a problem or situation
reset input lines).
+ Logic-gate inputs
We thus have all logic functions as well as memory—the
+ Logic-gate outputs
two requirements for computation—built up from a very simple
+ Final outputs
device, the nor gate (or Nano gate). With enough nor gates we can
If we assign each of these nodes a unique number, we
build any computer or implement any algorithm. All machine
can then describe any machine by a series of numbers that describes
intelligence (and, if we accept the Church-Turing thesis, all natural
the connections of the nodes. For example, if our machine is
intelligence) can be constructed from nor.
described as 10, 126, 4034, 28, 1, 12,..., this would mean that node 1
To solve a problem, it is not enough, however, to simply
is connected to node 10, node 2 is connected to node 126, and so on.
have a sufficient number of nor gates. If we dumped a million nor
Thus, any machine can be described as a series of numbers
gates on a table, they would not perform any function, useful or
describing its logical connections.
otherwise. Obviously, they need to be connected together and to the
There is one other type of information required before our
outside world in appropriate ways. The outside world can be
machine can solve a problem, and that is the initial state of the
modeled as a series of switches that present problems or situations.
memory cells. Even though the memory cells are constructed from
Even human sensory inputs, such as sight or hearing, can be modeled
logic gates, they have two different states: they hold either the value
in this way. The input switches could be thousands or even millions
true or the value false at any one time. We need to set an initial value
of light sensors arranged in an array or auditory sensors that respond
for each such cell. If we number each memory cell, we can describe
to particular frequencies. Even continuous-valued input can be
these initial values as a sequence of one-bit numbers (such as 0, 1, 1,
modeled by using multiple input switches to represent different
0, etc.). If we group the memory cells into “words” of more than one
levels of input.
cell each, then our list of initial values can use numbers other than
just0 and 1.

156
Hardware and Software
ene ee

The first list, which described the interconnections of our inputs,

logic gates, and outputs, might be called the hardware description.
The second list, which described the initial states of the memory
cells, can be called the software description. In designing such
systems, real engineers would substitute more descriptive symbols
for most of the numbers. The final result would be a string of
numbers, nonetheless. At this point we notice that the hardware
description of our machine looks very similar to the software
description. Both are simply lists of numbers. We might substitute the
word “information” or even “knowledge” for the expression “list of
numbers.” Seen in this light, the creation of hardware and software
designs are very similar: they both involve the creation of software-
like information. We might even regard them both as forms of
software.
This model of intelligence as a sea of nor gates provides
us with a simple theoretical model for understanding machines that
can implement an algorithm and for machine complexity. If we
regard the two lists of numbers described above as descriptions of
software, then we can say that we have a simple formula (i.e., nor) for
the hardware of intelligence; we will have to look elsewhere for a
Hooking up the nors. Node con-
nections described by the list
unifying model of the software. 10, 126, 4034, 28, 1, 12
e

we
§

157
The Analytical Engine (1833-1871,
replica), the world's first program-
mable computer. Although it never
ran, Babbage’s entirely mechanical
computer foreshadowed the
modern electronic computers of a
century later. (Courtesy of IBM
Corporation and Neuhart Donges
Neuhart Designers)
 Mechanical Hoots

What if these theories are really true, and we were magically shrunk and put into someone's brain while
he was thinking. We would see all the pumps, pistons, gears and levers working away, and we would be
able to describe their workings completely, in mechanical terms, thereby completely describing the
thought processes of the brain. But that description would nowhere contain any mention of thought! It
would contain nothing but descriptions of pumps, pistons, levers!
Wilhelm Leibniz (contemporary of and collaborator with Isaac Newton), commenting on theories that the brain was
“just” a complicated mechanical computer

The human imagination for emulating human thought by machine did not stop at
mere philosophical debate and thought experiment. From Platonic times, inventors
were anxious to apply whatever technology was available to the challenge of re-
creating human mental and physical processes.' Before the taming of the electron,
this meant harnessing the state of the art in mechanical techniques.”

* Early Automata and Calculating Engines

Automata
As far back as the times of ancient Greece, machines that could emulate the natural
movements of living creatures were built as a source of delight and apparent magic
They were also constructed by philosophers and their associates as a way of demon-
strating that the natural laws were capable of producing complex behavior. This in
turn fueled speculation that the same deterministic laws governed human behavior.°
Archytas of Tarentum (c. 400-350 s.c.), a friend of Plato, constructed a
pigeon whose movements were controlled by a jet of steam or compressed air.
Even more elaborate automata, including an entire mechanical orchestra, existed in
China at the same time.*
The technology of clock and watch making produced far more elaborate
automata during the European Renaissance, including human androids that were
notable for their lifelike movements.® Famous examples of these include the mando-
lin-playing lady, built in 1540 by Giannello Torriano (1515-1585), and a child of 1772
that was capable of writing passages with a real pen, built by P. Jacquet-Droz
(1721-1790).°
Mech

P. Jacquet Droz's 1772 automaton

LEcrivain. (Supplied by the Muséum
d'Art et Histoire, Neuchatel,
Switzerland)

LEcrivain's mechanical intelligence.

(Supplied by the Muséum d’Art et
Histoire, Neuchatel, Switzerland)

160
 The abacus
Perhaps of greater significance in the development of intelligent machines were
early attempts to reduce the laborious efforts required for calculation. The abacus
developed more than 5,000 years ago in the Orient, is of particular interest in its
similarity to the arithmetic processing unit of a modern computer. It consists of
movable beads on rods, which together implement a digital number store. Using
prescribed “algorithms,” a user can perform computations ranging from simple
addition to evaluation of complex equations. The algorithms are performed directly
by the user, not by the machine, but the methods are nonetheless mechanistic

The abacus. Algorithms are provided

and implemented by the user.
(Photo by Dan McCoy of Rainbow)

Napier's bones
In 1617 John Napier (1550-1617), who is generally considered the discoverer of
logarithms, invented a method of performing arithmetic operations by the manipula-
tion of rods, called “bones” because they were often constructed from bones and
printed with digits. Napier’s innovation was of direct significance for the subsequent
development of calculating engines in which the “algorithms” were implemented in
the mechanism of the device, rather than by human manipulation.®

Napier's Bones (1617). “I have

always tried, most noble sir,
according to my strength and to the
N°
I"

measure of my ability, todo away 2

with the difficulty and prolixity
of calculations, the tedium of 3
n

4
Ih

which deters most people from

studying mathematics”—John
p

Napier. (Courtesy of IBM 5 “

Corporation and Neuhart Donges
Neuhart Designers)
é'
i
@
b
The Pascaline (1647). “I submit to the
public a small machine of my own
invention, by means of which you alone
may, without any effort, per-form all the
operations of arithmetic, and may be
relieved of the work which has often
times fatigued your spirit when you have
worked with the counters or the pen"—
Blaise Pascal. (Courtesy of IBM
Corporation and Neuhart Donges
Neuhart Designers)

The Pascaline
Working from the age of 19 until he was 30, Blaise Pascal (1623-1662) attempted to
perfect his mechanical calculator. After more than 50 models were constructed and
discarded using materials ranging from wood and ivory to a variety of metals, Pascal
finally perfected the world’s first automatic calculating machine in 1642.9 The ma-
chine was considered automatic in that the algorithm was performed by the machine

and not the user, at least for addition and subtraction. The Pascaline, replicas of
which are still used by many school children, uses rotating wheels inscribed with the
digits and a ratchet mechanism that controls the overflow of one place position to
the next. The ratchet adds one digit to (or borrows one from) the next highest place
when there is a complete revolution of the lower decimal position
The perfection of this computing machine created a stir throughout Europe
and established the fame of its inventor. The device stimulated Pascal's own philo-
sophical reflections, and in his last (unfinished) major work, Pensées (“Thoughts”),
Pascal writes, “The arithmetical machine produces effects which approach nearer to
thought than all the actions of animals. But it does nothing which would enable us to
attribute will to it, as to the animals.”
The advent of an automatic calculating machine also led to controversy
about its impact and created fear that it would lead to the unemployment of book-
keepers and clerks. The excitement generated by the Pascaline encouraged Pascal
and his father to invest most of their money in an advertising campaign to market
the invention. Unfortunately, Pascal was a better philosopher than businessman, and
problems with reliability and service caused the venture to fail (apparently, many of
the production models required repair services to be performed by Pascal himself)

162
 Blaise Pascal, philosopher, scientist,
and entrepreneur. Problems with
customer service caused his
enterprise to fail. (Supplied by North
Wind Picture Archives)

The Pascaline. (Courtesy of IBM

Corporation and Neuhart Donges
Neuhart Designers)

163
Leibniz's own sketch of the
calculator mechanism. “Many
applications will be found for this
machine, since the elimination of
all errors and of almost all labor
Gola 4arbond
Beith te,
Cr icy eran
from calculations with numbers is
06 fiebbra hens
tremendously useful for the conduct
of government and science”—
Gottfried Wilhelm von Leibniz.
(Courtesy of IBM Corporation and
Neuhart Donges Neuhart Designers)

The Leibniz computer

Inspired by the Pascaline, Gottfried Wilhelm Leibniz attempted to add multiplication,
division, and the extraction of square roots to the capabilities of a machine. After
studying the largely unsuccessful attempts of Sir Samuel Morland (1625-1695),
master of mechanics to King Charles || of England, Leibniz was able to perfect a
multiplying machine based on repetitive additions, an algorithm still used in modern
computers. '2
Leibniz recognized that the complicated mechanisms of his calculator could
be greatly simplified if the decimal system were replaced with a binary notation."°
Leibniz's contemporaries resisted the idea, but his writings on binary arithmetic and
logic were the inspiration, almost two centuries later, for George Boole (1815-1864)
Leibniz's stepped-drum mechanism.
(Courtesy of IBM Corporation and
to develop the theory of binary logic and arithmetic, still the basis of modern
Neuhart Donges Neuhart Designers) computation. '*

164
+ Charles Babbage and the World's First Programmer

It is not a bad definition of man to describe him as a tool-making animal. His earliest contrivances to
support uncivilized life were tools of the simplest and rudest construction. His latest achievements in the
substitution of machinery, not merely for the skill of the human hand, but for the relief of the human
intellect, are founded on the use of tools of a still higher order.
Charles Babbage

One evening | was sitting in the rooms of the Analytical Society at Cambridge...with a table of logarithms
lying open before me. Another member, coming into the room, and seeing me half asleep called out,
“Well, Babbage, what are you dreaming about?” to which | replied, “| am thinking that all these tables
might be calculated by machinery.”
Charles Babbage

We may say most aptly that the Analytical Engine weaves algebraic patterns just as the Jacquard loom
weaves flowers and leaves.
Lady Lovelace

The Difference Engine

In 1821 Charles Babbage wrote a paper entitled “Observations on the Application of
Machinery to the Computation of Mathematical Tables.”'° His ideas were well
received, and he was awarded the first gold medal by the British Astronomical
Society, which hoped to use the technology Babbage had proposed to compute
astronomical tables.'° With funding from the Royal Society and the British govern-
ment, Babbage attempted to build his ambitious Difference Engine. He worked on
the project to the point of exhaustion, almost never seeing his children. The com-
plexity of the machine was on the edge of what was technically feasible, and it ex-
hausted Babbage’s financial resources and organizational skills. He ended up in a
dispute over ownership with the British government, had problems getting the
unusual precision parts fabricated, and saw his chief engineer fire all of his workmen
and then quit himself. He was also beset with personal tragedies, including the
deaths of his father, his wife, and two of his children.'7

The Analytical Engine

With the Difference Engine still not completed, Babbage was inspired by what he
regarded as another mental breakthrough. Rather than a machine that could only
perform specific calculations, he conceived of what he called the Analytical Engine,
which could be programmed to solve any possible logical or computational
problem.'® The Analytical Engine, although based entirely on the mechanical technol-
ogy of the nineteenth century, was a remarkable foreshadowing of the modern
computer
It had a random-access memory consisting of 1,000 “words” of 50
decimal digits each. In today’s terminology this is equivalent to about 175,000 bits. A
number could be retrieved from any location, modified, and stored in any other
location. It had a punched-card reader inspired by the Jacquard looms, automatic
weaving machines controlled by punched metal cards, which used a similar mecha-
nism.'? Babbage’s forward-looking design also included a printer, even though it
would be another 50 years before either typesetting machines or typewriters were
 The Jacquard loom punch card.
(Supplied by North Wind Picture
Archives)

Charles Babbage, a lonely and

obsessed man a century ahead of
his time. (Supplied by the Charles
Babbage Institute, University of
Minnesota)

The origin of the punch-card, The

Jacquard loom “remembered” the
pattern via a punched card that
directed the shuttles. (Supplied by
North Wind Picture Archives)
to be invented. It had an arithmetic “mill” (Babbage’s term) with registers that could
perform a variety of logical and arithmetic operations similar to the central processing
unit of a modern computer. Most important, it had a special storage unit for the
instructions, or program, with a machine language very similar to those of today’s
computers. One decimal field specified the type of operation, and another specified
the address in memory of the operand. Babbage recognized the critical importance
of the “conditional jump” instruction and provided for this capability.?°

Lady Ada Lovelace

Though Babbage was a lonely man obsessed with his vision of a programmable
computer, he developed a liaison with the beautiful Ada Lovelace, the only legitimate
child of Lord Byron, the poet. She became as obsessed as Babbage with the project
and contributed many of the ideas for programming the machine, including the
invention of the programming loop and the subroutine.'
Although Babbage was too busy to communicate his ideas to the rest of
the world, he allowed an associate, L. P. Menabrea (1809-1896), to describe the
machine and its principles. Lovelace translated Menabrea’s paper from French to
English and in her added notes (which were longer than the original paper) extended
Babbage’s ideas. She included in these notes extensive discussions on programming
techniques, sample programs, and the potential of this technology to emulate
intelligent human activities.22 She describes the speculations of Babbage and herself
on the capacity of the Analytical Engine and machines like it to play chess and
compose music. She finally concludes that though the computations of the Analytical
Engine could not properly be regarded as “thinking,” they could nonetheless perform
activities that would otherwise require the extensive application of human thought.”?
Ada Lovelace is regarded as the world’s first computer programmer and
has been honored by the United States Defense Department, which named its
primary programming language, Ada, after her. She died a painful death from cancer
at the age of 36, leaving Babbage alone again to pursue his quest.
The mechanics required for his machine were so complex that it required
Babbage to make significant advances in the machinist’s craft. Despite his ingenious
constructions and exhaustive effort, neither the Difference Engine nor the Analytical
Engine were ever completed. Near the end of his life he remarked that he had never
had a happy day in his life. Only a few mourners were recorded at Babbage’s funeral
in 1871.74
It is possible that if Babbage had had more enthusiastic sponsorship during
his life, his efforts may have borne fruit, but the Analytical Engine is generally re-
garded as having been beyond the means of nineteenth-century engineering to
realize. J. W. Mauchly (1907-1980) has commented that Babbage might have
succeeded had he been willing to “freeze” his design, but his continual attempts to
improve it doomed the project. Babbage’s concepts would be implemented seventy
years later, when the first American programmable computer, the Mark |, was
completed in 1944 by Howard Aiken (1900-1973) of Harvard University and IBM,
using an architecture very similar to Babbage’s.?° Babbage was a man distinctly
ahead of his time. Despite his failure to complete the implementation of his

167
 machine, his concepts of a stored program, self-modifying code, addressable
memory, conditional branching, and computer programming itself still form the basis
of computers today.

+ The Practical Path

This apparatus works unerringly as the mills of the gods, but beats them hollow as to speed.
The Electrical Engineerin a review of Hollerith’s Tabulating Machine

What is a computer?
| would define a computer as a machine capable of automatically performing (that is,
without human intervention) sequences of calculations, and of choosing between
alternate sequences of calculations based on the results of earlier calculations. The
description of the sequence of calculations to be performed, which includes all
alternate paths and the criteria for choosing among them, is called a program. A
programmable or general purpose computer is one in which we can change the
program. A special purpose computer is one in which the program is built-in and
unchangeable. A calculator is distinguished from a computer by its inability to
perform more than one (or possibly a few) calculations for each human intervention
and its inability to make decisions to choose from among multiple paths of computa-
tion. With the advent of today’s programmable calculators, the distinction between
calculators and computers has become blurred. The distinction was clear enough in
the 1940s: calculators were generally capable of only a single calculation for each
manually entered number. Tabulating machines, such as sorters, were capable of
multiple calculations (of a certain limited kind), but were not able to alter the se-
quence of computations based on previous results.” Note that these definitions say
nothing about the underlying technology, which, at least in theory, might be me-
chanical, electronic, optical, hydraulic, or even flesh and blood. Indeed, the era of
practical computation began not with electronics but with mechanical and electrome-
chanical automata

The age of calculating machines

Economic expansion in the late nineteenth century, made possible by innovations in
manufacturing and transportation, created a demand for the efficient and accurate
calculation of numbers. The same painstaking and error-prone process of human
calculation that originally motivated Babbage was threatening to block further eco-
nomic and scientific progress. Demands for census data were growing more
complex and an emerging interest in social research provided additional motivations
for an entire generation of inventors.?”
Yet producing a truly reliable mechanical calculator proved to be a daunting
task. Many ingenious devices were created, but like the Pascaline and Leibniz’s
Stepped Reckoner of earlier centuries, most were plagued by severe problems of
accuracy and reliability.2* One of the inventors selling unreliable calculators was
Lady Ada Lovelace, the world’s first
William Burroughs (1857-1898). His first production run of fifty Adding and Listing
computer programmer. (Supplied by
the Charles Babbage Institute, Machines quickly sold out, but all had to be recalled because of inconsistent
University of Minnesota) performance.”? Burroughs persisted, however, and after years of exhausting work,

169
Mechanical foot

AHollerith data processing installa-

tion (1918). (Photo by Howe & Arthur;
supplied by IBM Archives)
the world’s first dependable key-driven calculator was brought to market. The
machine had begun to receive widespread acceptance by the time of the inventor's
death from tuberculosis in 1898.°° While Burroughs’s machine was substantially
simpler than Babbage’s, and not programmable, the perfection of mechanical
calculation succeeded by the early twentieth century in transforming the conduct of
business, government, and scientific investigation

The age of tabulating machines

The taking of the national census every ten years is required by the U.S
Constitution, and by 1890 the Census Bureau was in a crisis. The 1880 census had
been completed only a couple of years earlier. The population had burgeoned from a
new wave of European immigration, and commitments had been made to Congress
to provide extensive new types of information about the population. It looked as if
the 1890 census would still be in full swing when it would be time to start again in
1900.5!
To address the problem, a competition was held. Among the many innova-
tive ideas for more efficient manual tabulation were schemes using multicolored
cards and coded paper chips. However, no manual method prevailed. The winner
was Herman Hollerith (1860-1929), a young Census Bureau engineer.*? Inspired by
an off-hand suggestion of his supervisor and borrowing the idea of using holes
punched in cards to represent information from Babbage’s Analytical Engine and the
loom of Joseph-Marie Jacquard (1754-1834), Hollerith was able to demonstrate a
solution at least eight times faster than the other finalists in the competition.°° The
key to Hollerith’s breakthrough was a paper card remarkably similar to modern punch
cards. As with modern computer cards, one corner was clipped so that the card
orientation could be quickly determined. Each card contained 288 locations for
possible holes to represent up to 288 bits of information.** The equipment included a
keypunch machine for encoding information and a card reader called a pin press. The
latter consisted of 288 spring-loaded rods, each of which would make contact with a
small container of mercury to complete an electric circuit if the corresponding hole
was punched. The machine provided for multiple counters and logic circuits that
could respond to relatively complicated patterns of information. The equipment
turned out to be relatively fast and reliable: cards could be passed through at a rate
exceeding one per second.*>
The 1890 census represented the first time that electricity was used for a
major data-processing project. The punched card itself survived as a mainstay of
computing until quite recently. In view of the scale of the project, Hollerith’s innova-
tion was implemented with remarkably few problems. Despite a 25 percent increase
in population from the 1880 census (to 63 million people) and a dramatic increase in
the complexity of the analysis provided, Hollerith's electromechanical information
machine did the job in only two and a half years, less than a third of the time
required for the previous census.
With the success of the 1890 census in hand, Hollerith set up the Tabulat-
ing Machine Company in 1896.°” The business grew rapidly, with his tabulators being
used for a variety of business applications, census analysis in numerous European
countries, and even for Russia's first census in 1897.°° For the 1900 U.S. census

171
Mechani al foots
Hollerith introduced another innovation, an automatic card feed.°? This turned out to
be Hollerith’s last contract with the Census Bureau. A dispute arose over rental
charges, and for the 1910 census, the Census Bureau once again sponsored an
internal project to develop an alternative technology. The result was improved
tabulating equipment and another commercial concern called the Powers Accounting
Machine Company.*?
The two concerns became fierce competitors, a competition which lasted
well into the modern era. Hollerith's company, the Tabulating Machine Company,
acquired several other firms by 1911 to form the Computing-Tabulating-Recording
Company (CTR).*! In 1914, offering a salary of $25,000 and a stock option for 1,220
shares, CTR hired a forty-year-old executive named Thomas J. Watson (1874-1956),
Would you buy stock in this com- who had built a strong reputation for aggressive marketing strategies at National
pany? This IBM promotional piece
from the 1930s features electric
tabulating and accounting machines
based on Hollerith’s patents.
(Supplied by IBM Archives)

A FEW OF IBM’S MORE THAN 700 AIDS TO MODER

 The Hollerith Tabulating System
(1890). “I was traveling in the West
and I had a ticket with what | think
was called a punch photograp!
the conductor... punched out a
description of the individual, as light
hair, dark eyes, large nose, etc. So
you see, | only made a punch
photograph of each person”—
Herman Hollerith. (Courtesy of IBM
Corporation and Neuhart Donges
Neuhart Designers)

ERPRISE
Cash Register.*? After only three months Watson was named president, and over the
COUNTING and TABULATING
et of secoting aed satin next six years the firm’s revenues more than tripled from $4 million to $14 million. In
mative toda of Eletic and
Machines, Electric Tabulation
gDchines, Alphabetic Accounting 1924 Watson was named chief executive officer, and he renamed the company
Accounting Machine. the Cor
International Business Machines (IBM), a reflection of Watson‘s ambition and
confidence.**
The Powers Accounting Machine Company also went through a series of
mergers by 1927 to become Remington Rand Corporation, which merged with
Sperry Gyroscope to become Sperry-Rand Corporation in 1955. Sperry-Rand was one
of IBM's primary competitors when computers really took off in the late 1950s.**

The mechanical roots of computation

As we have seen, automatic computation did not start with the electronic computer
The architecture of the modern computer, as well as some of the major players of
today’s computer industry, have their roots in the mechanical and electromechanical
Is
automata of the late nineteenth century. The design and computational theory
inherent in Babbage’s unrealized programmable computer provided the inspiration for
the first realized programmable computers, those of the twentieth century (although
not the first one). Babbage’s ideas were conceived in terms of mechanical techno!-
ogy and only realized when electromechanical and later all-electronic technology was
perfected during World War II. Howard Aiken, the developer of the first American
programmable computer, commented, “If Babbage had lived seventy-five years later,
| would have been out of a job.”*° The computer industry itself was also conceived
in an era of largely mechanical technology. It had to wait for modern electronics to
provide the price-performance ratio required for it to flourish. It is interesting to note
that Sperry Rand, the company to introduce the first commercial computer, as well

as IBM, the modern industry's leader and one of the largest industrial corporations in
the world, were both spin-offs of the U.S. Census Bureau.“

173
Electronics did not start out small:
the first transistor (1947). (Courtesy of
AT&T Archives)

oy
 Electronic Roots

+The First Computer

The German aircraft is the best in the world. | cannot see what we could possibly calculate to
improve on.
A German officer explaining to Konrad Zuse why the Third Reich would provide no further support for development of
Zuse's Z series of computers

| won't say that what Turing did made us win the war, but | dare say we might have lost it without him.
|. J. Good, assistant to Turing

| remember when we first got pulses circulating rather precariously. Every time we switched on the light
in the room, another pulse went in and that was the only input that in fact we had. Then later we made a
half-adder work and saw our binary pulses counting up on an oscilloscope screen. Then we got the
accumulator working and then the whole arithmetic unit, and so on.
We were a small, informal group in those days and whenever we had anything to celebrate,
we adjourned to the local pub. It was a pub well known to generations of workers in the Cambridge
University Laboratories. It was known as the Bun Shop, though | think they'd be very surprised indeed if
you tried to buy a bun there.
We had a special celebration on the 6th of May, 1949, when the EDSAC did its first calcula-
tion, which was a table of squares, and a few days later it did a table of primes.
Sam Alexander has reminded me about the first substantial program | wrote which was for
computing Airy functions by solving the Airy differential equation, and in connection with that, | made a
discovery. | discovered debugging. Now you probably won't believe me and | don't know whether my
colleagues around this table had the same experience, but it just had not occurred to me that there was
going to be any difficulty about getting programs working. And it was with somewhat of a shock that |
realized that for the rest of my life | was going to spend a good deal of my time finding mistakes that|
had made myself in programs.
Maurice V. Wilkes

Unlike such other epoch-making inventions as the telephone or the electric lightbulb,
most people cannot recite who invented the digital computer. One reason is that
who invented the computer is a matter of how one defines “computer.” Another is
that by most accepted definitions the first functioning computers were developed in-
dependently and virtually at the same time in three different countries, one of which
was at war with the other two.

ne C—s—“C:SstsSC“‘ ‘LCC
 It is clear, however, that during World War II the computer was an idea
whose time had come. The theoretical foundations had been laid decades earlier by
Babbage, Boole, Russell, and others, and many design principles had been worked
out by Babbage, Burroughs, Hollerith, and their peers.’ With the advent of reliable
electromechanical relays for telecommunications and electronic (vacuum) tubes for
radio communication, the building blocks for a practical computer had finally become
available

The German Z machines

It appears that the world’s first fully programmable digital computer was invented in
Germany during the Third Reich by a civil engineer named Konrad Zuse.? Zuse’s
original motivation was to automate what he later called those “awful calculations”
required of civil engineers.? He saw the computational bottleneck as bringing his field
of applied technology to a virtual standstill. With little theoretical background (Zuse
was unaware of the work of Babbage and other developments that were influential
in precomputer history) and great skill as a tinkerer, Zuse set out to ease the
unenviable task of calculating the huge tables of numbers required for civil engineer-
ing projects.
His first device, the Z-1 (the names of Zuse’s inventions originally used the
letter V, but this was later changed so as not to cause confusion between his ma-
Konrad Zuse, inventor of the world’s chines and the German rocket bombs), was an entirely mechanical calculator built
first operational programmable
from an erector set in his parents’ living room. The Z-2 was a more sophisticated
digital computer. (Courtesy of C. W.
Publishing Company) machine using electromechanical relays, and though not programmable, it was
capable of solving complex simultaneous equations. The Z-2 attracted the attention
of military aircraft developers, and Zuse received some level of support for further
development, although this was informal and not at a high level
Zuse’s most significant invention is the Z-3, the world’s first programmable
digital computer, completed in late 1941. It had a memory of 1,408 bits organized as
64 words of 22 bits each. In addition to 1,408 relays to support the random access
memory, it used another 1,200 relays for the central processing unit. As with all of
the relay-based computers, it was quite slow: a multiplication took more than 3
seconds.
Besides its importance as a fully functioning program-controlled computer,
it employed a number of other innovations. Zuse reinvented reverse Polish notation,
a method of expressing any formula no matter how complex by successive pairs of
numbers and operators, originally developed two decades earlier by the Polish
mathematician Jan Lukasiewicz.‘ It also introduced floating-point numbers, a form of
notation that can represent a broad range of magnitudes. Zuse went on to develop in
1945 the first high-level language, called Plankalkul, which foreshadowed modern
programming languages such as C in its lack of a GoTo statement and structured
methodology. The first programmer of the Z-3, and thereby the world’s first program-
mer of an operational programmable computer, was August Fast, a man chosen by
Zuse for both his talent as a mathematician and his blindness. Zuse apparently
reasoned that being blind, Fast would not be called for military service!
A variation of Zuse’s Z-3 was apparently used by the Germans in the
design and manufacture of one of the Nazi's flying bombs, the HS-293, although

176
Zuse has denied this.® After the war British Intelligence classified Zuse as an “ardent
Nazi,” which he has not convincingly denied.® What is perhaps most interesting
about the Zuse computers is the lack of importance given to them by the Nazi war
machine. The German military gave immensely high priority to several advanced
technologies, such as rocketry and atomic weapons, yet they seem to have put no
priority at all on computers. While Zuse received some incidental support and his
machines played a minor military role, there was little if any awareness of computa-
tion and its military significance by the German leadership. The potential for using
computers for a broad range of military calculations from plotting missile trajectories
and weapon designs to decrypting intelligence messages seems to have escaped
Nazi attention, despite the fact that the technology was first developed under their
aegis.’ The motivation for developing the world’s first programmable computer came
primarily from Zuse’s own intense focus as an inventor.®
Nor did anyone in other countries pay much attention to Zuse’s work
Credit for the world’s first programmable computer is often given to Howard Aiken,
despite the fact that his Mark | was not operational until nearly three years after the
Z-3. Since the computer industry developed largely in the United States after the
war, one might expect hesitation to recognize a German, an accused Nazi, as the
inventor of the programmable computer. On the other hand, allied pride did not stop
the United States from harnessing German know-how in the development of the
rocket, nuclear weapons, or atomic energy. Zuse, on the other hand, was largely
ignored by the Americans, just as he had been by the Germans: both IBM and
Remington Rand turned down offers of assistance and technology rights from Zuse
after the war. He started his own company in Germany, which continued to build
relay-based Z-series machines, although never in large quantities. The last, the Z-11,
is still being used.?

Ultra
Unlike the Germans, the British did recognize the military value of automatic compu-
tation, at least for the decryption of intelligence messages. As recounted earlier, the
ability of the machines built by Alan Turing and his associates to provide a constant
stream of decoded German military messages was instrumental in turning the tide of
war. The size of the effort, code-named Ultra, consisting of almost 10,000 men and
women, is a testament to the strategic priority given to computers by the British
during the war.'° Ultra created two series of machines. Robinson, completed in early
1940 and based on electromechanical relay technology, was powerful enough to
decode messages from the Germans’ first-generation Enigma enciphering machine
Was Robinson, completed in early 1940, the world’s first operational computer?
According to our definition, Robinson was a computer, although not a
programmable one: it had a single hard-wired program. We can consider it, therefore,
to be the world’s first operational computer. When the Germans increased the
complexity of their Enigma machine by adding additional coding rotors, Robinson
was no longer fast enough, and the Ultra team set out to build a computer using
electronic tubes, which were a hundred to a thousand times faster than the relays
used in Robinson. The new machine, called Colossus, required 1500 tubes, a
significant technical challenge in view of the short life span and lack of reliability of

177
vacuum tubes.'’ Colossus worked reliably nonetheless and was able to keep up with
the increasing complexity of the German messages. The Ultra group considered
Colossus to be the world’s first electronic computer, although they were unaware of
the earlier efforts of John Atanasoff, an obscure American inventor.

The Americans
Working for five years, a team of Harvard and IBM scientists led by Howard Aiken, a
navy commander, completed in 1944 what they thought was the world’s first |
programmable computer.'? Many still consider their creation, the Mark |, the first
general-purpose computer, despite Zuse’s having built such a machine, still awk-
ward, but three years earlier. The Americans were obviously unaware of Zuse (as we
have seen, the Germans were hardly aware of him), and they were only dimly aware
of the British efforts. The American high command did know that the British were
using electromechanical and electronic equipment to decode German messages, but
did not have details of the process. Turing visited the United States in 1942 and ap-
parently talked with John von Neumann (1903-1957), a mathematician, but had little
impact on the American development of the computer until after the war, at which
time there was extensive American-English discussion on the subject of computa-
tion.'? The formal name for the Mark |, the IBM Automatic Sequence Controlled
Calculator, is reminiscent of the original names of other pivotal inventions, such as
“wireless telephone” for the radio and “horseless carriage” for the automobile.'*
The United States Navy made Aiken available to the Mark | project, as they
realized the potential value of an automatic calculator to a wide range of military
problems. Realizing the potential commercial value of the machine, IBM provided
virtually all of the funding, $500,000. Harvard's motivation was to transform the
tabulating machine into a device that could perform scientific calculations. The
resulting machine was enormous: fifty feet long and eight feet high. It was affection-
ately called “the monster” by one of its first programmers, Navy Captain Grace
Murray Hopper.'®
The machine used decimal notation and represented numbers with 23
digits. Input data was read from punched cards, and the output was either punched
onto cards or typed. The program was read from punched paper tape and was not
stored but rather executed as read
Grace Murray Hopper has sometimes been called the Ada Lovelace of the
Mark |. Just as Ada Lovelace pioneered the programming of Charles Babbage’s
Analytical Engine (despite the fact that the Analytical Engine never ran), Captain
Hopper was the moving force in harnessing the power of the Mark |. On assignment
from the Navy and having lost her husband in the war, she devoted most of her
waking hours to programming the Mark | and its successors, the Mark II and Mark
Ill. She was one of the first to recognize the value of libraries of subroutines, is
credited with having written the first high-level language compiler, and led the effort
to develop the Common Business-Oriented Language (COBOL), the first language
not identified with a particular manufacturer.'® She is also associated with the origin
of the term “debug.” One problem with the Mark | was fixed by removing a moth
that had died inside one of its relays. “From then on,” she recalls, “whenever

178
Grace Murray Hopper, the “Ada
Lovelace” of the Mark |. “I got into
the computer business by being
ordered straight from Midshipman
Schaol to the Harvard computer, to be
greeted by a large and appalling
Commander known as Howard Aiken,
and to be—three hours after | arrived
there—instructed in the art of
programming the Mark | computer by
a very tall and definite young
Ensign known as Richard Bloch. He
frustrated me quite regularly because
even then | wanted to keep my
software and use it over again. |
didn’t want to keep reprogramming
things. But unfortunately, every time |
got a program running, he'd get in
there at night and change the circuits
in the computer and the next morning
the program would not run. What's
more, he was home asleep and
couldn't tell me what he had done to
the computer”—Grace Murray
Hopper. “One reason, maybe, that
information didn’t travel so rapidly
during World War II was that Mark |
was operating for the Navy and
ENIAC was operating for the Army
and we were all very classified and
the Army didn’t talk to the Navy very
much."—Grace Murray Hopper. (Sup-
plied by the UNISYS Corporation)

The first bug. Hopper taped this dead go

moth to a page in the Mark |
computer's log book and wrote, “First
actual case of bug being found.” 4/49
(Supplied by Photri)
Se On tan stack oe L700 «7.037 sy? OUT
° byt ~ autem 7 F087 FY IPS cor
Ive He -Ac Sas ee Y-OISF2SOG
Cr
D3 Pro ێ 2 thogreyis
cow 2 (3067CmS
Rdoys b-t —~ 033 J sos feb
im Tuto Wim 4o4 -
2a Rae
ve arted C = lap (Sine he vk}

iS4y 2

179
 Aiken and IBM's Mark |, the first
American programmable computer
(1944). (Courtesy of IBM Corporation)

181
The ENIAC “super brain,” the world’s
first electronic programmable
computer. (Courtesy of the Computer
Museum, Boston)
anything went wrong with the computer, we said it had bugs in it. If anyone asked if
we were accomplishing anything, we replied that we were ‘debugging’.”'
The Mark | generated considerable interest in many quarters, but a severe
limitation soon became apparent. With its electromechanical relay technology, it was
capable of only a few calculations each second. The only real solution was to build a
programmable computer without moving parts, which was accomplished when two
professors at the University of Pennsylvania's Moore School of Electrical Engineer-
ing, J. Presper Eckert, Jr. (b. 1902) and John W. Mauchly (1907-1980), completed
work on their famous ENIAC (Electronic Numerical Integrator and Computer) in 1946.
Using 18,000 vacuum tubes (12 times as many as Colossus, which had been
considered ambitious only five years earlier!), it was capable of 5,000 calculations per
second, making it almost a thousand times faster than Aiken’s Mark |. Early press
coverage included reports of the ENIAC accomplishing in a matter of hours what
would have required a year for a hundred scientists. It was from these reports that
terms such as “super brain” and “electronic genius” first became associated with
computers. Conceived in 1943, the ENIAC was originally intended for military
applications; however, it was not completed in time for service in World War II. It
ultimately served a wide variety of both military and civilian purposes, including
physics studies, designing atomic weapons, numerical analysis, weather prediction,
and product design.'®
ENIAC was undoubtably the world’s first fully electronic general-purpose
(programmable) digital computer, although this claim, as well as the validity of the
patents for ENIAC, was challenged in 1973 by John V. Atanasoff (b. 1903), a profes-
sor at lowa State College. Judge Earl Larson found in favor of Atanasoff, stating that
“Eckert and Mauchly did not invent the automatic electronic digital computer, but
rather derived the subject matter from Atanasoff.”'? Working with a part-time
graduate student Clifford Berry, Atanasoff did indeed build an electronic computer,
called ABC (for Atanasoff-Berry Computer), in 1940.2° Although a relatively unsophis-
ticated device with only 800 tubes, it could compute simultaneous linear equations
using binary representations of the variables. It was revealed during the trial that
Mauchly had visited Atanasoff while the ENIAC was being designed. The judge,
however, misunderstood the distinguishing feature and primary claim of the ENIAC,
which was its ability to be programmed—Atanasoff's ABC was not programmable.
Fortunately, despite the court's finding, credit for building the first computer that was
both programmable and electronic continues properly to be given to Eckert and
Mauchly. This is a prime example of the challenges faced by the courts in adjudicat-
ing issues with substantial technical content, particularly at times when the issues
have yet to be fully understood by the technical community.
The ENIAC represented the first time that the public became fascinated
with the potential for machines to perform mental functions at “astonishing”
speeds. Although many of today’s computers are a thousand times faster than
ENIAC (and our fastest supercomputers are almost a million times faster), the speed
problem exhibited by the Mark | seemed to have been solved by ENIAC. Yet this led
to yet another bottleneck. Loading a program into the ENIAC involved setting 6,000
switches and connecting hundreds of cables. The incredible tedium required to
change the program gave birth to a new concept called the stored program and to an

183
 John V. Astanasoff, a litigious inven-
tor, and his ABC computer. (Courtesy
of lowa State University)

184
 architecture that computers still use today.?! The stored program concept is generally
associated with the great Hungarian-American mathematician John von Neumann,
and we still refer to a computer with a single central processing unit (the primary
computational engine of a computer) accessing its program from the same memory
(or same type of memory) that holds the data to be manipulated as a von Neumann
machine. Although the first modern paper on the subject, published in 1946, carried
only von Neumann's name, the idea was not his alone. It clearly resulted from
discussions that von Neumann had with both Eckert and Mauchly. In fact, the first
reference to the stored program concept can be found in Babbage’s writings a
century earlier, although his Analytical Engine did not employ the idea
The idea of a stored program is deceptively simple: the program is stored
in the machine’s memory in the same way that data is stored.”? In this way the
program can be changed as easily as reading in new data. Indeed, the first stored
program computers read in their programs from the same punched-card readers
used to read in the data. A stored program is more than a convenience, it supports
certain types of algorithmic methods that would otherwise be impossible, including
the use of subroutines, self-modifying code, and recursion, all of which are essential
capabilities for programming most Al applications. It is hard to imagine writing a
chess program, for example, that does not make use of recursion

The von Neumann machine and the

stored program.

“P's represent
the Program
D's represent
the Data

185

EE ee”
Electra
Eckert and Mauchly set out to implement the stored-program concept in a
computer to be called EDVAC (Electronic Discrete Variable Automatic Computer)
while they were still at the Moore School. They subsequently left Moore to start
their own company, the Eckert-Mauchly Computer Corporation, and the completion
of EDVAC was left to others. This delayed the project, and it was not until 1951 that
EDVAC was completed. As a result, though EDVAC was the first computer designed
to incorporate stored-program capability, it was not the first to be completed. Eckert
and Mauchly’s new company brought out BINAC (Binary Automatic Computer) in
1949. BINAC was not the world’s first stored-program computer, either. Maurice
Wilkes, a professor at Oxford University, had taken a course on computers with
Eckert and Mauchly while they were still teaching at the Moore School. Returning to
England after completion of the course, he set out to build his own computer incor-
porating the stored-program concept that he had learned from Eckert and Mauchly’s
course. Wilkes’s machine, called EDSAC (Electronic Delay Storage Automatic
Computer), was completed a few months before BINAC, making /t the world’s first
operational stored-program computer.”?

The commercialization of computation

After building only two BINACs and without adequate financial backing to commer-
cially develop a complex new technology, Eckert and Mauchly sold their firm to
Remington Rand in 1950. By 1951 a computer called UNIVAC (Universal Automatic
Computer), which Eckert and Mauchly had started several years earlier, was intro-
duced as the world’s first commercially marketed electronic computer. The first
customer, appropriately enough, was the U.S. Census Bureau; the 1950 census was
the first to be handled by a programmable computer
Legend has it that Remington-Rand conducted a marketing study for their
new product, which indicated a potential worldwide market for only 50 computers.
This story is difficult to confirm, but it is true that UNIVAC was not a commercial
success. IBM, however, took the introduction of UNIVAC very seriously, viewing it
as a threat to their dominant position in tabulating equipment, and shortly thereafter
introduced the first of their 700 series of computers (the 701, 702, 704, and 705)
The 700 series became so popular that IBM was quickly positioned to dominate the
still nascent industry. The 700 series led to IBM's 1400 series (1401 and 1410) and
7000 series (7030, 7040, 7070, 7074, 709T for transistors, 7090, and 7094), which
became standards for large corporate customers and scientific installations. With the
introduction of IBM's 360 series in 1964, a computer architecture based on 32-bit
From vacuum tube to the transistor,
the beginning of miniaturization. words that continues to form the basis for IBM mainframes, IBM solidified its
(Photo by Dan McCoy of Rainbow) leadership of the computer industry and became one of the largest and most
profitable corporations in the world.”*

186
 This machine can keep track of your
recipes. J. Presper Eckert explains
his and John W. Mauchly's UNIVAC |
to Walter Cronkite. (Supplied by the
UNISYS Corporation)

187
Firsts in Computers

Name Inventor Completed Sponsor Hardware Programmable Stored program The first

Analytical Babbage Conceived in Originally the All Yes , No First program-

Engine 1835, never British gov't mechanical (punched cards) mable computer
completed although ever designed
support was
withdrawn

Robinson Turing et al Early 1940 British gov't Relays No No First operational

computer
(special purpose) ]
ABC Atanasoff 1940 lowa State Tubes & No No First electronic
(Atanasoff-Berry Research capacitors computer

Computer) Council (special purpose)

Z3 Zuse Late 1941 Partially Relays Yes No First program-
(Zuse-3) sponsored by mable computer
the German (actually built)
Aircraft Re-
search Institute

Colossus Turing et a 1943 British gov't Tubes No No First English

electronic
computer

Mark| Aiken 1944 IBM Mostly Yes No First American

(IBM Automatic relays (punched tape) programmable
Se computer
Controlled
Calculator)
ENIAC Eckert & 1946 U.S. Army Tubes Yes (patch No First electronic
(Electronic Mauchly cords & programmable
Numerical switches) computer
Integrator and
Computer)
EDSAC Wilkes Early 1949 Cambridge Tubes for Yes Yes First stored-pro-
(Electronic (based on infor- University logic gram computer
Delay Storage mation from mercury delay
Automatic a course given lines for
Computer) by Eckert & memory
Mauchly)

BINAC Eckert & 1949 Eckert- Tubes Yes Yes First American
Binary Mauch Mauchly stored-program
matic Computer computer
Computer) Corp.
EDVAC Begun 1951 Moore Tubes Yes Yes First stored-
(El by Eckert, School program
Discrete Mauchly, and computer
Variab von Neumann conceived
Autor, completed by
Computer) others at Moore
School

UNIVACI Eckert & 1951 Remington Tubes Yes Yes First

(Ur Mauchly Rand commercially
produced elec-
tronic computer
1BM 701 Nathaniel 1952 IBM Tubes First
Rocheste commercially
successful
computer

188
+ Welcoming a New Form of Intelligence on Earth: The Al Movement

Since Leibniz there has perhaps been no man who has had.a full command of all the intellectual activity
of his day... . There are fields of scientific work ... which have been explored from the different sides
of pure mathematics, statistics, electrical engineering and neurophysiology; in which every single notion
receives a separate name from each group, and in which important work has been triplicated or quadru-
plicated, while still other important work is delayed by the unavailability in one field of results that may
have already become classical in the next field.
Norbert Wiener, Cybernetics

Humans are okay. I'm glad to be one. | like them in general, but they're only human. . .. Humans aren't the
best ditch diggers in the world, machines are. And humans can't lift as much as a crane. . . . It doesn't
make me feel bad. There were people whose thing in life was completely physical—John Henry and the
steam hammer. Now we're up against the intellectual steam hammer. . . . So the intellectuals are
threatened, but they needn't be. .. . The mere idea that we have to be the best in the universe is kind of
far fetched. We certainly aren't physically.
There are three events of equal importance. . . . Event one is the creation of the universe. It's
a fairly important event. Event two is the appearance of Life. Life is a kind of organizing principle which
one might argue against if one didn’t understand enough—it shouldn't or couldn't happen on thermody-
namic grounds. . . . And third, there's the appearance of artificial intelligence.
Edward Fredkin

There are three great philosophical questions. What is life? What is conciousness and thinking and
memory and all that? And how does the universe work? The informational viewpoint encompasses all
three. .. . What I'm saying is that at the most basic level of complexity an information process runs what
we think of as physics. At the much higher level of complexity life, DNA—you know, the biochemical
functions—are controlled by a digital information process. Then, at another level, our thought processes
are basically information processii
| find the supporting evidence for my beliefs in ten thousand different places, and to me it's
just totally overwelming. It's like there’s an animal | want to find. I've found his footprints. I've found his
droppings. I've found the half-chewed food. | find pieces of his fur, and so on. In every case it fits one
kind of animal, and it's not like any animal anyone's ever seen. People say, Where is this animal? | say,
Well, he was here, he’s about this big, this that and the other. And | know a thousand things about him. |
don’t have him in hand, but | know he’s there. ... What I see is so compelling that it can't be a creature
of my imagination.
Edward Fredkin, as quoted in Did the Universe Just Happen by Robert Wright

Fredkin . . . is talking about an interesting characteristic of some computer programs, including many
cellular automata: there is no shortcut to finding out what they will lead to. This, indeed, is a basic
difference between the “analytical” approach associated with traditional mathematics, including
differential equations, and the “computational” approach associated with algorithms. You can predict a
future state of a system susceptible to the analytic approach without figuring out what states it will
occupy between now and then, but in the case of many cellular automata, you must go through all the
intermediate states to find out what the end will be like: there is no way to know the future except to
watch it unfold... . There is no way to know the answer to some question any faster than what's going
on... . Fredkin believes that the universe is very literally a computer and that it is being used by
someone, or something, to solve a problem. It sounds like a good-news/bad-news joke: the good news is
that our lives have purpose; the bad news is that their purpose is to help some remote hacker estimate pi
to nine jillion decimal places.
Robert Wright, commenting on Fredkin’s theory of digital physics

189
With at least a dozen inventors having some credible claim to having been “first” in
the field of computation, we can say that the computer emerged not from a lone
innovator’s basement, but rather from a rich Period of intellectual ferment on several
continents drawing upon a diversity of intellectual traditions and fueled by the
exigencies of war. And if there were a dozen fathers of the computer, there were at
least a couple dozen fathers of Al.
The notion of creating a new form of intelligence on earth emerged with
an intense and often uncritical passion simultaneously with the electronic hardware
on which it was to be based. The similarity of computer logic to at least some
aspects of our thinking process was not lost on any of the designers of the early
machines. Zuse, for example, applied his Z series of computers and Plankalkul
language to the problem of chess. Turing’s contributions to the foundations of Al are
extensive and well known. Turing’s 1950 classic paper “Computing Machinery and
Intelligence” lays out an agenda that would in fact occupy the next quarter century
of research: game playing, natural language understanding and translation, theorem
proving, and of course, the cracking of codes.”° Even Nathaniel Rochester, the
designer of IBM's first successful computer, the 701, spent several years developing
early Al technology; in fact, he was one of the principals in the 1956 Dartmouth
Conference, which gave artificial intelligence its name.”°
Perhaps the odd man out was John von Neumann, who found himself
unable to imagine that the cumbersome and unreliable vacuum tubes used to build
the first electronic computers could ever compete with the human brain.2”? Even
though the transistor had been invented when von Neumann expressed his skepti-
cism, its applicability to computing had not yet been realized. More significant than
the hardware limitations, however, was the hopelessness, according to von Neu-
mann, of ever describing natural human actions and decision making in the precise
language of mathematics. To some extent, von Neumann may have been reacting to
the unrealistic expectations that had been set in the popular media of the time
magazine covers were predicting that superhuman electronic brains were just around
the corner. Von Neumann did, however, show considerable interest in the idea of
expressing human knowledge using the formalism of a computer language. The
depth of his resistance to the idea that at least some behaviors we associate with
natural intelligence might ultimately be automated is difficult to gauge; von Neumann
died in 1957, shortly before he was to give a series of lectures at Yale on the
likelihood of machine intelligence.”°

Cybernetics: A new weltanschauung

The emergence of the computer, its early application to cognitive problems, the
development of the theoretical foundations of computation, and related speculation
on the implications of this new technology had a profound impact on fundamental
tenets of what we might call the scientific worldview. In Cybernetics, Norbert
Wiener’s seminal book on information theory, Wiener describes three ways in which
the world’s (and his own) outlook had changed forever.?°
Wiener, a prodigy who had studied with both Bertrand Russell and David
Hilbert and received his Ph.D. from Harvard at the age of 18, had mastered an
unusually broad range of intellectual fields from psychology and neurophysiology to
mathematics and physics.*° His book was intended to establish as a new science the
field of cybernetics, which he defines in his subtitle as control and communication in
the animal and machine. In sections that are altérnately addressed to the lay reader
and to his fellow mathematicians, Wiener expounds his three premises
First is the change from energy to information. Precybernetic reality
consists of particles and the energy fields that control them. Accordingly, the old
model of life was concerned primarily with the conversion of energy in its various
biochemical and physical forms.*' If the living cell was a heat engine in the early
twentieth century, it had now become a small computer. The new cybernetic model
treats information as the fundamental reality in living things as well as in intelligent
things, living or otherwise. In this new view, the most important transactions taking
place in a living cell are the information-processing transactions inherent in the
creation, copying, and manipulation of the amino acid strings we call proteins. Energy
is required for the transmission and manipulation of information in both animal and
machine, but this is regarded as incidental.
In recent times, with information-handling circuits becoming smaller and
using ever smaller amounts of energy, energy has indeed become incidental to the
process. The primary issue today in measuring information-handling systems is the
amount and speed of their data processing capabilities. Edward Fredkin has recently
shown that though energy is needed for information storage and retrieval, we can
arbitrarily reduce the energy required to perform any particular example of informa-
tion processing, and there is no lower limit to the amount of energy required. In
theory, at least, the energy required to perform computation can become arbitrarily
close to zero. Fredkin’s argument is significant in establishing that information
processing is fundamentally different from energy processing. Even without
Fredkin's theoretical argument, the energy required for computation is not a signifi-
cant issue today, although it was a bit of an issue in Wiener’s day. he lights in
Philadelphia were reported to have dimmed when ENIAC, with its 18,000 vacuum
tubes, was turned on, although this story is probably exaggerated.
The decoupling of information and energy is also important from an eco-
nomic point of view. The value of many products today is becoming increasingly
dominated by computation. As computation itself becomes less dependent on both
raw materials and energy, we are moving from an economy based on material and
energy resources to one based on information and knowledge.*
A second aspect of what Wiener saw as an epochal change in scientific
outlook is characterized by the trend away from analog toward digital.*® Wiener
argues that computation ultimately needed to be “numerical rather than on the
basis of measurement, as in the Bush Differential Analyzer.” When computing first
emerged, there was a controversy between analog and digital computing, with the
Differential Analyzer of Vannevar Bush (1890-1974), President Roosevelt's science
advisor, as a popular example of the former.*6 In an analog computer, mathematical
Operations are performed by adding, subtracting, or even multiplying electrical
quantities and then measuring the amount of voltage or current after the operations
are performed. A very simple example is the common thermostat. Through the use
of feedback loops, fairly complex formulas can be represented, although accuracy is
limited by the resolution of the analog component In a digital computer or

191

Lc etc ;eLC
Vannevar Bush and his differential process, numbers are represented not by amounts of voltage or current but rather by
analyzer, an analog computer. Bush's assembling numbers from multiple bits, where each bit is either on (generally
mechanical invention, built in
1930 at MIT, was used to calculate representing 1) or off (generally representing 0). To many observers at the time, it
artillery trajectories during World was not clear which type of computing device would become dominant. Today we
War Il. Although Bush later built a
generally do not find the need to use the word “digital” before the word “com-
more modern version using electrical
switches, analog computation was puter,” since analog computers are no longer common, although in Wiener's time,
rendered largely obsolete by the this was a necessary modifier. Wiener saw the limitations of analog computing in
digital computer. (Courtesy of the MIT
terms of both accuracy and the complexity of the algorithms that could be imple-
Museum)
mented. With a digital computer there is no theoretical limit on either accuracy or

computational complexity.**
The trend from analog to digital, which was just getting started in Wiener's
day, continues to revolutionize a growing number of technologies and industries. The

compact digital disk and digital audio tape are revolutionizing the recording of music
Digital technology is replacing the piano and other acoustic and analog musical

instruments. A new generation aircraft is being controlled by highly accurate digital

ol mechan sms replacing analog and mechanical methods. Phones and copiers
are becoming digital. There are many other examples

However, the question of whether the ultimate nature of reality is analog

or digital continues to b an important philosophical issue. As we delve deeper and
d eeper into both natural and artificial processes, we find the nature of the process
often alternate analog and digital representations of information. Consider,
for example, thr related to listening to a compact-disk recording: the

192
 reproduction of the sound from the compact disk recording, the musical understand-
ing of the audio signal by the listener, and the nature of the data structures in music
itself.*°
First consider the reproduction process. The music is communicated to the
listener by vibrations of the air, which are clearly an analog phenomenon. The electri-
cal signals sent to the loudspeaker are also analog. The circuits interpreting the
contents of the compact disk are, however, digital circuits that are computing the
analog values to be played as sound. The digital values computed by the digital
Circuits are converted into analog electrical signals by a device called a digital-to-
analog converter. Thus, the sound exists in a digital representation prior to it being
converted to analog electrical signals to be amplified, which is why we consider
compact disk technology to be a digital technology.*'
Let us now look at these digital circuits. As in many modern electronic
products, these circuits are packaged into tiny integrated circuits containing thou-
sands or tens of thousands of transistors each. It is interesting to note that while the
Circuits are digital, the transistors that comprise them are inherently analog devices.
The designers of the transistors themselves do not consider the transistor to be a
digital device but rather are very aware of its inherently analog nature. The transistor
is analog because it acts as a small electrical amplifier and deals with variable (con- |
tinuous-valued) amounts of current.‘ They are “tricked” into acting digital (with
values of on or off) by thresholding (comparing values to a constant) their otherwise
analog (continuous-valued) characteristics. But a transistor cannot model a digital
device perfectly; it will compute the wrong digital value some small fraction of the
time. This is not a problem in practice. Using the very information theory created by
Wiener and an associate of his, Claude Elwood Shannon, digital circuit designers can
make the likelihood that the analog transistor will malfunction in its digital role so low
that they can essentially ignore this possibility and thus consider the transistor to be
a reliable digital element.
On a deeper level we can understand the continuous levels of current
within one transistor to be ultimately discrete: they are made up of a very large
number of individual electrons with individual quantum states. So now we are back
to a digital view of things. But when we consider the Heisenberg uncertainty
principle, which tells us that the electrons have no precise locations but only
probability clouds of possible locations, we are back to a fuzzy (analog) model
of the world.

Analog and digital representations of

information.

193
 Elements of Music
timbre
(tone colon

oxo’
vibrato
tremolo

Auditory analysis by
the acoustic cortex and
other parts of the brain
\ combine both digital
. and analog techniques.

' >
Cochlea
‘The Cochlea (in the inner
ear) produces a digitally
ue oa
‘encoded spectral analysis
Loudspeaker Sound: an analog vibration of the air —

Analog
Circuits e
=
el .
ie

ae cee Roieeines
a,
information

Digital crexits process the

‘formation

But Digtal circuits are made

of Analog transistors

ae
But Analog transistors
‘amply large numbers
of discrete (i. digital)
‘electrons

But ofscrete (aga)

| electrons have no precise a
Jocatons (according tothe
‘Heisenberg Uncertainty
‘Prncipe), but rather proba:
bitty “clouds of possible
‘ocations, again an analog
‘conceptionof realty
* * *

194
Listening to a compact disk recording: As we approach finer and finer models of the world, physics flips several
digital and analog representations of
7 times between the digital and analog conceptions of reality. Consider now the path
information.
of the music sounds going toward the listener.-The analog sound waves vibrate the
ear drum and ultimately enter the cochlea, a natural electronic circuit that acts as a
spectrum (pitch) analyzer. The cochlea breaks up the sound waves into distinct
frequency bands and emits a digitally encoded representation of the time-varying
spectrum of the sound. This digital representation enters the acoustic cortex of the
brain, where processing takes place by techniques that are both analog and digital.**
Finally, consider the nature of music itself with its elements of melody,
rhythm, harmony, expression, and timbre. The first three elements, which are the
elements represented by musical notation, are clearly digital in nature. The availability
of music notation processors, which are to music notation what word processors are
to written language, are a clear testament to the inherently digital nature of these
three musical elements. Melodic, rhythmic, and harmonic structures are modeled in
music theory by the mathematics of digital logic rather than the mathematics of
analog calculus. Expression and timbre, on the other hand, though they certainly can
be represented using digital means, are nonetheless analog in nature.*°
It is clear that we use both digital and analog approaches in understanding
the world around us. As we view phenomena at different levels of specificity and
detail, we find their nature changes repeatedly. The ultimate nature of reality is still
being debated. Edward Fredkin has recently proposed what he calls a new theory of
physics stating that the ultimate reality of the world is software (that is, information).
According to Fredkin, we should not think of ultimate reality as particles and forces
but rather as bits of data modified according to computational rules. If, in fact,
particles are quantized in terms of their locations and other properties, then the
views of the world as made up of particles and as made up of data are essentially
equivalent. Though Fredkin's view has startled contemporary theoreticians, it is just
another way of postulating a fully quantized or digital world. The alternative analog
view would state that the position or at least one of the properties of a particle is not
quantized but is either continuous or uncertain.
Regardless of the ultimate nature of the world, Wiener's prediction of a
digital conception of nature was both profound and prophetic. From a technological
and economic point of view, the digital approach continues to replace the analog,
transforming whole industries in the process. The digital outlook has also permeated
our philosophical and scientific views of the world
An additional comment on the issue of digital versus analog concerns
another contemporary controversy. Some observers have criticized digital computing
as a means of replicating human cognition because of its all or nothing nature, which
refers to the fact that digital bits are either on or off with no states in between. This,
according to these observers, contrasts with the analog processes in the human
brain, which can deal with uncertain information and are able to balance decisions
based on “soft” inputs. Thus, they say, analog methods that can deal with grada-
tions on a continuous scale will be needed to successfully emulate the brain
My reaction to this is that the argument is based on a misconception, but |
nonetheless agree with the conclusion that we will see a return to analog comput-
ing, alongside digital techniques, particularly for pattern-recognition tasks. The
 argument that digital techniques are all or nothing is clearly misleading. By building
numbers (including fractional parts) from multiple bits, digital techniques can also
represent gradations, and with a greater degree of continuity and precision than
analog techniques—indeed, with any degree of precision desired. New knowledge-
engineering techniques that use a method called fuzzy logic can apply digital comput-
ing to decision making in a way that utilizes imprecise and relative knowledge in a
methodologically sound manner. Often the criticism of digital computing is really
aimed at the type of “hard” antecedent-consequent type of logic employed in the 4
first generation of expert systems. Overcoming the limitations of this type of logic
does not require resorting to analog techniques.
There is, however, a good reason that we are likely to see a return to
hybrid analog-digital computing designs. Analog computing is substantially less
expensive in terms of the number of components required for certain arithmetic
operations. Adding or even multiplying two quantities in an analog computer requires
only a few transistors, whereas a digital approach can require hundreds of compo-
nents. With integrated-circuit technology providing hundreds of thousands or even
millions of components on a single chip, this difference is usually not significant and }
' the greater precision of digital computing generally wins out. But consider the trend
toward massive parallel processing, in which a system performs many computations
simultaneously rather than just one at a time, as is the case in the classical von
Neumann computer architecture. For certain types of problems, such as visual image
analysis, it would be desirable to be able to perform millions of computations at the
same time (computing the same transformations on every pixel (point) of a high
resolution image, for example). Often these calculations do not require a high degree
of accuracy, so performing them with analog techniques would be very cost effec- :
tive. Doing a million computations simultaneously with analog methods would be !
practical with today’s semiconductor technology but still prohibitively expensive with
digital techniques. Evolution apparently found the same engineering trade-off when it
designed the human brain.
The third major theme of Wiener’s treatise concerns the nature of time. He
argues that we have gone from a reversible or Newtonian concept of time to an
irreversible or Bergsonian notion of time. Wiener regards Newtonian time as revers-
ible because if we run a Newtonian world backward in time, it will continue to follow
Newton's laws. The directionality of time has no significance in Newtonian physics.
Wiener used Bergson, a biologist who analyzed the cause and effect relationships in
biology and evolution, as a symbol for the irreversibility of time in any world in which
information processing is important.
Computing is generally not time reversible, and the reason for this is /
somewhat surprising. There are two types of computing transformations, one in
which information is preserved and one in which information is destroyed. The
former type is reversible. If information is preserved, we can reverse the transforma-
tion and restore the information to the format it had prior to the transformation. If q
information is destroyed, however, then that process is not reversible, since the
information needed to restore a state no longer exists. Consider, for example, a
program that writes zeros throughout memory and ultimately destroys (most of)

196
Two types of computing trans-
formations: the preserver and the
destroyer.

itself. We cannot conceive of a way to reverse this process, since the original
contents of memory are no longer anywhere to be retrieved. This is surprising
because we ordinarily think of computation as a process that creates new informa-
tion from old. After all, we run a computer process to obtain answers to problems,
that is, to create new information, not to destroy old information. The irreversibility of
computation is often cited as a reason that computation is useful: it transforms
information in a unidirectional “purposeful” manner. Yet the derivation of the proof
that computation is irreversible is based on the ability of computation to destroy
information, not to create it. Another view, however, is that the value of computation
is precisely its ability to destroy information selectively. For example, in a pattern-
recognition task such as recognizing images or speech sounds, preserving the
invariant information-bearing features of a pattern while “destroying” the enormous
amount of data in the original image or sounds is essential to the process. In fact,
intelligence is often a process of selecting relevant information carefully from a much
larger amount of unprocessed data. This is essentially a process of skillful and
purposeful destruction of information.
Wiener points out that a model of time that is irreversible is essential to
the concepts of computation, communication, and intelligence. One entertaining
example cited by Wiener is the impossibility of two intelligent entities attempting to
communicate when they are “traveling” in different directions in time. Neither one
will perceive the other as either responsive or intelligent.
Energy to information, analog to digital, and reversible time to irreversible
time are all facets of the same revolution in worldview known as computation. |t
was a revolution both for the world at large and for Wiener himself. It is interesting
to note that virtually all of the mathematics in Wiener's book is not that of the
logician and computationalist but rather the calculus of the Newtonian physicist,
which comprises Wiener’s original scientific tradition.
There are other significant insights in Wiener's classic book. He gives five
principles that should, in his opinion, govern the design of computing machinery; a
computer should

197
* be digital rather than analog,
+ be electronic rather than electromechanical,
* use binary rather than decimal notation, :
+ be programmable,
* contain a dynamic (erasable) random access memory.

Wiener says in his Introduction that he had written a letter to Vannevar

Bush in 1940 suggesting these principles and commenting on their importance to
military requirements. Bush, being firmly in the analog camp, apparently did not
respond.
Wiener proposed using computing machinery to provide effective prosthe-
ses for the sensory impaired. By converting visual information into an appropriate
auditory modality for the blind, or alternatively converting auditory information into a
visual display for the deaf, electronic technology could help overcome the primary
handicap associated with these disabilities. lt would be several decades before
computers would begin to master these tasks. Wiener ends the 1948 version of
Cybernetics with a description of a recursive strategy for building a computer chess
machine. He predicts that such a machine would be capable of playing at a level that
would be “not so manifestly bad as to be ridiculous.”
A theme that is echoed throughout this book, as well as in other writings |
of Wiener’s, is the need for collaboration among multiple disciplines if the difficult |
challenge of a machine's mastering cognitive problems is to be addressed. He |
articulates a need to reverse the trend toward increasingly narrow specialization in |
the sciences and predicts that solving the major technology problems of the second i
half of the twentieth century would require knowledge and expertise to be drawn {
and integrated from diverse disciplines. The greatest difficulty in achieving this type \
of cooperation, Wiener feels, is the fact that experts from different fields all use |
different terminologies often to describe the same phenomena. Though Wiener's
prediction was largely overlooked for decades, Al researchers just now seem to be
coming to the realization that further progress in creating machine intelligence |
requires exactly this sort of interdisciplinary collaboration |

The movement takes shape |

The 1940s saw the invention of the electronic programmable digital computer, major i
developments in computational theory, and the emergence of the idea that with the |
right software these new machines could simulate the human brain.*® After all, in |
some ways the new machines were substantially superior to human cognition,
solving in minutes what mere human mathematicians labored on for months. With
several important treatises by Turing (including his 1950 paper “Computing Machin-
ery and Intelligence,” in which Turing proposes his imitation game, which became |
known as the Turing test), Wiener’s Cybernetics, a 1943 paper on neural nets by [
Warren McCulloch and Warren Pitts, a proposal by Shannon in 1950 for a chess {
t
program, and several other influential papers, the Al agenda had been set.*” f|
In the 1950s concrete progress began to be made. Initial progress came so ij
rapidly that some of the early pioneers felt that mastering the functionality of the
human brain might not be so difficult after all. After developing IPL-II (Information
 4,
Who's in charge here? This cartoon, by
early computer pioneer Claude Shannon,
illustrates the potential advantage of Processing Language II), the first Al language, in 1955, Allen Newell, J. C. Shaw, and
human over machine: (1) human player
Herbert Simon created a program in 1956 called the Logic Theorist (LT), which used
loses; (2) vexed human reprograms
machine; (3) human wins. (From A Chess- recursive search techniques to solve problems in mathematics.* It was able to find
Playing Machine, copyright 1950 by proofs for many of the theorems in Whitehead and Russell's Principia Mathematica,
Scientific American, all rights reserved)
including at least one completely original proof for an important theorem that had
never been previously published.*? In 1957 Newell, Shaw, and Simon created a more
sophisticated problem solver called the General Problem Solver (GPS).5° As described
earlier, it used means-ends analysis, a variation of LT’s recursive technique, to solve
problems in a variety of domains. These early successes led Simon and Newell to
say in a 1958 paper entitled “Heuristic Problem Solving: The Next Advance in
Operations Research,” “There are now in the world machines that think, that learn
and that create. Moreover, their ability to do these things is going to increase rapidly
until—in a visible future—the range of problems they can handle will be coextensive
with the range to which the human mind has been applied.”®' The paper goes on to
predict that within ten years (that is, by 1968) a digital computer will be the world
chess champion. Such unbridled enthusiasm was to earn the field considerable
criticism and critics.5* While the chess prediction turned out to be premature, a
program completed in 1959 by Arthur Samuel was able to defeat some of the best
players of the time in the somewhat simpler game of checkers.°
In 1956 the first conference on Al was held. Organized by John McCarthy,
it included a number of the field’s future academic leaders, including Marvin Minsky
and Arthur Samuel, as well as Newell and Simon. Also participating were several
computer pioneers, including Oliver Selfridge, Claude Shannon, and Nathaniel
Rochester.* The conference gained its notoriety from its one identifiable accomplish-
ment, which was to give the field its name, artificial intelligence. McCarthy, who is
credited with the name, is not sure if he really made it up or just overheard it in a
conversation, but he does take credit for having promoted it. Some participants
argued against the name: Shannon felt it was too unscientific; Samuel criticized the
word “artificial” with the comment “It sounds like there is nothing real about this
work.” Perhaps just because the phrase is so startling and unsettling, it has out-
lasted many others, including Wiener's “cybernetics."°°
Other concrete accomplishments of the conference are hard to discern
although Minsky did write a first draft of his influential “Steps toward Artificial
Intelligence,” which was rewritten many times (as is Minsky’s style) and finally

199
 Thinking about thinking. Marvin
Minsky, cofounder of MIT's
Artificial Intelligence Lab, in the
1970s. Considered by many to
be the father of Al, his work in the
field has challenged and
inspired a generation of young
computer scientists and philoso-
phers. (Photo by Ivan Masser
of Black Star, courtesy of the MIT
Museum)

200
 Patrick Winston, director of MIT's
Arti | Intelligence Lab. His re-
search involves work on learning by
analogy and commonsense problem
solving. (Photo by Lou Jones)

The hard problems were easy. Al

pioneers Herb Simon and Allen
Newell. Their Logic Theorist (LT),
programmed with J. C. Shaw in 1956,
and General Problem Solver (GPS),
programmed in 1957, discovered
some | proofs for theorems in
Principia Mathematica, Whitehead
and Russell's seminal work on the
foundations of mathematics. There
were also some “easy” problems
that GPS was unable to solve. (Photo
by Dan McCoy of Rainbow)

201
 published in 1963.°° McCarthy refined his ideas for a language that would combine
list (treelike) structures with recursion, which was subsequently introduced in 1959
as LISP (List-processing language).*” It quickly hecame the standard for Al work and
has remained the principal Al language through several major revisions. Only recently
with Al having entered the commercial arena have other languages such as C begun
to compete with LISP, primarily on the basis of efficiency. Probably the major contri-
bution of the conference was to put a number of the leading thinkers in the field in
touch with one another. Progress, ideas, anda great deal of enthusiasm were
shared, although McCarthy left the conference disappointed that most of its specific
goals had not been met. He went on to found the two leading university Al laborato-
ries: one at MIT with Marvin Minsky in 1958 and one at Stanford in 1963.
By the end of the 1960s the full Al agenda was represented by specific
projects. A number of the more significant efforts of the decade are described in
Semantic Information Processing, edited by Marvin Minsky and published in 1968.
Included was Daniel G. Bobrow's Ph.D. project entitled Student, which could set up
and solve algebra problems from natural English language stories.** Student report-
edly rivaled the ability of typical high school students in solving story problems. The
same performance level was claimed for a program created by Thomas G. Evans
that could solve !Q test geometric-analogy problems.*?
Computer chess programs continued to improve in the 1960s, although not
nearly up to Simon’s expectations. A program that could achieve respectable tourna-
ment ratings was created in 1966 by Richard D. Greenblatt at MIT.
A new area for Al, expert systems, also got its start in the 1960s. With
leadership from Edward A. Feigenbaum, now the director of the Stanford Al Labora-
tory, a group started work on DENDRAL, a program based on a knowledge base
describing chemical compounds. It is considered the world’s first expert system.®
The 1960s also saw the creation of ELIZA, a natural language program
written by MIT professor Joseph Weizenbaum in 1966, which simulates a nondirec-
tive (i.e., Rogerian) therapist. ELIZA has continued for over two decades to receive a
high level of attention, including extensive criticism for its inability to react intelli-
gently in a variety of situations, from some Al critics, most notably Hubert Dreyfus.
Actually, ELIZA was never representative of the state of the art in natural-language
understanding, even at the time it was created. It was a demonstration of how
successful one could be in creating an apparently intelligent interactive system with
relatively simple rules. ELIZA is a good example of the principle of achieving complex
behavior from a simple system in a complex environment. People using ELIZA

Thomas Evans's program solves

analogy problems of the sort, Ais to
Bas Cis to___? (From Thomas G.
Evans, “A Program for the Solution of
A| IA ©
a Class of Geometric-Analogy Intelli-
gence-Test Questions,” in Minsky,
ed., Semantic Information Process-
ing, 1968, p. 271)

202
 Daniel Bobrow’s Student solves algebra story
problems. Note the stages of analysis from the (THE PROBLEM TO BE SOLVED IS)
original statement to the equations to be solved. (THE SUM OF LOIS'S SHARE OF SOME MONEY AND BOB'S SHARE IS $4.500.
(From Daniel G. Bobrow, “Natural Language LOIS’S SHARE IS TWICE BOB'S. FIND BOB'S AND LOIS'S SHARE.)
Input for a Computer Problem-Solving System,” (WITH MANDATORY SUBSTITUTIONS THE PROBLEM IS)
in Minsky, ed., Semantic Information Processing, (SUM LOIS’S SHARE OF SOME MONEY AND BOB'S SHARE IS 4.500 DOLLARS. LOIS'S
1968, pp. 212-213) ‘SHARE IS 2 TIMES BOB'S. FIND BOB'S AND LOIS'S SHARE.)
(WITH WORDS TAGGED BY FUNCTION THE PROBLEM IS)
(SUM / OP) LOIS’S SHARE (OF / OP) SOME MONEY AND BOB'S SHARE IS 4.500
DOLLARS (PERIOD / DLM) LOIS SHARE |S 2 (TIMES / OP 1) BOB'S (PERIOD / DLM)
(FIND/ QWORD) BOB'S AND LOIS'S SHARE (PERIOD / DLM))

(THE SIMPLE SENTENCES ARE)

((SUM
/ OP) LOIS'S SHARE (OF / OP) SOME MONEY AND BOB'S SHARE IS 4.500
DOLLARS (PERIOD / DLM))
(LOIS'S SHARE IS 2 (TIMES
/ OP 1) BOB'S (PERIOD / DLM))
((FIND
/ QWORD) BOB'S AND LOIS'S SHARE (PERIOD / DLM))
{THE PROBLEM TO BE SOLVED |S)
(IF THE NUMBER OF CUSTOMERS TOM GETS IS TWICE THE SQUARE OF 20 PER
(THE EQUATIONS TO BE SOLVED ARE)
CENT OF THE NUMBER OF ADVERTISEMENTS HE RUNS. AND THE NUMBER OF
ADVERTISEMENTS HE RUNS IS 45, WHAT IS THE NUMBER OF CUSTOMERS TOM (EQUAL G02519 (LOIS'S SHARE))
GETS Q) (EQUAL G02518 (BOB'S)
(WITH MANDATORY SUBSTITUTIONS THE PROBLEM IS) (EQUAL (LOIS'S SHARE) (TIMES 2 (BOB'S)))
(iF THE NUMBER (OF CUSTOMERS TOM GETS IS 2 TIMES THE SQUARE 20 PERCENT
(EQUAL PLUS (LOIS'S SHARE OF SOME MONEY) (BOB'S SHARE) (TIMES 4.500
OF THE NUMBER OF ADVERTISEMENTS HE RUNS, AND THE NUMBER OF
(DOLLARS)
ADVERTISEMENTS HE RUNS |S 45, WHAT IS THE NUMBER OF CUSTOMERS TOM
GETS Q) THE EQUATIONS WERE INSUFFICIENT TO FIND A SOLUTION
(WITH WORDS TAGGED BY FUNCTION THE PROBLEM |S)
(ASSUMING THAT)
{IF THE NUMBER (OF / OP) CUSTOMERS TOM (GETS/ VERB) IS 2 (TIMES / OP 1) THE
((BO8'S) iS EQUAL TO (BOB'S SHARE)
{SQUARE / OP 1) 20 (PERCENT/ OP 2) (OF / OP) THE NUMBER (OF / OP)
ADVERTISEMENTS (HE / PRO) RUNS, AND THE NUMBER (OF / OP) ADVERTISEMENTS (ASSUMING THAT)
{HE / PRO) RUNS IS 45, (WHAT/ QWORD) IS THE NUMBER (OF / OP) CUSTOMERS ((LOIS'S SHARE) IS EQUAL TO (LOIS'S SHARE OF SOME MONEY))
TOM (GETS / VERB) (QMARK / DLM))
(BOB'S IS 1.500 DOLLARS)
(THE SIMPLE SENTENCES ARE) (LOIS'S SHARE IS 3 DOLLARS)
(THE NUMBER (OF / OP) CUSTOMERS TOM (GETS / VERB) IS 2 (TIMES / OP 1) THE
(SQUARE / OP 1) 20 (PERCENT/ OP 2) (OF / OP) THE NUMBER (OF / OP}
(THE PROBLEM TO BE SOLVED |S)
ADVERTISEMENTS (HE / PRO) RUNS (PERIOD / DLM)
(MARY IS TWICE AS OLD AS ANN WAS WHEN MARY WAS AS OLD AS ANN IS NOW.
(THE NUMBER (OF / OP) ADVERTISEMENTS (HE / PRO) RUNS IS 45 (PERIOD / DLM)) IF MARY IS 24 YEARS OLD, HOW OLD IS ANN Q.)
(WHAT / QWORD) IS THE NUMBER (OF / OP) CUSTOMERS TOM (GETS / VERB) (WITH MANDATORY SUBSTITUTIONS THE PROBLEM IS)
(QMARK / DLM)) (MARY IS 2 TIMES AS OLD AS ANN WAS WHEN MARY WAS AS OLD AS ANN IS NOW
IF MARY IS 24 YEARS OLD, WHAT IS ANN Q.)
(THE EQUATIONS TO BE SOLVED ARE)
(WITH WORDS TAGGED BY FUNCTION THE PROBLEM |S)
{EQUAL G02515 (NUMBER OF CUSTOMERS TOM (GETS/ VERB) (MARY / PERSON) IS 2 (TIMES / OP 1) AS OLD AS (ANN / PERSON) WAS WHEN (MARY
{EQUAL (NUMBER OF ADVERTISEMENTS (HE / PRO) RUNS) 45) / PERSON) WAS AS OLD AS (ANN/ PERSON) IS NOW (PERIOD / DLM) iF (MARY
{EQUAL (NUMBER OF CUSTOMERS TOM (GETS / VERB) (TIMES 2 (EXPT (TIMES 2000 PERSON) IS 24 YEARS OLD, (WHAT / QWORD) IS (ANN / PERSON) (QMARK / DLM)
(NUMBER OF ADVERTISEMENTS (HE / PRO) RUNS) 2)))
(THE SIMPLE SENTENCES ARE)
(THE NUMBER OF CUSTOMERS TOM GETS IS 162) ((MARY / PERSON)'S AGE IS 2 TIMES / OP 1) (ANN / ERSON)'S AGE G02521 YEARS
AGO (PERIOD / DLM))
(G02521 YEARS AGO (MARY / PERSON'S AGE IS (ANN / PERSON)'S AGE NOW
(PERIOD / DLM))
(MARY/ PERSON)'S AGE IS 24 (PERIOD / DLM))
(WHAT/ QWORD) IS (ANN / PERSON)'S AGE (QMARK / DLM))

(THE EQUATIONS TO BE SOLVED ARE)

(EQUAL G02522 (ANN / PERSON)'S AGE))
(EQUAL ((MARY / PERSON)'S AGE) 24)
(EQUAL (PLUS ((MARY / PERSON)'S AGE) (MINUS (G02521))) (ANN
(EQUAL ((MARY / PERSON)'S AGE) (TIMES 2 PLUS ((ANN / PERSON)’S
(MINUS G02521)))))
(ANN'S AGE IS 18)

203

ee a eC C(‘CSC
Joseph Weizenbaum of MIT, Al
pioneer and critic. (Photo by Lou
Jones)

A conversation with ELIZA. (ELIZA’s Men are all alike

responses are in all caps.) IN WHAT WAYS.
(Courtesy of Avron Barr and Edward
Feigenbaum) They're always bugging us about something or other
CAN YOU THINK OF A SPECIFIC EXAMPLE

Well, my boyfriend made me come here.

YOUR BOYFRIEND MADE YOU COME HERE

He says I’m depressed much of the time.

| AM SORRY TO HEAR THAT YOU ARE DEPRESSED

It's true. I'm unhappy.

DO YOU THINK COMING HERE WILL HELP YOU NOT TO BE UNHAPPY

| need some help, that much seems certain

WHAT WOULD IT MEAN TO YOU IF YOU GOT SOME HELP

Perhaps | could learn to get along with my mother.

TELL ME MORE ABOUT YOUR FAMILY

WHAT ELSE COMES TO MIND WHEN YOU THINK OF YOUR FATHER

Bullies.
DOES THAT HAVE ANYTHING TO DO WITH THE FACT THAT YOUR BOYFRIEND
MADE YOU COME HERE

204

ee
provide it with very diverse inputs, which often lead to unexpected and seemingly
insightful responses despite its relative simplicity. A more substantial effort in the
area of natural language was Susumu Kuno’s pioneering work on English language
parsers: his 1963 program could understand the syntax of complex English
sentences.
Another attribute characteristic of Al work in the 1960s was the use of toy
worlds. A prime example was Terry Winograd’s Ph.D. thesis at MIT called SHRDLU,
which combined natural language understanding with a planning capability in a
simulated world (displayed on a terminal screen) that consisted only of different
colored and shaped blocks.® Users could ask SHRDLU to perform tasks with ques-
tions and commands phrased in natural language, such as “Take the red block in
front of the big cube and place it on the blue rectangular solid.” The system would
understand the English statements and could plan a strategy for performing the task.
Although several critics of Al jumped on the unrealistic nature of these toy worlds, it
was appropriate at the time to concentrate on problems of language understanding
and decision making without the vagaries of real-world complexity.“ The next step,
going from toy worlds to real worlds, continues to occupy the attention of Al
researchers.

The age of recursion

Through the end of the 1960s two schools of thought competed for the attention of
Al researchers and funding. The neural school, based on self-organizing neural nets,
started in the 1940s with a paper by McCulloch and Pitts and was heavily promoted
in the 1960s by Frank Rosenblatt of Cornell University.® The lure of neural nets is
the idea that a system of many simple elements, if set up in just the right way, could
actually teach (organize) itself to perform intelligent behavior. With the publication of
Minsky and Papert’s Perceptrons in 1969, this school of thought was eliminated
almost overnight. As mentioned earlier, the book proved several theorems that
demonstrated that certain types of simple problems could never be solved with this
approach. Now in the late 1980s there is renewed interest in a new type of neural
net to which the Minsky and Papert theorems do not apply.®
The school that survived, which one might call the recursive school, was
based on the idea that a relatively small set of powerful ideas would be sufficient to
capture at least some forms of intelligence. If one let these methods attack compli-
cated problems, one could obtain intelligent behavior. The ideas included methods of
exploring (searching) the “space” of possible solutions to a problem as well as
techniques for defining the rules that govern certain domains (e.g. the syntax rules of
natural language).
Foremost among these “powerful” ideas is recursion, the idea that the
statement of a problem can be used as a part, often the most important part, of the
problem's solution. It is a seductive concept; it implies that the essence of solving a
problem is often a matter of being able carefully and precisely to state the problem
This approach works surprisingly well in a variety of domains, and many Al method-
ologies make extensive use of it. It is one reason that the von Neumann computer
architecture that was first incorporated in the famous EDSAC and BINAC computers

205
was so important. Without the von Neumann capability for a stored program, recur-
sion is not possible.
We examined one illustration of the power of recursion earlier in the
context of game playing. To select our best move in a rule-based game such as
chess we call a program called Move. Move generates all of the legal moves for the
current board position and then calls itself to determine the opponent's best re-
sponse, to each possible move (which in turn calls itself again to determine our best
response to our opponent's best response, and so on). In this way as many move
and countermove sequences as we have time to compute are automatically gener-
ated. The most complicated part of implementing this technique is the generation of
the possible moves. Generating possible moves is a matter of programming the
rules of the game. Thus, the heart of the solution is indeed implementing a precise
statement of the problem.
A primary distinguishing feature of McCarthy's LISP, historically the Al
field's primary programming language, is its ability to easily represent recursive
procedures. Another example should illustrate how easily a recursive language can
be used to solve seemingly complex problems. Consider the problem called the
Tower of Hanoi. This famous children’s puzzle presents issues similar to many
important combinatorial problems found in mathematics and other fields.
We have three towers on which we can place round disks of various sizes.
We can move disks from tower to tower, but we can only move one disk at a time
and we cannot place a disk onto a disk of smaller size. Thus, a legal stack of disks
will have the largest disk on the bottom and the smallest disk on top. The problem is
to move a stack of disks from one tower to another. Consider first a stack of just
one disk. Here the answer is obvious: simply move the disk. It is also fairly easy to
determine the procedure for two disks. Let’s consider a larger stack. Try actually
solving the problem for a stack of seven or eight disks—it is rather challenging.
Now consider the general problem of describing a method that can quickly
determine the optimal procedure for a stack of any arbitrary height. This problem at
first appears very difficult, but a simple insight enables us to use the self-referencing
(recursive) paradigm to automatically generate the optimal procedure for any height
stack in three simple steps. Let us number the disks from 1 to n, n being the largest
(bottom) disk. The insight is this: if we are to move the entire stack from the original
tower to the destination tower, at some point we are going to have to move the
bottom disk. That's it! We have just solved the problem. The power of recursion is
that this simple observation is enough to solve the entire problem.
To wit, if the stack consists of only a single disk, then move that disk from
the original tower to the destination tower, and we're done. Otherwise,

Step 1 Since we know that we will need at some point to move the bottom disk,
we clearly need to move the disks on top of it out of the way. We therefore have to
move them away from both the original tower and the destination tower. Thus, we
have to move the stack consisting of all the disks except for the bottom disk (we
can call this the (n - 1) stack since it consists of disk 1 through disk n — 1) from the
original tower to the free tower. This is where the recursion comes in. Moving the fi
stack of (n — 1) disks is the same Tower of Hanoi problem that we started with, only {

206
 [ i Optimal solutions to the Tower of
Solution to the Tower of Hanoi puzzle for 1 disk Hanoi problem generated by our
recursive program

The solution of the Tower of Hanoi

problem for three disks.

Starting position. Tower2 is the Step1

destination tower.

Step2 Step3. We have now moved the (n—1)

stack to the free tower.
Solution to the Tower of Hanoi puzzle for 5 disks

Step 4. We have now moved the bottom Step 5. We now start moving the (n— 1) stack
disk (disk n) to the destination tower. from the free tower to the destination tower.

Step 6 Step 7. We're done!

207
for a smaller stack. The program solving the Tower of Hanoi problem simply calls
itself at this point to solve the problem of moving the (n — 1) stack to the free tower.
This does not lead to an infinite loop because of our special rule for a stack of only
one disk

Step 2 Now with all of the other disks out of the way, we can move the bottom
disk from the original tower to the destination tower.

Step 3 Finally, we move the stack of (n — 1) disks from the free tower to the
destination tower. This again requires a recursive call. Now we're done.

The power of recursive programming languages is that a program can call

itself with the language keeping track of all the tightly nested self-referencing. The
above three steps will not only solve the problem, it will produce the optimal solution
to the problem for any height stack. It generates exactly (2”- 1) moves for n disks. If
one attempts to solve the problem directly by moving disks, one quickly gets lost in
the apparent complexity of the puzzle. To solve the problem using recursion, all that
is required is enough insight to be able to state how the solution for n disks can be
built from the solution for (n — 1) disks. Recursion then optimally expands the
solution for any value of 7.
As the above example should make clear, recursion is a tool that can
unlock solutions to problems that otherwise would be enormously difficult to
understand. A variety of techniques based on this self-referencing paradigm found
widespread applicability by the 1960s. The initial success of recursive Al tech-
niques—finding, for example, an original proof for an important theorem in Principia
Mathematica—fueled much of the field’s early optimism

A recursive program for the Tower of Hanoi problem j

The following program, written in the C programming language, generates an optimal f
solution to the Tower of Hanoi problem. Try following the logic of this program from
the comments even if you are not familiar with programming in general or the C
language in particular. The program is written as a function that when called will print
out the solution. It is called with parameters specifying the original tower, the
destination tower, the free tower, and the number of disks. Note that in this func-
tion, anything that appears between /* and */ is considered to be a comment and is
not part of the program |

tower_of_hanoi (original, destination, free, number_of_disks)

/* This function will print out an optimal solution to the Tower of Hanoi problem. */
integer original; /* This parameter specifies the originating tower. */
integer destination; /* This parameter specifies the destination tower. */
integer free; /* This parameter specifies the free tower. */
integer number_of_disks; /* This parameter specifies the number of disks. The disks are '
numbered from 1 to n with n being the largest (bottom) disk. */
{
if (number_of_disks == 1) f
{

208 |
 print (“Move disk 1 from tower %d to tower %d\n”, original destination);
return;

}
/* \f the number of disks is 1, then this is the escape from recursion. We simply print
that disk 1 is to be moved from the originating tower to the destination tower and
then return. */
tower_of_hanoi (original, free, destination, number_of_disks — 1);
/* Here we have the first recursive call where the tower_of_hanoi function calls itself.
This call moves the (n — 1) stack (the stack consisting of all the disks except for the
bottom one) from the originating tower to the free tower. */
print (“Move disk %d from tower %d to tower %d\n”, number_of_ disks, original,
destination);
/* Now print that the bottom disk (disk n) is to be moved from the originating tower to
the destination tower. */
tower_of_hanoi (free, destination, original, number_of_disks — 1);
/* Move the (n — 1) stack from the free stack to the destination stack. */
return; /* We're done! */
} /* End of tower_of_hanoi function */

The age of knowledge |

Somewhere around the early 1970s a major conceptual change took place in Al
methodology (it is difficult to pick a hard date as there is no single paper one can cite
that ushered in the change).®” While work in the 1950s and 1960s concentrated
primarily on the mechanics of the reasoning process (search, recursion, problem
representation, etc.), it became apparent by the early 1970s that such techniques
alone were not sufficiently powerful to emulate the human decision-making process,
even within narrowly defined areas of expertise.®* Something was missing, and the
something turned out to be knowledge.
Some work on knowledge representation (knowledge about how to
represent knowledge) took place earlier.® In fact, two projects conducted during the
1960s and described in Minsky’s Semantic Information Processing were Bertram
Raphael's SIR program (Semantic Information Retrieval) and a theory of semantic
memory by M. Ross Quillian, both of which dealt with methods for representing
human knowledge (see figure). What changed in the early 1970s was the recognition
of the relative importance of knowledge versus method. In the 1950s and 1960s
there was an emphasis on the power of techniques, particularly recursive ones, for
emulating the logical processes associated with thinking. By the 1970s it was
recognized that programming reasoning techniques were relatively simple compared
to the task of creating a knowledge base with the depth required to solve real-world
problems.
Roger Schank points out the extensive knowledge required to understand
even simple stories. If we read a story about a restaurant, there is a vast body of
factual information about restaurants that we take for granted.’° Understanding the
statement “He paid the bill” requires understanding that in a restaurant we are
expected to pay for the food we order, we order food on credit, we are brought a
document (generally at the end of the meal) called the “tab,” “check,” or “bill,”
which itemizes the charges, we generally settle this debt prior to leaving the

209
Knowledge structures from M. Ross
Quillian’s theory of semantic
memory. (From M. Ross Quillian,
structure and
¥
“Semantic Memory,” in Minsky, ed., live get3
t ‘es trom 3 Fe :
\ Ma 55
Semantic Information Processing,
1968, p. 250)

thing
¥
has-to
+
to7

ie
keep

comfort “T)

ind or
———. a
make giveS5 make2

iectrd 5 iS less 2
ao Yesad

eq
restaurant, and so on. When we read a story, almost every sentence evokes vast
networks of similarly implied knowledge. The difficulty in mastering all of this
commonsense knowledge is that no one has ever bothered to write it all down, and
the quantity of it is vast.
Mastering knowledge has indeed turned out to be a far more difficult
process than mastering the logical processes inherent in deductive or even inductive
reasoning. First, we need to have a means of structuring knowledge to make it
useful. A simple list of all facts in the world, if such a list could be constructed,
would not help us solve problems, because we would have a hard time finding the
tight facts to fit the right situations. Douglas Hofstadter provides an amusing
example of the problem in Metamagical Themas. “How do | know,” he asks, “when
telling you I'll meet you at 7 at the train station, that it makes no sense to tack on
the proviso, ‘as long as no volcano erupts along the way, burying me and my car on
the way to the station,’ but that it does make reasonable sense to tack on the
proviso, ‘as long as no traffic jam holds me up?'” The objective of appropriate
knowledge structures is to quickly access the information truly relevant to any
particular situation. Once we have constructed suitable structures for knowledge
representation, we then need to actually collect the vast amount of information
required to solve practical problems. Finally, we have to integrate this knowledge
base with the appropriate decision-making algorithms
The early 1970s saw a number of pioneering efforts to address the knowl
edge-representation problem. Perhaps the most famous was Minsky’s theory of
frames, which he described in his 1975 paper “A Framework for Representing

210
Knowledge.””' A frame is a data structure that can include information of arbitrary
complexity about an object or type of object, and that allows for multiple hierarchies
for understanding relationships between classes of objects. For example, we can
have a frame of information about the concept of a dog, another for the concept of a
cat, and another for the concept of a mammal. The mammal frame is a higher-level
frame than those representing examples of mammals, and the relationship between
the levels (e.g., a dog is a type of mammal) is built into the frame structures. Each
frame allows for default information to be filled in or “inherited” from a higher-level
frame. For example, if the mammal frame says that mammals have four legs, this
information would not have to be repeated in the frames for dogs and cats. How-
ever, a human frame would have to indicate an exception to this default information
(one specifying only two legs). The frame methodology avoids redundancy, describes
hierarchical relationships, and allows for arbitrarily complex classifications. It also
helps us to make useful hypotheses. If we learn, for example, about another mam-
mal, we can assume that it has four legs until informed otherwise
Another important approach to representing knowledge and the interde-
pendency relationships between concepts was first described in another 1975 paper,
this one describing a project called SAM (Script Applier Mechanism) by Roger
Schank, Robert Abelson, et al. at Yale University. Schank’s methodology allowed for |
the development of “scripts” that provide the information implicit in everyday
situations such as restaurants.’?
The second part of the knowledge issue, actually collecting the knowledge,
has proved to be the greatest challenge. In developing modern expert systems
(computer-based systems that emulate the decision-making ability of human ex-
perts), the process of collecting the necessary knowledge is generally a painstaking
process involving a “knowledge engineer” interviewing the appropriate human
experts and literally writing down (in an appropriate computer language) all of the
relevant knowledge and decision-making rules used by that human expert. The sheer
volume of information involved is one problem, but a bigger one is that while human
experts are capable of solving problems within their domains of expertise, they
generally do not know how they accomplish these tasks. The skill required of the
knowledge engineer is to be able to extract the decision-making process from the
domain experts despite their not being consciously aware of many elements of this
process.”>
With a first generation methodology for building expert systems already
established, a number of ambitious projects were started in the 1970s. Internist
(now called CADUCEUS), an expert system that diagnoses a wide range of internal
diseases, was developed throughout the 1970s (and continued in the 1980s). In one
study, Internist was able to diagnose illnesses within at least one specialty with an
accuracy equal to or better than human physicians. MYCIN, a system that can make
diagnoses and recommend treatments for a wide range of bacterial infections, was
developed by Edward H. Shortliffe in the mid 1970s. Prospector, an expert system
that is capable of pinpointing energy and geology deposits, was initiated by R. O.
Duda and his associates at Stanford Research Institute in 1978. In at least one case
Prospector identified a number of important energy deposits overlooked by human

au
experts. Finally, XCON, probably the most successful expert system in commercial
use today, started operation in 1980 configuring complex computer systems for
Digital Equipment Corporation. This system, running on a single VAX computer, is
able to perform tasks that would otherwise require several hundred human experts,
and at substantially higher rates of accuracy. These systems and others, as well as
the problems of knowledge representation, will be discussed in greater detail in
chapter 8.7
The U.S. Department of Defense through its agency DARPA (Defense
Advanced Research Projects Agency), funded two major initiatives in the pattern
recognition area during the 1970s. The SUR (Speech Understanding Research)
project funded the development of several experimental continuous-speech-under-
standing programs aimed at machine recognition of normal human speech with a
large vocabulary. Though the most successful system from the SUR project did not
operate in real time and was limited to an artificial syntax, SUR did increase confi-
dence that practical, high-performance speech recognition was feasible. A similar
program called |UP (Image Understanding Program) attempted machine comprehen-
sion of visual images
The intensity of effort as well as the practical value of Al technology grew
enormously during the 1980s. Here | shall mention briefly two salient trends: the
commercialization and the internationalization of Al. The Al industry grew from just a
few million dollars at the beginning of the 1980s to $2 billion by 1988, according to
DM Data, a leading market-research firm. Many market analysts predict that the bulk
of the several-hundred-billion dollar computer and information processing market by
1999 will be intelligent, at least by today’s standards.
The 1980s began with a stunning challenge from Japan‘s powerful MITI
(the Ministry of International Trade and Industry) when they announced a plan to
design and build an intelligent fifth generation computer. This was seen by many as
an attempt by Japan to leapfrog its foreign competitors and establish dominance
over the international computer industry.”>
As | said earlier, the idea that human intelligence could be simulated
seems to have occurred to all of the pioneers who played a role in what | consider to
be the twentieth century's greatest invention, the computer. Though artificial
intelligence was not named until 1956, the concept was by no means an after-
thought. Despite the fact that the early computers were used primarily for numerical
calculation (as most computers still are today), these classic machines were not
thought of by their creators as mere number crunchers. They have been viewed
since their conception as amplifiers of human thought, what Ed Feigenbaum calls
“power tools for the mind.”76
Early success in the 1950s and 1960s with what were thought to be
difficult problems, such as proving theorems and playing chess, fueled a romantic
optimism that proved short-lived. It was an example of the “90-10” rule: solving the
first 90 percent of a problem often requires only 10 percent of the effort, and though
the remaining 10 percent then requires 90 percent of the effort, it generally repre-
sents 90 percent of the importance. With the realization in the 1970s that extensive

212
knowledge was required to solve practical problems, and with no easy way of
capturing that knowledge, the field gained a needed maturity
The 1980s saw the first practical solutions to real problems and the emer-
gence of a multibillion-dollar industry. The 1990s will, in my estimation, witness the
emergence of an industry valued at several hundred billion dollars and a generation
of ubiquitous intelligent machines that work intimately with their human creators

213
 What Is Intelligence?
aa eel

What is intelligence, anyway? It is only a word that people use to

name those unknown processes with which our brains solve
problems we call hard. But whenever you learn a skill yourself,
you're less impressed or mystified when other people do the same.
This is why the meaning of “intelligence” seems so elusive: it
describes not some definite thing but only the momentary horizon of
our ignorance about how minds might work. It is hard for scientists
who try to understand intelligence to explain precisely what they do,
since our working definitions change from year to year. But it is not at
all unusual for sciences to aim at moving targets. Biology explores
the moving frontier of what we understand of what happens inside
our bodies. Only a few decades ago the ability of organisms to
reproduce seemed to be a deep and complex mystery. Yet as soon as
they understood the elements of how our gene strings replicate
themselves, biologists wondered why it took so long to think of such
a simple thing. In the same way each era of psychology explores
what we don’t then know about processes in our brains.
Thoughts about Artificial Intelligence
Then, can we someday build intelligent machines? | take
the answer to be yes in principle, because our brains themselves are
Marvin Minsky, a founder of the machines. To be sure, we still know very little about how brains
MIT Artificial Intelligence
actually work. There is no reason for scientists to be ashamed of this,
Laboratory, Thinking Machines,
and Logo Computer Systems, considering that it was only a century ago that we began to suspect
is Donner Professor of Science that brains were made of separate nerve cells that acted somewhat
at the Massachusetts Institute
like computer parts and that it is only half a century since we began
of Technology. He was one of
developing technical ideas for understanding what such systems
the pioneers in computer
science, having made major could do. These ideas are still barely adequate for dealing with
contributions in establishing present-day serial computers, which have only thousands of active
the scientific foundation of components, and are not yet robust enough to deal with systems like
three important areas: the
those in the brain, which involve trillions of interconnected parts, all
mathematical theory of
computation, artificial intelli- working simultaneously.
gence, and robotics. His most Nor do we yet know how to make machines do many of
recent book is The Society of the things that ordinary people do. Some critics maintain that
Mind (1986).
machines will never be able to do some of those things, and some
skeptics even claim to have proved such things. None of those
purported proofs actually hold up to close examination, because we
are still in the dark ages of scientific knowledge about such matters.
In any case, we have not the slightest grounds for believing that
human brains are not machines. Because of this, both psychology and
artificial intelligence have similar goals: both seek to learn how
machines could do many things we can't yet make them do.
Why are so many people annoyed at the thought that
human brains are nothing more than “mere machines”? It seems to
me that we have a problem with the word “machine,” because we've
grown up to believe that machines can behave only in lifeless,

214

|
Marvin Minsky. (Photo by Lou Jones)

mechanical ways. This view is obsolete, because the ways we use somewhat shady reputation! This paradox resulted from the fact that
the word “machine” are out of date. For centuries words like whenever an Al research project made a useful new discovery, that
“machine” and “mechanical” were used for describing relatively product usually quickly spun off to form a new scientific or commer-
simple devices like pulleys, levers, locomotives, and typewriters. The cial specialty with its own distinctive name. These changes in name
word “computer” too inherits from the past that sense of pettiness led outsiders to ask, Why do we see so little progress in the central

that comes from doing dull arithmetic by many small and boring steps. field of artificial intelligence? Here are a few specialties that 1
Because of this, our previous experience can sometimes be a originated at least in part from Al research but later split into
handicap. Our preconceptions of what machines can do date from separate fields and, in some instances, commercial enterprises:
what happened when we assembled systems from only a few robotics, pattern recognition, expert systems, automatic theorem
hundreds or thousands of parts. And that did not prepare us to think proving, cognitive psychology, word processing, machine vision,
about brainlike assemblies of billions of parts. Although we are knowledge engineering, symbolic applied mathematics, and
already building machines with many millions of parts, we continue computational linguistics.
to think as though nothing has changed. We must learn to change For example, many researchers in the 1950s worked
how we think about phenomena that work on those larger scales. toward discovering ways to make machines recognize various sorts
of patterns. As their findings were applied to problems involved with
What Is Artificial Intelligence? vision, speech, and several other areas, those fields evolved their
‘er
own more distinct techniques, they organized their own technical
Even though we don’t yet understand how brains perform societies and journals, and they stopped using the term “artificial
many mental skills, we can still work toward making machines that intelligence.” Similarly, an early concern of Al was to develop
do the same or similar things. “Artificial intelligence” is simply the techniques for enabling computers to understand human language;
name we give to that research. But as | already pointed out, this this spawned a field called computational linguistics. Again, many
means that the focus of that research will keep changing, since as ideas from artificial intelligence had a large influence among
soon as we think we understand one mystery, we have to move on to psychologists, who applied those ideas to their studies of the mind
the next. In fact, Al research has made enormous progress in only a but used the title “cognitive psychology.”
few decades, and because of that rapidity, the field has acquired a I can illustrate how Al projects develop by recounting the
research of James Slagle, who, as a graduate student at MIT in 1960,

215
developed a program to solve calculus problems; he named it with nature of those calculus problems. Eventually ways were found to
the initials of “symbolic automatic integration.” Although there were replace all the trial and error processes in SAINT by methods that
many problems that SAINT couldn't solve, it surpassed the perform- worked without any search. The resulting commercial product, a
ance of average MIT students. When he first approached this subject, program called MACSYMA, actually surpassed the abilities of
most scientists considered solving those problems to require professional mathematicians in this area. But once the subject was
substantial intelligence. But after Slagle's work we had to ask so well understood, we ceased to think of it as needing intelligence.
ourselves instead why students take so long to learn to do such This area is now generally seen as belonging no longer to artificial
basically straightforward things. intelligence but to a separate specialty called symbolic applied
How did SAINT solve those problems? It employed about mathematics.
100 formulas from the domains of algebra and calculus and applied to
these about a dozen pattern-matching methods for deciding which Robotics and Common Sense
formula might be most likely to help solve a given problem. Since any
Sameera

particular attempt might fail, the program had to employ a good deal In the 1960s we first began to equip computers with mechanical
of trial and error. If one method did not work, the program automati- hands and television eyes. Our goal was to endow machines with the
cally went on to try another. Sometimes one of them would work, but sorts of abilities children use when playing with toys and building
frequently a problem was too hard for any single such method to blocks. We found this much harder to do than expected. Indeed, a
work. The system was programmed in that case to proceed on to scholar of the history of artificial intelligence might get a sense of
certain other methods, methods that attempted to split each hard watching evolution in reverse. Even in its earliest years we saw
problem into several simpler ones. In this way, if no particular method computers playing chess and doing calculus, but it took another
worked, SAINT was equipped with a great variety of alternatives. decade for us to begin to learn to make machines that could begin to
Now we can make an important point. For years the act like children playing with building blocks! What makes it easier
public has been told, Computers do only what they're programmed to to design programs that imitate experts than to make them simulate
do. But now you can see why that's not quite true: We can write novices? The amazing answer is, Experts are simpler than novices!
programs that cause the machine to search for solutions. Often such To see why it was harder to make programs play with toys than pass
searches produce results that greatly surprise their programmers. calculus exams, let's consider what's involved in enabling a robot to
The idea of making programs search greatly expanded copy simple structures composed of blocks: we had to provide our
their powers. But it also led to new kinds of problems: search robot with hundreds of small programs organized into a system that
processes could generate so many possible alternatives that the engaged many different domains of knowledge. Here are a few of the
programs were in constant danger of getting lost, repeating them- sorts of problems this system had to deal with:
selves, or persisting at fruitless attempts that had already consumed
The relation between the hand and the eye
large amounts of time. Much research in the 1960s was focused on
Recognizing objects from their visual appearances
finding methods to reduce that sort of fruitless search. Slagle himself
Recognizing objects partially hidden from view
experimented with some mathematical theories of how to take into
Recognizing relations between different objects
account both how much effort had been spent on each solution
Fitting together three-dimensional shapes
attempt and how much apparent progress had been made. Thus the
Understanding how objects can support one another to form stable
SAINT program worked as well as it did, not merely because of its
structures
specialized knowledge about calculus, but also because of other
Planning a sequence of actions to assemble a structure
knowledge about the search itself. To prevent the search from simply
Moving in space so as to avoid collisions
floundering around, making one random attempt after another, some
Controlling the fingers of a hand for grasping an object
of the program's knowledge was applied to recognize conditions in
It is very hard for any adult to remember or appreciate
which its other, more specialized knowledge might be particularly
how complex are the properties of ordinary physical things. Once
useful.
when an early version of our block-building program was asked to
When SAINT first appeared, it was acclaimed an
find a new place to put a block, it tried to place it on top of itself! The
outstanding example of work in the field of artificial intelligence.
program could not anticipate how that action would change the
Later other workers analyzed more carefully its virtues and deficien-
cies, and this research improved our understanding of the basic

216

eee
 situation. To catalog only enough fragments of knowledge to enable a problem is simply accumulating so much knowledge. But Al research

robot to build a simple blocklike house from an unspecified variety of also encountered a second, more subtle problem. We had to face the

available materials would be an encyclopedic task. College students simple fact that in order for a machine to behave as though it

usually learn calculus in half a year, but it takes ten times longer for “knows” anything, there must exist, inside that machine, some sort of

children to master their building toys. We all forget how hard it was structure to embody or “represent” that knowledge. Now, a special-

to learn such things when we were young. ized, or “expert,” system can usually get by with very few types of
what we call knowledge representations. But in order to span that
larger universe of situations we meet in ordinary life, we appear to
Expertise and Common Sense
SSS need a much larger variety of types of representations. This leads to a
second, harder type of problem: knowledge represented in different
Many computer programs already exist that do things most people
ways must be applied in different ways. This imposes on each child
would regard as requiring intelligence. But none of those programs
obligations of a higher type: they have to learn which types of
can work outside of some very small domain or specialty. We have
knowledge to apply to which kinds of situations and how to apply
separate programs for playing chess, designing transformers, proving
them. In other words, we have to accumulate not merely knowledge,
geometry theorems, and diagnosing kidney diseases. But none of
but also a good deal of knowledge about knowledge. Now, experts
those programs can do any of the things the others do. By itself each
too have to do that, but because commonsense knowledge is of more
lacks the liveliness and versatility that any normal person has. And no
varied types, an ordinary person has to learn (albeit quite uncon-
one yet knows how to put many such programs together so that they
sciously) much more knowledge about representations of knowledge,
can usefully communicate with one another. In my book The Society
that is, which types of representation skills to use for different
of Mind| outline some ideas on how that might be done inside our
purposes and how to use them.
brains.
If this sounds very complicated, it is because it actually
Putting together different ideas is just what children learn
is. Until the last half century we had only simple theories of mind, and
to do: we usually call this common sense. Few youngsters can design
these explained only a little of what animals could do in the
transformers or diagnose renal ailments, but whenever those children
impoverished worlds of laboratory experiments. Not until the 1930s
speak or play, they combine a thousand different skills. Why is it so
did psychologists like Jean Piaget discover how many aspects of a
much easier for Al programmers to simulate adult, expert skills than
child's mind develop through complicated processes, sometimes
to make programs perform childlike sorts of commonsense thought? |
composed of intricate sequences of stagelike periods. We still don't
suspect that part of the answer lies in the amounts of variety. We can
know very much about such matters, except that the mind is much
often simulate much of what a specialist does by assembling a
more complex than imagined in older philosophies. In The Society of
collection of special methods, all of which share the same common
Mind| portray it as a sort of tangled-up bureaucracy, composed of
character. Then so long as we remain within some small and tidy
many different experts, or as | call them, “agencies,” that each
problem world, that specialist's domain of expertise, we need merely
develop different ways to represent what they learn. But how can
apply different combinations of basically similar rules. This high
experts using different languages communicate with one another?
degree of uniformity makes it easy to design a higher-level supervi-
The solution proposed in my book is simply that they never come to do
Sory program to decide which method to apply. However, although
it very well! And that explains why human consciousness seems so
the “methods” of everyday thinking may, by themselves, seem simpler
mysterious. Each part of the mind receives only hints of what the
than those of experts, our collections of commonsense methods deal
other parts are about, and no matter how hard a mind may try, it can
with a great many more different types of problems and situations.
never make very much sense of itself.
Consider how many different things each normal child must learn
about the simplest-seeming physical objects, such as the peculiari-
Supercomputers and Nanotechnology
ties of blocks that are heavy, big, smooth, dangerous, pretty, delicate,
SS
ee
or belong to someone else. Then consider that the child must learn
Many problems we regard as needing cleverness can sometimes be
quite different kinds of strategies for handling solids and liquids;
solved by resorting to exhaustive searches, that is, by using massive,
Strings, tapes, and cloths; jellies and muds as well as things he is told
raw computer power. This is what happens in most of those
are prohibited, poisonous, or likely to cut or bite.
inexpensive pocket chess computers. These little machines use
What are the consequences of the fact that the domain of
commonsense thinking is so immensely varied and disorderly? One

217
programs much like the ones that we developed in the 1960s, using memory banks. Wouldn't it be better to keep more of the hardware in
what were then some of the largest research computers in the world. actual operation? A more active type of computer architecture was
Those old programs worked by examining the consequences of tens proposed in Daniel Hillis's Connection Machine (MIT Press, 1986),
of thousands of possible moves before choosing one to actually make. which describes a way to assemble a large machine from a large
But in those days the programs took so long to make those moves that number of very small, serial computers that operate concurrently and
the concepts they used were discarded as inadequate. Today, Pass messages among themselves. Only a few years after being
however, we can run the same programs on faster computers so that conceived, Connection Machines are already commercially
they can consider millions of possible moves, and now they play available, and they indeed appear to have fulfilled their promise to
much better chess. However, that shouldn't fool us into thinking that break through some of the speed limitations of serial computers. In
we now understand the basic problem any better. There is good certain respects they are now the fastest computers in the world.
reason to believe that outstanding human chess players actually This is not to say that parallel computers do not have their
examine merely dozens, rather than millions, of possible moves, own limitations. For, just as one cannot start building a house before
subjecting each to more thoughtful analysis. the boards and bricks have arrived, you cannot always start work
In any case, as computers improved in speed and memory simultaneously on all aspects of solving a problem. It would certainly
size, quite a few programming methods became practical, ones that be nice if we could take any program for a serial computer, divide it
had actually been discarded in the earlier years of Al research. An into a million parts, and then get the answer a million times faster by
Apple desktop computer (or an Amiga, Atari, IBM, or whatever) can running those parts simultaneously on that many computers in
do more than could a typical million-dollar machine of a decade parallel. But that can’t be done, in general, particularly when certain
earlier, yet private citizens can afford to play games with them. In parts of the solution depend upon the solutions to other parts.
1960 a million-bit memory cost a million dollars; today a memory of Nevertheless, this quite often turns out to be feasible in actual
the same size (and working a hundred times faster) can be purchased practice. And although this is only a guess, | suspect that it will
for the price of a good dinner. Some seers predict another hun- happen surprisingly often for the purposes of artificial intelligence.
dredfold decrease in size and cost, perhaps in less than a decade, Why do | think so? Simply because it seems very clear that our brains
when we learn how to make each microcircuit ten times smaller in themselves must work that way.
linear size and thus a hundred times smaller in area. What will Consider that brain cells work at very modest speeds in
happen after that? No one knows, but we can be sure of one thing: comparison to the speeds of computer parts. They work at rates of
those two-dimensional chips we use today make very inefficient use less than a thousand operations per second, a million times slower
of space. Once we start to build three-dimensional microstructures, than what happens inside a modern computer circuit chip. Could any
we might gain another millionfold in density. To be sure, that would computer with such slow parts do all the things that a person can do?
involve serious new problems with power, insulation, and heat. For a The answer must lie in parallel computation: different parts of the
futuristic but sensible discussion of such possi! s, recommend brain must do many more different things at the same time. True, that
Eric Drexler's Engines of Creation (Falcon Press, 1986). would take at least a billion nerve cells working in parallel, but the
Not only have small components become cheaper; they brain has many times that number of cells.
have also become faster. In 1960 a typical component required a
microsecond to function; today our circuits operate a thousand times Al and the World of the Future
ee
= ee
faster. Few optimists, however, predict another thousandfold
increase in speed over the next generation. Does this mean that even Intelligent machines may be within the technological reach of the
with decreasing costs we will soon encounter limits on what we can next century. Over the next few generations we'll have to face the
make computers do? The answer is no, because we are just problems they pose. Unless some unforeseen obstacles appear, our
beginning a new era of parallel computers. mind-engineering skills could grow to the point of enabling us to
Most computers today are still serial; that is, they do only construct accomplished artificial scientists, artists, composers, and
one thing at a time. Typically, a serial computer has millions of personal companions. Is Al merely another advance in technology, or
memory elements, but only a few of them operate at any moment, is it a turning point in human evolution that should be a focus of
while the rest of them wait for their turn: in each cycle of operation, a discussion and planning by all mankind? The prospect of intelligent
serial computer can retrieve and use only one of the items in its machines is one that we're ill prepared to think about, because it
raises such unusual moral, social, artistic, philosophical, and

218
 religious issues. Are we obliged to treat artificial intelligences as only mislead ourselves when we ask our machines to do those things
sentient beings? Should they have rights? And what should we do that we admire most. No one could deny that our machines, as we
when there remains no real need for honest work, when artificial know them today, lack many useful qualities that we take for granted
workers can do everything from mining, farming, medicine, and in ourselves. But it may be wrong to seek the sources of those
manufacturing all the way to house cleaning? Must our lives then qualities in the exceptional performances we see in our cultural
drift into pointless restlessness and all our social schemes heroes. Instead, we ought to look more carefully at what we ordinary
disintegrate? people do: the things we call common sense and scarcely ever
These questions have been discussed most thoughtfully consider at all. Experience has shown that science frequently
in the literary works of such writers as Isaac Asimov, Gregory develops most fruitfully once we learn to examine the things that
Benford, Arthur C. Clarke, Frederick Pohl, and Jack Williamson, who seem the simplest, instead of those that seem the most mysterious.
all tried to imagine how such presences might change the aspira-
tions of humanity. Some optimistic futurists maintain that once we've
Satisfied all our worldly needs, we might then turn to the worlds of
the mind. But consider how that enterprise itself would be affected
by the presence of those artificial mindlike entities. That same Al
technology would offer ways to modify the hardware of our brains
and thus to endlessly extend the mental worlds we could explore.
You might ask why this essay mixes both computers and
psychology. The reason is that though we'd like to talk about making
intelligent machines, people are the only such intelligence we can
imitate or study now. One trouble, though, is that we still don’t know
enough about how people work! Does this mean that we can't
develop smart machines before we get some better theories of
psychology? Not necessarily. There certainly could be ways to make
very smart machines based on principles that our brains do not use,
as in the case of those very fast, dumb chess machines. But since we
are the first very smart machines to have evolved, we just might
represent one of the simplest ways!
But, you might object, there’s more to a human mind than
merely intellect. What about emotion, intuition, courage, inspiration,
Creativity, and so forth. Surely it would be easier simply to under-
stand intelligence than to try to analyze all those other aspects of our
personalities! Not so, | maintain, because traditional distinctions like
those between logic and intuition, between intellect and emotion,
unwisely try to separate knowledge and meaning from purpose and
intention. In The Society of Mind | argue that little can be done
without combining elements of both. Furthermore, when we put them
together, it becomes easier, rather than harder, to understand such
matters, because, though there are many kinds of questions, the
answers to each of them illuminate the rest.
Many people firmly believe that computers, by their
nature, lack such admirable human qualities as imagination,
‘sympathy, and creativity. Computers, so that opinion goes, can be

only logical and literal. Because they can't make new ideas,
intelligent machines lie, if at all, in futures too remote for concern.
However, we have to be wary of such words as “creativity.” We may

219
The Moving Frontier
(Photo by Lou Jones)
 Pattern Aecognition: The Search for Order

The digitization of information in all its forms will probably be known as the most fascinating develop-
ment of the twentieth century.
An Wang, founder of Wang Laboratories

Most probably, we think, the human brain is, in the main, composed of large numbers of relatively small
distributed systems, arranged by embryology into a complex society that is controlled in part (but only in
part) by serial, symbolic systems that are added later. But the subsymbolic systems that do most of the
work from underneath must, by their very character, block all the other parts of the brain from knowing
much about how they work. And this, itself, could help explain how people do so many things yet have
such incomplete ideas of how those things are actually done.
Marvin Minsky and Seymour Papert, the 1988 epilogue to Perceptrons

+ Vision

Two types of thinking

Try not to think about elephants. For the next sixty seconds, do not let the image of
these huge mammals with their large ears and swaying trunks enter your mind. Now
look across the room and focus your vision on an object. Without closing your eyes
or turning them away, try not to determine what the object is. Finally, consider the
Tower of Hanoi problem described in the last chapter. For the next sixty seconds, do
not solve this problem.
You are undoubtedly having difficulty avoiding the mental image of an
elephant. Assuming that the object that you selected to look at is not unknown to
you, you were probably unsuccessful as well in not determining its identity. On the
other hand, unless you have an unusual passion for mathematics, you are probably
experiencing little difficulty in not solving the Tower of Hanoi problem.
Two types of thought processes coexist in our brains, and the above
exercises illustrate one of the profound differences between them. Perhaps most
often cited as a uniquely human form of intelligence is the /ogica/ process involved in
solving problems and playing games. A more ubiquitous form of intelligence that we
share with most of the earth’s higher animal species is the ability to recognize
patterns from our visual, auditory, and tactile senses. We appear to have substantial
control over the sequential steps required for logical thought. In contrast, pattern
recognition, while very complex and involving several levels of abstraction, seems to
happen without our conscious direction.’ It is often said that a master chess player
can “see” his or her next move without going through all of the conscious se-
quences of thinking required of less experienced players. It may be that after being
exposed to tens of thousands of board situations, the master player is able to
replace at least some of the logical processes usually used to play games with
pattern-recognition methods.”
There are several key differences between these two forms of intelligence,
including the level of success the Al field has had in emulating them.? Ironically, we
find it easier to create an artificial mathematician or master chess player than to
emulate the abilities of animals. While there are many animal capabilities that our
machines have not yet mastered, including the intricacies of fine motor coordination,
the most difficult barrier has been the subtleties of vision, our most powerful sense
and a prime example of pattern recognition.
One attribute that the two types of thinking have in common is the use of
imagination. The first example cited above, imagining an elephant, is a direct
exercise in imagination. The second example, identifying an object, also involves
imagination, particularly in the latter stages of the process. If part of the object we
are trying to identify is blocked or if its orientation toward us prevents the most
useful view, we use our imagination to visualize in our minds what the full object
might look like and then determine if the imagined object matches what we can see.
In fact, we almost always use our imagination to visualize an object, because
invariably there are sides we cannot see.* The technical term for this technique is
“hypothesis and test”; we use our imagination to hypothesize the answer and then
test its validity.> Hypothesis and test is also used in logical thought. We often
imagine an answer to a logical problem based on methods of intuition that are only
partially understood and then work backward to the original problem statement
If we examine the nature of our visual imagination, we can gain some
insight into the most important way in which pattern recognition differs from logical
rule-based thinking. Consider again your imagination of an elephant. Your mental
picture probably does not include a great deal of detail: it is essentially a line draw-
ing, probably one that is moving (I'll bet the trunk is swaying back and forth). The
phenomenon of the line drawing—the fact that we recognize a line drawing of an
object (a face, for example) as representing that object though the drawing is
markedly simpler than the original image—provides us with an important clue to the
nature of the transformations performed during the process of human vision. Very
important to the recognition of visual objects is the identification of edges, which we
model in two dimensions as lines. If we explore what is required to extract edges
from an image, we shall gain an appreciation of one major way in which visual
perception differs from logic
There are many aspects of visual perception that we do not yet under-
stand, but some understanding of the identification of edge location and orientation
has been achieved.® A particular set of computations has been discovered that is
capable of detecting edges with reasonable accuracy. There is some evidence that
similar techniques are used in visual processing by mammals. The technique is

224
 based on two observations. The first observation is that we need to smooth the
data; changes involving tiny regions can probably be considered to be non-informa-
tion-bearing visual noise. Thus, small defects in edges can be ignored, at least
initially, in locating all of the edges in an image. Second, we note that changes in the
visual information (across any spatial dimension) are more important than the
information itself. In other words, we are primarily interested in sudden and consis-
tent alterations in color or shading from one region to another
| shall now describe a method for inferring edges from visual images.’ The
following two paragraphs are somewhat technical. Yet it is not necessary to under-
stand all of these details to appreciate some of the implications of the method. The
image itself is represented by a two-dimensional array of pixels, or points of informa
tion. In a black and white image, each pixel can be represented by a single number
representing a shade of grey. In a color image, several numbers (usually three) are
required to represent the color and shade. We can take this initial raw image and
modify it to take advantage of the two observations cited above. The modification is
achieved by applying what is called a filter, in which each pixel has an influence on
its surrounding pixels. For example, a Gaussian filter designates certain pixels as
propagating pixels; it then increases the intensity of each pixel in the vicinity of each
propagating pixel on the basis of the intensity of the propagating pixel and the
distance to the neighboring pixel. The function of intensity to distance is based on
the familiar Gaussian (normal) curve, with the peak of the curve representing zero
distance (that is, the propagating pixel itself). A Gaussian filter is applied to an image
by making every pixel a propagating pixel; thus, all pixels bleed into their surrounding
pixels. This has the impact of smoothing the image, with the sharpness of the
resulting image being a function of the width of the Gaussian curve. A different filter,
the Laplacian, can then be applied to detect changes. This filter replaces the value of
every pixel with the rate of change of the rate of change (that is, the second deriva-
tive) of the pixel values
These two processes—smoothing and determining rates of rates of
change—can be combined into a single filter in which every pixel influences all of the
pixels within its vicinity. This filter, with the appropriate, if forbidding, name of
“Laplacian of a Gaussian convolver,” has a graph with the shape of an upside-down
Mexican hat, so it is often called a sombrero filter. As the figure shows, each pixel
has a positive influence on the pixels in its immediate vicinity and a negative

The sombrero filter. (Supplied by

Tomaso Poggio)

225
 influence on pixels in a band surrounding the immediate vicinity. Once the sombrero
filter has been applied, edges can be inferred by looking for zero crossings, places
where values change from negative to positive e
Let us consider some of the implications of this process. First, the tech-
nique is not particularly complicated. Second, experiments have shown that it is
reasonably successful. In general, edges are correctly inferred. False hypotheses are
generated, but these can be eliminated by later processing that incorporates knowl-
edge about the types of objects we expect to see in the environment and the nature
of their edges.? Third, there is evidence that the hardware exists in mammalian
brains to perform this type of transformation. For example, David H. Hubel and
Torsten N. Wiesel of Harvard Medical School have discovered specialized edge
detector cells in the outer (early) layers of the visual cortex of the human brain.'?
Most important is a conclusion we can draw regarding the amount of
computation required to perform edge detection. While it has not been proved that

Ascanned image and the edges detected

by an edge detection system developed by
J. F. Canny in 1983. (From J. F. Canny,
“Finding Edges and Lines in Images,” MIT
Artificial Intelligence Laboratory, TR-720,
figures 6.14a and 6.14b)

this precise filter, the Laplacian of a Gaussian convolver, is used in mammal vision, it
can be shown that any algorithm that could possibly perform edge detection with the
facility of human (and apparently most mammal) vision must use a center-surround
filter (a filter in which each pixel influences all pixels within a certain distance) that
requires a comparable amount of computation. This amount turns out to be vast and
is determined by a six-dimensional computation. First, the filter must be applied for
every pixel, and the pixels are organized in a two-dimensional array. For each pixel
we must apply the filter to all pixels in a two-dimensional array surrounding that
pixel, which gives us a four-dimensional computation. We noted earlier that the
sharpness of our edge analysis was a function of the size of the Gaussian (normal)
curve applied. In the combined sombrero filter, the size of the Mexican hat has the
same impact. A large sombrero will enable us to detect the edges of large objects; a |
small sombrero will detect smaller features. We thus need to perform this entire

computation several times, which is a fifth dimension. The sixth dimension is time;

226
since vision must be capable of dealing with moving images, this entire computation
must be repeated many times each second. Undoubtedly, some optimizations can
be applied. For example, if we note that portions of the image are not changing, it is
not necessary to repeat all of the computations. Nonetheless, the number of compu-
tations required is essentially determined by this six-dimensional array."'
Let us plug in some numbers to get a feeling for the orders of magnitude
involved. Human vision is estimated to have a resolution of 10,000 positions along
each of the two axes of vision, or about 100 million pixels (there are indeed about
100 million rod cells in each eye to detect shape and motion and 6 million cone cells
to detect color and fine detail).'* The diameter of typical sombrero filters used in
computer-vision experiments range from 10 to 30 pixels, but these experiments are
based on images of only 1,000 pixels on a side. A reasonable average size for a
human sombrero filter would be about 100 by 100 pixels. If we assume about 3
different sombreros for different size objects and a refresh rate of recomputing the
image of 30 times per second, we have the following number of multiplications per
second: 10,000 x 10,000 x 100 x 100 x 3 x 30, or about 100 trillion. Now, a
typical personal computer can perform about 100,000 multiplications per second.
Thus, we would need about a billion personal computers to match the edge detec-
tion capability of human vision, and that’s just for one eye!"?
Typical computer vision systems have somewhat less demanding specifi-
cations. Typically image resolution is about 1,000 by 1,000 pixels, which requires
smaller filters of about 25 by 25 pixels. With three filters of different sizes and a
refresh rate of 30 images per second, we have 1,000 x 1,000 x 25 x 25 x 3 x 30,
or only 60 billion multiplications per second, which could be handled in real time by a
mere 600,000 personal computers.
This brings us back to the issue of digital versus analog computation. As
mentioned earlier, the need for massive parallel processing (doing many computa-
tions at the same time) may reverse, at least partially, the trend away from analog
computing. While it is possible to achieve billions of digital computations per second
in our more powerful supercomputers, these systems are large and expensive. The
computations described above for the sombrero filter do not need high degrees of
accuracy or repeatability, so analog multiplications would be satisfactory. Multiplying
60 billion analog numbers per second (600,000 computing elements each performing
100,000 multiplications per second) could be achieved using VLSI circuits in a
relatively compact system. Even the 100 trillion multiplications per second required
for human vision, though out of the question using digital circuits, is not altogether
impractical using analog techniques. After all, the human brain accomplishes image-
filtering tasks using just this combination of methods: massive parallel processing
and analog computation.'*
The human visual system picks up an image with 100 million specialized
(rod and cone) cells. Multiple layers, each of a comparable number of cells, would
have the capability to perform transformations similar to the sombrero filter de-
scribed above. In fact, the visual cortex of the brain contains hundreds of layers, so
these filtering steps are but the first transformations in the long (but quick) journey
of processing that a visual image undergoes.'*

227
 The images from both eyes need to be processed, and then the two
images need to be fused into one through a technique called stereopsis. As a result
of having two eyes, we can detect depth; that is, we can determine the relative
distance of different objects we see.'® Because our eyes are a few inches apart, the
same object will be slightly shifted in the images they receive. The amount of shift is
determined by simple trigonometric relationships. Distant objects will have little shift,
whereas close objects will have larger shifts. However, before our visual system can
apply trigonometry to the problem of determining depth it needs to line up the
corresponding objects in the two visual fields. This is more difficult than it sounds.
Experiments indicate that matching the image of each object in the visual field of
one eye to the image of that object in the visual field of the other must take place
after the detection of edges.'? Once edge detection has taken place, the edges can
be matched using additional pattern-recognition techniques."®
Once the edges are detected and the dual images fused with correspond-
ing information regarding depth, it becomes possible for more subtle processes of
discrimination to begin. Edges and depths can be organized into surfaces, the
texture of the surfaces can be estimated, and finally the objects themselves
identified.'? In this process a great deal of knowledge about the types of objects we
expect to see in our environment is used. The paradigm of hypothesis and test is
clearly used here in that people typically see what they expect to see in a situation.
Visual experiments have shown that people often misrecognize objects that are not
expected if they appear to be similar to those that are anticipated. This indicates that
the testing of the hypotheses has given a positive result. If an unusual object does
not match our hypothesis (i.e., the test fails), then that object is likely to grab our
focus of attention
We have now described a fundamental way in which pattern recognition in
Picture 1 Picture 2
general, and vision in particular, differs from the logical processes of thought. The
essence of logic is sequential, whereas vision is parallel. | am not suggesting that
Stereo vision. The different views
provided by our two eyes enable us (in the human brain does not incorporate any parallel processing in its logical analyses,
most instances) to reconstruct the but logical thinking generally involves considering only one transformation and its im-
three-dimensional arrangement of plications at a time. When speaking of parallelism in human vision (and in any
objects. (Courtesy of Paul Cohen and
Edward Feigenbaum) attempt to truly emulate vision in a machine), we are speaking not of a few compu-
tations at the same time but rather of billions simultaneously. The steps after edge
detection also involve vast amounts of computation, most of which are also accom-
plished through massive parallelism.7° Only in the final stages of the process do we
begin to reason about what we have seen and thereby to introduce more sequential
logical transformations. Though vision involves vastly greater amounts of computa-
tion than logical processes, it is accomplished much more quickly because the
number of processing stages are relatively fewer. The trillions of computations
required for the human visual system to view and recognize a scene can take place
in a split second
This explains the relatively automatic (not consciously controlled) nature of
vision: these tremendously parallel circuits are constantly processing information and
piping their results to the next stage. It is not a process we can turn off unless we
close our eyes. Even then we have trouble preventing our imagination from present-
ing images for analysis.

228
 Logical thought appears to be a more recent evolutionary development
than pattern recognition, one that requires more conscious control over each
sequential step.?' The amount of computation required is not as vast, and less
massive parallelism appears to be involved. This is one reason that we have been
more successful in emulating these more “advanced” logical processes in our
“intelligent” machines. Despite the relatively slow speed of neuronal circuits, the
massive parallelism of the human brain makes it capable of vastly more computation
than today’s computers. Thus, the relative lack of computational capability of
computers to date (less parallel processing) have rendered them inadequate for a
level of visual processing comparable to human vision. On the less computationally
intensive (and more sequential) tasks of solving problems and playing games, even
the very early computers were sufficient to perform at credible levels. Conversely,
the brain’s capacity for massive parallel processing is at least one of the keys to the
apparent superiority of human versus computer thought in areas such as vision.”

Parallel processing
The realization of this superiority has focused attention on breaking the von Neu-
mann bottleneck of conventional, single-processor computers. W. Daniel Hillis’s
Connection Machine, for example, is capable of 65,536 computations at the same
time, and machines with a millionfold parallelism are on the way.” Billions of
simultaneous processes, particularly if analog methods are combined with digital, are
not out of the question.2*
The realization that certain critical mental processes are inherently mas-
sively parallel rather than sequential has also refocused attention on the neural net as
an approach to building intelligent machines.”° The 1960s concept of a neural net
machine incorporated very simple neuron models and a relatively small number of
neurons (hundreds or thousands) organized in one or two layers. They were provided
with no specific task-oriented algorithms and were expected to organize themselves
by rearranging the interneuronal connections on the basis of feedback from the
human trainer. These systems were capable of recognizing simple shapes, but
Minsky and Papert showed, in their classic Perceptrons, that the machines were
essentially just matching individual pixel values against stored templates. These early
neural nets were simply not capable of more sophisticated discriminations.”° As
noted earlier, the 1980s school of neural nets uses potentially more capable neuron
models that can incorporate their own algorithms.2’ Designers are targeting systems
with millions of such artificial neurons organized into many layers. Though the self-
organizing paradigm is still popular, its role can be limited. Predetermined algorithms
can be built into both the neuron models themselves and the organization of each
layer. For example, a layer designed to detect edges should be organized differently
from a layer designed to integrate edges into surfaces. Of course, this is still a far
cry from the human visual system, with its billions of neurons organized into
hundreds of layers. We still have very limited understanding of the algorithms
incorporated in most of the layers or even what their functions are. Greater insight
into these issues will be required before neural nets can solve real problems. Minsky
and Papert remain critical of the excessive reliance of the new connectionists on the
self-organizing paradigm of neural nets. In the prologue to a new edition of

229
The Connection Machine, 65,536
computers in one. Al Researchers
now realize that massive parallel
processing is one of the keys to
achieving human-level intelligence.
(Courtesy of Thinking Machines
Corp.)
 Perceptrons (1988) they state, “Our position remains what it was when we wrote
the book: We believe this realm of work to be immensely important and rich, but we
expect its growth to require a degree of critical analysis that its more romantic
advocates have always been reluctant to pursue—perhaps because the spirit of

connectionism seems itself to go somewhat against the grain of analytic rigor.”"

Another difference between logical and imaginal thinking is the issue of
gradual versus catastrophic degradation Nn animal vision the failure of any neuron

to perform its task correctly is irrelevant. Even substantial portions of the visual
cortex could be defective with relatively little impact on the quality of the end result
Leaving aside physical damage to the eyes themselves, the ability of the human

brain to process visual images typically degrades the same way that a holographic
(three-dimensional, laser-generated) picture degrades. Failure of individual elements
subtract only marginally from the overall result. Logical processes are quite different
Failure of any step in a chain of logical deductions and inferences dooms the rest of
the thought process. Most mistakes are catastrophic (in that they lead to an invalid
result). We have some ability to detect problems in later stages, realize that earlier
assumptions must have been faulty, and then attempt to correct them, but our ability
to do this is limited
The difference between parallel thinking and sequential thinking is signifi-
cant in skill acquisition. When we first learn to perform a pattern-recognition task
(learning a new type of alphabet, for example, or, on a higher level, a new language),
we use our rational facilities to reason through the decision-making tasks required
This tends to be slow, deliberate, and conscious.*° As we “master” the new task,

The promise of parallel processing.

Hidehiko Tanaka of the University of
Tokyo designs super fast computers
that incorporate thousands of tiny
processors working in unison on a
single problem. (Photo by Lou Jones)
Yuichiro Anzai, one of the leading Al
authorities in Japan. (Photo by Lou
Jones)

our parallel facilities take over and we no longer need to consciously think through
each step. It seems just to happen automatically. We have programmed our parallel
pattern-recognition systems to take over the job. The process of recognition be-
comes substantially faster, and we are no longer conscious of the steps in the
process. Visual-perception experiments have indicated that when we read, we do
not perform recognition on individual characters and then group the characters into
words but rather recognize entire words and even groups of words in parallel. lf we
had to reason through each discrimination (e.g., “Now there's a semicircle with a
straight line to the left of it, so that must be a p”), our reading speeds would be
extremely slow. Indeed, a child’s reading speed is very slow until the child has
succeeded in programming his parallel pattern-recognition facilities to recognize first
individual letters, then words, finally, after years, groups of words
There is a similar phenomenon on the output side of human intelligence
When we learn to perform a certain task that involves the coordination of our
muscles (learning a sport or even speaking a new language), we start out very

deliberately and conscious of each step in the process. After we “master” the new
skill, we are conscious only of the higher-level tasks, not of the individual steps. We
have gone from sequential to parallel thinking
One of the objections that philosophers such as Hubert Dreyfus have
made of Al is that computers appear to lack the ability for this type of parallel
thought (the objection is generally expressed in the much vaguer terms that comput-

232
ers lack intuition).*! It is true that the purely logical processes of most expert sys-
tems do not have the capacity for achieving this vital category of massively parallel
thought. It is not valid, however, to conclude that machines are inherently incapable
of using this approach.
One might point out that even massively parallel machines ultimately use
logic in their transformations. Logic alone, however, is not the appropriate level of
analysis to understand such systems. It is similar to trying to understand meteorol-
ogy using the laws of physics.*? Obviously, cloud particles do follow the laws of
physics, but it is hopeless to attempt to predict the weather by means of the physics
of particle interactions alone (not that we are very successful at weather forecasting
even with “appropriate” methods). As an example of the weakness of rule-based
methodologies in mastering certain intelligent tasks, consider the problem of describ-
ing how to recognize faces using logic alone. Face recognition is a process we are
very good at despite our having little awareness of how the process actually works
No one has been able to program a computer to perform this task, in part because
no one can begin to describe how we perform this feat. In general, we find it far
easier to reconstruct our mental processes for sequential thinking than for parallel
thinking because we are consciously aware of each step in the process.

Building a brain
We can draw conclusions from the above discussion regarding some of the capabili-
ties required to simulate the human brain (i.e., to emulate its functionality). Clearly,
we need a capacity for hundreds of levels of massively parallel computations (with
the parallelism of each stage potentially in the billions). These levels cannot be fully
self-organizing, although the algorithms will in some cases allow for “growing” new
interneuronal connections. Each level will embody an algorithm, although the algo-
rithms must permit learning. The algorithms are implemented in two ways: the
transformations performed by the neurons themselves and the architecture of how
the neurons are connected. The multiple layers of parallel neuronal analysis permit
information to be encoded on multiple levels of abstraction. For example, in vision,
images are first analyzed in terms of edges; edges form surfaces; surfaces form
objects; objects form scenes.
Another example is human written language. Lines and curves form
letters, which form words, which form phrases, which form sentences, and so on. In
spoken language, we have sounds forming phonemes, which form words, and so
on. Knowledge regarding the constraints of each level of abstraction is used in the
appropriate layer. The knowledge itself is not built in (though algorithms for manipu-
lating it may be) and methods need to be provided to acquire, represent, access, and
utilize the domain-specific knowledge.
Each level of analysis reduces information. In vision, for example, we start
with the signals received from the hundred million rod and cone cells in each eye
This is equivalent to tens of billions of bits of information per second. Intermediate
representations in terms of surfaces and surface qualities can be represented with
far less information. The knowledge we finally extract from this analysis is a reduc-
tion of the original massive stream of data by many orders of magnitude. Here too

233
we see the selective (i.e., intelligent) destruction of information discussed earlier as
the purpose of computation.*
The human brain has a certain degree of plasticity in that different areas of
the brain can often be used to represent the same type of knowledge. This property
enables stroke victims to relearn lost skills by training other portions of the brain that
were not damaged. The process of learning (or relearning) requires our sequential
conscious processes to repetitively expose the appropriate parallel unconscious
mechanisms to the knowledge and constraints of a pattern-recognition or physical-
skill task. There are substantial limits to this plasticity, however. The visual cortex, for
example, is specifically designed for vision and cannot be used for most other tasks
(although it is involved in visual imagination, which does impact many other areas of
thought).
We can also draw a conclusion regarding the type of physical construction
required to achieve human-level performance. The human brain achieves massive
parallelism in all stages of its processing measured in the tens or hundreds of billions
of simultaneous computations in a package substantially under one cubic foot, about
the size of a typical personal computer. It is capable of this immense level of per-
formance because it is organized in three dimensions, whereas our electronic circuits
are currently organized in only two. Our integrated-circuit chips, for example, are
essentially flat. With the number of components on each side of a chip measured in
the thousands, we are limited to a few million components per chip. If, on the other
hand, we could build three-dimensional chips (that is, with a thousand or so layers of
circuitry on each chip instead of just one), we would add three orders of magnitude
to their complexity: we would have chips with billions rather than mere millions of
components. This appears to be necessary to achieve hardware capable of human
performance. Evolution certainly found it necessary to use the third dimension when
designing animal brains.*° Interestingly, one way that the design of the human brain
uses the third dimension is by elaborately folding the surface of the cerebral cortex
to achieve a very large surface area
A primary reason that the third dimension is not utilized is thermal prob-
lems. Transistors generate heat, and multiple layers would cause chip circuitry to
melt. However, a solution may be on the horizon in the form of superconductivity:
because of their lack of electrical resistance, superconducting circuits generate
virtually no heat. This may enable circuit designers to further reduce the size of each
transistor as well as to exploit the unexplored third dimension for a potential million-
fold improvement in performance.*®
David Marr and Tomaso Poggio pointed out another salient difference
between human brains and today’s computers in their first paper on stereo vision in
1976.3” While the ratio of connections to components in a conventional computer is
about 3, this ratio for the mammalian cortex can be as high as 10,000. In a computer
virtually every component and connection is vital. Although there are special fail-safe
computers that provide a small measure of redundancy, most computers depend on
a very high degree of reliability in all of their components. The design of mammalian
brains appears to use a radically different methodology in which none of the compo-
nents or connections are crucial; massive redundancy allows major portions of the
process to fail with little or no effect on the final results.%°

234
 In summary, there are two fundamentally different forms of thinking
logical thinking and parallel thinking. Logical thinking is sequential and conscious. It
involves deliberate control over each step It tends to be slow and errors in early
stages propagate throughout the rest of the process often with catastrophic results.
The amount of computation required tends to be limited. Thus, computers lacking in
parallel-processing capabilities (nearly all computers to date) have been relatively
successful in emulating some forms of logical thought. Most Al through the mid
1980s has been concerned with emulating this type of problem solving, with parallel
thought processes often being overlooked.*° This has led to criticism of Al, often
with the unjustified conclusion that computers are inherently incapable of parallel
thought. Parallel thinking is massively parallel. It is capable of simultaneously
processing multiple levels of abstraction, with each level incorporating substantial
knowledge and constraints. It tends to be relatively fast because of its highly parallel
construction. It generally takes place without either conscious direction or even
awareness of the nature of the transformations being made. Skill acquisition gener-
ally involves the sequential mind repeatedly training the parallel mind.

The principles of pattern recognition

Building on the observations above, we can describe several principles that govern
successful pattern recognition systems. While specific implementations and tech-
niques will differ from one problem domain to another, the principles remain the
same.
It is clear that parallel processing is important, particularly in the early
stages of the process, since the quantity of information is greatly reduced by each
stage of processing. Pattern-recognition tasks generally require a hierarchy of deci-
sions. Each stage has its own manner of representing information and its own
methods for deriving the information from the previous stage. For example, in vision
we represent the original image data in terms of pixel intensity values. Hypothesized
line segments, on the other hand, are probably represented in terms of the coordi- |
nates of the ends of each segment along with additional information about the
characteristics of each segment (curvature, edge noise, shading, etc.). Similarly,
surfaces are represented by a large number of coordinates plus information regard-
ing the surface characteristics. A variety of methods have been devised for repre-
senting objects, including the primal sketch, the 2'/2D sketch, and the world model.*°
A key issue in analyzing each stage of representation is segmentation. In
speech recognition, for example, we need to divide a sample of continuous speech
into smaller segments such as words or perhaps phonemes (basic sounds).*' Choos-
ing the appropriate types of segments for each stage is one of the most important
decisions in designing a pattern-recognition system. Once segments in the data have
been located, they can be labeled (described). In vision, for example, once we have
segmented a scene into line segments, we can describe the nature of the seg-
ments. We then segment the edge representation into surfaces and go on to label
the surfaces with their characteristics.
After we have determined the stages of processing, the representation of
information contained in each stage, the segments to be extracted, and the type of

235
Pattern Recognition

How about this pattern? How many

faces are there? Can you find a hand?
What other body parts can you
identify? Human pattern recognition
is capable of many levels of
abstraction. (Girl before a Mirror by
Pablo Picasso, in the collection
of the Museum of Modern Art, New
York)

236
 One approach to describing an image
is the primal sketch, shown here
with the original image. The primal
sketch is not an image at all but
a data base that describes brightness
change, blobs and textures. (From
Winston, Artificial Intelligence, 1984,
p. 337)

ess change: 15

The next level of representation, the

t ¥/ 2'/2-dimensional sketch (sometimes
called a needle diagram for obvious
“> reasons), describes the location
and orientation of surfaces. (From

a J Winston, Artificial Intelligence, 1984,

p. 338)

Finally, the world model, which

Hh describes the construction of objects
a / | in the image by modeling them as
ee | Vy | generalized “cylinders.” (From
a / Winston, Artificial Intelligence, 1984, |
ae awL = zs FS%)

aT a. Amore complicated world model

consisting of a wedge-shaped
cylinder with protrusions describes
the shape of a telephone. (From
Winston, Artificial Intelligence, 1984,
p. 340)

237
labeling desired for each segment, we are still faced with the heart of the problem:
designing methods to make the segmentation and labeling decisions. The most suc-
cessful paradigm | have found for accomplishing this is that of multiple experts.*
Usually the only methods available to perform specific recognition tasks are very
imperfect ones. Information theory tells us that with several independent methods of
relatively low accuracy we can still achieve high levels of accuracy if we combine
them in a certain way. These multiple methods, called experts, are considered
independent if they have what are called orthogonal invariances, that is, independent
strengths. Another way of saying the same thing is that the different experts (some-
times also called knowledge sources) tend to make different types of mistakes. The
goal is to assemble a group of experts diverse enough that for each pattern that
arises, at least one of the experts will have the proficiency to respond correctly. (Of
course, we still have to decide which expert is right, just as in ordinary life! | shall
come back to this question.)

Character recognition
As an example, consider the recognition of printed letters.*7 One useful expert we
can call on would detect a feature called the loop, which is an area of white com-
pletely surrounded by black. The capital A, for example, has one loop; B has two.
Another useful expert would detect concavities, which are concave regions facing in
a particular direction. For example, A has one concavity facing south, F has one
concave region facing east, and E has two concavities facing east.
Our loop expert would be proficient at distinguishing an O from a Cin that
O has a loop and C does not. It would not be capable, however, of discriminating C
from / (no loops in either case) or O from 6 (each has one loop). The concavity expert
could help us here, since it can distinguish C from / and 6 from O by the presence of
an east concavity in C and 6. Similarly, the concavity expert by itself would be unable
to distinguish C from 6 (since they both have an east concavity), but the loop expert
could identify 6 by its single loop. Clearly, the two experts together give us far
greater recognition capability than either one alone. In fact, using just these two
experts (a loop detector and a concavity detector), we can sort all 62 sans-serif
roman characters, excluding punctuation (A through Z, a through z, and 0 through 9)
into about two dozen distinct groups with only a few characters in each group. For
example, the group characterized by no loops with north and south concavities
contains only the characters H and N. In other words, if the loop expert examined a
pattern and indicated it had found no loops and the concavity expert indicated
concave regions facing south and north, we could conclude that the character was
(probably) either an H or an N. Additional experts that examined the location and
orientation of line segments or angle vertices could then help us to make a final
identification
It is clear that in addition to a set of experts that can provide us with the
ability to make all of the discriminations necessary, we also need a process to direct
and organize the efforts of these experts. Such a system, often called the expert
manager, is programmed with the knowledge of which expert to use in each situ-
ation.** It knows the relative strengths and weaknesses of each expert and how to

238
 The loop feature.

The concavity feature.

239
 Aand n touching

from a fund’:
along to its in’
Tealize the lo:
Noise around r

Typical defects in real print.

combine their insights into making final decisions. It would know, for example, that
the loop expert is relatively useless in discriminating 6 from O but very helpful for
determining whether a character is a 6 or a C, and so on.
In a real system (one that deals with images from the real world), classifi-
cations are rarely as straightforward as the examples above suggest. For example, it
is entirely possible that an A as actually printed might not contain a loop because a
printing error caused the loop to be broken (see figure). An a (which should contain
one loop) might actually contain two loops if an ink smear caused the upper portion
to close. Real-world patterns rarely display the expected patterns perfectly. Even a
well-printed document contains a surprisingly large number of defects. One way to
deal with the vagaries of real-world patterns is to have redundant experts and
multiple ways of describing the same type of pattern. There are a number of
different ways of describing what an A should look like. Thus, if one of our experts
failed (e.g., the loop expert), we still have a good chance of correctly recognizing the
pattern
There are many sources of variability. One, called noise for obvious
reasons, consists of random changes to a pattern, particularly near the edges,
caused by defects in the pattern itself as well as imperfections in the sensing
mechanism that visualizes the pattern (e.g., an image scanner). Another source of
variability derives from the inherent nature of patterns defined at a high level of
abstraction. For example, the concept of an A allows for a great deal of variation (see
figure). There are hundreds of different styles of type in common use and many
more if ornamental styles are considered. If one considers only a single type style,
then one could obtain accurate recognition using a relatively small number of
experts. If, on the other hand, we attempt to recognize printed characters drawn
from a wide multitude of styles, then it is clear that a substantially more diverse set

240
Ces UAy-430 te
CW Vb sel
Babin

m4 an is
GEARPAAAR
ARNAA AA
AAARAAR
 The classification of roman well-
printed sans-serif characters by loop
and concavity features.
e x
+ | Concaviies
oops
ia || Concavities
eee | _Concavities
se toe
_Concoviles |
OOO OO eae
ao 0 0s sop tei
° OO Orie
20s ao. Te a ere
26 0 1 ° 1 $Zsz25
° ° 1 Ones
Oo 50a ee Dr eee
0 1 oO 0 0 | JUVYjruv
° 1 ° ° Vy
0. pS eo ee
0 1 Of 2) 0 Ma
Ocal 1 ea re
° eee e ae
0 1 oa 1 > Xx
OES Sea 1 | 0 Ww
ve Te pe ° 0 0 ___ DOPQbdepqo4
1 0 ° Oe Seis.
1 o ° 1] 0 Ae
SS eee ee O66
1 ° 1 PS OR
2 0 ° o T= [9
yas Br aed te ee) o
2 ° 1 ° 1 :

Disambiguating N from H using a line

segment expert.

Southeast
Une
Segment

to
Southwest D> Southeast
seo Line Segment

a
Midwest
0
Mideast
‘Une Segment.
(a "eoss-bar")

242
The loop and concavity features of 62
roman sans-serif characters. Some of
the concavities are ambiguous or
marginal. For example, the southern
concavity in the letter a is so small it
may be overlooked by the concavity
expert. Thus in a practical system a
would be classified in both the “has
one southern concavity” and “has no
southern concavity” categories. To
account for multiple type fonts most
characters will in fact have multiple
classi ations.
of experts is required.*° Allowing such variability in the patterns to be recognized also
complicates the task of the expert manager.
Since the classification of patterns in the real world is often not clear cut, it
is desirable for our experts to provide their “opinions” on a continuous scale. Rather
than stating that this pattern has a loop, it would be of greater value for the expert to
indicate its relative level of confidence in the presence of such a property (e.g.,
“There is a 95 percent probability of there being one loop in this pattern, a 3 percent
probability of there being two loops”). A less-than-certain result might indicate that
the loop expert almost found a loop, that the “loop” found is broken by a few pixels.
Even if the loop is entirely closed, there is always the possibility that it really should
not be there at all but is only an artifact of a printing or scanning error. If all of the
experts provide their analyses in terms of probabilities, then the expert manager can
use information theory to combine these results in an optimal way.
In cases of significant print distortion, even human perception can fail on
the level of individual letters. Yet we are often able to correct for printing defects by
using our knowledge of janguage context. For example, if we have trouble distin-
guishing a t from a c because of poor printing, we generally look (consciously or
unconsciously) at the context of the letter. We might determine, for example, that
“computer” makes more sense than “compucer.” This introduces the concept of
experts that use knowledge of the constraints of higher levels of context. Knowing
that “compucer” is not a word in English but that “computer” is enables us to
disambiguate an otherwise ambiguous pattern. Similarly, in the field of speech
recognition, the only possible way to distinguish the spoken word “to” from “too”
and from “two” (all of which sound identical) is from context. In the sentence “| am
going to the store,” we can eliminate “too” and “two” from consideration by relying
on our higher-level syntactic knowledge. Perceptual experiments indicate that human
pattern recognition relies heavily on such contextual discrimination. Attempting to
recognize printed letters without a word context, human speech without a sentence
context, and musical timbres without a melodic context sharply reduces the accuracy
of human perception. Similarly, machines dealing with highly variable types of
patterns require extensive use of context experts with substantial knowledge about
their domains. A word-context expert in a character-recognition system requires
knowledge of all the possible words in the language. A syntactic expert in a speech-
recognition system requires knowledge of possible word sequences. Again, an
expert that can say, “’Phthisis’ has a probability of .0001,” is more valuable than one
who can only say, “‘Phthisis’ is possible.”
All of the experts mentioned above deal with relatively abstract concepts.
Concavity is not a perfectly defined concept. Detecting this property is not straight-
forward and requires a relatively complex program. A very different category of
experts, low-level experts (as distinguished from the high-level experts described
above), deal with features that are simple transformations of the original input data.
For example, in any type of visual recognition we could have a low-level property
associated with every pixel whose value is simply the value of the pixel. This is, of
course, the simplest possible property. A slightly higher level property (but still low
level) could detect the amount of black in a particular region of the image. For

244
Varieties of low-level (minimal)
property sets.

Each pixel
can be
a minimal
property

The amount (number of

pixels) of “black” in the
top third of the image

The amount of black

in the middle third of
the image

The amount of black

in the bottom third of
the image

The amount of black

in the leftmost third of
the image

The amount of black

in the middle third of
the image
The amount of black
in the rightmost third of
the image

Shoot ten “arrows” from

left to right and count
the number of times
image changes from
white to black
Shoot ten “arrows” from
top to bottom and count
the number of times
image changes from
white to black

245
Pattern fecogs
example, a T will tend to have more black in the upper region of the image than an
L, which will tend to be more black in the lower region. In actual use, minimal prop-
erties tend to be more complex than in the above two examples but nonetheless
use straightforward and well-defined transformations of the original input.
It turns out that such low-level properties are quite useful in recognizing
patterns when the possible types of patterns are highly constrained. For example, in
character recognition, if we restrict the problem to a single style of type, then a
system built entirely with low-level property experts is capable of a very high level of
accuracy (potentially less than one error in over ten thousand printed characters).
This limited problem is often attacked with template matching, so called because it
involves matching the image under consideration to stored templates of every letter
in the character set.*° Template matching (and other methods of minimal-property
extraction) also work well for recognizing printed letters drawn from a small number
of type styles. If we are trying to recognize any nonornamental type style (called
omnifont, or intelligent, character recognition), then an approach using only minimal
property extraction does not work at all. In this case, we must use the higher-level
(more intelligent) experts that are based on such abstract topological concepts as
loops, concavities, and line segments. The minimal properties can still play an
important role, however. Fortunately, printed material does not generally combine
multiple type styles in anything like a random fashion. Any particular document (e.g.,
a book or magazine) will tend to use a limited number of type styles in a consistent
way.’”? When an omnifont character-recognition machine first encounters a new
document, it has no choice but to use its intelligent experts (its loop expert, concav-
ity expert, etc.) to recognize the characters. As it begins successfully to recognize
characters, its higher-level experts can actually train its lower-level experts to do the
job, and its expert manager (which directs the overall recognition process) can begin
to rely more heavily on the lower-level experts for recognition. The higher-level
experts train the lower-level ones by presenting actual examples of recognized
characters and telling them, in essence, “Here are examples of characters as they
actually appear in this document, and this is what we believe their correct identifica-
tions to be.” The advantages of such an automatic learning process include both
speed and accuracy. The lower-level experts are not only potentially much faster,
they can also be less sensitive to image noise
To return to the first theme of this chapter, the higher-level experts in such
a character-recognition system are representative of logical analysis, whereas the
lower-level experts represent a more parallel type of thinking. The lower-level
experts use much simpler algorithms, so they are more amenable to massive parallel
processing, which is a major reason for their potential speed advantage
Interestingly, perceptual experiments indicate that the human visual
system works in a similar way. When we first encounter a new typeface, to recog-
nize it, we rely on our conceptual understanding of print (a logical type of analysis),
and our recognition speeds are relatively slow. Once we get used to the style, our
recognition process becomes less analytic, and our speed and accuracy increase
substantially. This is another example of our logical mind training our parallel mind.
The paradigm of pattern recognition described above is common to most
serious recognition problems: multiple stages of processing based on a hierarchy of

246
 levels, massive parallel processing (particularly in the early stages), segmentation and
labeling, multiple experts on both high and low levels, expert management, disam-
biguation using the constraints of higher levels of context, and learning from actual
recognition examples.** The actual content of the paradigm, however, will differ
substantially from one problem area to another. Most of the technology of any
successful pattern-recognition system is domain specific; that is, it is based on the
detailed nature of the types of patterns to be recognized. Every so often one hears
claims regarding a general-purpose pattern-recognition system that can recognize any
type of pattern—printed characters, spoken words, land-terrain maps—regardless of
their source. As mentioned earlier, while such systems do recognize many types of
patterns, they perform these tasks poorly. To perform any specific pattern-recognition
task well with commercially acceptable rates of accuracy requires substantial knowl-
edge deeply embedded in the algorithms and specific to the domain of inquiry

+ The Real World

Looking at the real world

Attempts to emulate the general capabilities of human vision are being pursued at a
number of leading Al laboratories. One is an ambitious project to create an artificial
eye-head system at the MIT Vision Laboratory under the direction of Tomaso Pog-
gio.*® The MIT work includes edge detection (using the Laplacian Gaussian convolver
described above and other similar algorithms), fusing stereo images to provide
information on depth, understanding color perception, reconstructing surfaces and
their properties, tracking trajectories of moving objects, and the ultimate problem of
describing the content of what is seen. One of the most interesting aspects of the
MIT work is the development of a new type of computer that combines digital
control with massive analog parallelism.5° Experiments conducted by Poggio, his
associate Christof Koch, and others have already suggested that the human nervous
system appears to be capable of substantial parallelism (hundreds) of analog compu-
tations within a single neuron.
| This work is exemplary of the paradigm of multiple experts. A number of
| different systems, each with extensive knowledge of a specific aspect of the visual-
perception task, are combined in the MIT eye-head system.*! For example, an expert
being developed by Anya Hurlbert and Tomaso Poggio uses knowledge of the
spectral (color) reflectance of surfaces to help describe them.°*? The project also
addresses the issues of integrating visual perception with the mechanical control of a
robot and includes a head with two solid-state cameras for eyes (that is, cameras
with a special chip called a charge-coupled device as an electronic retina).
A major center for the development of vision systems and their application
to the field of robotics is the Robotics Institute (RI) at Carnegie-Mellon University
under the direction of Al pioneer Raj Reddy. A particularly ambitious project at Rl,
funded by the Defense Advanced Research Projects Agency (DARPA) is an autono-
mous vehicle called Terregator (Terrestrial Navigator) which combines a high-resolu-
tion vision system, parallel processing, and advanced decision-making capabilities.*°
In view of the strong Japanese commitment to the application of robotics
to production techniques, Japanese researchers have targeted vision as a priority

247

Se
Ne ll
 Raj Reddy, director of the Robotics
Institute of Carnegie-Mellon
University. Reddy has been a pioneer
in the development of voice
recognition, computer vision, and
robotics. Now working on the Terre-
gator (Terrestrial Navigator) for the
Defense Advanced Research
Projects Agency (DARPA), Reddy
predicts that future robotic- ision
systems will eventually revolutionize
driving and provide cars with
effective collision-control and road-
following capabilities. (Photo by Lou
Jones)

Computer vision pioneer Tomaso

Poggio at the MIT Vision Lab. (Photo
by Lou Jones)
 Takeo Kanade, director of the
Computer Vision Laboratory of
Carnegie-Mellon University. Kanade
is developing a highly parallel
computer for real-time processing of
images from robotic-vision systems.
(Photo by Mary Jo Dowling of the
Robotics Institute, Carnegie Mellon
University)

research topic. Building on the work of Poggio and his associates, Yoshiaki Shirai (of
the Electrotechnical Laboratory, !baraki, Japan) and Yoshiro Nishimoto (of the Re-
search Laboratory, Kobe Steel, Kobe, Japan) are attempting to build a practical
system for fusing stereo images. Based on parallel hardware, the Shirai-Nishimoto
system uses a Laplacian of a Gaussian convolver (a sombrero filter) as well as more
advanced pattern-matching techniques. Japanese development efforts are emphasiz-
ing the integration of vision with real-time robotic control to provide a new genera-
tion of robots that can see their environment, perceive and understand the relevant
features of objects, and reason about what they have seen. Hirochika Inoue and
Hiroshi Mizoguchi (of the University of Tokyo) have developed a system that can
detect, recognize, and track rapidly moving objects in real time
One promising approach to organizing the massive parallelism required for
pattern-recognition tasks is to develop specialized chips to perform those tasks
requiring the most computation. One researcher pursuing this approach is Carver A
Mead (of the California Institute of Technology), one of the original pioneers in the
development of design methodologies for large-scale integrated circuits. Mead and
his associates have developed an artificial-retina chip that performs such early-vision
tasks as edge detection and the adjustment of an image for the effects of varying
levels of illumination.** One of the innovations of Mead’s approach is his reliance on

massively parallel analog circuits to provide the bulk of the computation. Mead is
also working on an artificial-cochlea chip based on similar principles
While research is just beginning on systems that emulate the full range of
human visual processing, machines that perform more limited tasks of visual
perception have already found significant commercial applications. For example
optical character recognition (OCR) was a $100 million industry in 1986 and is
projected to grow to several hundred million dollars in 1990.°° Applications include
reading aloud for the blind, as well as scanning printed and typed documents for
entry into word processing, electronic publishing, transaction processing, and data-
base systems

249
 Seeing and believing. At the
Tsukuba Research Center in Japan,
Yoshiaki Shirai’s research in
robotics focuses on the development
of three-dimensional vision systems.
(Photo by Lou Jones)

Hirochika Inoue, a pioneer in robotic

vision systems, at the University of
Tokyo. (Photo by Lou Jones)
Makoto Nagao of Kyoto University
explores pattern recognition by
means of shadows and surfaces.
(Photo by Lou Jones)
 Systems using pattern-recognition techniques are revolutionizing the
= handling of fingerprints by law enforcement agencies. A system called the Auto-
mated Fingerprint Identification System (AFIS) developed by NEC of Japan enables
agencies across the United States to rapidly identify suspects from fingerprints or
even small fragments of fingerprints by intelligently matching them against the
stored prints of hundreds of thousands of previously arrested men and women. A
report by the U.S. Bureau of Justice Statistics stated, “AFIS may well have the
greatest impact of any technological development on law enforcement effectiveness
since the introduction of computers to widespread use in the criminal justice system
in the 1960s."°° AFIS is capable of identifying a suspect in several minutes; the
manual methods it replaces took months or even years
Similar techniques are being used in security devices. Systems manufac-
tured by Fingermatrix, Thumbscan, and other firms include a small optical scanner
into which a person inserts his finger.5” The device quickly reads the person’s finger
pattern and uses pattern-recognition techniques to match it against stored images.
The system can control entry to restricted areas and protect information in comput-
ers from unauthorized access. Such systems could eventually replace ordinary locks
and keys in homes and cars.
One of the largest applications of commercial vision systems so far can be
found in factories, where the systems are used for inspection, assembly, and
process control. Such systems typically use solid-state cameras with specialized

Systems for automated fingerprint

analysis identify distinguishing ridge
points that characterize each
individual. (Courtesy of Fingermatrix)
 Robotic vision. The robot's TV
camera eye sees a jumble
of disposable razors. None of the
objects can be seen in full. Many
crucial features are hidden or
otherwise obscured. (Photo by
David Lowe)

In this three-dimensional-object-
recognition system developed by
David Lowe at New York University
the first step is to extract line
segments by looking for sudden
changes in intensity. (Photo by
David Lowe)

Next the computer attempts to match

the two-dimensional projections
represented in the image with
its internal model of the three-
dimensional shape of a razor. Here
one set of lines are from the model,
and the other set of lines are
matching segments from the original
picture. (Photo by David Lowe)

Finally, the location and orientation

of each object are recognized. The
full models are shown here
superimposed on the original image.
The dotted lines represent hypothe-
sized line segments that are hidden
from view. (Photo by David Lowe)

253
Robert Shillman and the Cognex
Machine Vision System. (Photos by
Lou Jones)
electronics to digitize moving images and provide for the computationally intensive
early phases of processing. A general-purpose computer with custom software
provides for the higher levels of analysis. One of the more sophisticated of such
systems has been developed by Cognex Corporation, founded by Robert Shillman
and a team of MIT Al researchers in 1981. One Cognex product can scan manufac-
tured products streaming by on a conveyor belt and detect and recognize such
information as serial numbers embossed in metal or even glass. Other Cognex
products can identify specific objects and their orientation for inspection and to assist
robotic assemblers. Other major providers of vision systems include Automatix,
Defracto, Perceptron, Robotic Vision Systems, and View Engineering. A major player
has been General Motors, which has provided investments and contracts for several
of the players. According to DM Data, overall revenues for the factory-vision industry
were over $300 million in 1987 and are projected to hit $800 million in 1990.%?
Military systems account for another major application of artificial vision.
The ability to scan and recognize terrain at very low altitudes is a crucial element of
the cruise missile, which can be launched thousands of miles from its intended
target. Modern fighter planes have a similar ability to track terrain and provide pilots
with a continually updated display of the location and trajectory of the aircraft. Smart
weapons (bombs, missiles, and other munitions) use a variety of sensing mecha-
nisms including vision to locate, identify, and reach intended targets.
The advent of weapons that can see has resulted in profound changes in
military tactics and strategy. As recently as the Vietnam War, it was generally
necessary to launch enormous numbers of passive blind weapons in relatively
indiscriminate patterns to assure the destruction of a target. Modern battlefield
tactics emphasize instead the carefully targeted destruction of the enemy with
weapons that can recognize their objective. Intelligent missiles allow planes, ships,
and submarines to destroy targets from relatively safe distances. For example, a
plane can launch an intelligent missile to destroy a ship from tens or even hundreds
of miles away, well out of range of the ship's guns. A new generation of pilotless
aircraft use pattern-recognition-based vision systems to navigate and launch weap-
ons without human crews.®’ Vision systems and other pattern-recognition technolo-
gies are also deployed in defensive tactics to recognize an incoming missile, but
such defense is generally much more difficult than offense. The result is an increas-
ing degree of vulnerability for such slow-moving targets as tanks and ships.
An area of emerging importance is the application of pattern recognition to
medicine. Medical diagnosis is, after all, a matter of perceiving relevant patterns from
symptoms, test results, and other diagnostic data. Experimental systems can look at
images from a variety of imaging sources—X-ray machines, CAT (computerized axial
tomography) scanners, and MRI (magnetic resonance imaging) systems—and
provide tentative diagnoses. Few, if any, medical professionals are ready to replace
their own perceptions with those of such systems, but many are willing to augment
their own analysis. Often an automatic system will detect and report a diagnosis that
manual analysis would have overlooked.®
One particularly promising medical application is the analysis of blood-cell
images. Certain types of cancer can be diagnosed by finding telltale precursor

255
The Cruise Missile. Relying heavily
on image processing and pattern-
recognition technology, the missile
identifies the terrain below it from
computerized maps in its memory.
This capability permits it to fly only The B-1 Bomber. Computers assist
hundreds of feet from the ground and a wide range of functions, including
thus to evade radar-based air flying, navigating, and targeting.
defenses. (Supplied by Photri) (Supplied by Photri)
Can you find the airplanes in this
picture? Tom Binford, head of
Robotics at Stanford University, with
images from his aircraft-recognition
system. (Photo by Dan McCoy of
Rainbow)

257
 This image of the head of a normal
18-week-old fetus is revealed by
sonography, which uses computer-
enhanced sound-wave reflections.
(Photo taken by Howard Sochurek at
the Rush Presbyterian St. Luke's
Medical Center, Chicago)

Computers peer into the human brain.

Five new computer-based scanning
systems provide doctors with
dramatic and precise images of the
brain and its ailments. Magnetic
resonance imaging (MRI) uses radio
waves and a strong magnetic field.
This MRI image shows a herniating
finger of tissue from a brain slumping
into the base of the skull. (Photo
taken by Howard Sochurek at the
Mallinckrodt Institute of Radiology,
Saint Louis)

Shigeru Eiho, director of Engineering

at Kyoto University, with his medical
imaging system. (Photo by Lou Jones)
 malignant cells in a blood sample. Analysis of blood samples by human technicians
typically involve the examination of only about a hundred cells. By the time a malig-
nant cell shows up in a sample that small, the cancer is often too advanced to be
effectively treated. Unhampered by fatigue or tedium and able to operate at speeds
hundreds of times greater than human technicians, artificial “technicians” using
pattern-recognition techniques can search for signs of cancer in hundreds of thou-
sands or even millions of cells and thus potentially detect the presence of disease at
a treatable stage
Ultimately, medical applications of pattern recognition will have enormous
benefit. Medical testing comprises a major fraction of all of medicine and costs
several hundred billion dollars per year. Examining the images and other data
resulting from such tests is extremely tedious for human technicians, and many
studies have cited the relatively low level of accuracy that results. Many doctors
routinely order tests to be conducted in duplicate just to improve their accuracy.
Once computers have mastered the requisite pattern-recognition tasks, the potential
exists for a major transformation of medical testing and diagnosis.

A dangerous aneurysm is revealed in

an X ray processed by digital
subtraction angiography, a computer
enhancement technique. (Photo
taken by Howard Sochurek at the
University of Kansas Medical Center)

The positron emission tomography

images reveal damage from a
stroke. (Photo taken by Howard
Sochurek at the Mallinckrodt
Institute of Radiology, Saint Louis)

One area of medicine that is already being revolutionized is medical

imaging. Using a variety of computer-based image-enhancement techniques,
physicians now have access to unprecedented views inside our bodies and brains
Similar techniques also allow scientists to visualize such extremely small biochemical
phenomena as viruses for the first time
One of the more surprising results of image processing and recognition
took place recently when Lillian Schwartz observed the striking unity of the juxta-
posed halves of the “Mona Lisa” and the reversed “Self-Portrait” by Leonardo da
Vinci. Further investigation by Schwartz led her to identify Leonardo as the model
used to complete the “Mona Lisa,” thereby suggesting a remarkable conclusion to
the 500-year-old riddle of the identity of the celebrated painting.*

259
A glaring red tumor reveals itself in
this computer-enhanced three-
dimensional image from a computer-
aided tomography scanner. (Photo
taken by Howard Sochurek at the
University of Kansas Medical Center)

The computer imagines a tomato

virus. This virus is too small to be
seen even with an electron
microscope. However, scientists can
now “see” the virus for the first
time using computer-generated
images. To create these pictures, the
computer was provided information
about the virus’s molecular structure
and chemical composition. (Courtesy
of Lawrence Livermore National
Laboratory)

261
 fn Hecognitio

Who is this lady? Mona Lisa

revealed. With the help of computer
image processing, Lillian Schwartz
reached the controversial
conclusion that Mona Lisa was none
———

other than the artist himself.

(Mona-Leo by Lillian Schwartz. Dis-
cussed in The Computer Art Book,
W. W. Norton. Copyright 1987
by Computer Creations Corporation)

262
 Listening to the real world
Another human sense that computers are attempting to emulate is hearing. While
input to the auditory sense involves substantially less data than the visual sense
(about a million bits per second from both ears versus about fifty billion bits per
second from both eyes), the two senses are of comparable importance in our
understanding of the world. As an experiment, try watching a television news
program without sound. Then try listening to a similar broadcast without looking at
the picture. You will probably find it easier to follow the news stories with yours ears
alone than with your eyes alone. Try the same experiment with a situation comedy;
the result should be the same.
Part of the importance of our auditory sense is the close link of verbal
language to our conscious thinking process. A theory popular until recently held that
thinking was subvocalized speech.® While we now recognize that our thoughts
incorporate both language and visual images, the crucial importance of the auditory
sense in the acquisition of knowledge is widely accepted
Blindness is often considered to be a more serious handicap than deaf-
ness. A careful consideration of the issues, however, shows this to be a misconcep-
tion. With modern mobility techniques, blind persons with appropriate training have
little difficulty in travelling from place to place, reading machines can provide access
to the world of print, and the visually impaired experience few barriers to communi-
cating with other persons in groups and meetings large or small. For the deaf,
however, there is a barrier to engaging in a very fundamental activity—understanding
what other people are saying in person to person contact, on the phone, and in
meetings. The hearing impaired are often cut off from basic human communication
and feel anger at society's failure to accommodate or understand their situation.
We hear many things: music, speech, the varied noises of our environ-
ment. Of these, the sounds that are the most important in terms of interacting with

This is the CBS Evening News. A

speech spectograph provides a
picture of speech. In the large middle
portion labeled “Wide-Band
Spectogram” the vertical axis is
frequency (pitch) and the horizontal
axis is time. The darker the plot, the
more energy is represented for
that pitch at that moment. (Photo by
Dan McCoy of Rainbow)

263
 Jonathan Allen, director of the MIT
Research Laboratory of Electronics
and a pioneer in speech science.
Allen’s MiTalk System was a
forerunner of modern speech
synthesis programs. Allen has also
conducted important research in
speech understanding programs and
VLSI (very large scale integrated)
circuit design. (Photo by Dan McCoy
| of Rainbow)

Fred Jelinek, IBM researcher and

pioneer in the application of
statistical modeling techniques to
automatic speech recognition.
(Courtesy of Fred Jelinek)

The late Dennis Klatt, senior

research scientist at MIT, listening
to synthesized speech from his
KlatTalk program. Klatt’s algorithms
are used in two commercially
successful speech synthesizers,
DECtalk from Digital Equipment Cor-
poration and Prose 2000 from
Speech Plus. (Courtesy of the MIT
Research Laboratory of Electronics)

264
 and learning about the world are those of human speech. Approp!
aud! ry recogniti

°
tion in both human and machine.

know the most about a

Richard Goldhor, and others
other laboratories, we have some know
by the auditory nerve and early stages of the auditory cortex. As w
knowledge of the details of the later stages is relatively slight due mc
inability itis diff
analyze the human auditory system

recogn on to date has been devoted to teaching machines to und

As with vision, the success of such efforts may provide u: hw
to how the human auditory system might work, and these theor!

quently be verified by neurophysical experimentation

us examine automatic speech recognition (ASR) in

same way that the length and tautness of a violin or piano

We c the tautness of our vocal corc

the overtones produced by our vibrating voc:

and lips, which has the eff

act is a chamber

mber similar to the opening of another organ pipe

In addition to the pitched sound produ
produce a noiselike sound by the rush of air through t!

Heather Baker (age six) talks to a

speech recognizer developed by her
parents, Jim and Janet Baker, and
their associates at Dragon Systems.
When she says the correct word, the
bear smiles. (Photo by Hank Morgan of
Rainbow).

265
does not have specific overtones but is rather a complex spectrum of many frequen-
cies mixed together. Like the musical tones produced by the vocal cords, the spectra
of these noise sounds are also shaped by the changing resonances of the moving
vocal tract.©
This apparatus allows us to create the varied sounds that comprise human
speech. While many animals communicate with others of their species with sound,
we humans are unique in our ability to shape sound into language. Vowel sounds
(/a/, /e/) are produced by shaping the overtones from the vibrating vocal cords into
distinct frequency bands called formants. Sibilant sounds (/s/, /z/) are created by the
rush of air through particular configurations of tongue and teeth. Plosive consonants
(/p/, /k/, /t/) are transitory sounds created by the percussive movement of lips,
tongue, and mouth cavity. Nasal sounds (/n/, /m/) are created by invoking the
resonances of the nasal cavity.®”
Each of the several dozen basic sounds, called phonemes, requires an
intricate movement involving precise coordination of the vocal cords, alveolar flap,
tongue, lips, and teeth. We typically speak about 3 words per second. So with an
average of 6 phonemes per word, we make about 18 complex phonetic gestures
each second. We do this without thinking about it, of course. Our thoughts remain
on the conceptual (that is, the highest) level of the language hierarchy. In our first
two years of life, however, we thought a lot about how to make speech sounds (and
how to meaningfully string them together). This is another example of our sequential
(logical) conscious mind training our parallel (pattern recognition) mind.
The mechanisms described above for creating speech sounds—vocal cord
vibrations, the noise of rushing air, articulatory gestures of the mouth and tongue,
the shaping of the vocal and nasal cavities—produce different rates of vibration. A
physicist measures these rates of vibration as frequencies; we perceive them as
pitches. Though we normally consider speech to be a single time-varying sound, it is
actually a composite of many different sounds, each of which has a different fre-
quency. With this insight, most commercial ASR systems start by breaking up the
speech waveform into a number of different bands of frequencies. A typical com-
mercial or research ASR system will produce between three and a few dozen
frequency bands. The front end of the human auditory system does exactly the
same thing; each of the nerve endings in the cochlea (inner ear) responds to differ-
ent frequencies and emits a pulsed digital signal when activated by an appropriate
pitch. The cochlea differentiates several thousand overlapping bands of frequency,
which gives the human auditory system its extremely high degree of sensitivity to
frequency. Experiments have shown that increasing the number of overlapping
frequency bands in an ASR system (and thus bringing it closer to the thousands of
bands of the human auditory system) substantially increases the ability of that
system to recognize human speech.
Typically, parallel processing is used in the front-end frequency analysis of
an ASR system, although not as massively as in vision systems, since the quantity of
data is much less. (If one were to approach the thousands of frequency bands used
by the human auditory system, then massive parallel processing would be required.)
Once the speech signal has been transformed into the frequency domain, it is
normalized (adjusted) to remove the effects of loudness and background noise. At

266
 this point we can detect a number of features of frequency-band signals and
consider the problems of segmentation and labeling
As in vision systems, minimal property extraction is one popular technique
One can use as a feature set either the normalized frequency data itself or various
transformations of this data. Now, in matching such minimal-property sets, we need
to consider the phenomenon of nonlinear time compression.®° When we speak, we
change our speed according to context and other factors. If we speak one word
more quickly, we do not increase the rate evenly throughout the entire word. The
duration of certain portions of the word, such as plosive consonants, will remain
fairly constant, while other portions, such as vowels, will undergo most of the
change. In matching a spoken word to a stored template, we need to align the
corresponding acoustic events, or the match will never succeed. This problem is
similar to the matching of visuai cues in fusing the stereo images from our two eyes
A mathematical technique called dynamic programming has been developed to
accomplish this temporal alignment.”
As with vision systems, high-level features are also used in ASR systems
As mentioned above, speech is made up of strings of phonemes, which comprise
the basic “alphabet” of spoken language. In English and other European romance
languages, there are about 16 vowels and 24 consonants; Japanese primarily uses
only 5 vowels and 15 consonants. The nature of a particular phoneme (such as /a/) is
an abstract concept in the same way that the inherent nature of a printed character
(such as A) is: neither can be simply defined. Identifying phonemes in human speech
requires intelligent algorithms and recognition of high-level features (something like
the loops and concavities found in printed characters). The task of segmenting
speech into distinct time slices representing different phonemes is also a formidable
task. The time-varying spectrum of frequencies characterizing a phoneme in one

Nonlinear alignment of speech

events. The lines in the middle graph
show the optimal alignment of
the spectographs of two utterances
of the word “further” spoken by
the same female speaker. (Photo by
Vladimir Sejnoha and courtesy
of Kurzweil Applied Intelligence)

267
context may be dramatically different in a different context. In fact, in many in-
stances no time slice corresponding to a particular phoneme can be found; we
detect it only from the subtle influence it has op phonemes surrounding it.
As with vision and character recognition, both high-level and low-level
features have value in speech-recognition systems. For recognizing a relatively small
vocabulary (a few hundred words) for a single speaker, low-level feature detection
and template matching by means of dynamic programming is usually sufficient, and
most small-vocabulary systems use this approach. For more advanced systems, a
combination of techniques is usually required: generally, multiple experts and an
expert manager that knows the strengths and weaknesses of each.’!
High-level context experts are also vital for large vocabulary systems. For
example, phonemes cannot appear in any order. Indeed, many sequences are
impossible to articulate (try saying “ptkee”). More important, only certain phoneme
sequences will correspond to a word or word fragment in the language. On a higher
level, the syntax and semantics of the language put constraints on possible word
orders. While the set of phonemes is similar from one language to another, context
factors differ dramatically. English, for example, has over 10,000 possible syllables,
whereas Japanese has only 120.
Learning is also vital in speech recognition. Adaptation to the particular
characteristics of each speaker is a powerful technique in each stage of processing.
Learning must take place on a number of different levels: the frequency and time
relationships characterizing each phoneme, the dialect (pronunciation) patterns of
each word, and the syntactic patterns of possible phrases and sentences.
In sum, we see in speech recognition the full paradigm of pattern recogni-
tion that we first encountered in vision and character recognition systems: parallel
processing in the front-end, segmentation and labeling, multiple experts on both high
and low levels, expert management, disambiguation by context experts, and learning
from actual recognition examples. But while the paradigm is the same, the content is
dramatically different. Only a small portion of the technology in a successful ASR
system consists of classical pattern-recognition techniques. The bulk of it consists of
extensive knowledge about the nature of human speech and language: the shape of
the speech sounds and the phonology, syntax, and semantics of spoken language.”
Automatic speech recognition is receiving considerable attention because
of its potential for commercial applications.”? We learn to understand and produce
spoken language in our first year of life, years before we can understand or create
written language. Thus, being able to communicate with computers using verbal
language would provide an optimal modality of communication. A major goal of Al is
to make our interactions with computers more natural and intuitive. Being able to
converse with them by talking and listening is a vital part of that process.
For years ASR systems have been used in situations where users neces-
sarily have their hands and eyes busy, making it impossible to use ordinary computer
keyboards and display screens. For example, a laboratory technician examining an
image through a microscope or other technical equipment can speak results into the
microphone of an ASR system while continuing to view the image being examined.
Similarly, factory workers can verbalize inspection data and other information on the
production or shop floor directly to a computer without having to occupy their hands

268
with a keyboard. Other systems are beginning to automate routine business transac-

tions over telephone lines. Such telecommunications applications include credit-card

verification, sales-order entry, accessing data base and inventory records, conducting
banking and other financial transactions, and many others.
The applications mentioned are usually highly structured and thus require
the ASR system to recognize only a small vocabulary, typically a few hundred words
or tess. The largest markets are projected for systems that can handle vocabularies
that are relatively unrestricted and much larger, say, ten thousand words or more.
The most obvious application of large vocabulary ASR systems is the creation of
written documents by voice. The creation of written documents is a ubiquitous
activity engaged in by almost everyone in offices, schools, and homes. Just copying
all of the written documents created each year is a $25 billion industry. Tens of
billions of dollars are spent each year in the creation of original written works from
books to interoffice memoranda. Being able to dictate a document to a computer-
based machine and see the document appear on a screen as it is being dictated has
obvious advantages in terms of speed, accuracy, and convenience. Large-vocabulary
ASR-based dictation systems are beginning to see use by doctors in writing medical
reports and by many other professionals.
There are also significant applications of ASR technology to the handi-
capped. Persons who are unable to use their hands because of quadriplegia (paraly-
sis due to spinal-cord injury), cerebral palsy, or other neurological impairment can still
create written documents, interact with computers, and otherwise control their
environment by speaking to a computer equipped with ASR. Certain types of brain
damage cause a person's speech to be slurred and distorted. While such speech
patterns are unrecognizable by human listeners (unless they have received special
training), it is distorted in a consistent way and thus can be recognized by machine.
A system consisting of an ASR system to recognize the distorted speech connected
to a speech synthesizer can act as a translator, allowing the person to be understood
by others. Research has been conducted at Children’s Hospital, Boston, by Dr.
Howard Shane on assisting both the hand-impaired and the speaking-impaired using
ASR. Of perhaps the greatest potential benefit to the handicapped, and also repre-
senting the greatest technological challenge, would be an ASR-based device for the
deaf that could provide a visual read-out of what people are saying
There are three fundamental attributes that characterize a particular ASR
system: vocabulary size, training requirements, and its ability to handle continuous
speech. Vocabulary size is the number of different words that a system can handle
at one time. Text creation requires a large basic vocabulary as well as the ability to
add additional words to the active personal vocabulary of each individual user. For
most other applications, small vocabularies suffice
Most ASR systems require each user to train the system on their particular
pronunciation patterns. This is typically accomplished by making the user provide the
system with one or more spoken samples of each word in the vocabulary. However,
for large vocabulary systems, speaking every word in the vocabulary is often not
practical. It is preferable if the ASR system can infer how the speaker is likely to
pronounce words never actually spoken to the machine. Then the user needs to train
the system on only a subset of the full vocabulary

269
 Some small-vocabulary systems have been preprogrammed with all of the
dialectic patterns anticipated from the population expected to use the system and
thus do not require any prior training by each user. This capability, called speaker
independence, is generally required for telephone-based systems, where a single
system can be accessed by a large group of users.
Most commercial systems to date require users to speak with brief pauses
(usually around 100 milliseconds) between words. This helps the system make a
crucial segmentation decision: where words start and end. Speaking with such
pauses reduces the speed of a typical speaker by 20 to 50 percent. ASR systems
that can handle continuous speech exist, but they are limited today to small vocabu-
laries. Continuous-speech systems that can handle large vocabularies are expected in
the early 1990s
Other characteristics that are important in describing practical ASR sys-
tems include the accuracy rate, response time, immunity to background noise, re-
quirements for correcting errors, and integration of the speech recognition capability
with specific computer applications. In general, it is not desirable to simply insert a
speech recognition system as a front-end to ordinary computer applications. The
human requirements for controlling computer applications by voice are substantially
different from those of more conventional input devices such as keyboards, so the
design of the overall system needs to take this into account.
While ASR systems continue to fall far short of human performance, their
capabilities are rapidly improving, and commercial applications are taking root. As of
1989 ASR systems could either recognize a large vocabulary (10,000 words or more),
recognize continuous speech, or provide speaker independence (no user training),
but they could provide only one of these capabilities at a time. In 1990, commercial
systems were introduced that combined speaker independence with the abililty to
recognize a large vocabulary. | expect it to be possible in the early 1990s to combine
any two of these attributes in the same system. In other words, we will see large
vocabulary systems that can handle continuous speech while still requiring training
for each speaker; there will be speaker-independent systems that can handle
continuous speech but only for small vocabularies; and so on. The Holy Grail of
speech recognition is to combine all three of these abilities, as human speech
recognition does

Other types of auditory perception

There are applications of pattern recognition to auditory perception other than
recognizing speech. A field closely related to ASR is speaker identification, which
attempts to reliably identify the person speaking by his or her speech profile. The
techniques used are drawn from the ASR field. Common applications include entry
and access control
Military applications of auditory pattern recognition consist primarily of ship
and submarine detection. It is not possible to see through the ocean, but sound
waves can travel through the water for long distances. By sending sound waves and
analyzing the reflected patterns, marine vehicles can be recognized. One of the
challenges is to correctly characterize the patterns of different types of submarines
and ships as opposed to whales and other natural undersea phenomena.

270
 There are also a number of medical applications. Scanning the body with
sound waves has become a powerful noninvasive diagnostic tool, and systems are
being developed to apply pattern-recognition techniques to data from sonagram
scanners. Perhaps the most significant medical application lies in the area of listening
to the human heart. Many irregularities in heartbeat (arrhythmias) occur infrequently
So a technique has been developed that involves recording an electrocardiogram for
a 24-hour period. In such Holter monitoring the patient goes through a normal day
while a portable unit makes a tape recording of his heart pattern. The recording is
then reviewed on a special screen by a human technician at 24 times normal speed,
thus requiring an hour of analysis. But reading an electrocardiogram at 24 times real
time for an hour is extremely demanding and studies have indicated a significant rate
of errors. Using pattern-recognition techniques, computers have been programmed
to analyze the recording at several hundred times real time and thus have the
potential for providing lower costs and higher accuracy. Similar systems are used to
monitor the vital functions of critical-care patients. Eventually wristwatch systems
will be able to monitor our vital functions on a continuous basis. Heart patients could
be told by their wristwatches to slow down or take other appropriate action if it de-
termines they are overexerting themselves or otherwise getting into difficulty
Next to human speech, perhaps the most important type of sound that we
are exposed to is music. It might be argued that music is the most fundamental of
the arts: it is the only one universal to all known cultures. The musical-tone qualities
created by complex acoustic instruments such as the piano or violin have unique
psychological effects, particularly when combined with the other elements of music
(melody, rhythm, harmony, and expression). What is it that makes a piano sound the
way it does? Every piano sounds somewhat different, yet they are all recognizable as
the same type of instrument. This is essentially a pattern-recognition question, and
insight into the answer has provided the ability to recreate such sounds using
computer-based synthesizers. It also provides the ability to create entirely new
synthetic sounds that have the same richness, depth, and musical relevance as
acoustically produced sounds.”*
Analogues to the other human senses are being developed as well
Chemical-analysis systems are beginning to emulate the functions of taste and
smell. A variety of tactile sensors have been developed to provide robots with a
sense of touch to augment their sight
Taken together, applications of pattern recognition comprise fully half of
the Al industry. It is surprising that many discussions of Al overlook this area
entirely. Unlike expert systems and other areas that primarily emphasize logical rules
and relationships, the field of pattern recognition combines both parallel and sequen-
tial types of thinking. Because of their need for enormous amounts of computation,
pattern-recognition systems tend to require the cutting edge of advanced computer
architectures.
It is widely recognized that computers will require extensive knowledge
about the world to perform useful intelligent functions. It is not feasible for computer
scientists to explicitly teach our computers all there is to know about the entire
world. Like children, Al systems will need to acquire their own knowledge by
reading, looking, listening, and drawing their own conclusions based on their own
perceptions, perceptions based on pattern recognition.

271
 Success provides the opportunity for growth, and growth provides the
opportunity to risk at a higher level.
Eric Vogt

Most of the Al projects that | have been involved in personally are in

the pattern-recognition field. The following is a description of some
of these efforts from the point of view of technology. This section is
perhaps misnamed. This is not really a personal postscript; not
recounted here are the many exceptional people who have made
contributions to these projects, the early struggles of growing
companies, the efforts to attract both capital and talent, the ambigui-
ties and subtleties of understanding markets, the challenge of
establishing manufacturing facilities, the relationships with vendors,
suppliers, contractors, dealers, distributors, customers, consultants,
attorneys, accountants, bankers, investment bankers, investors and
the media, or the interpersonal challenges of building organizations.

Raymond Kurzweil Optical Character Recognition

ie

A Personal Postscript | founded Kurzweil Computer Products (KCP) in 1974. Our goal was to
solve the problem of omnifont (any type font) optical character
recognition (OCR) and to apply the resulting technology to the reading
needs of the blind as well as to other commercial applications. There
had been attempts to help the blind read using conventional OCR
devices (those for a single or limited number of type fonts), but these
machines were unable to deal with the great majority of printed
material as it actually exists in the world. It was clear that to be of
much value to the blind, an OCR machine would have to read any
style of print in common use and also deal with the vagaries of
printing errors, poor quality photocopies, varieties of paper and ink,
complex page formats, and so on. OCR machines had existed from the
beginning of the computer age, but all of the machines up to that time
had relied on template matching, a form of low-level property
extraction, and thus were severely limited in the range of material
they could handle. Typically, users had to actually retype printed
material using a specialized typeface before scanning. The principal
value of these devices was that typewriters were at that time more
ubiquitous than computer terminals.
It was clear to us that to produce an OCR device that was
font invariant as well as relatively insensitive to distorted print, we
would need additional experts beyond minimal property extraction.
Our solution was to develop software for multiple experts, including
topological experts such as loop, concavity, and line-segment
detectors, with an expert manager that could combine the results of
both high- and low-level recognition experts. The system was able to
 The flow of information in the
Kurzweil Reading Machine.
(Courtesy of the Kurzweil Reading
Machine Division of Xerox)

eked uid image isoiates Geometric Features Considered

. ‘a:
ec -= J DISAMBIGUATOR

oa ote yas
2 .

E a : a
Scanning e
s Cm OA Se a
oa _ oon
Sx convenes | =
‘etectromc image !
gnats: ‘Ennancement

ae
ec

= Direct Phoneme
& Conversion
s
= 7s
ee at Phonemes of words with
ez one
7 ee eee
combined into sentence assembly

2 EXcernON oICTIONAAY al PaO ——E Ake sms onmaroandcainpe

= = assicament Lnaest toesone
Foune
PHONETIC RULESintra:
sword
meeecontent inane

é ‘VOICE SYNTHESIS RULES | KURZWEIL

Sone ‘ou F COMPUTER
‘eration aoe inter SYNTHESIZER PRODUCTS
y AMPLIFICATION A Xerox Company

273
learn by having the high-level experts teach the low-level experts the Font invariance is a primary goal of
type faces found in a particular document. At a later point we added Kurzweil Computer Products’
intelligent character recognition
context experts by providing the machine with a knowledge of (ICR). (Courtesy of the Kurzweil
English (and ultimately several other languages). Reading Machine Division of Xerox)

The first Kurzweil Reading Machine (KRM), introduced in

Users can verify recognized
1976, consisted of an image scanner we developed ourselves that
characters against images of the
contained an electro-optical camera. The camera's eye consisted of a original characters if so desired.
(Courtesy of the Kurzweil Reading |
charge-coupled device containing 500 light-sensitive elements
Machine Division of Xerox)
arranged in a straight line. The camera was mounted on an electro-
mechanical “X-Y mover,” which could move the camera in both A few of the topological features
vertical and horizontal directions. The material to be read (a book, considered by Kurzweil Computer
Products’ ICR. (Courtesy of the
magazine, typed letter, etc.) lay face down on the glass plate that Kurzweil Reading Machine Division
formed the top of the machine. The camera automatically moved back of Xerox)
and forth scanning each line of print, transmitting the image
electronically to a minicomputer contained in a separate cabinet.
Using our omnifont OCR software, the minicomputer recognized the
characters, grouped them into words and computed the pronunciation
of each word. To accomplish this last task, several hundred
pronunciation rules were programmed in, along with several
thousand exceptions. The resulting string of phonemes was sent to a
speech synthesizer, which articulated each word. Subsequent
models of the KRM have been substantially improved, but they are
organized in a similar way.
Using the device is straightforward: the user places the
document to be read face down on the machine, presses start, and
listens. The KRM has a control panel to control the movement of the
scanner, back it up, make it reread sections or spell out words, and
provide a variety of other functions.

Jamal Mazrui uses the Kurzweil

Reading Machine. The Kurzweil
Reading Machine scans and
recognizes such text as books,
magazines, and memos and converts
it into synthesized speech and thus
is able to provide blind readers with
independent access to printed
material. (Photo by Lou Jones)

274
 The KRM has been called the first commercial product to
successfully incorporate Al technology. A recent survey showed that
most blind college students have access to a KRM to read their
educational materials. Nothing in my professional career has given
me greater satisfaction than the many letters | have received from
blind persons of all ages indicating the benefit they have received
from the KRM in enabling them to complete their studies or gain and
maintain productive employment.
Two years after the introduction of the KRM, we intro-
i ue zweil duced a refined version, the Kurzweil Data Entry Machine (KDEM),
re ee
0d ce ee
ee
designed for commercial applications. The KDEM, like the KRM,
could scan printed and typed documents and recognize characters
and page formats from a wide variety of sources, but rather than
speaking the words, it transmits them. It has been used to automati-
cally enter documents into data bases, word-processing machines,
electronic-publishing systems, and a variety of other computer-based
systems. For example, the KDEM was used to automatically scan and
recognize all of the contributed articles in this book for entry into a
computerized publishing system.
Many computerized systems move information from
electronic form onto the printed page. The KDEM allows it to move
back not just as an electronic image but in an intelligent form that a
computer can understand and process further. The result is to make
the printed page another form of information storage like a floppy disk
or tape. Unlike electronic media, however, the printed page can be
easily accessed by humans as well, which makes it the medium of
choice for both people and machines.
| find it interesting to review the rapidly improving price
performance of computer-based products in terms of the products of
my own companies. The 1978 KDEM were sold for $120,000, which,
adjusted for inflation, is equivalent to $231,000 in 1990 dollars. It had
65,536 bytes of memory and recognized print at about 3 characters per
second. In 1990 KCP offered a far superior product for under $5,000.
The 1990 version has 2 to 4 million bytes of memory, can recognize
between 30 and 75 characters per second, can recognize a substan-
tially wider range of degraded print, and is far more accurate than the
1978 KDEM. The 1990 version thus has 32 to 64 times as much memory,
is 10 to 25 times faster, and is more accurate and versatile than the
1978 version. If we conservatively assume that it provides at least 15
times the performance at 1/46.2 the price, it represents an overall
improvement in price-performance of 693 to 1. Since 2°*= 693, KCP
has doubled its price-performance 9.4 times in 144 months, which is a
doubling of price-performance every 15.3 months. That rate is
somewhat better than the computer industry at large, which is
generally considered to double its price-performance ratio only every
18 to 24 months.

275
 Speech Recognition
2sos

On July 1, 1982, | founded Kurzweil Applied Intelligence, (KAI). KAI's

goal has been to master automatic speech recognition (ASR)
technology and to integrate ASR with other Al technologies to solve
real-world problems. The long term goal is to establish ASR as a
ubiquitous modality of communication between human and machine.
In 1985 we introduced the Kurzweil Voice System (KVS), the first
commercial ASR device with a 1,000 word vocabulary. A refined
version called KVW, the first to provide a recognition vocabulary of
up to 10,000 words, was introduced in 1987."
The KAI speech recognizers follow the paradigm
described in the previous section. The KVW, for example, has seven
Viadimir Sejnoha, a speech scientist different experts, all of which attempt to recognize each spoken word
at Kurzweil Applied Intelligence,
looks at images of human speech. simultaneously. Several of the experts analyze the acoustics (or
(Photo by Hank Morgan of Rainbow) sound) of the word spoken, and the others are context experts
programmed with a knowledge of English word sequences (other
Say “baby.” Spectograms of two
languages will follow in the future). The system makes extensive use
utterances of the same word by the
same speaker show the high degree of user training and adaptation. It learns the phonological patterns
of variability in human speech. (that is, the frequency patterns) and the phonetic sequences (the
(Photo by Hank Morgan of Rainbow)
dialect) of each user. It also includes a user syntax expert that learns
the characteristic word-sequence patterns used by each speaker.
The KVW consists of specialized electronics, an industry-
standard personal computer and software. Before a user can begin
productive dictation with the system, he needs to enroll, which
involves speaking a sample of words to the machine to provide it with
initial phonological and phonetic models. Once that has been
accomplished, the user can dictate to the machine and watch each
word appear on the screen within a fraction of a second after
speaking it. The 1990 models require the user to provide a brief pause
between words, although later models are expected to accept
continuous speech.
The KVW actually improves its performance as the system
=

adapts to (that is, learns about) the user's pronunciation patterns and
syntax. This adaptation continues indefinitely. In addition to
displaying each recognized word, the system also displays its second
through sixth choices. If the KYW makes a mistake, one of these
alternate words is very often the correct choice. Thus, errors are
typically corrected by the user saying “Take two” or “Take three,”
which replaces the word originally displayed in the text with the
appropriate alternate word.
KVW speech-recognition technology has been integrated
with a variety of applications. One version includes a full-function
word-processor with the capability of entering text as well as issuing
all editing and formatting commands by voice. Several sections of
this book were written by voice using this version of the KVW. Other

276
 Sample reports from Kurzweil
VoiceRad (for radiology).
VoiceRad combines large-
vocabulary speech recognition
and knowledge engineering.
(Courtesy of Kurzweil Applied
Intelligence)

A doctor dictates medical reports

Routine mammography report to VoiceRad. (Courtesy of Kurzweil
Say Display Applied Intelligence)

“today’ 4/11/1987 Mammography (bilateral)

“mammography
“bilateral”
“routine” A low dose film technique was used. Cephalocaudad and
modifed lateral views were obtained Automobile accident report
“diffuse Both breasts show a very prominent duct pattern with
nodularity” diffuse nodularity throughout. Such diffuse nodularity 8/19/1987 Auto accident
makes the breasts difficult to evaluate 10 year old white male non-belted passenger in high vilocity accident struck form
“no focal lesions” There are no dominant masses or clustered calcifications rear not thrown from vehicle.
on either side Denies rolling vehicle. No glass fragments. Onset of pain in right hip.
“obtain outside The patient gives a history of previous mammography Physical exam vital signs: See nurse's notes.
study” elsewhere. If these could be obtained for comparison it Head: Atraumatic, without sinus, temporal or scalp tenderness
would be helpful to exclude any interval change Eyes: PERLA,EOMI, no discharge, injection, a-v nicking, hemorrhage, exudate or
“signature” John H. Jones, M.D. 9:18:32 4/11/1987 papilledema
Ears: TM without perforation, injection or bulging. External canals clear without
exudate.
Nose: No injection, exudate or septal hernatoma
Throat: Clear tongue, pharynx without injection, exudate or tonsillar hypertrophy
Word-by-word tation Teeth: Normal variant
Neck: Supple, non-tender; without HJR, JVD, thyromegaly, or lymphadenopathy.
Mammography (left side only) Lungs: CTA equal
Xerographic images were obtained. There is a 3 cm ill-defined mass 2 cm deep to Heart: RRR without murmurs, gallops, or rubs
the nipple. Findings compatible with malignancy. No other abnormality is seen at Chest: Non-tender, symmetrical; no retractions
this time. Abdomen: Soft, flat, non-tender, good bowel sounds. No hepatosplenomegaly,
Impression: Probable cancer rebound, guarding, or firm pulsatile mass
Advise biopsy.
Back: No costovertebral, paravertebral, intervertebral, or vertebral tenderness
Neurological: Alert and oriented X3. Cooperative. Cranial and cerebellar functions
John H. Jones, M.D. 9:22:51 4/11/87 normal. DTR equal and symmetrical
Extremity: Right proximal lateral hip tender. Neurovascular status intact. Joint
tenderness present, Pulses unequal.
Genitourinary: No penile ulcers or urethral discharge. Testes descended, vertical
and non-tender
Rectum: No mass or tenderness. Non-bloody stool
Xray: Right hip fracture, simple
Diagnosis: Acute fracture right hip
Consult specialty service. Admit to hospital
John H. Smith, M.D. 4:18:32 8/19/1987

Earache report
Earache
4 year old white female patient complains of right ear pain for 2 day(s)
Physical exam: See nurse's notes
Sample reports produced by doctors Right TM injected, without perforation, without bulging. External canals clear,
stating to Kurzweil VoiceEM, without exudate
a system for emergency medi
Diagnosis: Acute right otitis media
physicians that is similarto Discharge plan:
1, Rest at home. Drink lots of clear liquids for 24 hours
VoiceRad. (Courtesy of Kurzweil 2. Take amoxicillin 250 mg TID.
Applied Intelligence) 3. If your symptoms worsen or if you notice no improvement with
days, see your Doctor as soon as possible
4. Excuse from activity for 0 dayis)
5. Return to Emergency Department for any emergency
Charles Watson, M.D. 14:03 8/15/1987

277
versions of the KVW are integrated with knowledge-based systems instruments—the acoustic and the electronic—had developed with
that have expertise in the types of reports created in different no bridge existing between them.
professions. For example, VoiceRad integrates KAI's speech- On the one hand, acoustic instruments such as the piano,
recognition technology with knowledge of radiology reporting, violin, and guitar provided the musical sounds that were still the
allowing a radiologist to quickly dictate the results of an examination sounds of choice for most of the world’s musicians. While these
for instantaneous transcription. As with the word-processor version acoustic sounds were rich, complex, and musically satisfying, only
of the KVW, the radiologist can dictate a report word by word. In limited means were available for controlling or even playing these
addition, the system can automatically generate predefined sections sounds. For one thing, once a piano key was struck, there was no
of text based on its knowledge of radiology reporting. VoiceEM is a further ability to shape the note other than to terminate it: the initial
similar system for emergency medicine. A variety of similar systems velocity of the key strike was the only means for modifying piano
have been developed for medicine and other disciplines. This sounds. Second, most instruments could only play one note at a time.
approach combines the productivity gains of ASR-based dictation Third, there were no ways to /ayer sounds, that is, play the sounds of
with those of a built-in domain-specific knowledge base. These different instruments simultaneously. Even if you had the skills to play
products mark the first time that a commercially available large- both a piano and a guitar, for example, you could hardly play both at
vocabulary ASR product has been used to create written text by voice the same time. Even two musicians would find playing the same
in other than experimental situations. chords on a piano and guitar almost impossible. In any case, very few
KAI also has a major commitment to applying technology musicians, no matter how accomplished, could play more than a very
for the handicapped. Versions of KVS and KVW technology provide few instruments, as each one requires substantially different playing
means for text creation and computer and environmental control for techniques. Since the playing methods themselves were linked to the
quadriplegic and other hand-impaired individuals. A long-term goal physics of each acoustic instrument, many instruments required a
of the company is to develop a sensory aid for the deaf that would high level of finger dexterity. If a composer had a multi-instrumental
provide a real-time display of what someone is saying on the phone arrangement in mind, he had no way of even hearing what the piece
and in person. sounded like without assembling a large group of musicians. Then
The company’s long term objectives are two-fold. First, it making changes to the composition required laborious modification
intends to continue strengthening its core speech-recognition of written scores and additional rehearsal. | recall my father's
technology, to move toward the Holy Grail of combining large- lamenting the same difficulties.
vocabulary ASR with continuous-speech capability and minimal Steve pointed out that on the other hand there existed the
requirements for training the system for each user. Second, it intends electronic world of music in which most of the above limitations are
to integrate ASR with a variety of applications, particularly those overcome. Using just one type of playing skill (e.g., a piano-keyboard
emphasizing other Al technologies. Ultimately, our goal is to technique), one can activate and control all available electronic
establish voice communication as a desirable and widely used sounds. A wide variety of techniques exist for modifying many
means of communicating with machine intelligence. aspects of the sounds themselves prior to as well as during perform-
ance (these techniques have expanded greatly since 1982). One can
The Electronic Music Revolution layer sounds by having each key initiate different sounds simultane-
er
ously. Using sequencers, one can play one part of a multi-instrumen-
| founded Kurzweil Music Systems (KMS), also on July 1, 1982. The tal composition, then play that part back from memory and play a
inspiration for starting KMS came from two sources. One was my second part over it, repeating this process indefinitely. However,
lifelong interest in music, along with a nearly lifelong interest in electronic instruments at that time suffered from a major drawback,
computers. My father, a noted conductor and concert pianist, had told namely the sounds themselves. While they had found an important
me shortly before his death in 1970 that | would combine these two role in both popular and classical music, synthetic sounds were
interests one day, although he was not sure how. The other and more “thin,” had relatively limited diversity, and did not include any of the
immediate genesis of KMS was a conversation | had with Stevie desirable acoustic sounds.
Wonder, who had been a user of the Kurzweil Reading Machine from Steve asked whether it would be possible to combine
its inception. While showing me some new musical instruments he these two worlds of music to create in a single instrument the
had recently acquired, Steve noted that two worlds of musical capabilities of both. Such an instrument could produce music that
neither world of instruments alone could create. Accomplishing this

278
 The Kurzweil 250 Computer-Based
Synthesizer. (Courtesy of Kurzweil
Music Systems)

A musician plays the Kurzweil 250

using the keyboard and an electronic '
drum controller. (Courtesy of
Kurzweil Music Systems)

279
would, for example, enable musicians to play a guitar and a piano at process involves a form of painstaking tuning and attention to detail
the same time. We could take acoustic sounds and modify them to ironically reminiscent of old-world craftsmanship. The automatic
accomplish a wide variety of artistic purposes. A musician could play aspects of the process deal primarily with the issue of data compres-
a multi-instrumental composition (such as an entire orchestra) by sion. The original recorded data for even a single instrument would
himself using real acoustic (as well as electronic) sounds. A exceed the K250’s memory capacity. Thus, it is necessary to include
musician could play a violin or any other instrument polyphonically only the salient information necessary to accurately represent the
(playing more than one note at a time). One could play sounds of any original sounds.
instrument without having to learn the playing techniques of each. When the keyboardist strikes the K250's pianolike keys,
One could even create new sounds that were based on acoustic special sensors detect the velocity of each key’s motion. (Other KMS
sounds, and thus shared their complexity, but moved beyond them to a keyboards can also detect the time-varying pressure exerted by each
whole new class of timbres with substantial musical value. finger.) The K250's computer and specialized electronics extract the
This vision defined the goal of KMS. In June of 1983 we relevant information from the appropriate sound models in memory
demonstrated an engineering prototype of the Kurzweil 250 (K250), and then compute in real time the waveforms representing the
and we introduced it commercially in 1984.2 The K250 was considered selected instrument sound, pitch, and loudness for each note. The
the first electronic musical instrument to successfully emulate the varied control features, such as sequencing, layering, and sound
sounds of a grand piano and a wide variety of other instruments: modification, are provided by software routines stored in the unit's
orchestral string instruments (violin, viola, etc.), guitar, human voice, memory.
brass instruments, drums, and many others. In listening tests we In evolving our instruments at KMS, we have followed
found that listeners, including professional musicians, were two paths. First, the K250 has evolved into a comprehensive system
essentially unable to tell the K250 “grand piano” sound apart from that for creating complex musical works. It is essentially a digital
of a real $40,000 concert grand piano. A 12-track sequencer, sound recording and production studio in an instrument, and it has become
layering, and extensive sound modification facilities provide a full a standard for the creation of movie and television soundtracks and
complement of artistic control methods. professional recordings. KMS has also moved to bring down the cost
The essence of K250 technology lies in its sound models. of its sound-modeling technology. Its K1000 series, for example, is a
These data structures, contained in read-only memory within the line of relatively inexpensive products that provide the same quality
instrument, define the essential patterns of each instrument voice. We and diversity of sounds as the K250.
needed to create a signal-processing model of an instrument that will There is a historic trend taking place in the musical
respond to changes in pitch, loudness, and the passage of time in the instrument industry away from acoustic and mechanical technology
same complex ways as the original acoustic instrument. To create a and toward digital electronic technology. There are two reasons for
sound model, the starting point is to record the original instrument this. First, the price-performance of acoustic technology is rapidly
using high-quality digital techniques. Surprisingly, just finding the deteriorating because of the craftsmanship and labor-intensive
tight instruments to record turned out to be a major challenge. We nature of its manufacturing processes. A grand piano, for example,
were unable, for example, to find a single concert grand piano with has over 10,000 mostly hand-crafted moving parts. The price of the
an attractive sound in all registers. Some had a beautiful bass region average piano has increased by over 250 percent since 1970. At the
but a shrill midrange. Others were stunning in the high range, but same time, it is widely acknowledged that the quality of new pianos
mediocre otherwise. We ended up recording five different pianos, is diminishing. On the other hand, the price-performance of digital
including the one that Rudolph Serkin plays when he comes to electronics is, of course, rapidly improving. Furthermore, it is now
Boston. possible for an electronic instrument to provide the same sound
When capturing an instrument, we record examples of quality as an acoustic instrument, with substantially greater
many different pitches and loudness levels. When a particular key on functionality. For these reasons, electronic keyboard instruments
a piano is struck with varying levels of force, it is not just the have gone from 9.5 percent of the American market for keyboard
loudness level that changes but the entire time-varying spectrum of instruments in 1980 to 55.2 percent in 1986 (according to the
sound frequencies. All of these digital recordings are fed into our American Music Conference). It is my strong belief that this trend

sound analysis computer, and a variety of both automatic and manual will continue until the market is virtually entirely electronic. Our
techniques are used to shape each instrument model. Part of the long-term goal at KMS is to continue to provide leadership for this
emerging worldwide industry of intelligent digital music technology.

280
A Final Note
ss

| have tried to select projects that make it possible to build strong

companies while meeting social and cultural goals that are important
to me and others. | believe, for example, that there is a good match
between the capabilities of computer science and the needs of the
handicapped. It has been a personal goal of mine to apply Al
technologies to help overcome the handicaps associated with major
physical and sensory disabilities. | believe the potential exists in the
next couple of decades to largely overcome these major handicaps.
As amplifiers of human thought, computers have great potential to
assist human expression, improve productivity, and expand creativity
for all of us, in all areas of work and play. | hope to play a role in
constructively harnessing this potential.
All of the projects described above have been highly
interdisciplinary efforts and have required the dedication and talents
of many brilliant individuals in a broad range of fields. Inventing today
is very much a team effort and its success is a function of the quality
of the individual members of the team as well as the quality of the
group's communication. As Norbert Wiener pointed out in Cybernet-
ics, scientists and engineers with different areas of expertise often
use entirely different technical vocabularies to refer to the same
phenomena. Creating an environment in which a team of linguists,
speech scientists, signal-processing experts, VLSI designers, and
other specialists can understand each other's terminology and
effectively work together (as was required, for example, in the efforts
to develop the speech recognition technology described above) is at
least as challenging as the development of the technology itself. Once
developed, the technology (and the technologists) must be further
integrated into the equally well-developed disciplines of manufactur-
ing, marketing, finance, and the other management skills of a modern
corporation.
It is always exciting to see (or hear) a new product, to
experience the realization of a vision after years of hard collaborative
work. Perhaps my greatest pleasure has been the opportunity to share
in the creative process with the many outstanding men and women
who have contributed to these endeavors.

Notes
—

1. Raymond Kurzweil, “The Technology of the Kurzweil Voice Writer,” BYTE, March
1986.
2. Christopher Morgan, “The Kurzweil 250 Digital Synthesizer,” BYTE, June 1986.

281
A few bits of knowledge

Lr
Py

oad
ee
yee ula
+ has dimensionality

yous
Oe ee
eras

Cy
‘% usually mobile
Sea
dud
cle
ee ee ai)

Cet ea
ec ras
reproduces sexual, femalegi
mer)
eyo
ame ceca s

ee) ee eo)
er) © camows lage bran, especialy
rd peer
a er eee
Td Beet

Cat (Species)
ee
ny |
eee ed
aT
Pe ead

© brown har
eee
 The Search for Knowledge

A man is sent to prison for a ten-year term. The dining hall also serves as an auditorium, and there is a
stage at one end. After supper, one of the prisoners runs up onto the stage and hollers, “Four hundred and
eighty-seven.” Everyone starts laughing. The next day, it’s the same thing: After supper someone jumps
onto the stage, yells, “Two thousand six hundred and twenty-two,” and all the prisoners crack up. This
goes on for a couple of weeks, and finally the man asks his cellmate what's going on.
“Well,” says the cellmate, “It's like this. The prison library has a big fat book called The Ten Thousand
Best Jokes, and we've all been here so long that we know the book by heart. If somebody wants to tell a
joke, they just shout out the number of the joke in The Ten Thousand Best Jokes, and if it's a funny one,
everybody laughs.”
At dinner that night, the man decides to show the other prisoners that he’s a good guy.
Before anyone else can get to the stage, he dashes up there and shouts, “Five thousand nine hundred and
eighty-six!” But to his horror, nobody cracks a smile. There are even a few groans. He slinks back to his
cell to consult with his cellmate.
“Nobody laughed! Isn't the five thousand nine hundred and eighty-sixth joke a good one?”
“Sure it's a good one,” says the cellmate. “Old five thousand nine hundred eighty-six is
one of the best.”
“So why didn’t anyone laugh?”
“You didn’t tell it right.”
An old joke as retold in Mind Tools by Rudy Rucker

+ Knowledge and Expert Systems

Knowledge is not the same as information. Knowledge is information that has been pared, shaped,
interpreted, selected, and transformed; the artist in each of us daily picks up the raw material and makes
of it a small artifact—and at the same time, a small human glory. Now we have invented machines to do
that, just as we invented machines to extend our muscles and our other organs. In typical human fashion,
we intend our new machines for all the usual purposes, from enhancing our lives to filling our purses. If
they scourge our enemies, we wouldn't mind that either. . . .
The reasoning animal has, perhaps inevitably, fashioned the reasoning machine. With all the
risks apparent in such an audacious, some say reckless, embarkation onto sacred ground, we have gone
ahead anyway, holding tenaciously to what the wise in every culture at every time have taught: the
shadows, however dark and menacing, must not deter us from reaching the light.
Edward A. Feigenbaum and Pamela McCorduck, The Fifth Generation
What is knowledge?
Facts alone do not constitute knowledge. For information to become knowledge, it
must incorporate the relationships between ideas. And for the knowledge to be
useful, the links describing how concepts interact must be easily accessed, updated,
and manipulated. Human intelligence is remarkable in its ability to perform these
tasks. However, it is almost more remarkably weak at reliably storing the information
on which knowledge is based.' The natural strengths of computers are roughly the
opposite. They have, therefore, become powerful allies of the human intellect in their
ability to reliably store and rapidly retrieve vast quantities of information, but con-
versely, they have been slow to master true knowledge. The design of computer
data structures that can represent the complex web of relationships both within and
among ideas has been a quest of artificial intelligence from its inception. Many
competing approaches have been proposed. The following example illustrates
features of several approaches including the methodology of frames, first proposed
by Minsky in the mid 1970s.?
Each box (in the figure at the beginning of this chapter) represents a
concept sometimes referred to as an object or frame. One important relationship, is
a class of, refers to a group of entities that comprise a proper subset of a broader
set. The principle of inheritance tells us that the characteristics of a set apply to all of
its subsets (and all of their subsets, etc.) unless there is a specific indication to the
contrary. Thus, we can conclude that mammals, being a subclass of animals, ingest
proteins, even though this fact is not explicitly stated. We can also conclude that
mammals, being ultimately a subclass of objects, are visible or tangible. The
relationship is an example of refers to a subset with a single member. Thus, even
though little is explicitly revealed in the Ray Kurzweil frame, the principle of inheri-
tance lets us infer a great deal. We can conclude that Ray shares with the rest of the
human species a superior intellect (although after reading this book, one might wish
to enter an exception here). Since he is a male, we cannot conclude that he nour-
ishes his young with milk, but we can determine that he is warm blooded, is usually
mobile, and persists in time. We note that Fluffy, although of the cat species, does
not inherit the characteristic of a tail, but is nonetheless domesticated, carnivorous,
and so on.
Other types of relationships, types that do not imply the inheritance of
characteristics, are also shown on the chart. Belongs to and was eaten by are
examples of binary relationships that two singular objects or frames may have with
one another. These relationships can also be described using hierarchical structures.
For example, loves, hates, and is jealous of could all be examples of has feelings for.
Some areas of knowledge are more amenable to this type of hierarchical
analysis than others. Taxonomy, the study of the classification of organisms, is one
of the most effective. Biologists have been largely successful in constructing an
elaborate tree structure that incorporates virtually all of the many millions of known
earth species (including nearly a quarter million species of beetles alone).? The
system requires a minimum of seven levels:

Animalia kingdom
Chordata phylum
Mammalia class

284
* Primate order
* Homidae family
* Homo genus
+ Homo sapien species

More complex classifications are also possible:

+ Animalia kingdom
+ Metazoa subkingdom
+ Chordata phylum
- Vertebrata subphylum
+ Tetrapoda superclass
+ Mammalia class
+ Theria subclass
+ Eutheria infraclass
+ Ferungulata cohort
+ Ferae superorder
+ Carnivora order
- Fissipeda suborder
+ Canoidea superfamily
+ Canidae family
+ Caninae subfamily
+ Canis genus
* Canis lupus (wolf) species

Though some controversies persist (including, ironically, the exact place-

ment of our own species), the system has proved a remarkably useful tool for
organizing biological knowledge.* Other areas have more difficulty with hierarchic
classification. Consider the well-known Dewey Decimal system for classifying books
Through three levels of ten-fold expansion, Melvil Dewey (1851-1931) devised 999
categories for the Amherst College library in 1873.5 His system has since been
refined to provide a virtually unlimited number of subdivisions, for example,

-540 chemistry
+541 physical and theoretical chemistry
° 541.3 physical chemistry
+ 541.34 solutions

There is no question that the system is useful. Organizing libraries would

be immensely difficult without it or some similar system (say, the Library of Con-
gress system). Yet books do not always fall into one neat category, particularly with
the rapid development of new interdisciplinary topics. It will be interesting to see
how this book is classified: computer science, philosophy, history of technology,
humor? Since books often deal with complex subjects that may span multiple areas
of knowledge, a single position in a hierarchy is often an unsatisfactory compromise.
This is not to say that hierarchical classification of organisms works perfectly: it does
not. But it is interesting to note that every organism, even the lowest, has one or
more “parents,” which in turn had one or more parents, and so on. If we ignore

285
 interbreeding, the entire collection of all individual organisms that have ever lived
form one or more gigantic trees with trillions of leaves. There is obviously no such
equally fundamental tree organization for all baoks.
An ambitious attempt to organize all human knowledge in a single hierar-
chy is contained in the Propaedia section of the fifteenth edition of the Encyclopae-
dia Britannica, published in 1980. The Propaedia, which describes itself as an “outline
of knowledge,” is an 800-page attempt to codify all knowledge, at least that con-
tained in the remaining 30,000 pages of the encyclopedia. For example, money is
found under 534D1:

-5 Human society
+53 The production, distribution, and utilization of wealth
+534 The organization of production and distribution
*534D Institutional arrangements that facilitate production and output
*534D1 The nature and characteristics of money

Hierarchical relationships among

lateral relationships.

The Propaedia does allow for multiple classifications. Each entry, such as
534D1, will provide a number of references into the main portion of the encyclope-
dia. Conversely, any section in the encyclopedia is likely to have multiple references
to it from the Propaedia. The Propaedia takes time to understand, but it is surpris-
ingly successful in view of the vast scope of the material it covers.
Hierarchical classification schemes have provided a powerful organizing
tool for most of the sciences. It is clear, however, that we need to go beyond
treelike structures to represent most concepts. Key to the design of data structures
intended to represent concepts are the cross-links (i.e. nonhierarchical relationships).
Consider the structures, called semantic networks, depicted in the figures.° The
vertical lines continue to represent such hierarchical relationships as part and is a.
! The horizontal links, however, give the concepts their distinguishing shapes. Here
we see the same type of structure (with different shapes, of course) representing .
two very different ideas: the concept of an arch and that of a musical scale. An arch

286

ace
The concept of an arch in a semantic
network. (Figure reprinted with
permission from Curtis Roads,
“Research in Music and Artificial
Intelligence,” ACM Computing
Surveys 17 [1985], no. 2: 184; copyright
1985 by the Association for Computing
Machinery)

Does-not-touch
Brick

The concept of a musical scale in

a semantic network. (Figure re-
printed with permission from Curtis
Roads, “Research in Music and
Artificial Intelligence,” ACM
Computing Surveys 17 [1985], no. 2:
185; copyright 1985 by the Associa-
tion for Computing Machinery)

Phrase

Part Part Part Part Part

Next-scale-degree Next-scale-degree Next-scale-degree | Next-scale-degree

' Next-event Next-event Next-event

A B Cc

RT iesevent||| A taseeven| AS vasreven| | 4

Is-a Is-a Is-a Is-a

y y y y
Note Note Note Note

287
or scale can be implemented in a virtually unlimited number of ways, but all such
manifestations can share the same conceptual representation.
The horizontal links are themselves concepts and can obviously be quite
varied. They can also be represented by networks. The boxes, sometimes called
objects, can also be either simple entities or networks. Thus, each network may
refer to other networks to represent both the cross-links and the objects. The vertical
lines continue to represent simple hierarchical relationships.
Semantic networks and other similar systems are reasonably successful in
representing the shape and content of abstract ideas for use in computer knowledge
bases. Creating such networks is not easy, however, and this has proved to be a
major bottleneck. A major focus of current Al research is to create software that can
automatically build such structures from examples of a concept.”
Such data structures as semantic networks provide a formal methodology
for representing a broad class of knowledge. As they are easily stored and manipu-
lated in a computer, they are a powerful and practical tool for capturing and harness-
ing the patterns inherent in at least some types of ideas. Since humans routinely
deal with abstract concepts in a supremely subtle way, we can infer that the data
structures in our own minds must be at least as powerful as these networks (they
are, in fact, far more powerful). Though little is directly known about the data struc-
tures we use, we can draw a few hints from observations of our own behavior and
thought patterns.
First, it is clear that we rarely, if ever, model the relationships between
entities with single links.? Every time we experience or come into contact with a
concept, we add links that reinforce the structures inherent in a concept. For a
common concept we may have millions of links expressing the same or similar
associations. Indeed, the key relationships may not be explicitly coded at all but
rather implied by the general pattern of the data structures. This redundancy has
several implications. First, these knowledge structures are not subject to cata-
strophic failure if parts of the hardware fail. One estimate puts at 50,000 the number
of neurons that die each day in an adult human brain (and this process accelerates
with age), yet our concepts and ideas do not necessarily deteriorate with the
hardware. The massive number of links also helps us to appreciate both the unity
and the diversity of a concept. We probably have millions of links indicating or
implying that a-chair generally has four legs. The link between a chair and four legs js
thus strongly established. The links refer to (or evoke) experiences we have had with
chairs and so relate to memories of particular chairs we have known. The diversity
that is possible within the concept of a chair is thus also captured. The massive
redundancy also accounts for our ability (or inability) to deal with what is called
cognitive dissonance (a piece of information that appears to contradict a well-
established belief or understanding).° If we suddenly experience evidence, even
strong evidence, that an idea or concept that we have is invalid, we do not immedi-
ately update all of our knowledge structures to reflect this new insight. We are
capable of storing this apparently contradictory idea right alongside the concepts we
already had. Unless the new idea is reinforced, it will eventually die out, over-
whelmed by the large number of links representing our previous conceptions. There
is no evidence of a mechanism in our brains to avoid or eliminate contradictory

288
concepts, and we are sometimes quite comfortable with ideas that appear to be
incompatible. Hence, ideas that are presented repeatedly early in our life by our
parents or culture are not easily modified, even if our adult experiences are appar-
ently inconsistent. In general, it takes repeated exposure to a new idea to change
our minds. This is one reason that the media are so powerful. They have the ability
to reach large audiences on a repeated basis and are thus capable of having a meas-
urable effect on our data structures.'° Presenting an idea once, even if the idea is
powerful and true, does not generally have a significant impact.
There is a strong link between our emotions and our knowledge. If infor-
mation is presented in a way that elicits an emotional response, we are far more
likely to change our knowledge structures (and hence our minds). For this reason,
television, which has far greater potential than the print media to reach most people
emotionally, has a correspondingly greater impact. Television commercials are often
minidramas that attempt to engage us emotionally so that the underlying message
will have its desired impact."'
Our knowledge is also closely tied into our pattern-recognition capabilities.
People deal with the concept of a particular person's face as easily as with the
concept of an arch or a musical scale. But we have yet to devise computer-based
data structures that can successfully represent the unique features of a face, al-
though for simpler visual concepts (such as the shape of an industrial part, or the
shape of the letter A) representations similar to semantic networks are actively used.
One indication that we use structures similar to semantic networks is our
ability to jump from concept to concept via the crosslinks. The thought of taxonomy
may lead us to thoughts of primates, which may lead to monkeys, which may lead
to bananas, which may lead to nutrition, which may lead to dieting, which may lead
to the upcoming Thanksgiving party, which may lead to a particular family member,
and so on. We clearly have all of our semantic information, with its massively
redundant links, organized in a single vast network. Individual concepts are not
identified as such. It requires a great deal of disciplined thought and study to
explicitly call out (and describe) individual ideas. The ability to translate our mental
networks into coherent language is a difficult skill, one that we continue to struggle
with even as adults.
An interesting question concerns how much knowledge we are capable of
mastering. It is estimated that the human brain contains on the order of 100 billion
neurons.'? We now realize that individual neurons are each capable of remembering
far more than the one or several bits of data originally thought possible. One method
of storing information is in the strength of each synaptic connection. A neuron can
have thousands of such connections, each potentially storing an analog number.
There is also speculation that certain long-term memories are chemically coded in
the neuron cell bodies. If we estimate the capacity of each neuron at about 1,000
bits (and this is probably low by several orders of magnitude), that gives the brain a
capacity of 100 trillion (10'* bits). A typical computer-based semantic network
requires only a few thousand bits to represent a concept. Because of the redun-
dancy, however, our human semantic networks need much greater amounts of
Storage. If, as a rough guess, we assume an average redundancy factor of several
tens of thousands, this gives us about 100 million bits per concept, and this yields a
total capacity of 1 million concepts per human brain. It has been estimated that a

289
“master” of a particular domain of knowledge (chess, medicine, etc.) has mastered
about 50,000 concepts, which is about 5 percent of the total capacity, according to
the above estimate. s
Human intelligence is not, however, a function of the number of concepts
we can store, but rather of the coherency of our concepts, our ability to create
meaningful concepts from the information we are exposed to, the levels of abstrac-
tion we are capable of dealing with, our ability to articulate these concepts, and
perhaps most importantly, our ability to apply concepts in ways that go beyond the
original information that created them. This last trait is often regarded as a key
component of creativity.'* As Roger Schank and Christopher Owens point out in their
article in this book, we may be able to model this essential component of creativity
using Al techniques. We are beginning to understand some of the mechanisms that
enable us to apply a mental concept outside of its original domain and may ultimately
be able to teach computers to do the same. And while humans do excel at their
ability to recognize and apply concepts in creative ways, we are far from consistent
in our ability to do so. Computers may ultimately prove far more thorough in their
attempts to search all possibly relevant conceptual knowledge in the solution of a
problem."4

The knowledge-search trade-off

The human brain uses a type of circuitry that is very slow. Our neurons can perform
an analog computation in about 5 milliseconds, which is at least 10,000 times slower
than a digital computer. On the other hand, the degree of parallelism vastly outstrips
any computer architecture we have yet to design. The brain has about 100 billion
neurons each with about 1,000 connections to other neurons, or about 100 trillion
connections, each capable of a computation. If 1 percent of these are active, that
produces 1 trillion computations in 5 milliseconds, or about 200 trillion computations
per second. From this analysis Raj Reddy concludes that for such tasks as vision,
language, and motor control, the brain is more powerful than 1,000 supercomputers,
yet for certain simple tasks such as multiplying digital numbers, it is less powerful
than the 4-bit microprocessor found in a ten-dollar calculator.'®
It is clear that the human brain is not sequentially fast enough to perform
lengthy searches on the implications of its knowledge base. Yet with such a large
number of neurons and an even larger number of connections, it is capable of
storing a vast amount of highly organized knowledge and accessing this knowledge
in parallel. Thus, a typical strategy of the human brain is to access its memory of
previously analyzed situations, since it is not capable of performing a great deal of
analysis on a problem in real-time. The strategy is quite different for computers using
the conventional serial (that is, nonparallel) architecture. There is sufficient sequential
speed to perform extensive recursive search in the problem space, but often insuffi-
cient knowledge about a particular domain to rely too heavily on previously analyzed
situations.
A dramatic example of this difference is found in the game of chess, our
quintessential laboratory problem for exploring approaches to intelligence. Because
of the slow sequential speed of human thought, the chess master has time only to
consider perhaps 100 board positions, although he is able to take advantage of a

290
 The knowledge-search trade-off in chess

Human chess, master Computer chess master

Number of rules
or memorized situations 30,000-100,000 200-400*

Number of board positions

considered for each move 50-200 1,000,000-100,000,000

* Many computer chess programs do have extensive libraries of starting positions, so there can be an
appreciable amount of knowledge used in the early game. The figure of 200-400 rules refers to the mid and
end games.

large memory of previously analyzed situations.'° Because of his powerful pattern-

recognition capabilities, the chess master can recognize similarities in these previous
situations even if the match is not perfect. It is estimated that the chess master has
mastered 30,000 to 100,000 such board positions. The leading machine players,
while performing at comparable levels, have traditionally used very little knowledge
of chess beyond the rules. HiTech, for example, uses only a few hundred heuristic
rules to assist its search but is capable of examining millions of board positions for
each move (see table)
We thus find that there is a trade-off between knowledge and the compu-
tation required for recursive search, with the human and machine approaches to
chess being at the opposite extreme ends of the curve.'” As Raj Reddy says, “When
in doubt, sprout!” meaning that the sprouting of a recursive search tree by machine
intelligence can often compensate for its relative lack of knowledge. Conversely, our
human knowledge compensates for the brain's inability to conduct extensive se-
quential search in real-time. The figure illustrates the knowledge-search trade-off,
again with the human and machine approaches to chess at opposite ends."®
This brings up the issue of how much knowledge is enough. Interestingly,
in areas of knowledge as diverse as the size of an expert's vocabulary, the number
of symptom-illness correspondences known by a medical specialist, and the number
of board positions memorized by a chess master, the number of “chunks” of

The trade-off between knowledge 4010

\\\ Immediate knowledge
and searching. 10? LI \iinan. (Prepared)
gee
107 Equiperformance
isobars
10°
10°
Rules

10* Equicost
103 isobars

10?
Search
10! High tech——= knowledge
(deliberate)
10° 10' 10? 10% 104 10° 10® 10’ 108 10° 101°
Situations/tasks

291
knowledge possessed by a human master of any particular discipline appears to be
in the range of 30,000 to 100,000.'° We are also finding that a comparable number
of production rules are needed in our machine-based expert systems in order to
provide for sufficient depth of coverage and avoid the fragility that was evident in the
first generation of such systems. This realization about the long-term memory of a
human expert contrasts with observations of human short-term memory, which
appears to be about 10,000 times smaller.
These insights are encouraging with regard to the long-term outlook for
machine intelligence. Machine intelligence is inherently superior to human intelli-
gence in its ability to perform high-speed sequential search. As our machine meth-
ods for representing, learning, and retrieving chunks of knowledge improve, there is
no reason why our computers cannot eventually exceed the approximate 100,000
chunk limit of human knowledge within specific domains. Another advantage of
machine intelligence is the relative ease of sharing knowledge bases. Humans are
capable of sharing knowledge, but this requires a slow process of human communi-
cation and learning. Computers can quickly and efficiently pool their knowledge
bases
The so-called chunking of knowledge has become a major issue in Al
research. A recent system called SOAR created by Allen Newell and his colleagues
is able to automatically create its own chunks of knowledge and thereby learn from
experience.”° It is capable of performing recursive search but also takes advantage of
knowledge derived from a number of sources, including its own analysis of its own
previous searches, information provided directly by its human teachers, and correc-
tions of its own judgements. As it gains more experience in a particular problem
area, it is indeed able to answer questions more quickly and with greater accuracy.?!

The organization of an expert system

Expert systems are intended to replicate the decision making of a human expert
within narrowly defined domains. Such systems have three primary components: a
knowledge base, decision rules, and an inference engine.”
The knowledge base is intended to capture the ideas and concepts
inherent in the domain in question. As mentioned above, creating such knowledge
bases is a difficult and painstaking process. Although creating such knowledge bases
automatically (or semiautomatically) is a focus of Al research, the knowledge bases
of most expert systems in use today have been created link by link by human
“knowledge engineers.” Knowledge bases often incorporate structures similar to
those discussed above to represent the concepts and relationships between con-
cepts that are important to the domain. They may also include data bases of informa-
tion with a more uniform structure.
The decision rules describe the methods to make decisions. In XCON, a
system which configures computers for Digital Equipment Corporation, a typical rule
may State that if a particular computer model needs more than six expansion boards,
it requires an expansion chassis.”° If there is an expansion chassis, then there must
also be certain expansion cables, and so on. As of 1987 XCON incorporated over
10,000 such rules. It reportedly is doing the work of over 300 human experts with
substantially higher accuracy. XCON was developed as an expert system only after
 several earlier attempts using more conventional programming methodologies had
failed.**
The first generation of expert systems have used hard decision rules with
firm antecedents and certain consequences. Most human decision making, on the
other hand, is based on uncertain and fragmentary observations and less than certain
implications. If as humans we failed to consider information unless it was relatively
cértain, we would be left with very little basis on which to make most decisions. A
branch of mathematics called fuzzy logic has emerged to provide an optimal basis for
using probabilistic rules and observations.”®
The third component of the expert system, the inference engine, is 4
system for applying the rules to the knowledge base to make decisions. Many
expert systems in use today use standard serial computers (mainframes, minicom-
puters, and personal computers) with special software to perform the deductive and
inductive reasoning required by the decision rules. For the more sophisticated expert
systems now being created, this approach will not be fast enough in many cases
The number of inferences that must be considered (that is, the number of rules that
must be applied to each concept or datum in the knowledge base) explodes expo-
nentially as both the knowledge base and number of rules expands. Systems created
in the early 1980s typically included several hundred to several thousand rules with
similarly modest knowledge bases. For systems now being created with tens of
thousands of rules and far more extensive knowledge bases, serial computers are

ioneer of the compact

developing an intelligent
workstation called NEWS. (Photo by
Lou Jones)

293
 often too slow to provide acceptable response times. Specialized inference engines
incorporating substantial (and eventually massive) parallelism are being constructed
to provide the requisite computing power. As noted below the Japanese fifth-
generation computer project foresees the personal computer of the 1990s as
containing extremely high-speed, highly parallel inference engines capable of rapidly
manipulating abstract concepts.7°

+ Putting Knowledge to Work

Zeppo: We've got to think!

Chico (with a dismissive hand gesture): Nah, we already tried dat.
The Marx brothers, as quoted in Mind Tools by Rudy Rucker.

Knowledge has an important property. When you give it away, you don't lose it.
Raj Reddy, Foundations and Grand Challenges of Artificial Intelligence (1988)

The actual codifying of scientific and professional knowledge in terms that a com-
puter could understand began in the mid 1960s and became a major focus of Al
research in the 1970s. Edward Feigenbaum, Bruce Buchanan, and their colleagues at
Stanford University were early pioneers in the effort to establish the field now
known as knowledge engineering.”’ Their first effort, called DENDRAL, was one of
the first expert systems. Begun in 1965 and developed throughout the 1970s, it
embodied extensive knowledge of molecular-structure analysis, which was selected
simply as an illustration of the ability of a computer-based system to master an area
of scientific knowledge. A primary goal of the project was to investigate questions in
the philosophy of science, including the construction and validity of hypotheses. It
emerged nonetheless as a tool of practical value in university and industrial laborato-
ties. Much of the interest in, and methodology of, expert systems, which spawned a
major industry during the 1980s, can be traced to DENDRAL.#
A follow-on project, Meta-DENDRAL, was an early attempt to break the
knowledge-learning bottleneck. Presented with data about new chemical com-
pounds, Meta-DENDRAL was capable of automatically devising new rules for
DENDRAL. Problems that the Meta-DENDRAL team faced included dealing with
partial and often inaccurate data and the fact that multiple concepts were often
intertwined in the same set of data. The problem of learning was by no means
solved by the Meta-DENDRAL workers, but they did ease the job of the human
knowledge-engineer, and many of Meta-DENDRAL's techniques are still the focus of
learning research today.?°

Expert systems in medicine

With the success of DENDRAL and Meta-DENDRAL by the early 1970s, Feigen-
baum, Buchanan, and others became interested in applying similar techniques to a
broader set of applications and formed the Heuristic Programming Project, which
today goes by the name of the Knowledge Systems Laboratory. Perhaps their best
known effort, MYCIN, an expert system to diagnose and recommend remedial
treatment for infectious diseases was developed throughout the 1970s. In 1979 nine
researchers reported in the Journal of the American Medical Association a compari-

294
 Diabetic a pag
John
Tak
Insulin-medicine Aknowledge-base extract
describing diabetes. Causal links are
Pa\
shown with bold arrows. (From
Cause Blood Part Cause Winston, Artificial Intelligence, 1984,
Part p. 492)
___ Pancreas Insulin Sugar
Unhealthy<=
a Is |

Cause tow High

AKO Take
Diabetic <———. Tom ————» Insulin-medicine
Part

Blood
Part
Part

Pancreas Insulin Sugar

|High
son of MYCIN’s ability to evaluate complex cases involving meningitis to that of
human doctors.*° In what has become a landmark study, ten cases of infection were
selected, including three viral, one tubercular, one fungal, and one bacterial. For each
of these cases, diagnoses and treatment recommendations were obtained from
MYCIN, a Stanford Infectious Disease faculty member, a resident, and a medical
student. A team of evaluators compared the diagnoses and recommendations for all
of the cases, without knowledge of who (or what) had written them, against the
actual course of the disease and the actual therapy followed. According to the
evaluators, MYCIN did as well or better than any of the human doctors. Although the
domain of this evaluation was limited, the conclusion received a great deal of
attention from both the medical and computer communities, as well as the general
media. If a computer could match human intelligence in medical diagnosis and
treatment recommendation, albeit within a limited area of specialization, there
appeared to be no reason why the domains could not be substantially broadened. It
was also evident that such systems could eventually improve on human judgement in
terms of the consistent application of the vast (and rapidly increasing) quantity of
medical knowledge.
MYCIN’s success resulted in a high level of interest and confidence in
expert systems.*! A sizeable industry was created in the decade that followed.**
According to DM Data, the market research firm, the expert-system industry grew
from $4 million in 1981 to $400 million in 1988 and an estimated $800 million in
1990.28
Beyond the attention it focused on the discipline of knowledge engineering,
MYCIN was significant in other ways. It introduced the now-standard methodology of
a separate knowledge base and inference engine, as well as the recursive goal-
directed algorithms of the inference engine.** Further, MYCIN did not just give
diagnostic conclusions, it could also explain its reasoning and cite sources in the
medical literature. Of major significance was MYCIN’s use of its own version of fuzzy
logic, that is, reasoning based on uncertain evidence and rules, as shown in the
following rule, which also includes justification and reference:

295

eee ee LLhLLLTLTLTCCCLC“(‘#!S(‘COG
MYCIN Rule 280

IE
(1) The infection which requires therapy is meningitis, and
(2) The type of the infection is fungal, and
(3) Organisms were not seen on the stain of the culture, and
(4) The patient is not a compromised host, and
(5) The patient has been to an area that is endemic for coccidiomycoses, and
(6) The race of the patient is one of: Black, Asian, Indian, and
(7) The cryptococcal antigen in the csf was not positive
THEN:
There is suggestive evidence (.5) that cryptococcus is not one of the organisms
(other than those seen on cultures or smears) which might be causing the
infection
AUTHOR:
YU
JUSTIFICATION:
Dark-skinned races, especially Filipino, Asian and Black (in that order) have an
increased susceptibility to coccidiomycoses meningitis.
LITERATURE:
Stevens et al. Miconazole in Coccidiomycosis. Am. J. Med. 60:191-202, Feb 1976.

In the mid to late 1970s a variety of enhancements for MYCIN were

created, including TEIRESIAS, a knowledge-building tool, and EMYCIN (for Essential
MYCIN), a portable version (that is, applicable to other applications) of MYCIN’s
inference engine.*° These tools were applied to other tasks, including a system
called PUFF for interpreting pulmonary tests.*° During the 1980s the techniques
established by MYCIN were applied to more sophisticated systems, including
NeoMYCIN, a consultation system with a broader domain of medical knowledge than
MYCIN, and ONCOCIN, a system for the design of oncology (cancer therapy)
chemotherapy protocols.%” Innovations introduced by these later systems include
hierarchical knowledge bases that go beyond decision rules and the ability to engage
in a dialogue with the user to elicit symptoms, call for additional tests and input, and
clarify reasoning.°*
A particularly ambitious system called CADUCEUS (formerly Internist) was
developed at the University of Pittsburgh for diagnosis within the broad domain of
internal medicine. As of 1982 CADUCEUS contained over 100,000 associations
between symptoms and diseases and covered an estimated 70 percent of the
relevant knowledge in the field.*° The performance of CADUCEUS was compared to
that of human doctors on cases drawn from the New England Journal of Medicine.
In a wide range of diagnostic categories CADUCEUS was actually more accurate
than the human physicians.*°
It is not surprising that medicine would be the first professional domain to
attract significant knowledge-engineering attention. Medical knowledge is highly
organized, and the linkage of symptoms and test results to diagnoses and remedial
action is described in the medical literature in great detail. Although the knowledge-

296
 October 4, 1984 4:15PM in (EMYCIN) EMYCIN.EXE.20515 by J-Fox

4-Oct-84 61:14:42

PERSON-1

(1) What is the person called

** ANNIE BODDIE
(2) What sex is Annie Boddie
** FEMALE
(3) How old is Annie Boddie
**3
The first sample of Annie Boddie will be referred to as

SAMPLE-1

(4) What was SAMPLE-1 (blood, marrow, etc)

** MARROW
(5) What is the blast count in SAMPLE-1
** UNK
(6) What was the Tat result
720)
(7) What is the C-ALL result
**0
(8) What was the clinical diagnosis for Annie Boddie
* UNK
(9) What is the Sm Ig result |
== 50)
(10) What was the mouse rosette result
** UNK
(11) Is there a mediastinal mass
** UNK
(12) What was the Sheep rosette result
** UNK
(13) What was the HLA-DR result
** UNK
(14) What was the Kappa result
** WHY [l.e., why is it important to determine the percentage of Kappa-positive cells?]

Atypical dialogue with the EMYCIN [1.0] This will aid in determining the leukemic cell type. It has already been established that
leukemia diagnosis system. [1.1] the leukemic cell type is b-cell
(Courtesy of the Imperial Cancer Therefore, if
Research Fund Laboratories, London) [1.2] (1) The percentage of Kappa positive cells divided by the percentage lambda positive cells
is greater than or equal to 6, or
(2) The percentage lambda positive cells divided by the percentage of Kappa positive cells
is greater than or equal to 3
then there is weakly suggestive evidence (.4) that the leukemic cell type is monoclonal-b-cell
[RULE 032]

[Back to question 14.)

** WHY [Le., why is it important to determine the leukemic cell type?]

[2.0] The leukemic cell type is one of the 2 key parameters associated with a SAMPLE. The value
of this parameter is always determined as soon as a SAMPLE is mentioned.
[2.1] The leukenic cell type is very important in determining the diagnosis [as, for example, in
RULE 055]

[Back to question 14.]

** 62
(15) What was the Lambda result
762
(16) Are there any other samples of Annie Boddie
*NO
| was unable to make any conclusion about the diagnosis.

297
 TEIRESIAS The following investments seem to be the most suitable at the moment: Varian Corporation and
American Telephones & Telegraphs (AT&q). Is that correct?
Expert No. -
TEIRESIAS: Are there any missing?
Expert Yes, Digital Equiptment and Data General
TEIRESIAS: Are there any which should not be present?
Expert: Yes, AT&T.
TEIRESIAS: Would you like me to guide the error search?
Expert Yes
TEIRESIAS: Let's begin then with AT&T. | used rule 27 to reach that conclusion: As (1) the investment field
was not defined, (2) expected profits are above 10%, (3) the investment must be long term, so
AT&T is a plausible choice (0.4). Is the rule correct?
Expert Yes |
TEIRESIAS: Is there a premiss which should not have been evaluated as true?
Expert Yes, number 1
TEIRESIAS: | came to no conclusion on the issue of the field of investment. Should | have come up with
something precise?
Expert Yes. Advanced technology.
TEIRESIAS: Tests on all rules which might have concluded in favour of the advanced technology field all failed
because marked premisses were false. Were these two rules correct?
Expert Yes.
A dialogue with TEIRESIAS, a TEIRESIAS Should either of them not have been applied?
computerized investment counselor. Expert No.
(Taken from Jean-Louis Lauriere, TEIRESIAS Is there a rule missing?
“Knowledge Representation and Use. Expert Yes.
Part 1, Expert Systems,” Technology and TEIRESIAS: OK, so | need a rule which will allow me to deduce that the field should have been advanced
Science of Informatics1 [1983], no. 1: 15) technology. Would you like to give me a rule of that kind?
Expert If the client's tax bracket is 80%, if the client is following up on market trend carefully, then it is
obvious (0.8) that the investment should be made in high technology
TEIRESIAS The new tule is called rule 383. This is how | understand it: If (1) the client's tax payments are
50%, if (2) the market has followed an upward trend recently, if (3) the client manages his assets
carefully, then the investment should be made in high technology (0.8). Is that right?
Expert No.
TEIRESIAS: Should the premisses be changed?
Expert Yes, change (2).
TEIRESIAS: What if (2) were changed into ‘(2) the client follows the market carefully , would that do?
Expert Yes.
TEIRESIAS. Then you will doubtless want to eliminate (3). Right?
Expert Yes.
TEIRESIAS: (Repetition of R383, then:) | don’t like to criticize, Randy, but do you realize that usually, rules
affecting investment fields which mention the client's tax bracket and the way he follows the
market also mention the client's experience in making investments? Should | attempt to
formulate a third premiss?
Expert Yes
TEIRESIAS: What do you think of: the client has moderate degree of experience in investment?
Expert OK
TEIRESIAS. Most of those rules also give a conclusion on the desired level of risk for the investment. Should
| attempt to give that kind of conclusion here?
Expert: No.
TEIRESIAS. (Repetition of corrected R383, then:) Please tell me briefly why you have added this rule.
Expert: The system had not understood that it could specify the field of investment, which led to rule 27
being applied incorrectly.
TEIRESIAS R383 has been added to the knowledge base. The consultation will start again to test whether
this rule is in fact well founded or not. Make youself comfortable, this could take some time.
Results, suggested investments: Varian Corporation, Data General, Digital Equipment. Good, it
looks like everything has been corrected, doesn't it?
Expert: Yes.

298
engineering process remains painstaking, the knowledge is sufficiently well organ-
ized for expert-system techniques. The systems described above are among dozens
of medical expert systems in existence or under development. Some of these
systems are already contributing to medical care.*! PUFF, for example, is routinely
used at the Pacific Medical Center in San Francisco. Its reports are screened by an
attending pulmonary physiologist, who modifies no more than 5 percent of PUFF’s
reports. Many patients have had their cancer-treatment programs recommended by
ONCOCIN, which are also carefully reviewed. Although a large amount of develop-
ment of medical expert systems has been done and their actual use as advisors is
beginning, these systems have still had relatively little impact on medical practice in
general. There are four reasons for this. First, the medical community is understanda-
bly conservative with regard to any new technology, let alone an entirely new
approach to diagnosis. Even as consultants, these cybernetic diagnosticians will have
to prove themselves repeatedly before they are given major roles. Second, the
hardware these systems require has been expensive up until recently. This factor is
rapidly vanishing: the latest generation of personal computers has sufficient power to
run many of these applications. Third, many of these systems are not sufficiently
developed for practical use. They are often “brittle,” falling apart when cases fall
outside of their narrow areas of expertise. In order to provide a consistent and
predictable level of service, the knowledge bases require enormous refinement.*
The most advanced systems, such as CADUCEUS, are now beginning to reach
sufficient depths of medical knowledge. Finally, the issue of competition from a new
type of medical expert (one without a medical degree) is a real issue. Since the
medical profession has effective control over the delivery of medical services,
resistance to any new type of competition is not surprising. To succeed, the market-
ers of such services will need to position them clearly as tools that will make their
users more competitive with their human peers

Expert systems in other fields

Considering the gravity of many medical decisions, the slow pace of acceptance of
expert systems in medicine is understandable. In the financial field, the situation is
quite different. Here expert-system technology is often enthusiastically adopted. This
too is perhaps not surprising. Financial professionals are used to imperfect decision
making. They expect a certain fraction of decisions to be wrong. Practical demonstra-
tions that a financial expert system can match and often exceed the quality of
human decision making within a well defined area of expertise are more likely to be
readily accepted.*
For example, Syntelligence has created a system for the insurance industry
called Underwriting Advisor, which provides advice on risk analysis and pricing for
various lines of insurance, including workers’ compensation, property, liability, and
other types. A particularly successful system is the Authorizer’s Assistant, which
reduces the transaction time of an American Express credit authorizer by 20 to 30
percent and has reportedly reduced expensive bad judgements by about 75 percent.
Applications are proliferating in a broad variety of fields. A system devel-
oped at SRI International called Prospector received significant attention when it was

299
 1am considering the possibility of a (Type-A porphyry copper deposit—PCDA) in the target area
The following questions are intended to establish the nature of the petrotectonic setting,
1—To what degree do you believe that: (the target area is in a continental margin mobile belt)?
oar)
The following questions are intended to determine the nature of the regional environment.
2—To what degree do you believe that: (there are granitic intrusives in the region)?
#5
3—To what degree do you believe that: (the region contains an abundance of small stocks)?
+3 |

8—To what degree do you believe that: (igneous rocks in the region have porphyritic texture)?
“* Why
Aconversation with Prospector. |.am trying to establish whether some of the intrusive rocks in your area have textures suggestive of a
(Human responses start with a hypabyssal to subvolcanic environment. Porphyry textures resulting from the rapid cooling and crystallization
double asterisk.) (Courtesy of Edward of a silicate melt which contains large mineral grains are suggestive of relatively shallow levels of
emplacement.
Feigenbaum)
8—To what degree do you believe that: (igneous rocks in the region have porphyritic texture)?
m5

My certainty in (the texture and morphology of the intrusive system suggest a subvolcanic (hypabyssal)
regional environment) is now: 4.924.
Do you want to pursue this further?
** No

reported to have identified the location of a porphyry molybdenum deposit from

readily available predrilling exploration data.* Although this particular discovery had
more scientific than economic significance, the potential of expert systems to
improve the odds for the very expensive gambles involved in natural-resource
discovery has fueled substantial investment in knowledge engineering by the energy
and minerals industries.
Teknowledge is creating a system for General Motors that will assist
garage mechanics in quickly diagnosing and repairing engine-control problems in
GM's fuel-injected engines. With the increasing complexity of modern automobile
designs, keeping the skills and knowledge of mechanics up to date has become a
major problem. A computer-based system is an efficient way to provide continually
updated expertise to thousands of locations around the world. A similar system
called the Service Bay Diagnostic System, being developed by the Carnegie Group
for Ford, is described in the article by Jeff Pepper in this book
Digital Equipment Corporation's success with XCON has resulted in an
extensive development program of expert systems encompassing most aspects of
DEC's business, including XSEL (expert selling tool), an expert system that matches
DEC products to customer needs; ISA (Intelligent Scheduling Assistant), which i
schedules manufacturing and shop floor activity; IDT (Intelligent Diagnostic Tool), |
which assists human technicians in diagnosing manufactured products that fail
verification tests; and NTC (Network Troubleshooting Consultant), which diagnoses
computer network problems.*®

300
 A survey conducted by the Applied Artificial Intelligence Reporter showed
that the number of working expert systems swelled from 700 at the end of 1986 to
1,900 by the end of 1987. There were 7,000 systems under development at the end
of 1986 and 13,000 at the end of 1987. This has swelled to tens of thousands of
systems in 1990. The most popular application area is finance, with manufacturing
control second and fault diagnosis third. With the number of expert systems rapidly
expanding, a significant portion of the revenue of the expert-system industry comes
from selling the tools to create expert systems. These include specialized languages
like PROLOG and knowledge-engineering environments, sometimes called shells,
such as Knowledge Craft from the Carnegie Group, ART (Automated Reasoning Tool)
from Inference Corporation, and KEE (Knowledge Engineering Environment) from
IntelliCorp.4®
Some of the more advanced applications are being created by the U.S.
Defense Department as part of its Strategic Computing Initiative (SCl)—not to be
confused with the more controversial Strategic Defense Initiative (SDI), although
both SCI and SDI involve extensive application of pattern-recognition and knowledge-
engineering technologies.*” SCI envisions the creation of four advanced prototypes.
The Autonomous Land Vehicle is an unmanned robotic vehicle that can avoid ob-
stacles, conduct tactical maneuvers and carry out attack and defense plans while
traveling at speeds of up to 40 miles per hour. Its on-board expert system is ex-
pected to carry out 7,000 inferences (applications of rules) per second. This will
clearly require parallel processing, as typical expert systems implemented on serial
computers rarely achieve speeds of greater than 100 inferences per second. SCl’s
Vision System will provide real-time analysis of imaging data from intelligent weap-
ons and reconnaissance aircraft. It is expected to use computers providing 10 to 100
billion instructions per second, which will require massive parallel processing. The
Pilot's Associate will be an integrated set of nine expert systems designed to
provide a wide range of services to pilots in combat situations, including the planning
of mission tactics, monitoring the status of key systems, navigation, and targeting.*®
The system contemplates communication with the pilot via speech recognition and
advanced visual displays. Finally, SCl’s Battle Management System will assist the
commanders of naval aircraft in coordinating resources and tactics during conflicts.
One issue confronting defense applications is the complexity of the
software systems required and the inherent difficulty of testing such systems under
realistic conditions. The core avionics of a typical fighter aircraft in the late 1980s
used about 200,000 lines of software; the fighter aircraft of the early 1990s is
expected to require about a million lines. Altogether, it is estimated that the U.S
military in 1990 uses about 100 million lines of code, which is expected to double
within five years. Assuring the quality of such extensive software systems is an
urgent problem that planners are only beginning to address.
As we approach the next generation of expert systems, three bottlenecks
have been recognized and are being attacked by researchers in the United States,
Europe, and Japan. Inference engines that use parallel processing are being devel-
oped to process the substantially larger knowledge and rule bases now being
developed. The number of inferences per second that need to be handled often
grows exponentially with the size of the knowledge base. With expert systems being

301
 * User Library
* Group Library

Intelligent chip design. At an NEC

research lab in Kawasaki, Japan, a
computer-aided design (CAD) system
generates a blueprint for a VLSI chip.
(Photo by Lou Jones)
weapons
a
applied to such
systems,
real-time situations as monitoring
the need for rapid response
Perhaps the most important issue being addressed
nuclear plants and coordinating
is obvious.*?
is learning.*° Building
knowledge bases by hand is extremely painstaking. Methods for automating this
process have become a major focus of both academic and commercial development.
EURISKO, a system developed at Stanford's Knowledge Systems Laboratory and
now being refined at the Microelectronics and Computer Technology Corporation,
has demonstrated an ability to generate its own rules as it solves problems and
receives feedback from users.°' Several new expert systems are capable of growing
their own knowledge bases simply by observing human decision makers doing their
jobs. This is particularly feasible in financial applications, since in most cases financial
decisions, as well as the data they are based on, are already easily accessed by
computer communications
Finally, another major issue is the proper use of uncertain and fragmentary
information. It is well known that effective use of such information is vital to most
human decision making, but recent research by Robert Hink and David Woods at
Expert Systems Design has cast some doubt on just how skilled humans are at it.*
Hink and Woods found that for most people, including human experts within their
i domains, the ability to balance risks and make optimal decisions involving probabilis-
tic outcomes is surprisingly low.®? For example, in some instances subjects would
choose an opportunity for a large reward with a small probability over a smaller but
certain reward even though the expectant value (the reward multiplied by the
probability of receiving it) of the larger reward was substantially lower. One might

| ascribe this behavior to an enjoyment of risk taking and a general desire to reach for
i large rewards. In other instances these same subjects would select the more
 conservative option even though in these cases its expectant value was less. The
decision-making patterns were neither optimal nor consistent. Often the way in
which questions were worded had more effect on the choices than the underlying
facts. These findings are consistent with the commonsense observation that some
people display consistently better performance at certain types of tasks, even when
considerable uncertainty is involved. Classic examples are card games such as poker,
in which skilled players almost always come out on top, even though a significant
element of chance is involved. Such players are obviously able to apply probabilistic
rules in a more consistent and methodologically sound manner than their opponents.
Hink and Woods found that the validity and coherency of most people's decision
making, even within the persons’ areas of professional competence, was dramati-
cally lower than they had expected. The significance of this observation for the
knowledge-engineering community is mixed. The bad news is that devising sound
knowledge bases and rules will remain difficult, since even the human domain
experts are not very effective at applying them, apart from frequent lack of aware-
ness of their own decision rules. The good news is that there is considerable
opportunity for computer-based expert systems to improve on human decision-
making in those domains where the knowledge and decision-making processes can
be captured and applied.**

+ Language: The Expression of Knowledge

Gracie: A truck hit Willy.

George: What truck?
Gracie: The truck that didn’t have its lights on.
George: Why didn’t it have its lights on?
Gracie: \t didn’t have to. It was daytime.
George: Why didn’t the truck driver see Willy?
Gracie: He didn’t know it was Willy.
A comedy routine of Burns and Allen, quoted by Roger Schank to illustrate problems of language comprehension

This is the cheese that the rat that the cat that the dog chased bit ate.
Marvin Minsky, The Society of Mind (citing a valid sentence that humans have difficulty parsing)

Squad Helps Dog Bite Victim

Book of Newspaper Headline Gaffes

Birds can fly, unless they are penguins and ostriches, or if they happen to be dead, or have broken wings,
or are confined to cages, or have their feet stuck in cement, or have undergone experiences so dreadful
as to render them psychologically incapable of flight.
Marvin Minsky, The Society of Mind illustrating the difficulty of accurately expressing knowledge)

“Okay, where did you hide it?”

“Hide what?” ©
“You know.”
“Where do you think?”
“Oh.”
A “married conversation” cited by Nicholas Negroponte of the MIT Media Lab to illustrate the importance of shared
knowledge in interpreting natural language

303
No knowledge is entirely reducible to words, and
no knowledge is entirely ineffable.
Seymour Papert, Mindstorms <

Students of human thought and the thinking process have always paid special
attention to human language.*° Our ability to express and understand language is
often cited as a principal differentiating characteristic of our species. Language is the
means by which we share our knowledge. Though we have only very limited access
to the actual circuits and algorithms embedded in our brains, language itself is quite
visible. Studying the structures and methods of language gives us a readily acces-
sible laboratory for studying the structure of human knowledge and the thinking
process behind it. Work in this laboratory shows, not surprisingly, that language is no
less complex or subtle a phenomenon than the knowledge it seeks to transmit.
There are several levels of structure in human language. The first to be
actively studied was syntax, the rules governing the ways in which words are
arranged and the roles words play. Syntactic rules govern the placement of nouns,
verbs, and other word types in a sentence and also control verb conjugation and
other elements of sentence structure. Although human language is far more com-
plex, similarities in the syntactic structures of human and computer languages were
noticed by Al researchers in the 1960s. An early Al goal was to give a computer the
ability to parse natural language sentences into the type of sentence diagrams that
grade-school children learn. One of the first such systems, developed in 1963 by
Susumu Kuno of Harvard, was interesting in its revelation of the depth of ambiguity
in the English language. Kuno asked his computerized parser what the sentence
“Time flies like an arrow” means. In what has become a famous response, the
computer replied that it was not quite sure. It might mean (1) that time passes as
quickly as an arrow passes. Or maybe (2) it is a command telling us to time the flies
the same way that an arrow times flies; that is, “Time flies like an arrow would.” Or
(3) it could be a command telling us to time only those flies that are similar to
arrows; that is, “Time flies that are like an arrow.” Or perhaps (4) it means that the
type of flies known as “time flies” have a fondness for arrows: “Time-flies like
arrows. "°°
It became clear from this and other syntactical ambiguities that under-
standing language requires both an understanding of the relationships between
words and knowledge of the concepts underlying words.* It is impossible to
understand the sentence about time (or even to understand that the sentence is
indeed talking about time and not flies) without a mastery of the knowledge struc-
tures that represent what we know about time, flies, arrows, and how these
concepts relate to one another. An Al technology called semantic analysis attempts
to apply knowledge of the concepts associated with words to the problem of
language understanding. A system armed with this type of information would note
that flies are not similar to arrows (which would thus knock out the third interpreta-
tion above). (Often there is more than one way to resolve language ambiguities. The
third interpretation could also have been syntactically resolved by noting that “like”
in the sense of “similar to” ordinarily requires number agreement between the two

304
 **
t+ ae 77. J 35
HAIR FO a

Woo Le Wt, ere © 1 DSS

MO HHAC Lk D7 @

Biipos fo BCD) CCIE)

FBHOMAZATL&I EREICRNZ TRIER
S = sentence N=noun
NP= noun phrase
VP = verb phrase
PP = prepositional phrase © D = determiner
= verb
= preposition SV Oe UCHR,
$s
vP
(RO Ajo CARER (2# SS RUTS EAL Waa
.,
NP
NP PP pp

N Vv
Ai
D N eS 6
of N P
Je
a~
D N

John saw the woman in the park witha telescope

P
John saw the woman in the = with a bem

PP
NP NP NP A Japanese display at the Tokyo
Institute of Technology. (Photo by Lou
John saw the woman in the park witha telescope.
Jones)

oh
V
John saw the womanin the park witha — telescope

What did you mean by that? A

[> J.
z Ix
computer sentence-parsing program
reveals the ambiguity of language.
(Data from Skona Brittain, “Under-
standing Natural Languages,” Al
John saw the woman in the park witha a Expert2 [1987], no. 5: 32)

305
The Search for Know! objects that are compared.) It would also note that there is no type of flies known as
time flies (which would probably knock out the fourth interpretation). Another type of
analysis, known as pragmatic analysis, attempts to apply the vast wealth of practical
knowledge about the world to further resolve ambiguities. In applying this technique
to our previous results, we use such tidbits of knowledge as that flies have never
shown a fondness for arrows, and that arrows cannot and do not time anything,
much less flies, to select the first interpretation as the only plausible one.
The ambiguity of language is far greater than may be apparent. During a
language parsing project at the MIT Speech Lab, Ken Church found a sentence
published in a technical journal with over one million syntactically correct interpreta-
tions!
It is clear that a vast amount of knowledge is needed to interpret even the
simplest sentence. Indeed, some of the largest knowledge-based systems have
been developed in the area of language understanding and translation. Translation is
clearly impossible without understanding. Ever since Bar-Hillel’s famous paper of
1960, “A Demonstration of the Nonfeasibility of Fully Automatic High-Quality
Translation,” researchers have understood the necessity that the computer under-
stand both the syntax and semantics of the text in a language before attempting a
translation into another language.*® The Logos computer-assisted translation system,
for example, uses about 20,000 understanding and translation rules to translate
German technical texts into English and still provides results that are only about 80
percent as accurate as a human translator. Logos researchers estimate that it would
require about 100,000 rules to achieve human performance levels, even in the
restricted domain of technical texts.°?
Understanding human language in a relatively unrestricted domain remains
too difficult for today’s computers. Beyond the resolution of syntactic and semantic
ambiguities, there are many issues regarding unspoken assumptions and appropriate-
ness. If | say, “The bread is ready to eat,” one can assume that it is not the bread
that will do the eating; however, if | say, “The chicken is ready to eat,” it is less
clear who intends to eat whom. In this case, further contextual information is
required. Resolving such ambiguities in a general way requires extensive knowledge
and the ability to readily access the information most relevant. If | ask someone |
have just met, What do you do? | am probably not interested in hearing about things
which | obviously already know, such as eating, breathing, sleeping, thinking (al-
though if someone wanted to be evasive, these might very well be appropriate
responses). If | ask, Do you want something to eat or not? it would be technically
correct to answer yes even if you did not want something to eat.® Again, avoiding
overly literal interpretations of language requires vast collections of knowledge that
no one has yet bothered to collect. We are, in fact, only now beginning to know
what knowledge to collect and how to collect it.
The ability of humans quickly and accurately to resolve syntactic ambigui-
ties is not perfect, however, and a great deal of humor is based on this type of
confusion. Skona Brittain cites the following example:

John: | want to go to bed with Marilyn Monroe again tonight.

Jane: Again?
John: Yes, |'ve had this desire before

306
 A further complication in understanding language is the ubiquitous use of
idiomatic expressions. Consider the following: “She broke the ice by bringing up
their latest advertising client. ‘We're skating on thin ice with this one,’ she said, ‘the
problem in promoting their new product is just the tip of the iceberg.’” Although
there are three references to ice, these sentences obviously have nothing to do with
ice. Beyond the three ice-related idioms, the phrase “bring up” is also idiomatic,
meaning to introduce a topic by speaking about it. Each idiom introduces a story or
analogy that must somehow be integrated with the rest of the subject matter
Understanding language can thus require a knowledge of history, myths, literary
allusions and references, and many other categories of shared human experience.®!
When we talk to other human beings, we assume that they share with us
an enormous core of knowledge and understanding about both the world and the
subjects we plan to talk about. These assumptions vary according to who we are
talking to and how much we know we have in common. Talking to coworkers or
good friends obviously allows us to make more assumptions than talking to a
stranger on the street, although even here we often assume shared knowledge of
sports teams, current events, history, and many other topics. But in talking (or
typing) to computers, no understanding of human commonsense knowledge can yet |
be assumed. While the slow and arduous process of teaching basic world knowl- |
edge to computers has begun, the most successful strategy in creating viable
computer-based natural-language applications has been to limit the domain of
discourse to an area where virtually all of the relevant knowledge can be captured
One of the first examples of this approach was Terry Winograd’s 1970 MIT thesis
SHRDLU.® SHRDLU understood commands in natural English, as long as the
commands pertained only to an artificial world composed of different colored blocks.
While the limitations of the toy worlds of SHRDLU and other similar systems were
criticized in the 1960s and 1970s, it turns out that such sharply constrained domains
do have practical value.** For example, the world of computer data bases is no
more complicated than the world of colored blocks, but it happens to be one that
many business people do interact with daily. One of the more successful natural-
language-understanding programs, Intellect, from Artificial Intelligence Corporation,
allows users to ask questions in natural English concerning information in their data
bases.® Because the domain is sufficiently limited, the Intellect system can rely
primarily on syntactic rules, although semantic and pragmatic knowledge has also
been incorporated. Competitive systems developed by Cognitive Systems, a
company founded by Roger Schank, rely more heavily on explicit representations of
semantic and pragmatic knowledge. By means of Schank’s script methodology,
similar knowledge bases can be used for both language understanding and decision
making.® For example, an expert system called Courtier, created by Cognitive
Systems for a Belgian bank, can provide portfolio advice in response to natural-
language commands.®”
Perhaps the largest market for natural-language systems is translation.®
Translating (mostly technical) texts by means of traditional techniques is today a
multibillion-dollar business. While computerized translation systems are not yet
sufficiently accurate to run unassisted, they can significantly increase the productivity
of a human translator. One of the challenges in developing computerized translation

307
 Roger Schank, director of the
Institute for the Learning Sciences at
Northwestern University and a
pioneer in knowledge representation
and natural-language understanding.
(Photo by Dan McCoy of Rainbow)

Larry Harris, president of Artificial

Intelligence Corp. (AIC) and
developer of natural-language-
understanding systems. AIC's
Intellect program enables executives
to interact with their data base
systems using ordinary English
(Photo by Lou Jones)
systems is that each pair of languages represents a different translation problem, or
rather, each pair of languages represents a pair of problems. An interesting approach
to simplifying this difficulty is being pursued by DLT, a Dutch firm that is developing
translators for six languages to and from a standard root language. They use a
modified form of Esperanto, a century-old language originally proposed as a universal
language, as an intermediate representation. In their system, a translation from
English to French would be accomplished in two steps, English to Esperanto and
then Esperanto to French. Esperanto was selected because it is particularly good at
representing concepts in an unambiguous way, thus making it ideal as a root concep-
tual language. Translating among 6 different languages would ordinarily require 30
different translators (6 languages, each to be translated into 5 other languages), but
with the DLT approach, only 12 are required (6 translators from the 6 languages into
Esperanto and 6 more from Esperanto back to the 6 languages). Furthermore, as
DLT points out, the translating systems that go from Esperanto to another language
are relatively simple.®
Perhaps the most intensive work on automatic language translation is
being pursued at a number of Japanese research centers, including Kyoto University
and the Tokyo Institute of Technology. Japan's Ministry of International Trade and |
Industry has cited Japanese-English and English-Japanese translation systems as
vital to Japan’s economy.
One of the primary barriers to more widespread use of computer technol-
ogy is communication between human and machine. Most persons, even those with
technical training in fields other than computer science, find the specialized syntax
required by most computer applications to be intimidating and confusing. There is
considerable agreement that the optimal way to interact with computers would be
the same way we interact with our human colleagues and assistants: by talking
things over in a natural language. Providing this capability will require integrating
large-vocabulary speech recognition (recognizing what the words are from speech)
with a high level of natural-language understanding (understanding what the words
mean). Though these capabilities are not yet sufficiently well developed to replace
either the keyboard or the specialized computer syntaxes now used for most
computer applications, the natural-language market is beginning to take hold for a
variety of applications. DM Data estimates the natural-language market (not including
speech recognition) at $80 million in 1987 and projects a $300 million market in
1990.7°
Once again, we find that the relative strengths of machine and human
intelligence are quite different. Humans first learn to listen to and understand spoken
language. Later on we learn to speak. Computers, on the other hand, have been
generating printed natural-language output since their inception and have only
recently begun to understand it. Also, the ability of computers to speak predated
their ability to recognize spoken language. Another example: humans gain mastery of
written language at a much later age than verbal language. Here too computer
history has reversed this sequence. As a final example, computers today can
understand natural-language sentences dealing with “complex” financial inquiries yet
still stumble on the “simple” sentences that children can understand. This phenome-
non is widely misunderstood. The popular robot character R2D2 (of Star Wars fame)

309
 Language translation at present
involves bilingual humans making
copious use of a dictionary. (Photo
by Lou Jones)

Jun-Ichi Tsuji of Kyoto University,

an authority on the automatic
translation of languages. (Photo by
Lou Jones)

310
 Atale of two countries. The
keyboards on this computer displays
both English and Japanese charac-
ters. The upper board allows direct
input of Japanese ideographs.
Developed at the Tokyo Institute of
Technology, this machine translation
system can translate text from either
language almost instantaneously.
(Photos by Lou Jones)

SJ eS mT se or Apto op ek
Pal TE Ka (BE EE RE 29 ll BE PU V WX Y ZI+ —
Hr feat fa i) ike (8 So FP HB uv wx y zo.
RLS ERE OR RY Sf ET eCH
A WA a 2 fa RE HS RAN HA) Hes
ZCRAAGRSKVUADAHOOME
Sot FF BG OK 8A AL Eb fmm cm km cor mt kat
Be Fas SW FE AS AIL fice me de & ke
Ae PC 6 AE 8% ft IK fms ws ns p:
Se FS Wk 1 TK OAR UK i r
OR] RE ER URE a i
MS Fee Ac Ta Se, Hs i a it 1 f
& G8 7
BH
Br on] ft A tO
THE can HHL Je ie #7 SE SR he
fi EB ke A YK HB A PY /
INGER Az 5 2 BT A oh aK
re 2) #6 /4e Be Me te BRE 1 Oe Rk
SI Oh ihe $8 4 Pd Fe 2 ROK FR HE UK HALE A IM MKS 7
EKREGGMAABTAERZHOIKAMNEOT t
fem Ble Bro eFC nOrcKrauveéon
Te ZRIEITOXVQABBrREEXK3UM
THB dr boma i ar ne BK y

au
 The Search for Knowledge
is supposed to understand many human languages yet is unable to speak (other than
with “computerlike” squeaks and other noises), which gives the mistaken impres-
sion that generating human language (and speech) is far more difficult than under-
standing it.

+ Putting It All Together: The Age of Robots

| am incapable of making an error.

HAL, in Stanley Kubrick and Arthur C. Clarke's 1968 film 2001: A Space Odyssey

1. A robot may not injure a human being, or, through inaction, allow a human being to come to harm.
2. A robot must obey the orders given it by human beings, except where such orders would conflict with
the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First or
Second Law.
Isaac Asimov's three laws of robotics

In R.U.R., a play written in 1921, the Czech dramatist Karel Capek (1890-1938)
describes the invention of intelligent mechanical machines intended as servants for
their human creators. Called robots, they end up disliking their masters and take
matters into their own “hands.” After taking over the world, they decide to tear
down all symbols of human civilization. By the end of the play they have destroyed
all of mankind. Although Capek first used the word “robot” in his 1917 short story
“Opilec,” creating the term from the Czech words “robota,” meaning obligatory
work, and “robotnik,” meaning serf, R.U.R. (for “Rossum’s Universal Robots”)
introduced the word into popular usage. Capek intended his intelligent machines to
be evil in their perfection, their ultimate rationality scornful of human frailty. Although
a mediocre play, it struck a chord by articulating the uneasy relationship between
man and machine and achieved wide success on two continents.”' The spectre of
machine intelligence enslaving its creators, or at least competing with human intelli-
gence for employment and other privileges, has continued to impress itself on the
public consciousness
i Although lacking human charm and good will, Capek’s robots brought
together all of the elements of machine intelligence: vision, auditory perception,
touch sensitivity, pattern recognition, decision making, judgement, extensive world
knowledge, fine motor coordination for manipulation and locomotion, and even a bit
i of common sense. The robot as an imitation or substitute for a human being has
remained the popular conception. The first generation of modern robots was, how-
i ever, a far cry from this anthropomorphic vision.” The Unimation 2000, the most
popular of the early “robots,” was capable only of moving its arm in several direc-
tions and opening and closing its gripper. It had no senses and could move its arm
with only two or three degrees of freedom (directions of movement) of the six
possible in three-dimensional space. Typical applications of these early robots,
introduced during the 1970s, involved moving objects from one place to another (a
capability called pick and place)
More sophisticated devices, such as the American company Cincinnati
Milacron’s T3 (The Tomorrow Tool), the German KUKA robots, and the Japanese

312
Wabot-2, an anthropomorphic robot
with a penchant for jazz. (Courtesy of
Ichiro Kato of Waseda University,
Tokyo)

313
The dance of the robots. Their use for
welding cars was one of the earliest
applications of industrial robots,
shown here at a Chrysler manufactur-
ing plant. (Courtesy of Cincinnati
Milacron)
A computer-assisted design (CAD)
display. Here the computer designs a
robot (with some human help). (Photo
by Dan McCoy of Rainbow)

Hitachi robots, were introduced in the early 1980s. These second-generation robots
can move with five or six degrees of freedom, can effect more precise movements,
are faster, and have more delicate grippers. The motions of these robots can be
programmed, but they still had no provision for conditional execution, that is, opera-

ai
tions conditioned on some external event. Since these robots still have no way of
sensing their environment, there are no inputs on which to base any decision
making. These second-generation robots became well known for welding and spray
painting, primarily in the automotive industry.”
The third generation, introduced in the mid 1980s, began to display a
modicum of intelligence. Robots of this generation—Unimation’s PUMA, IBM's 7535
and 7565, and Automatix’s RAIL series—contain general-purpose computers inte-
grated with vision and/or tactile sensing systems. By 1987 robotic vision systems
alone had developed into a $300 million industry, with estimates of $800 million for
1990.”* Specialized programming languages, such as Unimation’s VAL and IBM's
AML, allow these robots to make decisions based on changes in their environment.
Such systems can, for example, find industrial parts regardless of their orientation
and identify and use them appropriately in complex assembly tasks.’
With the flexibility and sophistication of robots improving each year, the
population of industrial robots has increased from a few hundred in 1970 to several
hundred thousand by the late 1980s. Some are used in factories that are virtually
totally automated, such as Allen Bradley's facility for manufacturing electric motor
starters (see the companion film to this book).”” The only human beings in this
factory monitor the process from a glass booth, while computers control the entire
flow of work from electronically dispatched purchase orders to shipped pro!
Though the era of workerless factories has begun, the most significant sh
impact of this latest generation of robots is in settings where they wo
human coworkers. Increasingly, new factories are designed to incorporate b
human and machine assemblers, with the flow of materials mo
by computers.’®

315
 This man is making dishwashers. A
computer-controlled assembly line at
the GE dishwasher factory. (Photo by
|
Bill Strode).
!
|

Do not touch. Very few hands touch

this well-known personal computer
during its manufacture. (Photo by
Dan McCoy of Rainbow)

316
 With the arrival of the third generation, the diversity of tasks being accom-
plished by robots has broadened considerably.®° A robot named Oracle is shearing
sheep in western Australia. One called RM3 is washing, debarnacling, and painting
the hulls of ships in France. Several dozen brain operations have been performed at
Long Beach Memorial Hospital in California with the help of a robot arm for precision
drilling of the skull. In 1986 police in Dallas used a robot to break into an apartment
in which a suspect had barricaded himself. The frightened fugitive ran out of the
apartment and surrendered.®' The U.S. Defense Department is using undersea
robots built by Honeywell to disarm mines in the Persian Gulf and other locations
Thousands of robots are routinely used in bioengineering laboratories to perform the
extremely delicate operations required to snip and connect minute pieces of DNA.
And walking robots are used in nuclear power plants to perform operations in areas
too dangerous for humans. One such robot, Odetics’s Odex, looks like a giant spider
with its six legs.
The next generation of robots will take several more steps in replicating
the subtlety of human perceptual ability and movement, while retaining a machine's
inherent advantages in speed, memory, precision, repeatability, and tireless opera-
tion. Specialized chips are being developed that will provide the massively parallel
computations required for a substantially higher level of visual perception. Equally so-
phisticated tactile sensors are being designed into robot hands. Manipulators
with dozens of degrees of freedom will combine the ability to lift both very heavy
objects and delicate ones without breaking the latter. These robots’ “local” intelli-
gence will be fully integrated into the computerized control systems of a modern
factory.
Forerunners of these robots of the 1990s are beginning to compete with
human dexterity and intelligence on many fronts. A robot developed by Russell
Anderson of Bell Labs can defeat most human opponents at Ping-Pong.™ Two other
Ping-Pong playing robots, one English and one Japanese, recently met each other in
San Francisco for a match. A robot hand developed at the University of Utah can
crack an egg, drop the contents into a mixing bowl, and then whip up an omelette
mixture all at several times the speed of a master chef.®° The Stanford/JPL Hand,

It’s harder than it looks. A robot arm at

the Boston Museum of Science stack-
ing blocks. (Photo by Lou Jones)

317
 Tsuneo Yoshikawa, director of designed by Kenneth Salisbury and other MIT researchers, is a three-fingered robot
Robotics Engineering at Kyoto that can perform such intricate tasks as turning a wing nut. A collaborative effort
University, with his students. (Photo
v

by Lou Jones) now underway between the University of Utah Center for Biomedical Design and the
MIT Artificial Intelligence Laboratory aims at constructing a hand that will “exhibit
performance levels roughly equivalent to the natural human hand,” according to
Stephen Jacobsen, the chief designer of the project. A voice-activated robot to
provide quadriplegic patients such personal services as shaving, brushing teeth,
feeding and retrieving food and drinks is being developed by Larry Leifer and Stefan
Michalowski under a grant from the Veterans Administration (see the companion film
to this book).®°
A particularly impressive robot called Wabot-2 (Waseda Robot; see the
picture at the beginning of this section) was developed in the mid-to-late 1980s by
Waseda University in Tokyo and refined by Sumitomo Electric.®’ This human-size
(and humanlike) 200-pound robot is capable of reading sheet music through its
camera eye and then, with its ten fingers and two feet, playing the music on an
organ or synthesizer keyboard. It has a total of 50 joints and can strike keys at the
rate of 15 per second, comparable to a skilled keyboard player. Its camera eye
provides relatively high resolution for a robot. Using a charge-coupled device (CCD)
sen j array, it has a resolution of 2,000 by 3,000 pixels (by comparison, the eyes
of a person with good eyesight can resolve about 10,000 by 10,000 points). Wabot-2 ;
also has a sense of hearing: it can track the pitch of a human singer it is accompany-
ing and adjus the tempo of its playing accordingly. Finally, the robot has rudimentary

318
 Hurahiko Asada, a robotics expert
at Kyoto University. Asada pioneered
the application of the direct-drive
motor to improve robots’ fine motor
coordination. (Photo by Lou Jones)

af

A Japanese application of robotics.

(Photo by Fischer of Woodfin Camp
and Associates)

319
Ken Salisbury of the MIT Artificial speech-recognition and synthesis capabilities and can engage in simple conversa-
Intelligence Lab fine tunes the
tions. There are severe limitations on the complexity of the musical score that it can
dexterous Stanford-JPL robot hand.
(Photo by Lou Jones) read, and the music must be precisely placed by a human assistant. Nonetheless,
Wabot-2 is an impressive demonstration of the state of the robotic art in the late
1980s
Another state-of-the-art robot developed around the same time was a half-
scale model of a Quetzalcoatlus northropi (better known as pterodactyl, a winged
dinosaur that lived 65 million years ago). The replica, developed by human-powered-
flight pioneer Paul MacCready, could fly by flapping its robotic wings, much like its
reptile forebear. Unfortunately, in a demonstration for the press, MacCready’s
invention crashed, which caused a loss of public interest in it. It has flown success-
fully, however, and it represents a particularly sophisticated integration of sensors

and actuators with real-time decision-making by on-board computers.

Not surprisingly, the world’s largest supporter of robotic research is the
U.S. Defense Department, which foresees a wide range of roles for robotic fighters
in the 1990s and early twenty-first century. A walking truck with fat bent legs is
being developed by the U.S. Army for roadless terrains.®° The U.S. Air Force is
developing a number of pilotless aircraft, or flying robots, that can perform a variety
of reconnaissance and attack missions.°° Early versions of such robot craft played a
vital role in the Israeli destruction of 29 Russian surface-to-air missile (SAM) sites in
the Bekaa Valley in a single hour during its invasion of Lebanon in 1982
The field of robotics is where all of the Al technologies meet: vision,
pattern recognition, knowledge engineering, decision-making, natural-language

320
 ad

Anita Flynn of the MIT Artificial

Intelligence Laboratory with a robot
||
| friend. (Photo by Lou Jones)

|
| The pterodactyl returns to life. Paul
MacCready’s flying robot recreates
the Quetzalcoatlus northropi, a flying
dinosaur that became extinct nd
millions of years ago. (Photo by 0. C.
Carlisle)

321
 understanding, and others. As the underlying technologies mature and as the
growing corps of robot designers gets better at integrating these diverse technolo-
gies, robots will become increasingly ubiquitous.*' They will tend our fields and |
livestock, build our products, assist our surgeons; eventually they will even help us |
clean our houses. This last task has turned out to be one of the most difficult. As we |
have seen with other Al problems, machine intelligence has first been successfully
deployed in situations where unpredictable events are held to a minimum. It was not
surprising, therefore, that manufacturing was the first successful application for |
robotic technology, since factories can be designed to provide predictable and |
relatively well-organized environments for robots to work in. In contrast, the environ- |
ments of our homes change rapidly and present many unpredictable obstacles.” So,
effective robotic servants in the home will probably not appear until early in the next
century. By that time, however, robotic technology will have dramatically trans-
formed the production and service sectors of society.%

+ An International Affair

In 1981 Japan’s powerful Ministry of International Trade and Industry (MITI) an-
nounced plans to develop a new kind of computer. This new computer would be at
least a thousand times more powerful than the models of 1981, would have the
intelligence to converse with its human users in natural spoken language, would be
programmed with vast arrays of knowledge in all domains, would have human-level
decision-making capabilities, and would sit on a desktop. They called this new type
of machine a fifth-generation computer. The first four generations were characterized
by the type of electronic components they used, the first being vacuum tubes, the
second transistors, the third integrated circuits, and the fourth VLSI (very large scale
integrated) chips. The fifth generation of computers, on the other hand, would be
characterized by something different, by its intelligence.** With MITI's track record of
having led Japanese industry to dominance in consumer electronics and a broad
range of other high-tech fields, the announcement was a blockbuster. MITI formed
the Institute for New Generation Computer Technology (ICOT) to carry out its
project. ICOT began active development in 1982 with funding of approximately $1
billion (half from MITI and half from industry) for ten years.2°
In the United States and Europe the spectre of the loss of the strategically
important computer industry led to urgent consultations at the highest levels of
government and industry.%A few months after ICOT began development, a major
response by American industry had been initiated. Twenty-one leading American
computer and electronics companies, among them Control Data, Digital Equipment,
Honeywell, NCR, Sperry, and Bell Communication Research, had formed a new
company called Microelectronics and Computer Technology Corporation (MCC). This
collaboration was intended to pool money and research talent to overcome several
bottlenecks in advanced computer research that no single member of the consor-
tium had the resources to solve alone.’’ IBM was not invited to join because of
concerns regarding antitrust laws. Special legislation was still required and was
signed by President Reagan in 1984.°° MCC's research budget of about $65 million
per year is targeted at a wide variety of Al, human-interface, and computer-architec-
ture problems. Primary goals of MCC research are the development of integrated-

322
Circuit packaging techniques and computer-assisted design tools that will give
American companies a practical edge in the worldwide electronics marketplace.
MCC hopes to develop a chip-design station that will enable a small group of
engineers to completely design an advanced custom VLSI chip in under a month,
whereas one to two years is now required. MCC has also identified parallel process-
ing as the most effective way to achieve the massive increases in processing power
required for future Al applications
In addition to MCC, the Defense Advanced Research Projects Agency
(DARPA), which has historically funded a major portion of American university-based
Al research, has increased its Al funding. It is now investing over $100 million per
year.
The English response to Japan's challenge was a $500 million program
called the Alvey Program, after John Alvey, who had proposed it and became its
chairman.%? Unlike MCC, which conducts 97 percent of its research in house, Alvey
has no research laboratory of its own. With funding primarily from the government,
Alvey has provided money for over 100 colleges, companies, and laboratories to
conduct research in a wide variety of Al, VLSI, software engineering, man-machine
interfaces, and other advanced computer technologies. Alvey has planned a number
of “demonstrators” to integrate the results of its research, but its primary emphasis
is to encourage advanced research laboratories in England to put high priority on
information technologies and to train a new generation of knowledge engineers and
computer scientists.'°
In 1984 the European Economic Community (EEC) formed the European
Strategic Program for Research in Information Technology (ESPRIT).'°' This $1.5
billion program has funded companies, universities, and government laboratories
throughout Europe in virtually all areas of computer technology, including office
automation, robotics, and computer-aided manufacturing.
'°
In many ways the original MIT| announcement was perfectly timed to
create an intense response. In addition to growing concern in the United States and
Europe over Japanese competition for trade markets, there was a growing aware-
ness that Al technology, which had been a focus of largely academic research for
over 25 years, was now poised to radically transform the way computers are used
and to have a far-reaching impact on society. The MCC, DARPA, Alvey, and ESPRIT
responses to the MITI challenge were just the tip of the iceberg. The enormous
publicity that Japan's fifth-generation computer project received and the ensuing
conferences, books, congressional inquiries, and other forms of debate helped to set
the priorities of industry and university research and development throughout the
world.‘
The timing turned out to be right. According to DM Data, commercial
revenue in the U.S. from Al-related technologies (not including robotics), was only
$52 million in 1981, grew to $1.4 billion by 1987, and is projected to be $4 billion in
1990. The Al industry as a whole is expected to hit tens of billions of dollars per year
during the 1990s. Very likely the worldwide information and computer industry will
be over one trillion dollars per year in the year 2000 and will be largely intelligent by
today’s standards. Indeed, the ability to effectively harness information and knowl-
edge is expected to be the key to wealth and power in the decades ahead

323
 bea igenbaum

Knowledge Processing:
From File Servers to Knowledge Servers It has been said that when people make forecasts, they overestimate
what can be done in the short run and underestimate what can be
achieved in the long run. | have worked in the science and technol-
| Edward Feigenbaum is
| professor of computer science ogy of artificial intelligence for twenty years and confess to being
at Stanford University, where chronically optimistic about its progress. The gains have been
| he is also scientific director of substantial, even impressive. But we have hardly begun, and we must
the Heuristic Programming
not lose sight of the point to which we are heading, however distant it
| Project, a leading laboratory
for work in knowledge may seem.

engineering and expert We are beginning the transition from data processing to
systems. He is cofounder of knowledge processing. The key tool of our specialty is the digital
two major applied Al
computer, the most complex and yet the most general machine ever
companies: IntelliCorp and
Teknowledge. His most recent invented. Though the computer is a universal symbol-processing
books are The Fifth Generation: device, we have exploited to date only its mundane capabilities to
Artificial Intelligence and file and retrieve data (file service) and to do high-speed arithmetic.
Japan's Computer Challenge to
Researchers in artificial intelligence have been studying techniques
the World (1983), coauthored
with Pamela MeCorduck, and for computer representation of human knowledge and the methods by
The Rise of the Expert which that knowledge can be used to reason toward the solution of
Company: How Visionary problems, the formation of hypotheses, and the discovery of new
Companies are using Artificial
concepts and new knowledge. These researchers have been
Intelligence to Achieve Higher
Productivity and Profits (1988), inventing the knowledge servers of our future.
coauthored with Pamela Like all creators, scientists and technologists must dream,
McCorduck and Penny Nii. He must put forth a vision, or else they relegate their work to almost
is widely regarded as a pioneer
pointless incrementalism. My dream is about the future of Al research
in the theory and methodology
of expert systems. and development over the next several decades and the knowledge
systems that can be produced thereby to assist the modern knowl-
edge worker.

324
Edward Feigenbaum. (Photo by Lou
Jones)

The Beginnings of the Dream (or consultation) system. The core of MYCIN was its knowledge base
SS ==
of rules for the diagnosis and therapy of infectious diseases. Its
Fifty years ago, before the modern era of computation began, Turing's reasoning process was simple (backward chaining), even ad hoc in
theorems and abstract machines gave hint of the fundamental idea parts. But MYCIN was built as an integrated package of intellectual
that the computer could be used to model the symbol-manipulating abilities. It could interact with a professional in the professional
processes that make up the most human of all behaviors: thinking. jargon of the specialty. It could explain its line of reasoning. And it
More than thirty years ago the work began in earnest (1991 will mark had a subsystem that could aid in the acquisition of new knowledge
the thirty-fifth anniversary of the Dartmouth Summer Conference on by guiding an expert to find defects in the stored knowledge. Overall,
Artificial Intelligence). The founding principle of Al research is really MYCIN provided strong confirmation to the knowledge-is-power
an article of faith that the digital computer has the necessary and hypothesis.
sufficient means for intelligent action. This first principle is called the At nearly the same time other efforts in medical problem
physical-symbol-system hypothesis. solving were providing similar results. At the University of Pittsburgh
The early dreaming included dreams about intelligent the focus of the Internist project was the construction of an enormous
behavior at very high levels of competence. Turing speculated on electronic textbook of the knowledge of internal medicine. With its
wide-ranging conversations between people and machines and on current knowledge base of 572 diseases, nearly 4,500 manifestations,
chess playing programs. Later Newell and Simon wrote about and hundreds of thousands of links between them, Internist has
champion-level chess programs and began their work toward that provided the strongest confirmation yet of the knowledge-is-power
end. Samuel (checker playing), Gelernter (geometry-theorem proving), hypothesis.
and others shared the dream. in the late 1970s an explosion of expert systems was
At Stanford, Lederberg and | chose reasoning in science taking place in fields other than medicine: engineering, manufactur-
as our task and began work with Buchanan and Djerassi on building ing, geology, molecular biology, financial services, diagnostic
a program that would elucidate chemical structure at a high level of servicing of machinery, military signal processing, and many other
competence: the DENDRAL program. What emerged from the many areas. There is little that ties these areas together other than this: in
experiments with DENDRAL was an empirical hypothesis that the each, high-quality problem solving is guided by experiential,
source of the program's power to figure out chemical structures from qualitative, heuristic knowledge. The explosion of applications
spectral data was its knowledge of basic and spectral chemistry. For created a new type of professional, the knowledge engineer (now in
DENDRAL, knowledge was power. Obvious? In retrospect, perhaps. extremely short supply) and a new industry, the expert systems
But the prevailing view in Al at the time ascribed power to the industry (now rapidly expanding ). One generalization from the frenzy
reasoning processes—in modern terms, to the inference engine, not of activity is simply massive additional confirmation of the knowl-
the knowledge base. Thus, in the late 1960s the knowledge-is-power edge-is-power hypothesis. The reasoning procedures associated
hypothesis stood as a counter-hypothesis awaiting further tests and with all of these systems are weak. Their power lies in their
the accumulation of evidence. knowledge bases.
Much evidence came in the 1970s. Medical problem Other areas of Al research made shifts to the knowledge-
solving provided the springboard. The MYCIN program of Shortliffe base viewpoint. It is now commonplace to say, A program for
and others at Stanford was the prototype of the expert-level advisory understanding natural language must have extensive knowledge of
its domain of discourse. A vision program for image understanding

325
must have knowledge of the world it is intended to see. And even, stylistic expression, vocabulary, and concepts. For example,
learning programs must have a substantial body of knowledge from programs rarely accept synonyms, and they cannot interpret and use
which to expand (that is, learning takes place at the fringes and metaphors. They always interact in a rigid grammatical straitjacket.
interstices of what is already known). Thus, the dream of a computer The need for metaphor to induce in the user a feeling of naturalness
that performs at a high level of competence over a wide variety of seems critical. Metaphorical reference appears to be omnipresent
tasks that people perform well seems to rest upon knowledge in the and almost continuous in our use of language. Further, if you believe
task areas. that our use of language reflects our underlying cognitive processes,
The knowledge-is-power hypothesis has received so then metaphor is a basic ideational process.
much confirmation that we can now assert it as the knowledge In the second era we shall see the evolution of the natural
principle: interface. The processes controlling the interaction will make greater
use of the domain knowledge of the system and knowledge of how to
A system exhibits intelligent understanding and action at a high level
conduct fluid discourse. Harbingers of naturalness already exist; they
of competence primarily because of the specific knowledge that it
are based to a large extent upon pictures. The ONCOCIN project team
contains about its domain of endeavor.
at Stanford invested a great effort in an electronic flow sheet to
A corollary to the knowledge principle is that reasoning
provide a seamless transition for the oncologist from paper forms for
processes of an intelligent system, being general and therefore weak,
patient data entry to electronic versions of these forms. The commer-
are not the source of power that leads to high levels of competence in
cially available software tools for expert-system development
behavior. The knowledge principle simply says that if a program is to
sometimes contain elegant and powerful packages for creating
perform well, it must know a great deal about the world in which it
pictures that elucidate what the knowledge system is doing and what
operates. In the absence of knowledge, reasoning won't help.
its emerging solution looks like (for example, IntelliCorp’s KEE
The knowledge principle is the emblem of the first era of
Pictures and Active Images).
artificial intelligence; it is the first part of the dream. It should inform
Naturalness need not rely upon pictures, of course. The
and influence every decision about what it is feasible to do in Al
advances in natural-language understanding have been quite
science and with Al technology.
substantial, particularly in the use of knowledge to facilitate
understanding. In the second era it will become commonplace for
The Middle of the Dream
knowledge systems to interact with users in human language, within
SS RENE
the scope of the system's knowledge. The interaction systems of the
Today our intelligent artifacts perform well on specialized tasks
second era will increasingly rely on continuous natural speech. In
within narrowly defined domains. An industry has been formed to put
person-to-person interactions, people generally talk rather than type.
this technological understanding to work, and widespread transfer of
Typing is useful but unnatural. Speech-understanding systems of
this technology has been achieved. Although the first era of the
wide applicability and based on the knowledge principle are coming.
intelligent machine is ending, many problems remain to be solved.
At Stanford we are beginning experiments with an experimental
One of these is naturalness. The intelligent agent should
commercial system interfaced with the ONCOCIN expert system.
interact with its human user in a fluid and flexible manner that
A limitation of first-era systems is their brittleness. To mix
appears natural to the person. But the systems of the first era share
metaphors, they operate on a high plateau of knowledge and
with the majority of computer systems an intolerable rigidity of
competence until they reach the extremity of their knowledge; then

326
they precipitously fall off to levels of utter incompetence. People assumptions, and you will understand what | mean by a large
suffer from the same difficulty (they too cannot escape the knowledge knowledge base of commonsense knowledge.
principle), but their fall is more graceful. The cushion for the soft fall As knowledge in systems expands, so does the scope for
is the knowledge and use of weaker but more general models that modes of reasoning that have so far eluded the designers of these
underlie the highly specific and specialized knowledge of the systems. Foremost among these modes are reasoning by analogy and
plateau. For example, if an engineer is diagnosing the failure of an its sibling metaphorical reasoning. The essence of analogy has been
electronic circuit for which he has no specific knowledge, he can fall evident for some time, but the details of analogizing have not been. An
back on his knowledge of electronics, methods of circuit analysis, analogy is a partial match of the description of some current situation
and handbook data for the components. The capability for such with stored knowledge. The extent of the match is crucial. If the
model-based reasoning by machine is just now under study in many match is too partial, then the analogy is seen to be vacuous or
laboratories and will emerge as an important feature of second-era farfetched; if too complete then the “analogy” is seen as hardly an
systems. The capability does not come free. Knowledge engineers analogy at all. Analogizing broadens the relevance of the entire
must explicate and codify general models in a wide variety of task knowledge base. It can be used to construct interesting and novel
areas. interpretations of situations and data. It can be used to retrieve
Task areas? But what if there is no “task”? Can we knowledge that has been stored, but not stored in the “expected”
envision the intelligent program that behaves with common sense at way. Analogizing can supply default values for attributes not evident
the interstices between tasks or when task knowledge is completely in the description of the current situation. Analogizing can provide
lacking? Common sense is itself knowledge, an enormous body of access to powerful methods that otherwise would not be evoked as
knowledge distinguished by its ubiquity and the circumstance that it relevant. For example, in a famous example from early twentieth-
is rarely codified and passed onto others, as more formal knowledge century physics, Dirac made the analogy between quantum theory
is. There is, for example, the commonsense fact that pregnancy is and mathematical group theory that allowed him to use the powerful
associated with females, not males. The extremely weak but methods of group theory to solve important problems in quantum
extremely general forms of cognitive behavior implied by common- physics. We shall begin to see reasoning by analogy emerge in
sense reasoning constitute for many the ultimate goal in the quest for knowledge systems of the second era.
machine intelligence. Researchers are now beginning the arduous Analogizing is seen also as an important process in
task of understanding the details of the logic and representation of automatic knowledge acquisition, another name for machine
commonsense knowledge and are codifying large bodies of learning. In first-era systems, adding knowledge to knowledge bases
commonsense knowledge. The first fruits of this will appear in the has been almost always a manual process: people codify knowledge
later systems of the second era. Commonsense reasoning will and place it in knowledge structures. Recent experiments by Douglas
probably appear as an unexpected naturalness in a machine's Lenat have shown that this laborious process can be semiautomated,
interaction with an intelligent agent. As an example of this in facilitated by an analogizing program. The program suggests the
medical-consultation advisory systems, if pregnancy is mentioned relevant analogy to a new situation, and the knowledge engineer fills
early in the interaction or can be readily inferred, the interaction in the details. In the second era we shall see programs that acquire
shifts seamlessly to understanding that a female is involved. Magnify the details with less or no human help. Many other techniques for
this example by one hundred thousand or one million unspoken automatic learning will find their way into second-era systems. For
example, we are currently seeing early experiments on learning

327
 apprentices, machines that carefully observe people performing journals sit on shelves waiting for us to use our intelligence to find
complex tasks and infer thereby the knowledge needed for competent them, read them, interpret them, and cause them finally to divulge
performance. The second era will also see (| predict) the first their stored knowledge. Electronic libraries of today are no better.
successful systems that couple language understanding with Their pages are pages of data files, but the electronic pages are
learning, so that knowledge bases can be augmented by the reading equally passive.
of text. Quite likely these will be specialized texts in narrow areas at Now imagine the library as an active, intelligent
the outset. knowledge server. It stores the knowledge of the disciplines in
To summarize, because of the increasing power of our complex knowledge structures (perhaps in a knowledge-representa-
concepts and tools and the advent of automatic-learning methods, we tion formalism yet to be invented). It can reason with this knowledge
can expect that during the second era the knowledge bases of to satisfy the needs of its users. These needs are expressed naturally,
intelligent systems will become very large, representing therein with fluid discourse. The system can, of course, retrieve and exhibit
hundreds of thousands, perhaps millions, of facts, heuristics, (i.e., it can act as an electronic textbook). It can collect relevant
concepts, relationships, and models. Automatic learning will be information; it can summarize; it can pursue relationships. It acts as a
facilitated thereby, since by the knowledge principle, the task of consultant on specific problems, offering advice on particular
adding knowledge is performed more competently the more solutions, justifying those solutions with citations or with a fabric of
knowledge is available (the more we know, the easier it is to know general reasoning. If the user can suggest a solution or a hypothesis,
more). it can check this and even suggest extensions. Or it can critique the
Finally, in the second era we will achieve a broad user viewpoint with a detailed rationale of its agreement or disagree-
reconceptualization of what we mean by a knowledge system. Under ment. It pursues relational paths of associations to suggest to the user
the broader concept, the “systems” will be collegial relationships previously unseen connections. Collaborating with the user, it uses
between an intelligent computer agent and an intelligent person (or its processes of association and analogizing to brainstorm for remote
persons). Each will perform tasks that he/she/it does best, and the or novel concepts. More autonomously, but with some guidance from
intelligence of the system will be an emergent of the collaboration. If the user, it uses criteria of being interesting to discover new
the interaction is indeed seamless and natural, then it may hardly concepts, methods, theories, and measurements.
matter whether the relevant knowledge or the reasoning skills The user of the library of the future need not be a person. It
needed are in the head of the person or in the knowledge structures may be another knowledge system, that is, any intelligent agent with
of the computer. a need for knowledge. Thus, the library of the future will be a network
of knowledge systems in which people and machines collaborate.
The Far Side of the Dream: The Library of the Future Publishing will be an activity transformed. Authors may bypass text,
SS SR YS
adding their increment to human knowledge directly to the knowl-
Here's a “view from the future,” looking back at our “present,” from edge structures. Since the thread of responsibility must be main-
Professor Marvin Minsky of MIT: “Can you imagine that they used to tained, and since there may be disagreement as knowledge grows,
have libraries where the books didn’t talk to each other?” The the contributions are authored (incidentally allowing for the
libraries of today are warehouses for passive objects. The books and computation of royalties for access and use). Maintaining the
knowledge base (updating knowledge) becomes a vigorous part of
the new publishing industry.

328
 At the far horizon the dream can take many forms and
dimensions. | have briefly sketched only a few, and | invite you to
exercise your imagination to sketch your own. At the far horizon, the
question is not “whether,” but only “when.”

Power Tools for the Mind: A Statement of the Encompassing Vision

‘Seam
SS SS ee Ss

The hard work of the farmer was revolutionized by agricultural

machinery. The labor of the industrial worker was revolutionized by
engines and heavy machinery. As we move toward the postindustrial
period of the twenty-first century, as work becomes increasingly the
work of professionals and knowledge workers, the power tools are
and will be digital computers. The economic and social well-being of
the advanced societies is increasingly the result of working
“smarter” rather than working harder, and computers are the agents
of that change. Knowledge is power, in human affairs, and knowledge
systems are amplifiers of human thought and action. As Carlo De
Benedetti, Chief Executive of Olivetti, has said, “For the first time in
the history of mankind, innovation is the fundamental raw material.
Real strategic resources are no longer represented by coal, steel, or
oil but by the cleverness and cognitive capability of man.”

329
 Big changes are coming to the world of automobile repair. In the Ford
dealership of the not too distant future, mechanics will be tracking
down and fixing problems in your car with the help of a powerful Al
diagnostic tool called SBDS, under development by a joint project
team at Ford Motor Company, the Carnegie Group, and Hewlett
Packard. SBDS, the Service Bay Diagnostic System, will revolutionize
the way cars are fixed. When it is completed in the next year or two,
SBDS's knowledge base will contain the expertise of Ford's top
diagnosticians, and it will make their diagnostic skills available to
mechanics in every Ford dealership in North America. This “expert in
a box” will guide a human technician through the entire service
process, from the initial customer interview at the service desk to the
diagnosis and repair of the car in the garage.
Imagine how this expert system will affect the way your
car is serviced. When you first walk up to the service desk, a service
writer interviews you to find out the details of your car's problem. But

| Baas most of the questions he asks are suggested by SBDS, which is trying
to gather important information from you before you leave the
dealership. The questions are tailored to your particular situation and
An Expert System for Automotive Diagnosis
depend on your car, your complaints, your car's repair history, and
Ford's knowledge about possible problems with cars similar to yours.
Jeff Pepper is product After the interview SBDS prints a repair order, and your
marketing manager of the
diagnostics division of the car is taken to the shop. There a technician begins his diagnosis. In

Carnegie Group. His current early versions of the system, the technician will use a computer
research interests include all terminal and a touch screen to communicate with the computer. But
aspects of knowledge-base in one future scenario we are working on, there is no terminal at all.
design, construction,
Instead, the technician dons a headset with a built-in microphone
validation, and maintenance.
He has published several and begins the diagnosis by simply saying “Let's go” to the expert
articles on building expert system. In this vision of the future, SBDS requests information and
systems and integrating them provides instructions by generating synthesized speech and sending
into service organizations. He
it into the technician's headset. This speech is provided in the
has also taught computer
programming at Carnegie- technician's preferred language and at a level of detail appropriate to
Mellon University. his education and depth of experience. The technician responds to
test requests by saying the result (such as “Pass” or “Twelve volts”)
into the microphone.
Whenever possible, SBDS avoids asking the technician to
perform physical tests. Instead, it goes directly to the car for the
information, using a specially designed Portable Vehicle Analyzer
connected to the on-board computer that controls much of the
operation of Ford cars. Raw data such as voltages and vacuum-
pressure readings are processed by the vehicle analyzer and passed
back to the expert system to help guide the diagnosis. If the techni-
cian does not understand why SBDS asks for a test or makes a
certain conclusion, he can respond by asking “Why?” The expert
system responds to this request by describing, in ordinary spoken
language, the reasoning that led to the test request or conclusion. The
 ES

technician can ask “How?” to listen to an explanation of how to ance of a car. The traditional methods for diagnosis, such as listening
perform a repair, or say “Show me” to trigger the display of a to the engine, adjusting engine controls and observing response, are "
schematic or animated illustration on a nearby TV monitor. rapidly losing their usefulness. Many mechanics, overwhelmed by the
SBDS does not rigidly control the session. It simply electronics of the car, are forced into a swap and test strategy, simply
suggests tests and asks questions according to its knowledge of the replacing a suspected component and checking afterwards to see if
problem and what it needs in order to proceed with its diagnostic the problem went away. Although there are a few experts who can
strategy. Yet the technician can override this line of reasoning if he reliably diagnose failures in today's electronics-laden vehicles, they
notices something unusual about the car or feels that the program is are a very scarce and expensive resource.
ignoring a likely cause for the failure. Whenever the technician In trying to overcome this expertise bottleneck, our
volunteers new information, SBDS responds immediately. It interrupts development team at Carnegie used existing techniques for Al-based ‘
its diagnosis, draws whatever conclusions it can from the new diagnosis and modified them to deal with the very large, unstructured
information, and refocuses its strategy to take advantage of what it domain of automobile repair. We soon learned that expert automobile

w
has learned. Thus, the technician can let SBDS control the diagnostic diagnosis is a very complex task. Skilled auto mechanics, we ter
session completely, or he can use it as an expert advisor, depending discovered, rely on several different kinds of knowledge about the
on his skill level and how he wants to use the system. domain and use several completely different reasoning techniques
SBDS is part of a major effort by Ford Motor Company to during the course of a diagnosis. Our task was to design a knowl-
introduce artificial intelligence into the design, manufacture, and edge-base structure that would be rich enough to hold all the kinds of
service of their products. It is the cornerstone of Ford's commitment to knowledge that human experts bring to bear on the problem but
bring Al into the automotive world. With an expected installed base simple enough that it could be built and maintained by non-Al
of over 5,000 dealerships throughout North America, SBDS may experts.
become the largest application of artificial intelligence in the world. All human diagnosticians, whether they work in nd
Ford asked the Carnegie Group to help them develop the automotive repair or medicine, have certain characteristics in e
expert-system component of SBDS, because they recognized that the common. Both groups have an internal mental model of the task
complexity of today’s cars is rapidly exceeding the capability of domain. This model is a body of knowledge about the parts of the
human mechanics to fix them. Today's technicians are generally mechanism or organism they are trying to fix and about how those
unable to tell what is wrong with a car just by listening to it, because parts fit together. This model is closely tied to two additional
af
a maze of electronics controls virtually every aspect of the perform- knowledge sources: the expert's formal understanding of the laws of

Jeff Pepper. (Photo by Patrick Terry)

331
the domain (such as electrical theory) and a large loosely structured This knowledge-base design is sufficient to describe the
body of knowledge consisting of common sense and experience common ways in which cars fail. But just like human experts, it
gained simply by living in the world. Taken together, these three needs a way to modify its behavior when confronted with exceptions
knowledge sources are very powerful and enable human beings to and unusual conditions. To replicate this important aspect of problem
solve new problems by reasoning them through. This process is solving, we added the ability to attach exception rules to the failure
called causal reasoning or reasoning from first principles. schemata. These rules modify the knowledge base to handle special
We discovered an interesting fact about causal reason- conditions. For example, a rule might say that if the car has a very
ing: experts rarely solve problems this way. They only perform causal high odometer reading, then there is a higher likelihood that clogged
reasoning when absolutely necessary, such as when confronted with fuel injectors will cause stalling. This rule doesn’t actually conclude
a new situation. Most of the time experts work from a different anything, it simply modifies the knowledge base so that the reasoning
representation, a much simpler model derived from the original strategy can reach a correct conclusion faster in this case.
causal model but much easier to work with. This second model, In the world of automotive diagnosis, knowledge is
called a troubleshooting or weak causal model, consists of failures constantly changing. A technical-service bulletin may be issued from
and their relationship to each other. The troubleshooting model the manufacturer, or mechanics in the field may discover a new kind
retains most of the diagnostic power of the original causal model but of problem, a new diagnostic, or a new repair procedure. The
does not require such difficult reasoning. traditional way to handle this was to mail technical-service bulletins
The knowledge base of SBDS is built using a modified to dealerships. These bulletins would be posted on bulletin boards
troubleshooting model. It was written using a tool kit for expert- and discussed among the technicians. But with SBDS, knowledge
system development called Knowledge Craft, and it consists of engineers at Ford simply insert the new information into the
thousands of chunks of information called schemata. There is one knowledge base. This makes the new information available not just
schema for each known way that the vehicle can fail. These modes of for reference but to actually improve the reasoning strategy. SBDS
failure are connected to each other by predefined relations such as gets smarter with every new chunk of knowledge.
due to and always leads to, which enables a knowledge engineer to The actual reasoning about repairs is done by a program
represent the fact, for example, that an empty gas tank always leads called the Diagnostic Interpreter. This interpreter is completely
to a no-start. Taken together, these relationships among failures form separate from the knowledge base and knows nothing about cars or
a richly interconnected hierarchy ranging from visible, identifiable any other object. Yet it knows a great deal about how to do machine
customer concerns at the highest level down to specific component diagnosis. In some ways the interpreter is like a detective who is
failures at the lowest level (see figure). The treelike structure of the suddenly asked to fix a broken water heater. He doesn't know
failure-mode network is augmented with large amounts of supporting anything about water heaters, but knows a great deal about
information that assist the expert system in diagnosis and repair. deductive-reasoning techniques. If we give him a manual that
Each failure module has links to the failures that it can cause, failures describes the water heater, he can apply his reasoning skills to this
that are possible causes for it, and many other kinds of schemata. new domain. In much the same way, an expert system makes a strict
These include test procedures that can confirm or reject the separation between “pure” knowledge in the knowledge base and
hypothesis that the failure actually occurred, repair procedures that the procedural information in the diagnostic interpreter. This makes it
can fix it if it does occur, documentation that describes how to fix it, much easier to create and update the knowledge base and makes it
and guidelines on how to proceed if the repair fails.

332
a

Atypical knowledge base for Ford's

Service Bay Diagnostic System Failure #1
Rule #1
showing failures, the relations among
NO START Ifcar doesn't crank
them, and two rules that can modify remove NO FUEL
the knowledge base. Arrows indicate from possible causes
a “due to” relation between two for NO START
failures. (Supplied by Jeff Pepper)

Failure #2 Failure #3 Failure #4

Rule #2
NO SPARK AT If car has ot NO VOLTAGE
NO FUEL rm
IGNITION 50,000 miles, move TO STARTER
CLOGGE D FUEL FILTER
ahead of EMPTY FUEL
TANK as. a cause for Ze ~~
NO FUEL

Failure #5 Failure #6 Failure #8

EMPTY FUEL CLOGGED DEAD BAD

TANK FUEL FILTER BATTERY CABLE

Atypical display from the expert

portion of Ford's Service Bay Diagnostic
System. (Courtesy of the Carnegie Group)

Ford Service Bay Diagnostic System

9. How often does the vehicle stall?
ALWAYS
10. Does the vehicle restart immediately (within 15 seconds) after the stall?
YES
11. Is the vehicle in neutral (park) or in gear?
NEUTRAL OR PARK
12. What is the engine temperature when the vehicle stalls?

COLD, DURING WARM UP

OPERATING TEMP ALL TEMPS

UNKNOWN

HELP WHY HOW MENU BACKUP} | OASIS PRINT QUIT

333
possible to reuse the SBDS diagnostic interpreter for different makes the program performs as expected, that it can accurately and
and models of cars by simply replacing appropriate portions of the efficiently guide a human user to the correct diagnosis of a vehicle
knowledge base. fault. But in looking back on what we've done, two questions remain:
The diagnostic interpreter doesn't know anything except Did we accurately replicate the expert's mental map when we built
how to reason, and it can apply its different reasoning strategies to the SBDS knowledge base? And is the program intelligent? The
the contents of any knowledge base. It knows how to reason by answer to the first question is no, and is likely to remain no for expert
process of elimination, how to modify its behavior to make best use of systems in the near future. We know that a human expert's under-
unexpected or volunteered information, and so on. It also knows how standing is rooted in real-world experience. A human understands
to use all the various tricks of the trade contained in the knowledge that an empty gas tank prevents a car from starting, just like our
base, such as recognizing a failure from a certain combination of expert system does. But his understanding is based on observations
symptoms. And it knows how to perform backward chaining, or goal- that he makes in terms of his commonsense understanding of how the
driven reasoning, the inference method that forms the backbone of world works. Our expert system has no substratum of experience in
this and many other expert systems. which to root its knowledge. It doesn’t know why an empty gas tank
After SBDS reaches a conclusion, another difficult task prevents a car from starting, it doesn’t really know what gas is, or
remains: it must make sure that the conclusion is correct and that the even what a car is. When confronted with situations outside its
recommended repair actually fixes the car. This validation process or limited knowledge base, like a car with two gas tanks, it can't be of
repair strategy is much more difficult than it might seem. Many much help. The best it can do is to recognize the situation as being
factors complicate the picture: A single customer concern may have outside its domain and gracefully admit that it doesn't know what to
several co-occurring causes, so finding one cause doesn't solve the do.
customer's problem. The diagnosis may lead to the conclusion that a As to the second question, Raymond Kurzweil has devoted
part is bad, but that part may not have anything to do with the much of this book to pointing out that the question Is it intelligent? is
customer's original problem. The technician might be unable or meaningless unless we first clearly define what we mean by
unwilling to perform the suggested repair, and another must be intelligence. If we are concerned merely with the ability to perform
substituted. There may be a likelihood for certain repairs to cause symbolic reasoning, then of course SBDS is intelligent, because it is
other things to break, and these other, secondary problems need to be extremely adept at manipulating its own knowledge structures. But at
investigated if the repair fails to fix the car. These are problems that a deeper level, the answers become more elusive.
human experts intuitively know how to handle. But they must be Probably the best contribution | can make to this
studied, formalized, and converted into computer software in order for discussion is to digress a little and take you to the Carnegie Museum
SBDS to solve them. As we built SBDS, we found that the task of of Natural History. We go inside the museum, and we see an exhibit
verifying a repair is just as difficult as the original diagnosis, and of the evolution of the horse. Over there is a row of horse skeletons
much less understood. The repair-strategy component of SBDS is ranging from the small doglike creatures of the ice ages to the
perhaps the first large-scale attempt to intelligently verify the modern thoroughbred racehorse. The small creature there on the far
correctness of computer-assisted diagnosis and repair. left doesn’t look much like a horse at all. It is Eohippus, the “dawn
As of this writing, SBDS is undergoing testing in Ford horse.” It was evolution’s first experiment with this type of animal,
garages, and results have been very encouraging. We know now that and it was successful enough to survive and provide the seeds of

334
evolution that have led to the magnificent creatures we know as
horses today.
Perhaps SBDS and other similar systems are the dawn
horses of intelligent machines. Although they are primitive in
comparison to their human counterparts, they contain the seeds of
intelligence. They are useful enough to ensure that they will spawn
future generations, and their descendants will improve and deepen
their methods of reasoning, perception, and understanding. Natural
selection and strong evolutionary pressures will encourage this
growth, one experiment giving way to another, until they evolve into
the racehorses of Al, a future generation of truly intelligent machines.
The following describes the machine-user interaction
during two diagnostic sessions for a car that does not start.

Session 1 Session 2

SBDS: Does the car crank? SBDS: Does the car crank?
User: Yes. User: No.
Result: Rule1 does not fire. Result: Rule 1 fires, removing no-Fuet from possible causes.
SBDS: \s there spark at the ignition coil? SBDS: Is there spark at the coil?
User: Yes. User: Yes.
Result: No SPARK AT IGNITION is ruled out. Result: NO SPARK AT IGNITION is ruled out.
SBDS: \s the fuel pressure OK? SBDS: Is there voltage at starter?
User: No. User: No.
Result: no Fue.is confirmed. Result: NO VOLTAGE AT STARTER is confirmed.
SBDS: Does the car have over 50,000 miles? SBDS: Is the battery OK?
User: Yes. User: Yes.
Result: Rule 2 fires, moving cLocceo Fuel Fitter ahead of EMPTY FUEL TANK. Result: DEAD BATTERY is ruled out.
SBDS: \s the fuel filter clogged? SBDS: Is there a short in the battery cable?
User: Yes. User: Yes.
Result: cuocceo ruet river is confirmed. Result: BAD CABLE is confirmed.
SBDS: Diagnosis is complete. SBDS: Diagnosis is complete.

335
 Kazuhiro Fuchi. (Courtesy of the Insti-
tute for New Generation Computer
Technology, Tokyo)

The Fifth-Generation Computer System (FGCS) Project was inaugu-

rated in 1983. | would like to report on the significance of the project
today. The word “significance” is quite hard to define. It will probably
be given different definitions by different people. To put it most
simply, if | say the project will be useful or profitable, some people
may think it is very significant. Since, however, profit is a very tough
subject for me to deal with, | would like to see the significance of the
project from a different perspective.
The Significance of Fifth-Generation
Computer Systems Computer History Will Change
el
a

Since 1982 Kazuhiro Fuchi has The greatest significance of the FGCS Project, as I see it, is that
been director of the Research computer history will be changed by the successful completion of
Center of the Institute for this project. Will this change be for the good? | feel that the computer
New Generation Computer
must evolve to the next generation if it is really to take root in society.
Technology (ICOT, also known
as the Fifth-Generation To make such evolution possible is the aim of the project. | think that
Computer System Project). He if the aim is achieved, that will be the significance of the project.
is editor of the journal New More simply, the project, as | see it, is aimed primarily at changing
Generation Computing and
the basic design principle that has given us existing computers. |
author of numerous articles on
cutting-edge computer shall expand on this later. Bringing computer technology to a new
technology in Japan. stage by changing its basic design philosophy, then, is the aim of the
project. Has a new computer-design principle been established yet?
The answer is no. | think that changing the basic design means
establishing new concepts and translating them into practical basic
technologies. Establishing basic technologies is the goal of this ten-
year project.
As we work on the project, we have visitors from various
fields, such as journalists and researchers. One visitor wanted to
have a look at a fifth-generation computer. When | told him that we
did not have one as yet, he joked, “You've hidden it somewhere,
haven't you?” Another serious visitor said this, to my embarrassment:
“We have a plan at our company to introduce a new computer system
three years or so in the future. Your fifth-generation computer seems
very good, and we'd like to install it.”

336
Ten Years Are Needed to Establish Basic Technologies the very intensive discussions we had with various people and on
er reas Se Sa SS SE =
insights gained from past trends.
The ten-year time frame of the project is rather long, but | feel it takes Establishing new technology is the primary aim of the
ten years or so to establish basic technologies. If the project project. Japan is not very experienced in developing new technology
succeeds, it will still take several more years to build commercial where there is an idea or goal but no example to go by. First anew
products on the basis of it. So its realization will take a dozen years methodology for implementation of the project must be developed. If
from inception, perhaps too distant a future for some people. But | we could develop new technology by just following traditional
believe it is still very worthwhile to pursue the project. methods, that would be best. But that is not possible, | think. As is
When | am asked if there is any blueprint for the fifth- often pointed out, the Japanese traditionally prefer stability to
generation computer system kept in some vault, | say no, at least not innovation. But just developing old themes will not create new
for now. | should rather say that drawing up such a basic blueprint is technology. That is why we need a new method for carrying out the
the goal of the project. The most difficult part of the project is to make project. To sum up, | think that making an effort to demonstrate where
the idea of the project understood. Once a product is physically Japan is going with new technology is the nature and significance of
available, we can readily make it understood. But we don't have that the project.
yet. There is now no example anywhere in the world that we can cite
to show what our projected computer system will be like. This is a The Basis Is a Logic Machine
aD TS ERS Se +
characteristic of the project. If what we aim at could be explained in
terms of increasing the speed of some known process by ten times or As |see it, there is emerging a situation that calls for Japan to strive
reducing its cost by a certain percentage, it could be understood very to make great contributions at what Professor Moto-oka aptly called
easily. As it is, there is no such process. a precompetitive stage. The FGCS Project is a project that responds to
But that does not mean we have stepped into an entirely just that situation.
nebulous field. And | think that this too is a characteristic of the Though the primary goal of the project was just now
project. Planning for the project was preceded by three years of explained in terms of changing the design philosophy of computers, it
research and study and to tell the truth, it incorporated discussions can also be described as developing an easy-to-use computer or a
held in various places even further back. For instance, at the computer with intelligence. When | talked about this with one
Electrotechnical Laboratory, where | was before | came to ICOT, we gentleman, he suggested that the goal might be to seek a new
had had discussions for five or six years on what the next age would paradigm. The term “paradigm” is normally used to indicate an
be like. One of the motives behind the project is to integrate such example, tangible or intangible, that provides the basis for evolution
various discussions from various places. In the planning process we of a cultural or scientific theory. Let me explain this is my own
also discussed where we should go on the basis of diverse leading- parlance. The computer, as |see it, is a logic machine.
edge research under way throughout the world. | myself made efforts From that standpoint, the basis of the FGCS Project can be
in that direction, and so did other researchers. So the project is built traced to logic. But what logic or system of logic is the present
on research conducted in the past and not on just a collection of computer based on? At the beginning there was Turing-machine
casual ideas. However, when it comes to shaping an image out of theory. It is not entirely wrong to say that the computer has evolved
various research, you can do it by merely gathering and processing on the basis of that theory. While the present computer is not a Turing
the data statistically. | may say the project represents a refinement of

337
 R&D objectives
Antificial-intelligence technology
- Natural-language processing (for interence/ knowledge bases)

Support for decision making Architecture technology (Non-von Neumann,

and intelligent activities highly parallel processing technology;
hardware implementation of inference/
Improvement of programming knowledge-retrieval function)
productivity
Software engineering technology
(programming technology)

machine per se, its basis can be traced back to the Turing theory. So Logic programming is closely related to various fields.
it is still operating in the Turing paradigm. Is there, then, a different Take the field of data bases, for example. In the world of data bases,
paradigm from the Turing paradigm? There is. The Turing theory was relational data bases are now accepted. The basic concept of
published in 1936, which happens to be the year in which | was born. relational data bases is relations, and the relations form a concept
According to the texts, those days were the golden age of logic. that is based on predicate logic. Data bases are becoming a very
Various logic systems were devised to pursue computability, etc., large proportion of computer systems, but the relational-data-base
which resulted in the establishment of the concept of computability. model is not consistent with the programming world at present.
From this came the Turing theory. Programs are based not on relations but on procedures.
From the standpoint of logic, however, the Turing theory Relations and procedures are fundamentally different from each
was a very special one. The mainstream of logic is a system of logic other. Data-base languages are theoretically more advanced, while
called predicate logic. There were a number of great men involved in the programming languages in current use are based on old concepts.
establishing predicate logic. This began with Frege in the nineteenth | feel that the present situation is that we have no alternative but to
century, who was followed by Godel in the 1930s. Von Neumann also connect the two in unnatural ways. Data-base and programming
had his name recorded in the history of predicate logic. With a long languages are similar in that logic programming may be regarded as
history dating way back to Aristotle, predicate logic is, in a sense, an intended to bring programming up to the level of relational data
ordinary, more natural logic. So when it comes to logic machines, bases, but not the other way around. | thus see the possibility that use
there might have been predicate-logic machines rather than of logic in programming will allow it to connect very beautifully with
machines modeled after the Turing machine. But history did not turn the world of relational data bases, though not the reverse.
out that way. In the field of software engineering, research was
The computer has followed the route as it has. Some conducted on a variety of subjects such as new styles of programs,
people say that this may have been a gigantic historical detour. | program verification, and synthesis. From these also came logic
partly agree with them. | think the time is approaching to return to programming. When considering program verification or synthesis or
predicate logic as a paradigm. In the past ten years or so there have a very efficient debugging system, which | think will be a future
been moves to restore predicate logic in the field of programming too. challenge, we have great difficulty in theoretically dealing with the
Called logic programming, the movement is aimed at developing basic computer model of today, which is based on the coro statement
programming languages based on systems of logic like predicate and assignment. The concepts of assignment and the coro statement
logic and using them for programming. are basic and easy to understand but are not suitable for proving
Looking back on the planning for the FGCS Project, | may properties, because they are oriented toward functions. By contrast,
say the project is based on the concept of logic programming. This machines based on predicate logic are suited for such purposes,
could be interpreted as redesigning software and applications within since proving is built into them. This is not just a theoretical argument
that concept, or it could be viewed as building machines with a new but is somewhat in line with the trend some ten years ago toward
type of architecture to support the concept of logic programming. avoiding the use of coro statements wherever possible so as to write
Though logic was our starting point, | would say that logic was not neat programs. That fact suggests that programming based on
taken into consideration from the outset. Rather, it came as a predicate logic is better suited to our purposes than the ordinary
conclusion after the analysis of various research projects, as | languages we are currently using.
mentioned a short while ago.

338
R&D objectives for fifth-generation
computers to meet the needs of the
coming information society. (Courtesy
of the Institute for New Generation
Computer Technology, Tokyo)

Fifth-generation computer Von Neumann type

Parallel
processing, Sequential processing,
Non-von Neumann associative
Fourth generation (VLSIs) addressable memory,
type memory. 3th generation (LSIs) numerical calculation
inference/ Third generation (ICs)
Knowledge- | knowledge Abacus
Second generation (transistors)
processing | processing First generation (vacuum tubes)
schemes
Calculators

Electronic
methods

Mechanical
Stored-program methods
schemes
Program-
commaled Mechanization of Four
Twentieth century n:
arithmetic operations basic

functions
Eighteenth century

The positioning of fifth-generation

computers in the hierarchy of
computer technology. (Courtesy of
the Institute for New Generation
Computer Technology, Tokyo)

Hardware/architecture Software-engineering Antificial-intelligence technology

technology technology
———— —————————— 7 SS eS u oe 7

| + Parallel inference + Problem-solving/inference software

| machine Programming - Knowledge-base-management software] |
languages
| - Knowledge base god + Intelligent interface !
I machine I
i + Intelligent programming f
1 ee ee SS SS ee a
Applications software
The research and development of
fifth-generation computers.
(Courtesy of the Institute for New
Generation Computer Technology,
Tokyo)

339
 Close Relations with Artificial Intelligence
ES

The FGCS Project is also closely related to the field of artificial

intelligence. Knowledge-engineering systems, or expert systems,
which have been drawing considerable attention recently, are based
on the idea of making a consultation system by representing
knowledge in the form of rules and processing the rules. Roughly,
organizing knowledge into rules is a return to predicate logic. If we
go into detail, we shall find diverse arguments on this subject. |
might add that Al experts tend to be concerned with even minuscule
differences. For knowledge representation there are numerous
proposals that are substantially the same with minor differences.
In connection with artificial intelligence, there is the
question of natural-language processing. Grammar, for instance, is
very closely related to predicate logic. So is semantics. This may be
only natural, because logic itself came from a desire to formulate
part of the mechanism of natural language. Historically, logic pro-
gramming originated from proposals by researchers of natural
The aims of the fifth-generation
language.
computer. (Courtesy of the Institute
for New Generation Computer Historical background aside, if we go with the predicate-
Technology, Tokyo) logic paradigm instead of the Turing paradigm, we can expect that
all the problems we now face will be put in order as far as software
is concerned.

Linkage with Software Concepts

SA Oe OTE EEE EE DEE

The resurgence of predicate logic is not limited to the area of

software. Diverse research in computer architectures has yielded
new interesting architectures. They are interesting not just in
hardware configuration but in that they are intended to connect with
software concepts. They include frequently cited data-flow
machines and their variations, reduction machines, and more
| Conventional Fifth-generation computer recently proposed advanced parallel machines. These represent a
computer arithmetic { t logic t new trend, because they are aimed at linking architecture to
A-—B.B-C
A=C software engineering concepts.
Conversely, previous bottom-up concepts alone, such as
ies >> s.,
Data processing’ <_|_ Knowl
Knowledge parallelism and associative memory, are not sufficient. They must be
rs processing
Numeric tos connected with high-level software. Recent notable moves take the
calculation” 7/ standpoint that bringing in such software concepts may make it
JApplications| possible to organize parallelism well.
In this context, functional programming is often dis-
cussed. Predicate-logic programming, again in macro terms, is an
expanded concept that encompasses functional programming. From
the viewpoint of architecture too, logic programming may provide a
better base for parallel processing, because it contains more
parallelism than functional formulation. This thinking underlies the
FGCS Project.

340
Productivity improvements in Decision-support Corporate decision making
low-productivity fields, systems
business,
public services Formation of knowledge industries
administration, Development of
distribution sector office automation improvement of medical weltare
Various expert systems Formation of health
insurance industry
Development of Voice Solving social problems
database industry typewriters Qualitative Medical consultation associated with an aging
information Medical expert systems society
Automatic creation Analog information output
of databases
Strengthened ability in
Speech recognition engineering industries
Rosolution of Image recognition
international friction Machine
translation Conservation of energy
Accumulation of Image processing and resources, greater
know how Natural-language manufacturing productivity
Improvement of intellectual processing
productivity in R&D and other areas Intelligent Diversification and
Be pecan Intelligent programming customization of computers
computer inleriace pues
program
Improvedie productivity of
Inference
the computer industry
synthesis
Computers for
individual instruction Great leaps in software
Knowledge productivity
Japanese language bases
machines Ultra-high-speed
devices (Josephson
Home computers, junctions, gallium Optical fibers
hobby computers arsenide |Cs)
SS — VLSIs wis

The social impacts of fifth-

generation computers. (Courtesy of
the Institute for New Generation
Computer Technology, Tokyo)

The approach for developing a fifth- Knowledge/

generation computer. (Courtesy of information
processing
the Institute for New Generation
Computer Technology, Tokyo) Ee Xs

Logic programming Knowledge bases

= a
Highly parallel! computer
architecture

i
n computer systems

The basic configuration of fifth-

generation computer systems. (Courtesy
of the Institute for New Generation
Computer Technology, Tokyo)

Expert system for Intelligent Machine-transiation Demonstration

speech and office-automation system system for basic
image system software
understanding

t i
Knowledge-programming Language that processes natural
language language, speech, and pictures

Software module for

knowledge-base Software Basic software
Software management f
enues system
module for for
intelligent intelligent
preoreninn Software module for Toe
problem solving and inference

@ System-description language
( Fifth-generation kernel language (logic-programming language)

Hardware system

Intelligent
Parallel interface
Knowledge inference pees
operation mechanism
Relational database mechanism
mechanism
Network system

VLSI architecture

342
 Intelligent interface software
Problem-salving and
inference software
Inference mechanism

Des
Intelligent programming software

Intelligent interface software

Knowiedge-base management
software
nowledge-base mechanis

rs
Intelligent programming software
| eraser eases Kn al
| [Befinate modites forknowiedge-base managemer’
| [Software modules for intelligent interface !
|
| [Batis edies for iristigent rogramming I

- —
|
!
|
!
=
Our plan to consider such parallel machines is supported
Stages of fifth-generation computer by the progress of VLSI technology. Let us look again at the history.
research and development. (Courtesy
What was the primary reason for going with the Turing paradigm in
of the Institute for New Generation
Computer Technology, Tokyo) the first place? In the 1940s memory elements were very expensive.
So it was necessary to use as simple a hardware configuration and as
few vacuum tubes as possible. A great idea born in those circum-
stances is what is now called the von Neumann computer concept.
But computer history has reached a point where the conventional
concept of making software do everything on simple hardware
presents various problems. The so-called software crisis is taking
place in some fields. For this reason | feel we may see computer
history moving out of a phase characterized by strict adherence to a
basic philosophy born in the days of costly hardware and into a new
phase.

FGCS as a Return to Predicate Logic

aS >

For these reasons | see the basic thrust of FGCS as a return to

predicate logic. This is not an innovation. Rather it is a return to
something old. To people adverse to the word innovation, | explain it
as restoration. To people fond of innovation, on the other hand, |
explain it as a venture into a new field. | am not lying to either: we are
in a transition phase of the cycle of history. What | mean is that
history has not yet reached a turning point but will in the 1990s. It will
 An overview of research and devel-
opmentin the initial stage. (Courtesy
Demonstration system for basic software of the Institute for New Generation
Computer Technology, Tokyo)

Experimental system for

knowledge utilization

Software for a]
Software module for intelligent interface Software module for intelligent programming sequential
Pilot model of a semantic-analysis sytem
ry Program for program-verification management inference
Pilot model of a dictionary system machine
(SIM)
High-level parsing Modular programming
program system

Software module for problem solving Software module for knowledge-base

and inference management
Basic software for problem solving Knowledge representation system
Basic software
for parallel
Program for large-scale relational
inference database management
kak na

Sequential
Kernel inference
language 1 Kernel language 0 machine
(pilot models
for software
development)
Basic software system

SS
Inference subsystem Knowledge-base subsystem
Modules for individual functional Modules for individual functional
mechanisms mechanisms
Basic mechanism for + Base mechanism for basic knowledge
parallel inference + Operation mechanism for parallel
Data-tlow mechanism relations and knowledge Hardware tor
+ Abstract-data mechanism + Relational database mechanism
a sequential
inference
macine
niques for integration in VLSI Techniques for integration in VLSI (SIM)

344
be a little more than a half century since the computer as we have We have made a fairly good start on the project. This
known it was born. The time span may also justify this scenario. project is very ambitious in that it is aimed at ushering in a new
So far | have explained the FGCS Project from my own computer age rather than at developing products in the near future.
point of view. Behind this view is, of course, lots of discussions. | What we have to do to make that happen and what we sought to do is
have boiled down the results of the discussions | have had with to employ the framework of logic programming with the aim of
various experts over the years and recapitulated them as | understand building a new hardware architecture and new-type software and
them. The ICOT Research Center started activities with researchers applications within that framework and thereby establish a basic
sent on loan from numerous organizations. Though they were like a body of computer technology for a new age.
scratch team, these people were very quick to blend themselves Some people say that our commitment to logic program-
harmoniously into the center. Very quickly a sense of togetherness ming is simple-minded and may place us under restraint. But
prevailed in the center, and all the people have done better than evidence indicates that things may go in the direction of logic
expected in their respective research fields. This is attributable partly programming. Moreover, we have no intention of excluding anything
to the enthusiasm of the researchers and those supporting them and else, though we think that the various good ideas suggested so far
partly to the goal of the project, which, if somewhat vague, is will fit naturally within that framework. We selected logic program-
ambitious enough to stimulate the young researchers. ming as the basic idea in the expectation that it would increase
Overseas reactions to the project are now very different freedom in hardware and software design rather than limit it.
from what they were when we started discussions on the project. At The project requires more than group activities on a
first there was some criticism that our project had little chance of limited scale. The wisdom of Japan must be combined, and in a
success. But there have since been increasing numbers of people broader perspective, the wisdom of computer scientists all over the
overseas who support the project's objective of ushering in a new world must be marshalled to usher in a new age for mankind. So far
age of computers. | think that thi: is evidenced:by the start of new we have received much more support and cooperation than is usual
programs in a number of countries. Behind these startups is the with other projects. But we have a long, long way to go, and there
support of researchers in the respective countries. will be numerous difficulties on the way.
Allow me to sum up. The FGCS Project is considerably
different in nature from a number of other projects under way. It is
aimed at something entirely new. For this reason | think we need to
make greater efforts than in the past. ICOT has to play a central role in
these efforts. But that alone will not be enough. We must encourage
research activities throughout Japan and all over the world. The bud
is present, and now we need to make it grow. By doing so, we can
usher in a really new computer age. But global rather than merely
local efforts are needed to make this happen, and after all, the new
age will benefit all mankind. As we exert efforts toward that end, we
look for cooperation and support from all those concerned every-
where.

345
 For reasons that will be made clear, the term “artificial intelligence”
is not much used in the U.K. “IKBS” (intelligent knowledge-based
systems) tends to be the term employed to cover work in expert
systems, natural language, and logic languages. There is a long
history of Al work in the U.K., dating back to Alan Turing’s pioneering
work just before and after World War Il. That strange genius foresaw
the use of stored-program machines for much more than mathemati-
cal calculation, and in the early 1950s he enabled us to compare
human and machine intelligence by postulating a test by which one
could determine whether true artificial simulation of human
intelligence had been achieved. He proposed that an observer in one
room should try to tell if he is conversing with a man or a machine in
another room. And he also predicted that this test for Al would be
achieved by the year 2000. Well, there are still some years to go, and
it remains an open question whether his prediction will be proved
right; most of his other predictions have been amply fulfilled. One

| Sia might add that some people believe Colby’s program PARRY has
achieved Al, because it fooled psychiatrists into believing they were
dealing with the output from paranoid human beings. If some people
Intelligent Knowledge-Based Systems:
are not impressed with this, it may be because they are not impressed
Alin the UK.
with psychiatrists.
Alan Turing worked in various places in the U.K.,
Brian Oakley, a physicist who
including the National Physical Laboratory, Bletchley Park, and the
started his career working on
microwaves and computer University of Manchester. After his time the center of progress in Al
applications in what is now the work shifted to the University of Edinburgh, where Donald Michie
Radar& Signals Research built up a considerable team and formed one of the first centers in the
Establishment in England, is
world for the study of Al. The university has produced excellent work
the former director of the Alvey
Programme, a British national in the language field, though perhaps its most renowned output has
cooperative program in been its students, who have gone on to populate Al centers through-
information technology formed out the world. Donald Michie himself has moved on to found the
in response to the Japanese
Turing Institute at Strathclyde University, which provides an advisory
Fifth-Generation Computer
System Project. He is currently and training service to industry. But the very broadly based team at
chairman of Logica Cambridge. Edinburgh, with groups ranging from linguistics through logic
programming to speech recognition and computer science, still
continues as perhaps the dominant center for Al work in the U.K.
Another much smaller center for Al work in the U.K.
academic world is at Sussex University, where the POPLOG language
environment is being developed under Professor Aaron Sloman.
POPLOG contains compilers for PROLOG and Common LISP as well as
one for POP, a simulation language that has its adherents in the U.K.
community. After Edinburgh, perhaps the main centers for Al work are
at Imperial College, London and Cambridge. Bob Kowalski heads the
team at Imperial, where work is in progress on logic programming,
expert systems, and Declarative Language architectures. One of the
interesting recent applications of Al techniques there has been to the
construction of a rule base for the interpretation of legislation, the

346
 chosen example being the British Nationality Act, perhaps not one of
the most logically constructed pieces of British legislation.
Computer architecture work flourishes at many university
centers. Imperial College and Manchester University are cooperating
to produce a general-purpose parallel architecture, and a develop-
ment stage of this machine, called Alice, has been installed at
Imperial College and also at the computer center at Edinburgh
University, where it is possible to make comparisons with various
other machines. Other work in parallel architectures is in progress at
universities at Cambridge, Reading, Bath, Southampton, East Anglia,
Glasgow, and St. Andrews as well as at University College and Queen
Mary College, London.
Work in natural language is relatively weak in the U.K., in
comparison with the amount of work on logic programming and
advanced architectures. But there is some work being done at
Cambridge and the Open University, and as a part of speech
recognition, work is in progress at various centers, most notably at
Edinburgh.
It is generally believed that the Lighthill Study of Artificial
Intelligence, which was commissioned by the Science Research
Council toward the end of the 1960s, was a major setback for Al work
in the U.K. In this report Sir James Lighthill suggested that one
ime Gakiey (Courtesy of Brian needed a major breakthrough before it would be possible to gain
Oakley) much by tackling the subject of Al as a coherent whole. He recom-
mended that robotic developments should be pursued in their own
right, just as the study of human intelligence should be, but that it
would not be very profitable, for the moment, to expect an under-
standing of the way the mind operated to lead to practical applica-
tions of so-called artificial intelligence. In a sense, time has
supported his contention, for with the exception of low-level vision,
there is little evidence that the development of Al practical applica-
tions have received much benefit from the study of how the human
mind works. Interestingly, it would appear that the study of human
behavior has benefited rather more from developments in computer
science.
It was probably not intended that the Lighthill report
would result in a cutback in work on the component parts of artificial
intelligence in the U.K., but this is generally said to have been the
result. It probably also intensified the feeling in the U.K. that the term
“artificial intelligence” is unfortunate, with its implications that the
products are a replacement or rival to human intelligence. For this
reason, when work on Al was intensified in the early 1980s, the term
“knowledge-based systems” was employed, usually preceded by the
word “intelligent” as a sop to the Al community.
If the Lighthill Report of the early 1970s was paradise lost
for the Al community, the Alvey Report of the early 1980s was

347
paradise regained. The Alvey Report was triggered by the Japanese expert systems in the Large Demonstrator parts of the program, which
announcement of their fifth-generation program at a conference in are designed to pull the work together into some practically
Tokyo in the autumn of 1981. In practice, this program is not very large orientated major projects.Overall, Al work probably constitutes
by Japanese standards. The Japanese Al research center, ICOT, still nearly half of the whole program, appearing in all parts. For example,
does not number more than 60 workers. But the fact that Japan the software-engineering program includes a set of integrated
announced this belief in the importance of Al work to the future of the project-support environments, software-project tool sets. The most
Japanese economy was enough to launch renewed support for Al in advanced of these will be based on the use of Al techniques. But it is
the U.S., Europe, and elsewhere. In the U.K. the growth of interest in only fair to add that the Al and software-engineering communities in
Al, and indeed in information technology generally, actually preceded the U.K. are not really converging as fast as one might like to see.
the Japanese announcement. Much of the Al work in the U.K. has suffered from a lack of
The Science and Engineering Research Council (SERC), suitable experts. So part of the Alvey program is devoted to building
the body in the U.K. that has the responsibility for funding academic up the skilled manpower. When the research-trained postgraduate
research in the physical and biological sciences and in engineering, workers from the Alvey projects soon start becoming available for
had already initiated a specially promoted program for distributed employment in the general community, this will make a very
computing, which led to much of the flowering of work in parallel considerable impact on the available skilled manpower. The program
computing and declarative and logic languages. A study of the has initiated a Journeyman scheme under which good people from
computer-architecture requirements for IKBS was initiated, and this industry with skills in information technology but lacking in Al
in practice formed the background for the strategy for the IKBS or Al expertise go to sit at the feet of masters in the academic centers for
part of the Alvey program. six months or so while working on projects of interest to their
The Aivey Committee, which planned the program, saw it employers.
as a way of marshaling all the research resources of the U.K. to an The Awareness part of the Alvey program has proved to
attack on what were then seen, and are still largely seen today, as be a great success. The Alvey Tapes are a set of videotaped lectures
the enabling or underlying technologies required to support the of distinguished experts on artificial intelligence prepared for the
whole of information technology. The committee's report, accepted by Alvey program by the Open University. The 16 one-hour tapes cover
the government in May 1983, called for a program costing 350 million subjects like logic programming, dealing with uncertainty, image
pounds over 5 years, of which industry would pay for about half, or understanding, machine learners, natural-language processing, and
150 million pounds. The government contribution of 200 million various aspects of expert systems. One of the most popular tapes,
pounds would come from the Department of Trade and Industry, the designed to counter the oft repeated canard that expert systems are
Ministry of Defence, and the SERC. not really used, shows examples of such systems at work in British
The program was seen as one of cooperative research, industry.
and in practice the work is being carried out by some 200 project Another way of helping British industry to get started on
teams drawn from industry, the universities and polytechnics, and the building expert systems has been through the Expert Systems Starter
research institutions, which are largely government establishments. Pack put together by the National Computing Centre. These consist of
Typically, teams from two or three firms and one or two universities computer tapes for four different expert-system design tools, together
take part in each project. Universities are involved in over 85 percent with training guides. These tools provide elementary experience of
of the projects, and there is a special “uncle” project class for work such techniques as backward- and forward-chaining inference
of a long-range or speculative nature in the universities supported by mechanisms and the use of dialogue generators.
some company that takes an avuncular interest in the work. To spread the knowledge of expert systems widely in
Altogether, the program employs some 2,300 research workers and British industry and commerce, a set of Community Clubs has been
has probably doubled the effort in the fields covered. Sixteen clubs established. A Community Club typically consists of 10 to 30 like-
exist to bring together all the workers in a common field; for example, minded firms who join together to study the development of an expert
the Speech Club covers workers in the 10 projects in that field. system in an area of particular interest to them. They work with a
Al work largely falls into the IKBS and man-machine- team from the computer industry or the universities that has had
interface (MMI) parts of the program, the other parts being VLSI and experience in building expert systems, and together they draw up a
software engineering. MMI covers human factors and speech and specification, recruit staff to build the system, construct the system,
pattern recognition. There are also considerable developments in and then test it. Often one or more of the firms acts as the guinea pig
for trying out the system. The Alvey program has sponsored nine such

as
Community Clubs in fields like financial services (Small Company States in terms of quality if not in terms of quantity. The recognition
Health Adviser), insurance, data processing (Help Disk Adviser for that yet to come are the true breakthroughs, both in understanding
data-processing installations), econometric modeling, planning for and in application, spurs on the Al community in Europe. And the
real-time manufacturing projects, quantity surveying, real-time cooperative programs of the last few years have served to build up
quality control of processing plants, transport-route planning, and significant, multidisciplinary teams that bid to make a real impact on
water-industry construction planning. Altogether, some 200 organiza- the world Al scene. The Europeans who have led and inspired many
tions are participating in these clubs and learning how to build and of the Al developments of the last forty years will be followed up by a
use expert systems for a purely nominal fee. They have greatly helped growing army of experts. These experts recognize that Al is going to
to spread awareness of the power and limitations of expert systems make a great impact on the world in the future but that the break-
in the U.K. The initiative is being copied by others, so that knowledge throughs that will be needed will demand considerable teams drawn
about expert systems has grown very fast in the U.K. It is not from a very wide range of disciplines.
accidental that the expert-systems group of the British Computer
Society is perhaps the most flourishing such society in the U.K. today.
It is probably true to say that there is more true awareness of the
implications of expert systems in the U.K. than elsewhere, even if the
magnitude of some of the applications in the U.S. is far larger.
No picture of the Al scene in the U.K. today would be
complete without mention of the part played by the EEC’s ESPRIT
program. ESPRIT is a program very like the Alvey program except that
the cooperative teams are drawn from throughout the European
Community of 12 nations. In scale it is, for the U.K., about half the size
of the Alvey program, which means that in total for Europe it consists
of about twice the number of workers. Al is about a fifth of the
program, being largely represented in the Advanced Information
Processing part of the program. Of course, ESPRIT has the crucial role
of bringing the research workers of Europe together. Perhaps for the
first time the Al community in the U.K. is coming into active contact
with the other research workers in Europe not just to carry out project
work but also to plan, evaluate, and implement the program. And in
several countries in Europe there are also nationally sponsored
programs, so the total Al effort in Europe is far from negligible and its
historic problem of fragmentation is being overcome.
Looking to the future, a committee in the UK is now
planning an After Alvey program, and the second phase of ESPRIT is
also under discussion. Though one cannot be certain at this stage just
what will emerge, it seems very probable that some planned and
coordinated programs will continue, perhaps with a rather larger bias
toward actual applications than the precompetitive research
programs of the first wave. What is certain is that the buildup of
interest and work in artificial intelligence in Europe will continue.
The overall scene in the U.K. and Europe can be characterized by
saying that there is now immense awareness of the potential to be
unlocked by applying the techniques developed under the banner of
Al work. Though much of the work is at a rather elementary stage,
characterized by a shortage of expert workers, there are nevertheless
some true centers of expertise to rank with the best in the United

349
 The Science of Art

The great discovery of the twentieth century in art and physics alike, is a recoil from and transformation of
the impersonal assembly-line of nineteenth century art and science.
Marshall McLuhan, The Gutenberg Galaxy

+ The Musical Arts

Ata time like ours, in which mechanical skill has attained unsuspected perfection, the most famous
works may be heard as easily as one may drink a glass of beer, and it only costs ten centimes, like the
automatic weighing machines. Should we not fear this domestication of sound, this magic that anyone can
bring from a disk at his will? Will it not bring to waste the mysterious force of an art wiuch one might have
thought indestructible?
Claude Debussy, La Revue S..M. (1913)

Collaboration with machines! What is the difference between manipulation of the machine and collabora-
tion with it? | have sometimes experienced a state of dynamic tension rising in me out of what would
seem to be a state of mutual responsiveness between the machine and myself. Such a state could require
hours of concentrated preparatory exploration, coaxing of machines, connecting, so to say, one’s own
sensibilities, one’s own nerve endings to the totality of the tuned-up controls. And, suddenly, a window
would open into a vast field of possibilities; the time limits would vanish, and the machines would seem to
become humanized components of the interactive network now consisting of oneself and the machine,
still obedient but full of suggestions to the master controls of the imagination. Everything seemed possible:
one leaned on the horizon and pushed it away and forward until utter exhaustion would set in and, one by
one, the nerve endings ceased to connect, the possibilities contracted, and an automatic reversal to
routine solutions was a sure danger signal to quit. An affectionate pat on a control here and there was not
to be resisted. Switches and lights off! If there is an unfinished bit of conversation between you and the
machines, either take note of all the controls or leave them alone until tomorrow. Recapturing the exact
circumstances of such periods as just described is not easy. Tomorrow it may seem all cold steel, copper
and colored plastic. The coaxing may have to start all over again.
Vladimir Ussachevsky (composer and early pioneer of electronic music), Electronic Tape Music

Computer technology is now having a major impact on all fields, including the creative
arts. While the visual arts are just beginning to feel the impact of advances in co
puter graphics and imaging technologies, the computer revolution in music is already
well under way.' One reason that the transformation of music is further along has to
do with the substantially greater “bandwidth” (communication and memory capacity)
required for images as compared to sounds. With digital technology we can already
store, analyze, modify, and recreate sounds with an accuracy equal to that of the
human auditory system.” To do the same with visual images requires an amount of
memory and processing power that is still quite challenging for today’s computers.
Another reason has to do with the fact that music theory is more highly developed
and quantitative than theory in the visual arts.
As discussed earlier, a historic transformation is taking place in the musical-
instrument industry away from the acoustic technology of the piano and violin and
toward the computer-based technology of the synthesizer.* The advent and now
enthusiastic acceptance of digital instruments follows a long tradition. Music has
always used the most advanced technologies available: the cabinet-making crafts of
the eighteenth century, the metalworking industries of the nineteenth century, and
the analog electronics of the 1960s. This latest wave—the digital electronics and
artificial intelligence of the 1980s—is again making historic changes in the way music
is created.
Up until recently, instrument-playing technique was inextricably linked to the
sounds created. If you wanted flute sounds, you had to use flute-playing technique.
The playing techniques derived from the physical requirements of creating the
sounds. Now that link has been broken. If you like flute-playing technique (or just
happened to have learned it), you can now use an electronic wind controller that plays
very much like an acoustic flute yet creates the sounds not only of the flute but also
of virtually any other instrument, acoustic or electronic. In fact, there are now control-
lers that emulate the playing technique of most popular acoustic instruments,
including piano, violin, guitar, drums, and a variety of wind instruments. Since we are
no longer linked to the physics of creating sounds acoustically, a new generation of
controllers is emerging that bears no resemblance to any conventional acoustic
instrument but instead attempts to optimize the human factors of creating music with
our fingers, arms, feet, mouth, and head.
Music controllers and sound generators (or synthesizers) can be linked
together by an industry-standard electronic interface called MIDI (Musical Instrument
Digital Interface).* MIDI, which employs an inexpensive communication link, has
allowed independently developed synthesizers, controllers, sequencers (computer-
based devices that can remember and replay sequences of notes), and other sound-
modification devices to communicate and interact with each other. MIDI has even
been used to control stage lighting and other visual effects. Another industry protocol
called SMPTE Time Code (Society of Motion Picture and Television Engineers) allows
interfacing sound-manipulating devices to video. MID] and SMPTE Time Code
together are facilitating the development of computer-based systems that are
revolutionizing video and audio production in the same way that electronic publishing
software, personal computers, and laser printers have transformed the way printed
documents are created.
In addition to new music controllers, new sounds to be controlled are being
created at an accelerating pace. While the sounds of the piano and other acoustic
instruments continue to be sounds of choice, they are routinely mixed with new
sounds that approach the richness and complexity of acoustic sounds but have never
been heard before.® The creation of a musical work such as a pop song used to

352
 Breath
eg i)

Ahome music studio that involve the creation of a few fixed elements: melody, rhythm, harmony, lyrics, and or-
uses Musical Instrument Digital
chestration, say. Now an entirely new element, the invention of new timbres, has
Interfaces (MIDIs)
been added. This has led to the appearance of a new type of musician, one who
specializes in sound design. Rare these days is the successful pop song that does not
introduce some new timbre to the music world’s palette of sounds.
Sound modification techniques are being developed as rapidly as new
sounds and control methods. It is now possible with personal computers and appro-
priate software to take a musical note, break it into all of its frequency components,
reshape the amplitude (or loudness) envelope of each such component, and then
reassemble the sound.® Like most experiments, many such attempts fail to produce
anything worthwhile, but the experiments that do succeed are creating an ever
expanding repertoire of new, musically relevant timbres. A wide variety of sound-
modification techniques can be used off-line when editing a multi-instrumental
composition and controlled in real-time during performance
Digital technology has overcome many of the conventional limitations of
acoustic instruments. All sounds can now be played polyphonically and be layered
(i.e., played simultaneously) or sequenced with one another.’ Also, it is no longer
necessary to play music in real time.® Traditionally, music performance has often
depended on nearly superhuman feats of finger acrobatics. While technical playing
skills are still being used to good effect by the virtuosos, absence of such skills no
longer represents a barrier to the creation of music. Music can be performed at one
speed and played back at another, without changing the pitch or other characteristics
of the notes. Further, it is possible to edit a recorded sequence by inserting, changing
and deleting notes in much the same way that one edits a text document with a word
processor. In this way, music can be created that would be impossible to perform in
real-time.?

The computer as composer

Expert systems and other Al techniques are also providing new methods for

353
 composing music. A musician can provide his own original ideas for a composition
and allow the computer to do the rest of the work using systems that are pro-
grammed with extensive knowledge about the music composition process.'° One
such system that can compose music automatically, the Cybernetic Composer, was
developed by Charles Ames and Michael Domino. The Cybernetic Composer com-
poses entire pieces automatically, while similar systems allow the musician to
contribute his own musical ideas. For example, one can write the melody and rhythm
and allow the expert-system software to generate the harmonic progression, the
walking bass line, the drum accompaniment, or any combination of these parts. Such
systems will be of growing value in teaching music composition and theory, as well as
in allowing composers to concentrate on those aspects of the composition process
where they can add the most creativity.'' Also increasing in popularity are systems
that allow the musician to interactively change parameters of a piece in real time—
essentially, computer-assisted improvisation. For example, for the Macintosh there is
Jam Factory, M, Music Mouse, and Ovaltine.'? Other computerized aids for compos-
ers include systems that can automatically generate high-quality music notation. Ex-
amples include Finale, Professional Composer, and an advanced system now under
development by Don Byrd.'*
Al systems are likely to have a major impact on music education. In place of
the tedious and simpleminded auto-play type of features found on home organs,
future musical instruments for the home will contain intelligent music accompanists
that can help teach music and provide both children and adults with a richer musical
experience at early stages of keyboard skill development.'*

Computer-generated music notation.

(Courtesy of Donald Byrd; copyright String Quartet
1989 by Donald Byrd)
Donald Byrd (1967)
Moderato

4 z “ge i? je

psa
 Mirroring the broadening choices in sounds and musical control is a broad-
ening appreciation in popular music for diverse musical traditions. Initiated in part by
the Beatles almost two decades ago and fueled by many other musical innovators,
popular music has expanded from its Afro-American roots to include today the
thythms, styles, melodies, and harmonies of folk and classical traditions from around

the world.
Thus today’s musicians are confronted by a staggering array of choices: an
ever expanding set of sounds, music controllers, sound-modification techniques,
sequencing methods, composition tools, and even music traditions. All of this does
not necessarily make the musician's job easier: it is generally not desirable to use a
plethora of sounds and processing techniques in a single work. Indeed, much of
popular music has been criticized for the overuse of electronic techniques
The goal of music, however, remains the same: the communication of
emotions and ideas from musician to listener through sound, using the elements of
melody, rhythm, harmony, timbre, and expression. The challenge for the musical artist
is the same as for all artists: to make choices, to select the right timbres and melodies
to express their musical ideas.

+ The Visual Arts

An image can be placed somewhere between these two antipodes:

Realism = Abstraction
Abstraction = Realism
Wassily Kandinski, The Problem of Form(1912)

As | mentioned above, the computer revolution in the visual arts lags well behind the
major transformation already taking place in the musical arts. In music, computers are
no longer a novelty but are now the strongest driving force in the rapidly changing
face of music creation. In the visual arts there is still a widespread tendency to view
the computer as an interloper, as something alien to the creative process.'® This
inclination is reinforced by the fact that computer art has until recently been con-
strained by the limited resolution of most computer display screens and hard-copy
output devices.'® For several years now computer-based devices have been able to
create, modify, and manage sounds without compromising the quality and accuracy of
the result. The comparable situation in the visual arts is only now becoming possible
The result is that computer art, while a creative and rapidly changing field, is not yet in
the mainstream. Recognition that the computer is a viable and powerful tool for
artistic expression with unique capabilities should become more widespread in the
early 1990s now that the latest generation of personal computers is providing an
ability to create graphics comparable to other artistic media.'”
Some of the advantages of computer-based art techniques over manual
methods apply to any domain—e.g., ease of revising, of making backup and archival
copies, of making many types of global transformations (in music, transposing; in art,
changing size or color). The computer also permits the use of techniques that would
otherwise be impossible to realize.
Methods for using the computer to create visual art vary. The most straight-
forward approach is to simply use the display screen as a window onto a simulated

355
Art by AARON by Cohen. (Photo by canvas. The Macintosh MacPaint program and many more-recent applications popular-
Becky Cohen)
ized the computer as a sketchpad but also made obvious some of its limitations,
including an inability to represent fine detail, the jagged appearance of curved lines,
and for many of these systems, the lack of color. More recent high-resolution color-
graphics displays are now beginning to provide image quality comparable to that of
good 35-mm slides.'® Foremost among the advantages of using a computer screen as
a canvas is the availability of image-processing techniques that would be impossible
using ordinary paints and pencils. Among the dozens of techniques available, users
can alter shadings, reconstruct surfaces, synthesize natural and artificial backgrounds,
create reflections, and distort shapes.'?
One simple technique is the automatic repetition and rotation of images. If
the original image is complex, this would require enormous skill, not to mention time,
to perform manually. Using a computer, artists can easily view any object from any
orientation, complete with accurate perspectives if so desired.?° But few artists would
be happy with these exotic techniques if the more ordinary effects they have always
used were not available. Just as each musical instrument provides its own unique set
of methods for sound modification (vibrato, pizzicato, and others on a violin, for
example), each method of visual art creation provides techniques for creating and
modifying color and shape. Oil paints provide an ability to blend and fuse colors,
pastels provide certain types of shadings, and so on. With improvements in display
technology and image processing techniques, artists at a computer canvas will soon
be able to simulate all of these techniques in a single medium, while taking advantage
of methods that would be impossible with real paints and solvents.?'
Grass Series Five, a computer-
generated drawing programmed by
Colette Bangert and Charles J.
Bangert, 1983. Their mathematical
program was particularly effective in
depicting natural forms, such as
these interwoven patterns of blades
of grass. (Hardware: Intertrec
Superbrain computer, Wanatabe WX
4671 plotter. Courtesy of Colette
Bangert and Charles Bangert)

Large Landscape II: Ochre and Black,

a drawing plotted by a computer
programmed by Colette Bangert and
Charles Bangert, 1970. This drawing
was done using a Honeywell 635
digital computer and a Benson-
Lehner Draft-O-Matic plotter at the
University of Kansas. (Courtesy of
Colette Bangert and Charles Bangert)
Computer graphics from Tokyo. Perhaps the greatest advantage the computer gives to the artist is the
ieee et tna doves) ability to experiment and change one’s mind. Doing this is quite difficult with real
paints, but it is one of the inherent benefits of the computer canvas

Artificial life
An unusual way of applying computers to visual art is called artificial life.?* This
technique uses the computer to create a simulated environment with simulated
“organisms” controlled by a “genetic” code.” The artist provides the genetic code
and the rules for procreation and survival. The computer then simulates dozens or
even thousands of generations of simulated evolution. The resulting “creatures” and
environment can be quite beautiful. The artist is thus like a deist god who creates a
starting point and then unleashes a recursive process of repetitive re-creation.* The
artistic value of this technique should not be surprising. After all, real evolution has
certainly created a myriad of beautiful forms. In fact, aesthetics itself is often consid-
ered to be rooted in the beauty of the natural creation. The artificial-life approach to
creating art is to imitate this ultimate design force

I life, Przemysiaw
Prusi icz and one of his
computer-grown flowers.
(Photo by Gail Russell and
courtesy of the New York Times)

Fractals
Artificial life is considered a recursive technique because each new generation of
artificial life begets the next.’° The ultimate application of recursion to design involves
the mathematics of fractals. Fractal images are derived from a branch of mathematics
called fractal geometry, devised in the 1970s by IBM Research Fellow Benoit Mandel-
brot.?° One method for generating a fractal image is through a recursive procedure in
which specific parts of a picture (e.g., every straight line) are repeatedly replaced with
more complex parts. By using various starting figures and transformations, a
variety of patterns can be generated
There are actually several types of fractals. | have been discussing
ric fractals. Instead of using exactly the same transformation at each stage, v
allow a random process to control the transformations (randomly skipping som
steps, for example). In these chaotic fractals, the patterns are less regular an
enough, more natural.?” This observation led Mandelbrot to realize the po’
application of fractals to describing a wide variety of natural phenom

361
A fractal dragon in the process of
being generated. (Photo by Lou Jones)

A fractal dragon. (Image by Richard

F. Voss; copyright 1982 by B. B.
Mandelbrot; programming by Mark
Laff and Alan Norton.)

Generation of a random fractal.

(Photo by Lou Jones)
Fractal mountains. These mountains
economic trends, the progression of epidemics, the organization of proteins, and the
were created by a computer using
fractal geometry. (Image by Richard F. structure of music.?¢
Voss; copyright 1982 by B. B. Fractals have a number of interesting properties. Unlike most mathemati-
Mandelbrot)
cally generated shapes, the nature of a fractal is the same at ali scales of magnifica-
tion. Magnify a fractal a thousand times, and it still looks the same. Magnify it another
thousand times, and again, no essential change. This paradoxical quality describes a
surprisingly large number of natural phenomena quite well. The most famous example
is that of coastlines. Maps of coastlines have the same jagged appearance regardless

the same whether one looks at a cloud of 10 feet or a cloud of 100 miles
Indeed, perfect fractals keep the same appearance at all scales
like shapes in the real world have this property, called self-similarity, only
In a graphic display the limit is the granularity of the display, which is know
pixel size .2?1n the universe as a whole, which turns out to be a spongelike
pixel corresponds to the galaxy. Within galaxies, a different type of fr
fractal whorl governs the organization of stars
 nee
Point Reyes. This elaborate scene
was created entirely by computer
using a variety of techniques,
including chaotic fractals. (Courtesy
of Lucasfilm; copyright 1983 by Pixar)

Self-similarity. These treelike

geometric patterns were generated
by repetitive application of a fractal
generator. (Copyright 1982 by B. B.
Mandelbrot)

364
 Apteimannchen, a graphic display
of the so-called Mandelbrot set, a set
of formulas by Benoit B. Mandelbrot
drawn from the mathematics of
Julia sets. This famous fractal looks
the same at all scales of magnifica-
tion. Magnify this image a million
times, and the whorls and snowmen
will look exactly the same. (Image by
Heinz-Otto Peitgen and Peter H.
Richter of the University of Bremen;
copyright 1984 by MapArt)

365
 The word “fractal” derives from the phrase “ fractional dimension."*4
Because of its infinitely fine detail, a perfect fractal shape actually has infinite length
even though it resides on a finite two-dimensional surface. According to fractal
geometry, it therefore has a dimensionality somewhere in between that of an ordinary
line segment, which has a dimensionality of 1, and a surface, which has a dimension- |
ality of 2.°° For example, the Koch snowflake, pictured above, has a dimensionality of
1.26, just slightly greater than the coastline of Great Britain, which is considered to
have 1.25 dimensions. The galactic sponge has been estimated to have about 2.2
dimensions.°°
For the sciences and engineering disciplines, fractal geometry provides a
powerful tool for understanding the dynamics of turbulence, chaos, and other unpre-
dictable patterns.®’ For the visual arts, fractal geometry provides an equally powerful
methodology for creating images that can be natural, spectacular, or both. Fractals are
capable of generating realistic clouds, trees, bushes, lakes, mountains, shorelines,
landscapes, and other similar images.°° For example, George Lucas used the tech-
nique to create the moons of Endor in his film The Return of the Jedi. \t can create
environments that, while very alien-looking, appear to be the result of natural
processes.*?

The computer as artist

Another use of the computer to create artistic images is to provide a computer with a
set of rules and other knowledge structures that explain the processes of drawing,
composition, and layout and to let it be the artist. Probably the leading exponent of
this approach is Harold Cohen. Cohen has programmed over 1,000 rules for the
creation of complex drawings; his computer uses these to direct a robotic drawing
machine to create the drawings.*° These can be of people, plants, or abstract objects.
Cohen clearly intends the pictures as art, and indeed, his work has been highly
acclaimed. Cohen, who is considered a master colorist, often colors in his robot's
drawings manually. His latest ambition is to build a robot colorist programmed with his
own theory of coloring.
The computer as artist raises a number of interesting questions. First, who
is the artist? Cohen claims that he is, and his computer has not been programmed to
complain. Second, why bother? Why not simply make the drawings directly? Both of
these questions also pertain to the computer as composer.
These are important questions, and there are, in my view, several answers.
First, computer art is a perfectly legitimate means of creating art. The computer is
simply a tool, a medium through which the artist (or musician) is expressing himself
Second, the work of art is, in a sense, more interesting. We can regard the
program itself as the work of art that manifests itself differently every time we choose
to look at it (or listen to it). It is a work of art that can change its appearance as quickly
as the program can generate a new drawing. Since the style is unmistakably similar in
each such manifestation, all manifestations can reasonably be regarded as a single
work. It can be pointed out, however, that it is not unusual for multiple works of one
artist to exhibit such a degree of similarity. We can, therefore, consider a program
such as Cohen's to be either an artist itself, or a work of art, or both

366
 Planetrise Over Labelgraph Hill
(Souvenir from a Space Mission That
Never Was.) A Fractal-generated
planet and landscape by Benoit
B. Mandelbrot. (Image by RichardF.
Voss; copyright 1982 by B. B
Mandelbrot)

367
 Third, the rules themselves give us an objective and rigorous statement of
the theory of the artwork and thus an unprecedented degree of insight into its struc-
ture. Understanding the boundary between the rules themselves and the expressive
aspects of art that we might feel cannot (yet) be expressed with such precision also
provides a valuable perspective.
The rules also have educational value. In art, a great deal is known about
perspective, composition, the drawing of different types of shapes, and other facets
of visual art. These facets can be expressed in precise terms. Similarly, a great deal
can be said with mathematical precision regarding the nature of rhythm, harmonic pro-
gressions, the structures of different genres of melodies, and the other elements of
music. Creating rules that can in turn create entire satisfactory works of art (or music)
helps us to understand, and therefore to teach, the objective aspects of artistic
creation.
Finally, such systems can have practical value. In music they can perform
compositional chores by automatically creating walking bass lines, rhythmic accompa-
niments, and so on. In the visual arts, after the artist indicated his intentions, such
cybernetic assistants could finish a drawing, performing chores of shading, coloring,
creating perspective, balancing the sizes of objects for proper compositional balance,
and other tasks. In this mode of operation, such computer-based assistants would
always be working under the careful eye (or ear) of the artist.
Such automatic assistants have already been actively harnessed in the
creation of animation, where the chores of creating hundreds of thousands of images
(over a thousand per minute of film) is formidable. Computer-based animation
systems have substantially boosted the productivity of human animators.*' Another
area of commercial art where computer design stations are already of substantial
practical value is in fashion design and layout.
As in the world of music, the role of the computer is not to displace human
creativity but rather to amplify it. It is a tool, like a paintbrush, but one of unique and
virtually unlimited potential. Clearly, the great artists of old must have had many ideas
beyond the ones they had the time to actually express. By reducing the many chores
involved, computers can give artists the opportunity to realize more of their artistic
VISIONS.

Also, many computer techniques allow the expression of forms that are
simply impossible to realize in any other way. Because the results of some of these
methods, such as recursive fractal generation, are unpredictable until they are tried,
the computer can be a partner with the artist in exploring the artist's imagination.
Finally, the computer can open the world of artistic expression to more
participants. Many of us have rich visual imaginations and a good sense of aesthetic
judgement but are lacking in the technical skills to express our imaginations using
conventional artistic materials and techniques. The inability to draw a straight line (as
the ads used to say) may no longer be a barrier to becoming an effective and success-
ful artist.*?

368
 Could I see this in a blue paisley?
| Computer-aided design stations
allow clothes designers to create
new fashions on digital mannequins.
Garments can be viewed from
! different perspectives with seam
allowances, tucks, notches, and
other design elements automatically
computed. Computerized pattern
graders then generate the different-
sized patterns needed for manufac-
turing. (Courtesy of Bobbin
International magazine)

Automated animation in Kawasaki,

Japan. (Photo by Lou Jones)

369
+ The Literary Arts

Atall events my own essays and dissertations about love

and its endless pain and perpetual pleasure will be -
known and understood by all of you who read this and
talk or sing or chant about it to your worried friends
or nervous enemies. Love is the question and the subject
of this essay. We will commence with a question: does
steak love lettuce? This question is implacably
hard and inevitably difficult to answer. Here is
a question: does an electron love a proton,
or does it love a neutron? Here is a question: does
aman love a woman or, to be specific and to be
precise, does Bill love Diane? The interesting
and critical response to this question is: no! He
is obsessed and infatuated with her. He is loony
and crazy about her. That is not the love
of steak and lettuce, of electron and proton and
neutron. This dissertation will show that the
love of a man and a woman is not the love of
steak and lettuce. Love is interesting to me
and fascinating to you but it is painful to
Bill and Diane. That is love!
A poem from The Policeman's Beard Is Half Constructed (the first book ever written entirely by a computer) by RACTER
(a program by William Chamberlain and Thomas Etter)

Bill sings to Sarah. Sarah sings to Bill. Perhaps they

will do other dangerous things together. They may eat lamb or stroke
each other. They may chant of their difficulties and their
happiness. They have love but they also have typewriters.
That is interesting.
RACTER, The Policeman's Beard Is Half Constructed

eons deep in the ice

| paint all time in a whorl
bang the sludge has cracked

eons deep in the ice

| see gelled time in a whorl
pffftt the sludge has cracked

all green in the leaves

| smell dark pools in the trees
crash the moon has fled

all white in the buds

| flash snow peaks in the spring
bang the sun has fogged
Haiku poems written by a program by Margaret Masterman and Robin McKinnon Wood

In the literary arts, computers are already of substantial practical benefit. Of greatest
impact is the simple word processor. Not an Al technology per se, word processing
was derived from the text editors developed during the 1960s at Al labs at MIT and
elsewhere.* Also assisting the craft of writing text are increasingly sophisticated

370
David Boucher, Harry George, and
the Interleaf Desktop Publishing
System. (Photos by Lou Jones)

nnn
mcm

spelling checkers, intelligent syntax checkers, style checkers, on-line dictionaries,

thesauri, and other linguistic data bases.“ Of practical value to poets is the on-line
rhyming dictionary that provides rhymes and half-rhymes with particular rhythmic
patterns.
Writers of nonfiction are beginning to utilize the rapidly growing array of
data bases available on almost every subject. As the world’s knowledge bases
gradually shift from paper to on-line information utilities that can be intelligently
accessed through telecommunications, the productivity of research will be greatly
improved. Ultimately, computers will be able to conduct entire research projects on
verbal requests from their human bosses.**
Magazine and newspaper writing has benefited for the past decade from
on-line systems, such as those provided by Atex and other companies, that facilitate
all aspects of material creation, including text creation and editing, graphics prepara-
tion, and layout. The new industry of desktop publishing, pioneered by Xerox

svi
he Science of Art

Mitch Kapor, founder of Lotus

Development Corp. and developer of
Lotus 1-2-3 and Agenda. (Photo by
Lou Jones)
Interleaf, Apple, and others, provides writers with their own personal publishing
capability through the use of an ordinary personal computer and a laser printer.*°
A new genre of expert systems for personal computers are now appearing
that provide assistance in the organization of ideas. These systems allow users to
explore a set of thoughts, notice patterns and similarities, and develop outlines that
can later be fleshed out into text. Some of these systems are intended to be used in
real time as thoughts develop and include abilities to organize a variety of activities. A
prime example of this type of software is Lotus’s Agenda, developed by Mitch
Kapor.*”
Expert systems can also be applied to tracking the complex histories,
characterizations, and interactions of characters in such extended works of fiction as
long novels, series of novels, and television drama series. (A system that can assist
with the development of characters and plots for soap operas is described in Michael
Lebowitz, “All Work and No Play Makes HAL a Dull Program” in this chapter.)
Most challenging of all would be a system that could actually write or
substantively assist with the writing of prose or verse in the same way that expert
systems are now beginning to assist the composer and the graphic artist. Efforts have
been made to accomplish this, and some are represented in the epigraphs at the
beginning of this section and in the article “A (Kind of) Turing Test” in this chapter.
The computer undoubtedly has a role to play in the generation of text, and computer-
generated prose and verse is a small but flourishing field. The results, however, are
significantly less successful to date than in the musical and visual arts. RACTER’s
prose has its charm, but is somewhat demented-sounding, due primarily to its rather
limited understanding of what it is talking about.“® Deeply imbedded in written
language is knowledge, extensive, diverse, and intricately organized knowledge, the
vast bulk of which has yet to be captured. With language translation systems currently
providing only 80 to 90 percent accuracy on technical texts, it is not yet realistic (until
well into the first half of the next century) to expect computers to have sufficient
facility with language to generate literary quality text without assistance.*? The
objection might be raised that there is a comparable depth of knowledge in works of
music and of visual arts. This is certainly true. The reason why computers have had
greater success in assisting with the creative process in these areas is that we have a
better understanding of the surface structures of music and visual art (particularly the
former) than we do for literature. But even in these fields, the ultimate responsibility
for making an artistic statement is still the province of the human artist
The role of the computer is the same for all of the arts. First is to provide an
effective tool for improving the productivity of the artist, to enable him to realize more
of his imagination. Second, the computer can provide forms of expression that were
previously unavailable. Finally, to whatever extent we are able to capture even a
portion of the creative process in a program, we learn something about the art form
as well as the process of creating it. The computer can be a powerful partner in
exploring our thoughts and emotions and finding new ways of expressing them

373
 No ideas but in things.
William Carlos Williams

As discussed in several of the contributed articles in this book, the

Turing test was devised by Alan Turing as a way of certifying
machine intelligence. Turing described a situation in which a human
judge communicates with both a computer and a human using a
computer terminal. The judge's task is to determine which is which.
The judge cannot see the computer or the human and must make his
or her determination by interviewing both. The computer attempts to
trick the judge into selecting it as the human.
The essence of the Turing Test is that the computer
attempts to act like a human within the context of an interview over
terminal lines. A narrower concept of a Turing test is for a computer
to successfully imitate a human within a particular domain of human
intelligence. We might call these domain-specific Turing tests. One
such domain-specific Turing test, based on a computer's ability to
Raymond Kurzweil
write poetry, is presented here.
The Kurzweil Cybernetic Poet is a computer program
A (Kind of) Turing Test (written by the author) and provided with an input file of poems
written by a human author or authors. The program analyzes these
poems and creates a word-sequence model based on the poems it
has just read. It then writes original stanzas of poetry using the model
it has created. Some of the following stanzas of poetry were written
by the Kurzweil Cybernetic Poet. Some were written by human
authors (in fact the same human authors that were read and analyzed
by the Kurzweil Cybernetic Poet). See if you can tell which are which.
On a piece of paper, write down the numbers 1 through 28. Then put a
C by that number if you believe that the corresponding stanza was
written by the computer. Put an H if you believe the stanza was
written by a human poet. The answers are in a footnote.' Following
the answers are the results obtained when the test was taken by 16
human (both adult and child) judges.
. is beauty itself 10. The morning and already

that they were walking there. All along the new world naked, a perfect if slightly paled
cold, familiar wind— old park turned with young women
seized in amber.
. Pink confused with white
flowers and flowers reversed 11. “Interesting book?”

take and spill the shaded flame she sits

darting it back dancing by the electric typewriter,

into the lamp’s horn bloodless revolution of meats

strings of use,
. The winds of the oozy woods which wear Politic, cautious, and the fact
the ocean, with azure moss and flowers she is the fact
So sweet, the purple even she is calling them all—
I sleep in the arrows The children at his feet
Of the dome of death. he is always time
To roll it was dark,
. O thou, damp, jagged, like the voice
Who moved among some fierce Maenad, even among noise Because of love ends.
and blue
Between the bones sang, scattered and the silent seas. 12. Men with picked voices chant the names
of cities in a huge gallery: promises
. She eyes me with an ingrown eye, that pull through descending stairways
in the rhythm of teacup tapping to a deep rumbling.
thinks of sweeping away crumbs
13. Where were thou, sad Hour, selected from whose race is
. Atsix | cannot pray: guiding me,
Pray for lovers, Lured by the love of Autumn's being,
through narrow streets Thou, from heaven is gone, where was lorn Urania
And pray to fly When rocked to fly with thee in her clarion o'er the arms of death.
But the Virgin in their dark wintry bed
14. Lady of Autumn's being,
. What seas what shores what granite islands towards my timbers Thou, from the day, having to care
And woodthrush calling through the fog Teach us now thoroughly small and create,
My daughter. And then presume?
And this, and me,
. Imagine now a tree in white sails still whirled And place of the unspoken word, the unread vision in Baiae’s bay,
About the leaves And the posterity of Michelangelo.
will be of silences
Calm and angels 15. | am lonely, lonely.
I slap an answer myself
. —and the sun, dipping into the avenues she hides deep within her
streaking the tops of yet plays—
the irregular red houselets, Milkless
and
the gay shadows dropping and dropping.

375
16. O my shoulders, flanks, buttocks with herself alone
against trespassers, and then dividing and over
against thieves, and splashed and after you are
storms, sun, fire, listening in her eyes :
against thieves,
storms, sun, fire, 24. All along the road the reddish
against flies, against weeds, storm-tides, purplish, forked, upstanding, twiggy
neighbors, weasels that waken stuff of bushes and small trees
The silent seas. with dead, brown leaves under them
leafless vines—
17. the days, locked in each other's arms,
seem still 25. Pray for those who are branches on forever
so that squirrels and colored birds
go about at ease over 26. Like a sod of war;
the branches and through the air. houses of small
white curtains—
18. | am watching ants dig tunnels and bury themselves Smell of shimmering
they go without water or love ash white,
an axe

19. Lady is sick,

perhaps vomiting, 27. By action or by suffering, and whose hour
perhaps laboring Was drained to its last sand in weal or woe,
to the usual reign So that the trunk survived both fruit or flower;—

20. Rain is sweet, brown hair; 28. is a steady burning

Distraction, music in passageways. the road the battle’s fury —
Six o'clock. clouds and ash and waning
The time. Redeem sending out
The world and waking, wearing young people,

21. Wipe your hand across your mouth, and laugh;

The worlds revolve like ancient women
Gathering fuel in vacant lots.

22. | should have been a pair of ragged claws

Scuttling across the floors of silent seas.

23. patches of all

save beauty
the rigid wheeltracks.
The round sun
the bed.
She smiles, Yes
you please first
then stays

376
 The above 28-question poetic Turing test was admini- Adult scores on poem stanzas composed by a computer
stered to 16 human judges with varying degrees of computer and (13 adults, % correct)
poetry experience and knowledge. The 13 adult judges scored an Level of computer experience
Level of poetry
average of 59 percent correct in identifying the computer poem
experience Little Moderate Professional Average
stanzas, 68 percent correct in identifying the human poem stanzas,
and 63 percent correct overall. The three child judges scored an Little 56 44, 69, 75 63, 75 64
Moderate 50, 56, 63 56, 63 15 61
average of 52 percent correct in identifying the computer poem
stanzas, 42 percent correct i identifying the human poem stanzas, Alot 25 25

and 48 percent correct overall. Average 56 61 59 59

The charts show the actual scores obtained by the 16
human judges as broken down by adult/child, computer experience,
Children’s scores on poem stanzas composed by a computer
and poetry experience. As can be seen from the charts, there were no
trends based on level of computer experience or poetry experience (3 children, % correct)
clearly discernible from this limited sample. The adults did score Scores 38, 50, 69
somewhat better than the children. The children scored essentially at Average 52
chance level (approximately 50 percent) and the adults achieving
slightly better than chance.
Adult scores on poem stanzas composed by a human
The next chart shows the number of correct and incorrect
(13 adults, % correct)
answers for each of the 28 poems or stanzas. While the adult judges
scored somewhat better than chance (63 percent), their answers Level of poetry Level of computer experience |

were far from perfect. The computer poet was able to trick the human experience Little Moderate Professional © Average
judges much of the time. Some of the computer poems (numbers 15
Little 83 58, 58, 100 50, 67 69
and 28, for example) were particularly successful in tricking the
Moderate 60, 67,83 58, 83 92 4
judges.
Alot a 25
We can conclude that this domain-specific Turing test
Average 73 72 59 68
has achieved some level of success in tricking human judges in its
poetry-writing ability. A more difficult problem than writing stanzas of
poetry is writing complete poems that make thematic, syntactic, and
Children’s scores on poem stanzas composed by a human
poetic sense across multiple stanzas. A future version of the Kurzweil
(3 children, % correct)
Cybernetic Poet is contemplated that attempts this more difficult task.
Scores 33, 42, 50
To be successful, the models created by the Cybernetic Poet will
Average 42
require a richer understanding of the syntactic and poetic function of
each word.
Even the originally proposed Turing test involving Adult scores on all poem stanzas (13 adults, % correct)
terminal interviews is notably imprecise in determining when the
Level of poetry Level of computer experience
computer has been successful in imitating a human. How many
experience Little Moderate Professional Average
judges need to be fooled? At what score do we consider the human
judges to have been fooled? How sophisticated do the judges need to Little 68 50, 64, 86 57,71 66

be? How sophisticated (or unsophisticated) does the human foil need Moderate 55, 61,71 61, 68 82 66

to be? How much time do the judges have to make their determina- Alot 25 25

tion? These are but a few of the many questions surrounding the Average 64 66 59 63
Turing test. (The article “A Coffeehouse Conversation on the Turing
Test” by Douglas Hofstadter in chapter 2 provides an entertaining
discussion of some of these issues). It is clear that the era of Children’s scores on all poem stanzas (3 children, % correct)

computers passing the Turing test will not happen suddenly. Once Scores 39, 43, 61
computers start to arguably pass the Turing test, the validity of the Average 48

377
 Numbers of right and wrong answers for each poem stanza tests and the testing procedures will undoubtedly be debated. The
Computer or same can be said for the narrower domain-specific Turing tests.
human poem We have not yet reached the point at which computers
Poem stanza No. right No. wrong stanza can even arguably pass the originally proposed terminal-interview
type of Turing test. This test requires a computer to master too many
1 9 7 computer
high-level cognitive skills in a single system for the computer of today
3 11 a computer P i
| | to succeed. As Dan Dennett points out in his article, the unadulterated
4 8 8 computer
Turing test is far more difficult for a computer to pass than any more
6 9 7 computer
restricted version. We have, however, reached the point where
8 11 5 computer
computers can successfully imitate human performance within
11 1 5 computer
narrowly focused areas of human expertise. Expert systems, for
13 8 8 computer , ims
example, are able to replicate the decision-making ability of human
14 10 6 computer
professionals within an expanding set of human disciplines. In at
15 6 10 computer
least one controlled trial, human chess experts were unable to
16 10 6 computer
distinguish the chess-playing style of more sophisticated computer
19 9 7 computer
chess players from that of humans. Indeed, computer chess programs
20 12 4 computer ‘
are now able to defeat almost all human players, with the exception
23 9 7 computer
of a small and diminishing number of senior chess masters. Music
25 8 8 computer
composed by computer is becoming increasingly successful in
26 11 5 computer
passing the Turing test of believability. The era of computer success
28 6 10 computer
in a wide range of domain-specific Turing tests is arriving.
Average 58% 42%

Z 9 7 human
Note
5 9 7 human
7 9 7 human
1. Four human poets were used: three famous poets (Percy Bysshe Shelley, T. S.
9 13 3 human
Eliot, and William Carlos Williams) and one obscure poet (Raymond Kurzweil). In the
10 8 8 human
case of the famous human poets, stanzas were selected from their; most famous
12 10 6 human ‘i ars
published work. In all cases, the stanzas selected did not require adjacent stanzas to
v7 8 7 human make thematic or syntactic sense. The computer stanzas were written by the
18 4 2 human Kurzweil Cybernetic Poet after it had read poems written by these same human
2 " 5 human authors. The answers are as follows:
2 11 5 human Poem stanza 1 written by the Kurzweil Cybernetic Poet after reading poems by
24 " 5 human William Carlos Williams
2 8 8 human Poem stanza 2 written by William Carlos Williams
Poem stanza 3 written by the Kurzweil Cybernetic Poet after reading poems by Percy
Average 64% 36%
Bysshe Shelley
Overall average 61% 39%

378
Poem stanza 4 written by the Kurzweil Cybernetic Poet after reading poems by T. S.
Eliot and Percy Bysshe Shelly
Poem stanza 5 written by Raymond Kurzweil
Poem stanza 6 written by the Kurzweil Cybernetic Poet after reading poems by T. S.
Eliot, Raymond Kurzweil, Percy Bysshe Shelley, and William Carlos Williams
Poem stanza 7 written by T. S. Eliot
Poem stanza 8 written by the Kurzweil Cybernetic Poet after reading poems by
Raymond Kurzweil and T. S. Eliot
Poem stanza 9 written by William Carlos Williams
Poem stanza 10 written by Raymond Kurzweil
Poem stanza 11 written by the Kurzweil Cybernetic Poet after reading poems by
Raymond Kurzweil and T. S. Eliot
Poem stanza 12 written by William Carlos Williams
Poem stanza 13 written by the Kurzweil Cybernetic Poet after reading poems by
Percy Bysshe Shelley
Poem stanza 14 written by the Kurzweil Cybernetic Poet after reading poems by T. S.
Eliot and Percy Bysshe Shelley
Poem stanza 15 written by the Kurzweil Cybernetic Poet after reading poems by
Raymond Kurzweil and William Carlos Williams
Poem stanza 16 written by the Kurzweil Cybernetic Poet after reading poems by T. S.
Eliot, Raymond Kurzweil, Percy Bysshe Shelley, and William Carlos Williams
Poem stanza 17 written by William Carlos Williams
Poem stanza 18 written by Raymond Kurzweil
Poem stanza 19 written by the Kurzweil Cybernetic Poet after reading poems by T. S.
Eliot, Raymond Kurzweil, Percy Bysshe Shelley, and William Carlos Williams
Poem stanza 20 written by the Kurzweil Cybernetic Poet after reading poems by
Raymond Kurzweil and. S. Eliot
Poem stanza 21 written by T. S. Eliot
Poem stanza 22 written by T. S. Eliot
Poem stanza 23 written by the Kurzweil Cybernetic Poet after reading poems by
Raymond Kurzweil and William Carlos Williams
Poem stanza 24 written by William Carlos Williams
Poem stanza 25 written by the Kurzweil Cybernetic Poet after reading poems by T. S.
Eliot, Raymond Kurzweil, Percy Bysshe Shelley, and William Carlos Williams
Poem stanza 26 written by the Kurzweil Cybernetic Poet after reading poems by
Raymond Kurzweil and William Carlos Williams
Poem stanza 27 written by Percy Bysshe Shelley
Poem stanza 28 written by the Kurzweil Cybernetic Poet after reading poems by
William Carlos Williams

379
 This is the year 1300. Brother Giorgio, scholar-monk, has the task of
making a map of Australia, a big island just south of India. Maps must
record what is known about the places they represent, and Giorgio
has been told about astrange Australian animal, ratlike, but much
bigger, with a long thick tail and a pouch. He draws it, and it comes
out like this:

A year later a world traveller is visiting Giorgio’s monastery, and he

tells our cartographer that he has the animal wrong. For one thing, it
isn't carryinga pouch; the pouch is actually part of its belly. (Mercy!
says Giorgio.) For another, it doesn't walk on all fours like a rat but on

| ER uy its hind legs, which are much bigger than its front legs. Giorgio
redraws his picture:

Brother Giorgio’s Kangaroo

Harold Cohen is a professor in

the Visual Arts Department at
the University of California at
San Diego. In recent years he
and AARON have been shown
at the San Francisco Museum
of Modern Art, the Stedelijk
Museum in Amsterdam, the
Tate Gallery in London, the

Brooklyn Museum, the Ontario But the tail rests on the ground. Giorgio tries once more. The traveller
Science banter, and ihe Basen screws up his face in concentration, his eyes closed. | don't think
Science Museum. He has
lectured widely on the subject that's quite right, he finally says , but | guess it's close enough.

of AARON, and AARON has

reciprocated by providing
Cohen with several thousand A

drawings, including designs for a

a series of mural projects. & 7
Cohen describes their relation- eal ie
ship as symbiotic. Cohen is
regarded as a pioneer in the SAS we
application of Al to the visual v E
arts. =
 The year is 1987. AARON, a computer program, has the
task of drawing some people in a botanical garden—not just making
a copy of an existing drawing, you understand, but generating as
many unique drawings on this theme as may be required of it. What
does it have to know in order to accomplish such a task? How could
AARON, the program, get written at all?
The problem will seem a lot less mystifying, though not
necessarily less difficult, if we think of these two stories as having a
lot in common. AARON has never seen a person or walked through a
botanical garden. Giorgio has never seen a kangaroo. Since most of
us today get most of our knowledge of the world indirectly and
heavily wrapped in the understanding of other people from grade
school teachers to television anchor persons, it should come as no
surprise that a computer program doesn’t have to experience the
world itself in order to know about it.
How did Giorgio know about kangaroos before the visitor
started to refine his knowledge? He had been told that the animal was
ratlike, but how much good would that have done him if he had never
seen a rat? For people, the acquisition of knowledge is cumulative, as
it clearly has to be. Nothing is ever understood from scratch. Even the
new-born babe has a good deal of knowledge “hard-wired” before it
starts. And when we tell each other about the world, it isn't practical
or even possible to give a full description of something without
referring to some thing else. That's as true for computer programs as
it is for people. There is an important difference, though. For people,
knowledge must eventually refer back to experience, and people
experience the world with their bodies, their brains, their reproduc-

Harold Cohen, computer artist. (Photo tive systems, which computers don’t have.
by Lou Jones)

381
 With this in mind, we might guess that AARON’s
knowledge of the world and the way AARON uses its knowledge are
not likely to be exactly the same as the way we use what we have.
Like us, its knowledge bes been acquired cumulatively. Once it
understands the concept of a leaf cluster, for example, it can make
use of that knowledge whenever it needs it. But we can see what
plants look like, and AARON can’t. We don't need to understand the
principles that govern plant growth in order to recognize and record
the difference between a cactus and a willow tree in a drawing.
AARON can only proceed by way of principles that we don't
necessarily have. Plants exist for AARON in terms of their size, the
thickness of limbs with respect to height, the rate at which limbs get
thinner with respect to spreading, the degree of branching, the
angular spread where branching occurs, and so on. Similar principles
hold for the formation of leaves and leaf clusters. By manipulating
these factors, AARON is able to generate a wide range of plant types
| fs and will never draw quite the same plant twice, even when it draws a
Athletes, a hand-colored, computer- number of plants recognizably of the same type. Interestingly enough,
generated drawing by Harold Cohen.
(Photo by George Johnson) the way AARON accesses its knowledge of plant structure is itself
quite treelike. It begins the generation of each new example with a
general model and then branches from it. “Tree” is expanded into
“big-tree/small-tree/shrub/grass/flower,” “big tree” is expanded into
“oak/willow/avocado/wideleaf” (the names are not intended
literally), and so on, until each unique representation might be
thought of as a single “leaf,” the termination of a single path ona
hugely proliferating “tree” of possibilities.
Obviously, AARON has to have similar structural
knowledge about the human figure, only more of it. In part, this extra
knowledge is demanded by AARON’s audience, which knows about
bodies from the inside and is more fussy about representations of the
body than it is about representations of trees. In part, more knowl-
edge is required to cope with the fact that bodies move around. But it
isn’t only a question of needing more knowledge; there are three
different kinds of knowledge required—different, that is, in needing
to be represented in the program in different ways.
First, AARON must obviously know what the body consists
From the Bathers series of hand-
of, what the different parts are, and how big they are in relation to
colored, computer-generated
drawings by Harold Cohen. (Photo by each other. Then it has to know how the parts of the body are
George Johnson) articulated: what the type and range of movement is at each joint.

Finally, because a coherently moving body is not merely a collection

of independently moving parts, AARON has to know something about
how body movements are coordinated: what the body has to do to
keep its balance, for example. Conceptually, this isn’t as difficult as it
may seem, at least for standing positions with one or both feet on the
ground. It's just a matter of keeping the center of gravity over the base
and, where necessary, using the arms for fine tuning.

382
 We started by asking what AARON would need to know
to carry out its task. What I've outlined here constitutes an important
part of that necessary knowledge, but not the whole of it. What else
is necessary? Let's go back to Giorgio. Has it struck you that
whatever Giorgio eventually knew about the relative sizes of the
kangaroo’'s parts and its posture, he had been told nothing at all
about its appearance?
Yet his drawings somehow contrived to look
sort of like the animal he thought he was representing, just as
AARON's trees and people contrive to look like real trees and real
people.
That may not seem very puzzling with respect to Giorgio.
In fact, it may seem so unpuzzling that you wonder why | raise the
issue. Obviously, Giorgio simply knew how to draw. | suspect that
most people who don’t draw think of drawing as a simple process of
copying what's in front of them. Actually it's a much more compli-
cated process of regenerating what we know about what's in front of
us or even about what is not in front of us: Giorgio's kangaroo, for
example. There's nothing simple about that regeneration process,
though the fact that we can do it without having to think much about
it may make it seem so. It is only in trying to teach a computer Ahand-colored, computer-generated
drawing of figures and trees with
program the same skills that we begin to see how enormously
rocks in the foreground, by Harold
complex a process is involved. Cohen. (Photo by Linda Winters)

383
Cohen | Brother Giargio’s kangaroo

How do humans learn to draw? To some degree,

obviously, we learn about drawing by looking at other peoples’
drawings. That's why we are able to identify styles in art, and why
most of the drawings coming out of Giorgio’s monastery would have
had a great deal in common and be distinguishably different from, say,
the drawings made in a Zen Buddhist temple in Japan. At the same
time, all children make very much the same drawings at any one
stage of cognitive development without learning from each other or
from adults. They don’t need to be told to use closed forms in their
drawings to stand for solid objects, for example. That equivalent is
universal; all cultures have used closed forms to stand for solid
objects. In short, knowledge of drawing has two components. Giorgio
learned about style, about what was culturally acceptable and what
was not, from his peers. But before cultural considerations ever arise,
drawing is closely coupled to seeing—so closely coupled that we
might guess all major visual modes of representation in human history
to have sprung directly from the nature of the cognitive system. So
Giorgio never had to be told how to draw or how to read drawings. He
Black and White Drawing, a Couldises!
computer-generated drawing of
He had to be told about kangaroos, not about how to draw
figures and trees, by Harold
Cohen. (Photo by Becky Cohen) kangaroos. Knowledge of drawing isn’t object specific; if Giorgio

384
could draw a kangaroo, he could also draw an elephant or a castle or core, and AARON ensures a sufficiently responsive embodying line
an angel of the Annunciation. If one can draw, then anything that can by sampling its relation to the core more frequently.

be described in structural terms can be represented in visual terms. Nothing has been said here about how AARON’s
That generality suggests that rather than thinking of knowledge of knowledge of the world is stored internally, about how its knowl-
drawing as just one more chunk of knowledge, we should think of it edge of drawing is actually implemented, or about its knowledge of
as a sort of filter through which object-specific knowledge passes on composition, occlusion, and perspective. AARON’s success as a
its way from the mind to the drawing. program stands or falls on the quality of the art it makes, yet nothing
Like Giorgio, AARON had to be told about things of the much has been said about art and nothing at all about the accultur-
world. Unlike Giorgio in having no hard-wired cognitive system to ated knowledge of style, for which its programmer, like Giorgio’s
provide a built-in knowledge of drawing, it had to be taught how to monastic peers, must admit or claim responsibility. All the same,
draw as well, given enough of a cognitive structure (the filter just there are interesting conclusions to be drawn from this abbreviated
referred to) to guarantee the required generality. If provided with account. It should be evident, for example, that the knowledge that
object-specific knowledge, AARON should be able to make drawings goes into the making of a visual representation, even a simple one, is
of those objects without being given any additional knowledge of quite diverse. | doubt that one could build a program capable of
drawing. manipulating that knowledge and exhibiting the generality and
AARON’s cognitive filter has three stages, of which the flexibility of the human cognitive system other than by fashioning the
first two correspond roughly to the kinds of knowledge described program as an equivalent, artificial cognitive system. If nothing much
above in relation to the human figure: knowledge of parts, articula- has been said about art, it is because remarkably little of the program
tion, and coordination. The third stage generates the appearance of has anything to do with art: it constitutes a cognitive model of a
the thing being drawn. Neither of the first two stages results in reasonably general kind, and | even suspect that it could be adapted
anything being drawn for the viewer, though they are drawn in to other modes without too much distortion. But the lack of art
AARON’s imagination, so to speak, for its own use. First AARON specificity isn't as puzzling as it may seem at first glance. The
constructs an articulated stick figure, the simplest representation that principal difference between artists and nonartists is not a cognitive
can embody what it knows about posture and movement. Then around difference. It is simply that artists make art and nonartists don’t.
the lines of this stick figure it builds a minimal framework of lines
embodying in greater detail what it knows about the dimensions of
the different parts. This framework doesn’t represent the surface of
the object. In the case of a figure, the lines actually correspond quite
closely to musculature, although that is not their essential function.
They are there to function as a sort of core around which the final
stage will generate the visible results. Quite simply, AARON draws
around the core figure it has “imagined.” Well, no, not quite so
simply. If you look at one of its drawings, it should be clear that the
final embodying stage must be more complicated than | have said if
only because AARON apparently draws hands and leaves with much
greater attention than it affords to thighs and tree trunks.
AARON’s embodying procedures are not like the
preliminary edge-finding routines of computer vision, which respond
to changes in light intensity without regard to what caused them.
AARON is concerned with what it is drawing and continuously
modifies the performance of this final stage with respect to how much
knowledge has already been represented in the core figure. The
greater the level of detail already present, the more AARON relies
upon it and the closer to the core the embodying line is drawn. Also,
greater detail implies more rapidly changing line directions in the

385
 Charles Ames (Photo by Lou Jones)

| en ee eee How Can Intelligent Computers Help Compose Music?

Artificial Intelligence and Musical Composition Consider the young music student who wants to learn about chord
progressions. Traditionally, he would be required to try out some
Charles Ames studied progressions of blocked chords on the piano. But if keyboard skills
mathematics aed eee were lacking, the student would have to put off learning about
composition at Pomona College,
| steal hicwniy wt Uitoriland progressions for several years until these skills had been acquired.

| composition at the State With an appropriately designed expert music-tutoring program, such
University of New York at a student might type a harmonic plan into a computer and have the
Buffalo, where he received his
program quickly derive a full-blown realization with rhythmic chords,
Ph.D. in 1984. He is currently
Director ef the Kurzweil a bass line, even a drum part.
Foundation’s Automated Consider the musical theorist who desires a truly
Composition Project, which empirical means of testing assumptions about how music is made.
created the Cybernetic Whereas in the past, musical theories have as often as not been
Composer, an Al based software . bee i
accepted on the basis of intrinsic logical elegance, Al techniques
‘system that composes music
in a variety of popular styles. now make it possible to implement a theory as a program whose
Ames is regarded as a pioneer in products may be evaluated empirically against authentic musical
the application of Al to music
examples.
composition.
Consider also the composer of soundtracks for recorded
dramas, documentaries, and advertisements. It is already common to
rely heavily on standard background patterns when producing large
volumes of precisely timed and thematically consistent music in the
face of production deadlines. It is very likely that tomorrow's
professional composers will be relying even more heavily on
software modules for accompaniments and perhaps even for leads.
For the art-music composer, artificial intelligence
provides a mechanical extension to human intellect. For example, a |
composer might wish to set up criteria for comparing solutions to a
compositional problem (e.g., the problem of choosing pitches for a
melody) and then let the computer evaluate every possible solution in
order to choose the best one.

386
 Perhaps the most radical potential for automated The basic compositional strategy for these programs was
composition resides in the capability of a single program to generate “left-to-right,” with some “top-down” influences. Wholly left-to-right
many distinct pieces of music. One might think of such a program as a programs selected musical details directly as they were to appear in
new kind of “record” that plays a different piece every time. The the music, conditioning later decisions upon earlier choices. Many
possibilities range from the mundane, such as computer-composed programs had top-down influences to the extent that they divided the
music for fast-food restaurants and supermarkets, to the sublime. We music into sections and then chose the details of each section ina
have now seen the appearance of a new type of composer: the left-to-right manner. Each new section was distinguished by unique
metacomposer, who shapes compositional principles as fluently as parameters that affected the statistical makeup of these details. One
traditional composers have shaped compositional material. For such of the composer-programmers of this period even interposed
artists the originality of the music will clearly depend on the subsections between sections and details.
originality and subtlety of the programming.
Rote Processing When computers went on-line during the 1970s, a
The situations described above are not at all speculative. number of musicians took advantage of the new technology to
All of them can become realities within five years if developers are
develop programs that would process musical information either in
willing to expend the necessary energy to make them happen. Indeed, real time or within the context of an interactive dialog. The desire for
successes already achieved indicate that creativity can be modeled
rapid interaction led to an emphasis on procedures that were simple
with much greater precision than conventional wisdom once
enough that they could respond at the touch of a button, and
suggested. A number of programs of my own are already producing
especially to an emphasis on rote operations on themes. Among the
credible emulations of human-composed music in popular styles, and most familiar of these operations are those that come from traditional
a program by Kemal Ebcioglu (see figure) for harmonizing chorales in
the style of J. S. Bach has managed on occasion to duplicate Bach's
solutions exactly.’ Although the styles in all of the current programs
have been rather narrowly defined, the insights they provide lead
toward ever-more-general representations of compositional
procedure.

A Survey of Decision-Making Tactics and Compositional Strategies

Sete arene aS 65 eee

From their beginnings in the mid 1950s up until a few years ago,
programs for automatic musical composition have been developed by
a small handful of musicians working in relative isolation from the
mainstream Al movement and, for that matter, from other computer-
music fields such as digital sound synthesis. This isolation has been
harmful to the extent that most early practitioners were ignorant of
recursion, linked data structures, and basic search methods. Yet it
has also led to some unique approaches that might have been passed
over by a more orthodox methodology.

Statistics Up to the late 1960s the most pervasive decision-making

tactics were statistical. Statistics provided an economical way of
representing trends of musical behavior from short-term
dependencies (e.g., how often G’ chords are followed by C major
chords, by A minor chords, etc.) to long-term distributions of material
(e.g., the relative usage of scale degrees). A statistical composing Kemal Ebcioglu’s 1984 program for duplicate Bach's own solutions
program from these early years selected options randomly; the harmonizing chorales guided its exactly. However, the solution

options assigned the greatest statistical weight by the programmer

left-to-right note selection with top- shown here is more typical.
down planning of harmonic goals and (Courtesy of Kemal Ebcioglu)
had the greatest probability of selection. subgoals. For a few chorale tunes
Ebcioglu’s program managed to

3387
 canons and fugues. They include transposition (playing a theme with
every note shifted equally up or down in pitch), inversion (playing a
theme upside down), retrograde (playing a theme backwards), and
diminution (decreasing the note lengths in a theme so that it goes by
twice as quickly). The user of one of these programs would typically
build a composition using a “bottom-up” strategy: he could either
enter an originally composed theme or have one randomly generated,
he might then derive variations on this theme using one or more
operations, and he could cut up, paste together, and edit this thematic
material in various ways until a whole composition had been
produced.
Searches Intelligent composing programs such as Ebcioglu’s and my
own are distinguished from statistical and rote programs by their
ability to discriminate among solutions to musical problems. The
secret behind this ability lies in constrained-search techniques
drawn directly from Al.‘ As an illustration of how a constrained
search works, consider the problem of composing a melody when the
rhythm, range, harmonic context, and style are given. Solving this
kind of problem means choosing a pitch for each note in a way that
conforms to the given style. Yet although one might have a very good
idea how a melody in this style should sound, that's not sufficient
understanding to develop a program. One must be able to describe the
style in terms meaningful to a computer.
To do this, the programmer needs to make some general
Top-down productions begin with a
observations concerning how pitches behave in the style. He might
very general archetype of a musical
form and recursively elaborate observe, for example, that the melodies never move in parallel
upon this archetype until a complete octaves or fifths with the bass, that they a/ways employ chord roots at
description of the musical
details has been obtained. (From cadence points, that nonchord tones a/ways resolve by step to chord
Charles Ames, “Crystals: Recursive tones, and that dissonant chord tones (e.g., chord sevenths) always
Structures in Automated
resolve downward by step in the next chord. The programmer might
Composition,” Computer Music
Journal 6 [1982], no. 3; courtesy of observe as well that scalewise motion is preferable to leaps, that
Charles Ames) leading tones tend to lead upward, and that the different scale
degrees are usually in balance. Observations characterized by such
In one stage of producing Charles
words as “never” and “always” can be implemented directly as
Ames's Protocol for solo piano,
the composer decided that
groupings of chords (themselves
computer-composed) would be most
appropriate to the musical spirit
of the piece when constituent pairs
of chords shared few degrees of the
chromatic scale in common. The
program scored the left group less
highly than the right one because
chords 7 and 13 on the left share five
out of seven degrees. (From Charles
Ames, “Protocol Motivation,
Design, and Production of a Compo-
sition for Solo Piano,” Interface:
Journal of New Music Research 11
[1982], no. 4: 226)

388
constraints. Constraints keep a search within the musical ballpark. to interact with a search-driven composing program on its own terms.
Though many are drawn from the rules of musical pedagogy, they Interactive feedback from a search consists at present of three
should in no sense be taken as standards of “good music.” If a melody phases: setting up constraints and preferences, leaving the search to
violates a constraint, we cannot say it is wrong, only that it is out of do its thing (searches are most effective when left to run with a
style. The basic mechanism for composing a melody subject to minimum of interference from the human user), and accepting or
constraints is as follows: For each note, the search steps through the rejecting the results. If the results are unsatisfactory, one might make
available pitches until it finds one conforming to all the constraints. adjustments in the constraints and preferences before running the
Whenever it finds an acceptable pitch, it moves forward to an search again.
uncomposed note; whenever it exhausts all the pitches available to a
given note, it backtracks to the most recent composed note that Conclusions
oe
caused a conflict, revises this note’s pitch, and begins working its
way forward again. The entry of Al into musical creativity is by no means as radical as it
Observations characterized by words such as “tend,” might seem to laymen in view of the active and long-standing tradition
“preferable,” and “usually” generally cannot be implemented as of composer involvement with technical theory about music. This
hard-and-fast constraints. The only alternative is to bias the search tradition reaches from Pythagoras and Aristoxenus of antiquity,
toward solutions with more preferable attributes. Should the through numerous medieval writers, through such Renaissance
programmer wish to encourage scalewise motion, for example, the theorists as Gioseffe Zarlino and Thomas Morley, through the Baroque
search might be designed to try neighboring pitches before others. If composer Jean-Philippe Rameau, through more recent composers as
such motion is more critical at the end of a phrase than at the diverse as Arnold Schoenberg, Henry Cowell, Paul Hindemith, Harry
beginning, it may be more effective to start at the end of the phrase Partch, and Joseph Schillinger to contemporaries such as Pierre
and work backward; this will prevent early phrase-notes from Boulez and lannis Xenakis. Each has built stylistic models from
forming a context that precludes scalewise motion later on. The constraints, preferences, and procedural descriptions of the act of
official Al jargon for this kind of procedural biasing is “heuristic making a composition, not as a way of codifying what went on in the
programming.” past, but as an intellectual aid to his craft. However, the ability of Al
The abilities to seek the more preferable solutions, to programs to generate actual music from such models brings them out
apply constraints, and to backtrack in the face of an impasse are the of the speculative realm and makes them accessible to all: to
basic advantages that constrained searches have over statistical and composers seeking an augmented working environment in which the
rote procedures. The method applies just as well to musical problems content, form, and style of entire compositions can be adjusted at the
radically different from melody writing, such as top-down generation touch of a button; to theorists wishing to determine where the
of musical details from forms. Searches are a complement, not a strengths and weaknesses of new models lie; and to music students
substitute, for statistical and rote procedures. If the programmer seeking expert assistance in the realization of their musical projects.
desires to maintain a sense of unpredictability within given stylistic
bounds, then randomness can be incorporated into the mechanism by Notes
—
which the search assigns priorities to options. If long-term statistical
distributions are of concern, then a search can be designed to favor 1. Ebcioglu’s results have great implications for pedagogy, since Bach's chorales are
options that best conform to these distributions. Finally, if rote the paradigm for traditional studies in musical harmony.
thematicism is desired, then themes generated by such operations as 2. Programs for automatic music composition began in 1956 with Hiller and
the ones described above can themselves be treated as options by Isaacson’s Illiac Suite, created slightly later than the first chess-playing and
searches evaluating many different thematic variations to choose the
theorem-proving programs. An exhaustive survey of composing programs, replete
with names, musical examples, and bibliography, is available in Charles Ames.
best one for the context.
“Automated Composition in Retrospect: 1956-1986,” Leonardo: Journal of the
One trade-off with searches is that their ability to
International Society for Science, Technology, and the Arts, 1987.
backtrack renders them compositional rather than improvisational. A
3. These statistical composing programs anticipated by some 20 years the current Al
composer in a pinch can throw things out, but an improvisor cannot
fashion for “fuzzy logic.”
turn back the clock on music that has already been played. As a
4, Rigorous searches have also been used by Stanley Gill, Marilyn Thomas, and
result, it is impractical to implement search-driven compositional possibly David Levitt.
procedures in real time. Another trade-off is that one must be willing
 Introduction
a

The clear majority of applied work in artificial intelligence (Al) has

involved practical problems in such areas as business, medicine,
defense, and so forth. This is as it should be, but there is also room for
the application of Al to the arts. Al has the potential to allow the
creation of compelling new entertainment forms and to improve the
creative process in existing art forms. In addition, such work can lead
to important new insight into the creative process. In this article |
shall discuss how Al can be applied to the area of story telling: both
how it can be used in intelligent tools for writers and how it might
lead to new forms of intelligent, interactive stories where a writer/
creator can intimately involve a reader/user in the story.
At Columbia University my colleagues and | investigated
the problem of developing systems that generate extended stories,
those that continue over time. We selected the domain of “interper-

Tica stows sonal melodrama,” one example of which is television soap opera.
We felt that this is a good domain to look at both for scientific and
practical reasons. Television melodramas are watched by large
All Work and No Play
numbers of people over very long periods of time. From a scientific
Makes HAL a Dull Program
point of view, stories based on interpersonal relations have received
much less attention than those that are more action oriented. Finally,
Michael Lebowitz was a stories of this sort combine complexity and creativity with a good
\ faculty member from 1980 to
deal of stereotypy. They help us get a handle on which parts of
1987 in the Department of
Computer Science at Columbia creativity are not too creative. Here is an outline of events in a typical
University. His primary melodrama (NBC's “Days of Our Lives”).
research interests lie in the Liz was married to Tony. Neither loved the other, and
areas of machine learning,
indeed, Liz was in love with Neil. However, unknown to either Tony
natural-language processing,
memory organization, and or Neil, Stephano, Tony's father, who wanted Liz to produce a
cognitive modeling. He also led
a research group at Columbia
that designed intelligent
information systems that could
read, remember, and learn from
natural-language text. He has
published many articles ina
wide range of areas in artificial
intelligence. He is currently a
vice president of the Analytical
Proprietary Trading unit at
Morgan Stanley & Co. in New
York City.

Michael Lebowitz (Photo by Betsy

Malcolm)
grandson for him, threatened Liz that if she left Tony, he would kill Enter trait and value (cr = quit): niceness 3
Neil. Convinced that he was serious by a bomb that exploded near Enter trait and value (cr = quit): self-confidence 4
Neil, Liz told Neil that she did not love him, that she was still in love Enter trait and vaiue (cr = quit): intelligence 3
with Tony, and that he should forget about her. Neil was eventually Enter trait and value (cr = quit):

convinced and married Marie. Later when Liz was finally free from Optimal occupations are: sleazy-doctor
Tony (because Stephano had died), Neil was not free to marry her, Enter occupation:
and their troubles went on. Selecting: sleazy-doctor
As part of our research we developed a simple, prototype Trait (intelligence 3) already satisfied
story-telling program, Universe. The program can create sets of For trait (self-confidence 4)
characters appropriate for interpersonal melodrama and can Choices (life-guard, swinger, big-eater): swinger
generate simple plot outlines of about the complexity of this outline For trait (niceness 3)
using the characters that it generates.’ Choices (taxidriver):

The Al aspects of the program arise in the representation

Assisting Writers
of the information and where the program suggests possibilities for
Se
the character being created. These suggestions are intended to help
In the immediate future it is unlikely that we will be able to create
make the character more believable and interesting by appropriately
programs that produce melodrama with the richness of human
motivating the characteristics that the user wants. In the first section
writers. However, we can hope to provide tools to assist in the
of the display, Universe prompts the user for a few vital statistics of
creation of such works. In particular, | feel that the creation of an
the character being created: sex, name, age group (young adult,
automated assistant that can suggest possibilities to a writer is well
middle age, old). These are used to begin filling in a frame for the
within reach. We can create programs that help a writer develop
character John Jones.’ In all cases, if no answer is provided, the
meaningful characters and that suggest plot possibilities. | shall
program will pick one.
illustrate a simple version of one such program below.
After getting the basic facts, the program begins to
In a sense, we can view a program of the sort | envision
“cycle” the character's life, that is, to prompt the user for major life
as a kind of intelligent story editor, the next step beyond text editing.
events, concentrating on marriages, divorces, the births of children,
Such a system would have to be embedded in a complete develop-
and deaths. Other parts of the program allow other sorts of history to
ment system that would also keep track of the history of the story over
be added. Note that by including past events and not just the current
time. The display below illustrates the kind of behavior | have in
state of the world, Universe helps the user build a consistent and
mind. It shows Universe running in an interactive mode in which
coherent set of characters. In the display the user specified that John
instead of creating characters itself, it prompts a user for the relevant
Jones had gotten married (m) in 1970. The program knew that a
information. Items in italics were entered by the user.
spouse was necessary, and the user let the program pick her name
Enter sex (m/f): m (Kathy Cole).
Enter name: John Jones The final section of the display is the most Al oriented. It
Enter age group (ya/ma/old): ma involves filling in a personality profile for the character. Universe
Cycling person John Jones knows about a variety of different personality traits. Here the user
Enter year of next life event (cr = quit): 1970 suggests a set of values to describe John Jones: he is not very nice, is
Next event (m/d): m moderately self-confident, and moderately intelligent. Just specifying
Try to find spouse in universe (u/n): n values, however, is not very satisfying, so the program assists the
Enter name (list): user in selecting various prototypical character types that will help
Cycling marriage between John Jones and Kathy Cole explain the trait values. It allows the user to either accept a proffered
Enter year of next marriage event (cr = quit): prototype or provide one. It suggests that sleazy doctor would be a
Beginning finish up for John Jones character type that meshes with these traits. It also suggests other
Pick traits (p/r): p prototypes that help explain the traits. Between the program and the
Possible traits to set: wealth, promiscuity, competence, niceness, user they come up with a character that is a swinging sleazy doctor
self-confidence, guile, naivete, moodiness, physical attractiveness, and taxidriver. Needless to say, the quality of the system's help
intelligence
depends on the quality of the prototype library. Such a description is action oriented (even though at least one was modeled after a
the sort the writer would have to come up with anyway to make the television melodrama).
character one a reader/user will believe? It is important to note that what we are after is more than
Tools like the one described here (with plot management simply games. We seek the depth and quality of fiction along with the
included) can be of great use to writers of all sorts. They serve as interactive nature of games. Although the stories themselves would
partners by suggesting ideas but leave the real creativity to the be generated by computer, the knowledge bases from which they
writer. They will be particularly important in developing various worked— information about people, places, and plot devices—would
kinds of interactive fiction but should also be of use in more presumably be built by human writers, although not by specifying
traditional forms. One can easily imagine how an Al tool could greatly every plot turn, as is done with interactive games.
help the writers of a television serial that has been on the air for To date there has not been a great deal of work on
many years keep everything consistent and at the same time suggest automatic story generation. Perhaps the best known work is that
new possibilities. done by Jim Meehan in the mid 1970s when he developed Tale-Spin,
a program that told children’s stories in the style of Aesop's fables.‘
Personalized Stories Other workers have also been working on Al story telling.> Our own
ase
work on Universe has taken us to the point where it can generate
Beyond the creation of intelligent story-writing tools, the obvious simple plot outlines by applying planning techniques to author goals.°
next step in applying Al to story telling is the automatic generation of Below is a simple plot outline created by Universe. It uses
stories. With human writers available, it may not be clear why we characters created by the automatic version of the interactive
would want to do this (beyond the goal of understanding how program described in the previous section after being given as input
creative processes work). The reason is the same as for many other an author goal, here to “churn” a relationship, that is, to keep two
tasks: the application of Al techniques to story telling will lead to an people in love from being happy together (a staple of melodrama).
art form that is personalized and interactive. Imagine a soap opera Since Teddy Bryks, the husband of one of the lovers, has an evil
with characters that are exactly the kind you would find most father, Universe selects a plot fragment that involves the father's
interesting, not necessarily a set of characters that you create (after threatening the women not to betray her husband (which is loosely
all, we are not all writers), but one molded to your own tastes. In based at an abstract level on the outline above). The various elements
addition, imagine that you can influence the unfolding of events—not of this plot are expanded using other plot fragments. By the time the
necessarily directing the characters in detail, but influencing the first fragment ends, the spurned lover, Steven Kades, is involved with
story in more subtle ways. It is a compelling vision of a new fictional someone else, which creates further complications. Universe runs
form. Stories in textual form should be exciting, and the addition of through a couple more of its library of ways to keep people apart.
graphics and other computer-manipulated images should make it The setting generated by the program as part of a set of
even more impressive. characters is as follows: Nadine Burton is married to Teddy Bryks but
Stories of this kind will evolve from two current lines of is in love with Steven Kades. Gerald Bryks is Teddy's father, and none
entertainment: interactive computer games and role-playing games. too nice a guy. (Elaborative comments are enclosed in parentheses.)
Several manufacturers are currently producing interactive computer
games in which users work their way through a branching story, Churn Nadine and Steven
een
typically solving a series of puzzles along the way. Many such games
are quite intriguing. However, all the various outcomes have to be Gerald Bryks threatens Nadine Burton Bryks: forget it (her affair with
handcrafted by a programmer and as a result are relatively limited in Steven). Nadine Burton Bryks tells Steven Kades that she doesn't love
the world to be explored. In addition, programmers have tended to him. Norma Ryan Bryks is worried about Steven Kades. Linda
focus on action-oriented and mystery domains with little character Einbinder Kimball seduces Steven Kades. Joe Kimball decides to
development. expose Gerald Bryks. Joe Kimball tells the world about Gerald Bryks
Such role-playing games as Dungeons and Dragons come (and his evil doings). Nadine Burton Bryks and Teddy Bryks get
closer to the model | have in mind. Here with a human game master a divorced. Teddy Bryks tries to seduce Nadine Burton Bryks. Steven
story can move through a very wide range of possibilities. Yet the Kades gets frustrated (by this). Steven Kades tells Nadine Burton
game master is required, and these games too have tended to be quite Bryks that he doesn’t love her. Nadine Burton Bryks seduces Steven

Kades.

392
 This story is certainly not intended to be great fiction. 3. For further details on Universe's methods of character creation, see M. Lebowitz,

However, our program has shown us that even with a very limited “Creating Characters in a Story-Telling Universe,” Poetics 13 (1984): 171-194.
library of plot fragments (about 60), through interactions of plots and 4, See J. R. Meehan, “The Metanovel: Writing Stories by Computer,” Yale University
characters, Universe can produce some rather clever plot outlines. A
Department of Computer Science, technical report 74, 1976.
much larger library (perhaps by two orders of magnitude) should be
5. See N. Dehn, “Memory in Story Invention,” Proceedings of the Third Annual
Conference of the Cognitive Science Society (Berkeley, Calif, 1981), pp. 213-215; M.
practical and would be able to produce a large number of interesting
Yazdani, “Generating Events in a Fictional World of Stories,” University of Exeter
stories, particularly if there are techniques to automatically create
Computer Science Department, technical report R-113; and S. R. Turner and M. G.
new fragments.’ No small part of the effect is based on the same trick
Dyer, “Thematic Knowledge, Episodic Memory, and Analogy in MINSTREL, A Story
that authors of standard fiction use: the reader's imagination will
Invention System,” Proceedings of the Seventh Annual Conference of the Cognitive
enhance what is actually presented. Science Society (Irvine , Calif., 1985), pp. 371-375.
Our work on Universe has also given us some interesting 6. See M. Lebowitz, “Story-Telling as Planning and Learning,” Poetics 14 (1985):

insights into creativity, which | can only touch upon here. It appears 483-502.

that many parts of creativity involve primarily determining clever 7. See M. Lebowitz, “Story-Telling as Planning and Learning,” Poetics 14 (1985):

ways to apply previous situations (or abstracted versions of them) to 483-502.

new settings. In story telling, it is rarely necessary to create whole 8. See K. J. Hammond, “Planning and Goal Interaction: The Use of Past Solutions in

new ideas: one can take old ideas and combine them in new ways or Present Situations,” Proceedings of the Third National Conference on Artificial

apply new twists to them. This is certainly evident in television Intelligence (Washington, D.C., 1983), pp. 148-151.

melodrama and is by no means bad: even plot turns that are familiar
in the abstract can be quite interesting when applied to new
situations. Although there are many other aspects to creativity, a
crucial part is storing previous experiences in a way that they can be
efficiently retrieved and applied in the future. This applies to both
story telling and general day-to-day planning.’
We are just beginning the process of automatically
generating stories. There are many problems involved in creating plot
outlines, and beyond them there are issues in language generation,
knowledge representation, knowledge-state assessment (who knows
what when), memory organization and access, and user interaction,
among many others that must be dealt with to achieve the sort of
system envisioned at the beginning of this section.
Itis no doubt appropriate for most Al work to address
practical considerations. Indeed, most of our own research does just
that. However, people should know that Al will enhance other
aspects of their lives than just the workplace. Al can help creative
people make better use of their talents and create interesting and
entertaining new art forms. After all, if Al is going to help improve
productivity, then it had better also help fill the leisure time thus
created—and it can.

Notes
—

1. For further details, see M. Lebowitz, “Creating Characters in Story-Telling

Universe, ” Poetics 13 (1984): 171-194; and M. Lebowitz, “Story-Telling as Planning
and Learning,” Poetics14 (1985): 483-502.
2. For more on frames, see M. Minsky, “A Framework for Representing Knowledge,”
in P. H. Winston, ed., The Psychology of Computer Vision (New York: McGraw-Hill,
1975).
 What exactly is creativity? Could a machine ever be creative? These
are questions psychologists, philosophers, and Al researchers would
all like to be able to answer. But what kind of answers are we
looking for? The search for a rigorous philosophical definition of
creativity has been overworked, and we don't intend to further pursue
that course here. On the other hand, redefining the question in Al
terms and applying Al research methods might result in a new and
useful kind of answer, or at least an interesting set of new questions
to consider.
A persistent criticism of Al work has centered around
arguments like, “By its very nature, a machine could never really be
creative.” Since a basic assumption underlying work in computer
science is that a machine can perform any task that can be described
via sufficiently specific rules, people who make statements like the
above mean that no rules can ever be found that will account for
creativity and other quintessentially human behavior, that there is

I Roger Schank and Christopher Owens something inherently mystical in these abilities that cannot be
expressed via rules and procedures. Or else they might mean that
even if such a set of rules and procedures could be found, a machine
The Mechanics of Creativity
that was obeying them would only seem to be creative. Its behavior,
they say, would be a kind of elaborate parlor trick; it would be
Roger C. Schank directs the Institute for the achieving its effect merely by fooling us.
Learning Sciences at Northwestern University,
where he is also John Evans Professor of But as Al researchers and cognitive scientists, our work

Electrical Engineering and Computer Science, is based upon the assumption that rules and procedures underlying
Psychology, and Education. Previously, he was human behavior can be found. Our job is to define problems in such a
Chairman of the Computer Science department at
way as to maximize our chances of succeeding in this endeavor. Our
Yale University and Director of the Yale Artificial
Intelligence Project. In addition, he was Assistant goal is to come up with an algorithmic definition of creativity, a set of

Professor of Linguistics and Computer Science at processes and steps that can account for the kind of creative thinking
Stanford University. Schank holds a Ph.D in that we observe in people. Although the idea of a human or machine
Linguistics from the University of Texas at Austin. exhibiting creativity by following a set of rules seems on the face to
He is the founderof two businesses, Compu-Teach,
an educational software company, and Cognitive be a contradiction, this is not necessarily so. If we can agree on some

Systems, a company specializing in natural kinds of behavior that constitute creative thinking and can develop an
language processing. An internationally acclaimed algorithmic model that accounts for these behaviors, then we have an
researcher in artificial intelligence, Schank is the
algorithmic theory of creativity and hence a first step toward creative
author of numerous articles and books, including
Dynamic Memory; Scripts, Plans, Goals, and machines. Whether or not a philosopher would agree that the

Understanding with Robert Abelson; and The resulting machine truly embodied creativity is almost irrelevant to us:
Cognitive Computer and The Creative Attitude with building machines that act in ways that appear
to be creative would
Peter Childers. Christopher Owens is engaged in Al
be a significant enough step to undertake.
research at the Artificial Intelligence Lab at Yale
University, where he is currently completing a Creativity is often associated with problem solving,

Ph.D. His primary interests are studying the science, and the arts. People often view creative thinking as
organization of human memory and applying the something out of the ordinary, as a mode of reasoning in which
principles thereof to the task of making machines completed thoughts suddenly spring to mind without being cued,
smarter. His work focuses on people's ability to
thoughts perhaps having nothing at all to do with what the thinker
reuse old knowledge to solve new problems,
specifically the kind of frozen, culturally shared was working on at the time the thought occurred. Often people
knowledge typified by the planning advice given in, implicitly assume that creativity represents some divine, uncon-
proverbs and other folk adages. scious, or other inspiration out of the control of one’s ordinary thought

394
processes. Actual case studies of scientific and artistic creativity,
however, support the idea that creativity springs not from any
mystical source but from a certain set of cognitive skills. There is no
principled distinction between creative and less creative thinking
other than the degree to which this set of skills is applied. Highly
creative individuals simply have these skills better developed than do
the less creative. What, then, are these cognitive skills? How are they
used? How can we program a computer to exhibit them? These are
questions that we can study more fruitfully than the open-ended type
of question, “What is Creativity?” with which this chapter opened.
In our view, the basic cognitive skill underlying creativity
is the ability to intelligently misapply things. A creative solution to a
problem is one that uses an object, technique, or tool in a useful and Roger Schank and Christopher
Owens (Courtesy of Charles Martin
previously undiscovered way. A creative work of art, similarly, might
of the Yale Al Lab)
use some material or image in a new way. A creative piece of
scientific research might involve, for example, applying a principle
from one field to a problem in another. At Yale we are studying the
cognitive skills underlying one particular type of creative behavior,
the creation of novel explanations for unexpected events. Explana-
tion is a kind of problem solving in which the problem is of a mental
nature: “How can | connect this new and strange piece of knowledge
with the rest of my knowledge so that it all makes sense,” or “What
pattern of events do | know about into which | can fit this new fact.”
Of course, by this definition, many kinds of understanding
can be seen as explanation, in that all understanding consists of
integrating new facts into existing knowledge. But what we are
interested in here is the kind of explanation that requires conscious
work, the explanation of events that are at first puzzling. This kind of
explanation may require searching for a missing piece of knowledge,
or it may require finding some previously unseen causal connection.
Often an explanation can be found by seeing one object or event as
similar to another in a way that was not previously noticed, in other
words, by making up and using a novel analogy. For example, when
we asked people in the Yale Al lab to try to explain the unexpected
death of the successful three-year-old race horse Swale, one student
was reminded of the death of Jim Fixx, the runner. He reasoned that
Swale was like Fixx in that both participated in regular strenuous
activity, and that possibly Swale, also like Fixx, had a congenital
heart defect.
This kind of reasoning from prior examples is very
important to our approach to explanation. Although an understanding
system could conceivably explain each new situation by reasoning
from first principles, chaining together all the rules and pieces of
knowledge it needed to build the entire explanation from small
elements, this probably does not happen very often. One reason is that

395
the computational complexity of this task is unreasonably large. which weird remindings can be generated. Creativity, according to
Another is that people seem to be able to use remindings and this view, consists of being tolerant of inappropriate remindings, in
analogical reasoning to construct explanations: they can use the being slow to discard an erroneous idea. But that tolerance is only
same explanation over and over to cover a range of similar and half the process. Because processing power is finite, the key is to be
thematically related situations. intelligent about choosing which inappropriate XP to misapply.
For a second example of this kind of reasoning, consider Creativity does not lie in floundering through memory, trying one
the folk use of proverbs, which can be viewed as a kind of extremely randomly selected idea after the next; it lies in finding near misses
abstract pattern used by people to explain unfamiliar situations in a that are reasonably close to the right idea and fixing them to fit.
familiar way. When someone standing in the rain beside his disabled This approach puts a large demand on memory, since the
car and wishing he had had that overdue tune-up analyzes the retrieval task is no longer simply to find the closest fit from among a
situation by saying “A stitch in time saves nine,” he has placed the library of XPs (which selection, by the way, is an important task itself,
situation in a context in which knowledge about topics like and we do not mean to denigrate its difficulty here). The task of
prevention and the bad etrects of not taking precautions is available. memory is now to fail gracefully: to find the closest fit, or if a close fit
The causal reasoning represented within this proverb is available is not available, to find a near miss that nevertheless captures some
without the effort of building an analysis or explanation from scratch. important and relevant aspects of the situation being explained. This
In a manner similar to the way people might use proverbs, kind of near miss is the most likely candidate for modification.
our systems store and reuse old explanations using a knowledge Along with a means for getting reasonable near misses,
structure called an Explanation Pattern, or XP. Like a proverb, an XP this approach also requires that the system, when presented with an
is a frozen piece of causal reasoning. Because it is designed to be inappropriate XP, be able to analyze what is wrong with it and to
retrieved, modified, and applied by a computer program, it has the select an appropriate repair strategy based upon that analysis. The
following basic components: student who explained Swale’s death in terms of Jim Fixx's knew
that Jim Fixx was not a race horse, that the explanation would not
A characterization of a set of situations to which it is likely to apply,
directly apply without making the connection between a runner and a
for example, deaths of athletes
race horse. People do this so easily that we hardly think about it, yet
A characterization of a broader set of situations to which, even if it
the task is difficult. This modification of inappropriate XPs in order to
does not apply, it is likely to be at least relevant, for example,
adapt them to new situations, which we have been calling “tweak-
unexpected bad outcomes
ing,” is central to being a creative explainer.
A causally annotated description of the event being explained. For Our algorithm for creativity must therefore embody three
example: Running involves physical exertion. Physical exertion processes: a means of searching through memory for applicable
strains the heart. Straining the heart combined with a heart defect patterns that returns a reasonable set of near misses, a means of
can cause a heart attack. A heart attack can cause death. evaluating the near misses and seeing what is wrong with them, and
a means of modifying those inappropriate patterns to suit the current
Since we are viewing explanation as a kind of problem
situation. Of course, for any of this to work reasonably well, our
solving, and since a creative solution to a problem is one that uses an
creative machine must have a rich memory of facts, experiences, and
object, technique, or tool in a useful and previously undiscovered
relationships it can draw upon as starting points for new explana-
way, a creative explanation is one that uses an XP in a novel and
tions. Searching for and adapting patterns is a reasonable strategy
previously unencountered way. If the basic idea of an explanation
only if the library of patterns is large enough that the near misses will
system is to select a relevant XP from memory and apply it to a new
nevertheless have at least something in common with the episode
episode, then the basic idea of a creative explanation system is to
being explained. Our fourth requirement, therefore, is to have a large
intelligently misapply XPs; to try, if no relevant XP can be found, to
and richly indexed memory of explanation patterns and other
modify and apply an XP that, although at first seemingly irrelevant,
knowledge gained from experience. How these patterns are learned
might nevertheless bear some thematic relationship to the episode
is another interesting problem to be attacked.
being explained.
So we have transformed the questions “What is creativ-
The idea of using near misses is an important one. Often
ity?” and “Can a machine ever be creative?” into “What cognitive
people associate creativity with a simple relaxation of constraints on
skills underlie creative behavior?” and “How can we program these
retrieval and pattern matching, with some kind of random process by
skills into a computer?” Further refining the questions, we have
discussed creativity, which at first seems mystical and quintessen-

396
tially human, in terms of knowledge structures, search, retrieval, and
adaptation, which seem quintessentially mechanistic. Creative
explanation, we claim, is not so mysterious. It depends upon having a
stock set of explanations, some heuristics for finding them at the
right time, and some rules for tweaking them after they have been
found. These problems are, of course, not as simple as this quick
description might imply; each of them represents an important
research area that is yet to be properly explored. Although trying to
reduce creativity to mechanical processes might seem to be a kind of
cheapening or demystification of human capabilities, that is not the
case. The philosophy underlying Al research is that there must
necessarily be mechanically understandable processes underlying
intelligent behavior, and that our general purpose should be to
deepen our understanding of human capabilities by defining and
studying these mechanisms.

397
Visions of the Future
(Photo by Lou Jones)
 Visions

I never think of the future, it comes soon enough.

Albert Einstein

What is possible we can do now, what is impossible will take a little longer.
A modern-day proverb

Every electronic product being sold today is obsolete.

Fred Zieber, Senior Vice President, Dataquest

The problems of the world cannot possibly be solved by skeptics or cynics whose horizons are limited by
the obvious realities. We need men who can dream of things that never were.
JohnF. Kennedy

* Scenarios

Since the founding of the computer industry almost half a century ago, one of its most
salient and consistent features has been change. Functionality per unit cost has been
increasing exponentially for decades, a trend that shows no sign of abating. When |
attended MIT in the late 1960s, thousands of students and professors shared a single
computer, an IBM 7094 with 294,912 bytes of core storage (organized as 65,536
words of 36 bits each) and a speed of about 250,000 instructions per second. One
needed considerable influence to obtain more than a few seconds of computer time
per day. Today, one can buy a personal computer with ten times the speed and
memory for a few thousand dollars. In Metamagical Themas, Doug Hofstadter cites
an actual job that took 10 people with electromechanical calculators ten months to
perform in the early 1940s, was redone on an IBM 704 in the early 1960s in 20
minutes, and now would take only a few seconds on a personal computer.' David
Waltz points out that memory today, after adjustment for inflation, costs only one one-
hundred millionth of what it did in 1950.2 If the automotive industry made as much
progress in the past two decades, a typical automobile today would cost about two
dollars (the doubling of price performance every 22 months on average has resulted in
an improvement factor of about 2,000 in 20 years; this is comparable to the difference
 between the 7094 of the late 1960s and a personal computer with a Intel 80386 chip
today).° If we go back to the first relay-based computers, a personal computer today is
nearly a million times faster at a tiny fraction ofthe cost. Many other examples of
such progress abound.*
In addition to the basic power of computation as measured by speed and
memory capacity, new hardware and software technologies have greatly improved
our ability to interact with computer devices. Through the 1940s and 1950s most
communication with computers was through boards with plug-in cables; through the
1960s and 1970s, with reels of punched paper tape, stacks of punched paper cards,
and print-outs from line printers. Today the advent of high resolution graphic displays, |
the mouse, graphics tablets, laser printers, optical cameras, scanners, voice recogni-
tion, and other technologies have provided a myriad of ways for humans and ma-
chines to communicate
Advances in software have harnessed these increasingly potent hardware
resources to expand the productivity of most professions. Twenty years ago comput-
ers were used primarily by large corporations for transaction processing and by
scientists (occasionally by computer scientists to explore the power of computing)
Today most workers—professionals, office workers, factory workers, farmers—have
many occasions to use methods that rely on the computer. | can recall that fifteen
years ago even thinking about changing my company’s business projections was
regarded as a very serious endeavor; it would take the finance department days to
grind through the numbers to examine a single scenario. Today with spreadsheet
programs it is possible to consider a dozen alternative plans and determine their
implications in less than an hour. Twenty years ago the only people interacting with
computers were computer experts and a small cadre of students learning the some-
what arcane new field of computation. Today computers appear ubiquitously on office
desks, in kitchens, in play rooms, in grocery stores, and in elementary schools.

“The computer chip: inscribing

worlds on grains of sand"—George
Gilder (Photo by Dan McCoy of
Rainbow)

402
Computer-aided design (CAD) for
computer chips. A chip with photo
Will these trends continue? Some observers have pointed out that an
sensors being designed at Cal Tech
using technology from Synaptics. exponential trend cannot continue forever. If a species, for example, happens upon a
(Photo by Dan McCoy of Rainbow) hospitable new environmental niche, it may multiply and expand its population
exponentially for a period of time, but eventually its own numbers exhaust the
available food supply or other resources and the expansion halts or even reverses. On
this basis, some feel that after four decades, exponential improvement in the power
of computing cannot go on for much longer. Predicting the end of this trend is, in my
view, highly premature. It is, of course, possible that we will eventually reach a time
when the rate of improvement slows down, but it does not appear that we are
anywhere close to reaching that point. There are more than enough new computing
technologies being developed to assure a continuation of the doubling of price
performance (the level of performance per unit cost) every 18 to 24 months for many
years.
With just conventional materials and methodologies, progress in the next
ten years, at least in terms of computing speeds and memory densities, seems
relatively assured. Indeed, chips with 64 million bits of RAM (random access memory)
and processor chips sporting speeds of 100 million instructions per second are on the
drawing board now and likely to be available in the early 1990s. Parallel-processing
architectures, some including the use of analog computation, are an additional means
of expanding the power of computers. Beyond the conventional methods, a broad
variety of experimental techniques could further accelerate these trends. Supercon-
ducting, for example, while challenging to implement in a practical way, has the
potential to break the thermal barrier that currently constrains chip geometries. As |
mentioned earlier, the resulting combination of smaller component geometries with
the effective utilization of the third dimension could provide a millionfold improvement
in computer power. A variety of new materials, such as gallium arsenide, also have
the potential to substantially improve the speed and density of electronic circuits.>
And optical circuits—computing with light rather than electricity—may multiply
computing speeds by factors of many thousands.®
Will software keep up? It is often said that the pace of advances in software
engineering and applications lags behind that of the startling advance of hardware
technology. Advances in software are perhaps more evolutionary than revolutionary,
but in many instances software techniques are already available that are just waiting
for sufficiently powerful hardware to make them practical. For example, techniques
for large-vocabulary speech recognition can be adapted to recognize continuous
speech but require substantially greater computational speed. Vision is another
application with the same requirement. There are many techniques and algorithms
that are already understood but are waiting for more powerful computers to make
them economically feasible.” In the meantime, our understanding of Al methods, the
sophistication of our knowledge bases, the power of our pattern-recognition technolo-
gies, and many other facets of Al software continue to grow.
Where is all this taking us? People in the computer field are accustomed to
hearing about the rapidly improving speed and density of semiconductors. People in
other professions inevitably hear reports of the same progress. Numbing are the
extremely small numbers used to measure computer timings and the enormous
numbers used for memory capacity, time measured in trillionths of a second and
memory in billions of characters. What impact are these developments going to have?
How will society change? How will our daily lives change? What problems will be
solved or created?
One can take several approaches in attempting to answer these questions.
Perhaps most instructive is to consider specific examples of devices and scenarios
that have the potential to profoundly change the way we communicate, learn, live,
and work. These concrete examples represent only a few of the ways that computer
and other advanced technologies will shape our future world. These examples are
based on trends that are already apparent. In my view, it is virtually certain (barring a
world calamity) that all of these scenarios will take place. The only uncertainty is
precisely when. | will attempt to project current trends into the future and estimate
when we are likely to see each example.
Obviously, the further into the future we look, the more uncertain the timing
of these projections become. The history of Al is replete with examples of problems
that were either underestimated or (less often) overestimated. A great irony in early Al
history is that many of the problems thought most difficult—proving original theo-
rems, playing chess—turned out to be easy, while the “easy” problems—pattern-
recognition tasks that even a child can perform—turned out to be the most challeng-
ing.* Nonetheless, | believe that we now have a more sophisticated appreciation of
the difficulty of many of these problems, and so | will attempt the thankless task of
making specific projections. Of course, by the time you discover that my predictions
were altogether wrong, it will be too late to obtain a refund for the purchase price of
this book.

404
 As | mentioned, these projections are based on trends that are already
evident. What is most difficult to anticipate are breakthroughs. Any attempts to have
predicted the future at the beginning of this century would have almost certainly
overlooked the computer, as well as atomic energy, television, the laser, and indeed,
most of electronics. After going through the scenarios, | shall discuss some possible
breakthroughs that may result from current research. In the following chapter | offer a
discussion of the overall impact these developments are likely to have on our educa-
tional, social, political, medical, military, and economic institutions.

The translating telephone

Koji Kobayashi, chairman of the powerful Japanese corporation NEC, has a dream.
Someday, according to Kobayashi, people will be able to call anyone in the world and
talk, regardless of the language that they speak. The words will be translated from
language to language in real time as we speak.°
Three technologies are required to achieve Kobayashi’s dream: automatic
speech recognition (ASR), language translation (LT), and speech synthesis (SS). All
three exist today, but not nearly in sufficiently advanced form. Let us consider each of
these requirements.
ASR would have to be at the “holy grail” level, that is, combining large (rela-
tively unrestricted) vocabulary, accepting continuous speech input, and providing
speaker independence (no training of the system on each voice). Conceivably, the last
requirement could be eased in early versions of this system. Users of this capability
may be willing to spend fifteen minutes or so training the system on their voice. Such
enrollment would be required only once. Combining the first two elements—large
vocabulary and continuous speech—will take us to the early to mid 1990s. Adding
speaker independence will take us another several years
LT requires only the ability to translate text, not speech, since ASR technol-
ogy would be translating speech input into written language. The LT capability would

The translating telephone.

 not require literary-quality translations, but it would have to perform unassisted. LT
systems today require human assistance. Completely automatic LT of sufficient
quality will probably become available around the same time that the requisite ASR is
available. It should be noted that every pair of languages requires different software
(going from French to English is a different problem from going from English to
French). While many aspects of translation will be similar from one set of languages
to another, LT technology will vary in quality and availability according to the lan-
guages involved
SS is the easiest of the three technologies Kobayashi requires. In fact, it is
available today. While not entirely natural, synthetic speech generated by the better
synthesizers is quite comprehensible without training. Naturalness is improving, and
SS systems should be entirely adequate by the time the necessary ASR and LT
systems are available
Thus, we can expect translating telephones with reasonable levels of
performance for at least the more popular languages early in the first decade of the
next century. With continuing improvements in performance and reductions in cost,

Ih such services could become widespread by the end of that decade. The impact will
be another major step in achieving the “global village” envisioned by Marshall
McLuhan (1911-1980) over two decades ago.'? Overcoming the language barrier will
result in a more tightly integrated world economy and society. We shall be able to talk
more easily to more people, but our ability to misunderstand each other will remain
undisturbed

The intelligent assistant

You are considering purchasing an expensive item of capital equipment and wish to
analyze the different means of financing available. Issues to consider are your
company’s current balance sheet, other anticipated cash flow requirements, the state
of various financial markets, and future financial projections. You ask your intelligent
assistant to write a report that proposes the most reasonable methods to finance the
purchase and analyzes the impact of each. The computer engages you in sufficient
spoken dialogue to clarify the request and then proceeds to conduct its study. In the
course of its research, it accesses the balance sheet and financial projections from the
data bases of your company. It contacts the Dow Jones data base by telephone to
obtain information on current financial markets. It makes several calls to other com-
puters to obtain the most recent financing charges for different financial instruments
In one case, it speaks to a human to clarify certain details on a lease-repurchase plan.
It speaks with your company’s vice president of marketing to obtain her level of confi-
dence in achieving the sales projections for the following two years. It then organizes
and presents the information in a written report complete with color charts. The report
is presented to you the following day, since it took that long to reach the two humans
involved. (Had it been able to conduct the research through communication only with
other computers, it would have required only a few minutes.) Other services provided
by your computerized assistant include keeping careful track of your schedule,
including planning your travel from one appointment tc another. The system plans
your work for you, doing as much of it itself as it is capable of and understanding what
portion of it you need to do yourself.

406
 When we shall see the above system depends on how intelligent an
assistant we would like to have. Crude forerunners exist today. Large-vocabulary ASR
has been integrated with natural-language understanding and data-base-management
programs to provide systems that can respond to such commands (posed by voice) .
as, Compare the sales of our western region to our three largest competitors. Such
systems are, of course, highly limited in their problem-solving ability, but efforts to
integrate ASR with data-base access have already begun.
The most challenging aspect of the vision is problem solving, having
sufficient commonsense knowledge and reasoning ability to understand what infor-
mation is required to solve a particular problem. Required are expert systems in many
areas of endeavor that are less narrowly focused than the expert systems of today.
One of the first intelligent assistants is likely to be one that helps get information from
data bases through telecommunications."' It has become clear to a number of soft-
ware developers that a need exists to improve substantially the ease of accessing
information from such data-base systems as Compuserve, Delphi, The Source, Dialog,
Dow Jones, Lexis, Nexis, and others. Such data-base systems are greatly expanding
the volume and diversity of information available, but most persons do not know
where to find the appropriate information they need. The first generation of office-
assistant programs are now being developed that know how to obtain a broad variety
of information without requiring precisely stated requests. | expect that within several
years such systems will be generally available, and some of them will take ASR for
input.
Thus, in the early to mid 1990s we shall see at least part of the above vision
in use: flexible access to information from increasingly varied information services
around the world, accessed by systems that understand human speech as well as the
syntax and (at least to some extent) the semantics of natural language. They will
support their own data bases and be able to access organization-specific knowledge
You will be able to obtain information in a flexible manner without having to know
which data-base service has what information or how to use any particular information
utility. As the 1990s progress, these systems will be integrated with problem-solving
expert systems in many areas of endeavor. The level of intelligence implied in the
above scenario describing a capital-equipment purchase will probably be seen during
the first decade of the next century.

The world chess championship

As noted earlier, the best machine chess players are now competing successfully at
the national senior-master level, regularly defeating all but about 700 players.'* All
chess machines use some variant of the recursive algorithm called minimax, a
strategy whose computational requirements are multiplied by some constant for each
additional move ahead that is analyzed. Without a totally new approach, we thus need
to make exponential progress in computational power to make linear gains in game-
playing performance (though we are indeed making exponential progress in hard-
ware). The analysis | gave before estimated that the requisite computer power to
achieve world-championship chess playing should become available between 9 and
54 years from now. This estimate was based on the continuing gains anticipated in
the speeds of individual microprocessors. If we factor in the increasing popularity of

407
An endangered species? Two human
world chess champions, Gary
Kasparov (right) and Anatoly Karpov,
battle it out in 1986. The best human
chess playing has remained relatively
constant in performance, while
computer chess playing is rapidly
improving. How much longer will the
world chess champion
be a human? (Supplied by AP/Wide
World Photos)

parallel-processing architectures, the result will be much closer to the short end of
this range. Some of the other scenarios in this section require significant advances in
both hardware power and software sophistication. In my view, the ability of a machine
to play championship chess is primarily a function of the former. Some of the possible
breakthroughs in electronic hardware discussed below will be directly applicable to
the chess issue. For example, if we are successful in harnessing the third dimension
in chip fabrication (that is, building integrated circuits with hundreds or thousands of
layers of active circuitry rather than just one), we will see a major improvement in
parallel processing: hundreds or thousands of processors on a single chip. Taking into
consideration only anticipated progress in conventional circuit-fabrication methodolo-
gies and continued development of parallel-processing architectures, | feel that a
computer world chess champion is a reasonable expectation by the end of the
century

What will be the impact of such a development? For many, such as myself,
it will simply be the passing of a long anticipated milestone. Yes, chess is an intelli-
gent game (that is, it requires intelligence to play well), but it represents a type of in-
telligence that is particularly well suited to the strengths of early machine intelligence,
what | earlier called level-2 intelligence (see “The Recursive Formula and Three Levels

The march of machine intelligence 3006

toward the world chess champion- pion HiTech
2600 senior r Belle
Ht 2300 23595) 2400
ship. Computers are improving in 240 0 master Chess 46 =o _ ]
their chess-playing prowess. In 1987 expert | nec, Ohess 30 1900
HiTech defeated a series of national z 1500 “
senior masters who were rated at a
over 2,400. Computers now have less 000 Ps |
than 400 points to go to capture the |
world championship. (Data from The
Computer Museum Report [Boston], Dl Gt 4t bo bo ols ea ee
summer-fall 1987, p. 7) E 2 3 8 Soe e oo

40s
of Intelligence”). While level-3 intelligence will certainly benefit from the increasing
power of computer hardware, it will also require substantial improvements in the
ability of computers to manipulate abstract concepts.
Defenders of human chess playing often say that though computers may
eventually defeat all human players, computers are not able to use the more abstract
and intuitive methods that humans use."? For example, people can eliminate from
consideration certain pieces that obviously have no bearing on the current strategic
situation and thus do not need to consider sequences of moves involving those
pieces. Humans are also able to draw upon a wealth of experience of previous similar
situations. However, neither of these abilities is inconsistent with the recursive
algorithm. The ability to eliminate from consideration branches of the expanding tree
of move-countermove possibilities not worth pursuing is an important part, called
pruning, of any minimax program. Drawing upon a data base of previous board
Positions is also a common strategy in the more advanced chess programs (particu-
larly in the early game). It is estimated that human chess masters have memorized be-
tween 20,000 and 50,000 chess boards.'* While impressive, it is clear that this is
again an area where machines have a distinct edge. There is little problem in a
computer mastering millions of board positions (each of which can have been
analyzed in great depth in advance). Moreover, it is feasible for computers to modify
such previously stored board positions to use them even if they do not precisely
match a current position .
It may very well be that human players deploy methods of abstraction other
than recalling previous board positions, pruning and move expansion. There is little
evidence, however, that for the game of chess such heuristic strategies are inherently
superior to a simple recursive strategy combined with massive computational
power.'® Chess, in my view, is a good example of a type of intelligent problem solving
well suited to the strengths of the first half century of machine intelligence. For other
types of problem solving (level-3 problems), the situation is different.
Not everyone will cheerfully accept the advent of a computer chess cham-
pion. Human chess champions have been widely regarded as cultural heroes, espe-
cially in the Soviet Union; we regard the world chess championship as a high intellec-
tual achievement. If someone could compute spreadsheets in his head as quickly as
(or faster than) a computer, we would undoubtedly regard him as an amazing prodigy,
but not as a great intellect (in fact, he would actually be an idiot savant). A computer
chess championship is likely to cause a watershed change in how many observers
view machine intelligence (though perhaps for the wrong reasons). More construc-
tively, it may also cause a keener appreciation for the unique and different strengths
(at least for the near future) of machine and human intelligence

The intelligent telephone-answering machine

The intelligent telephone assistant answers your phone, converses with the calling
party to determine the nature and importance of the call (according to your instruc-
tions), interrupts you if necessary, and finds you in an emergency. The latter may
become relatively easy once cellular-phone technology is fully developed. If cellular
phones become compact and inexpensive enough, most people would be able to
carry them, perhaps in their wristwatches. This raises some interesting issues of
privacy. Many people like the ability to be away from their phones; we do not neces-
sarily want to be accessible to phone calls at alt times. On the other hand, it could be
considered irresponsible to be completely unavailable for contact in the event of an
emergency. But addressing this issue will be more a matter of evolving social custom
than artificial intelligence.
The other aspects of the above scenario require the same machine skills as
the intelligent assistant: two-way voice communication, natural-language understand-
ing, and automated problem solving. In some ways this application may be more
challenging than the office assistant. For a cybernetic assistant that we interact with,
many people would be willing to spend time learning how to use such technology if it
really helped them to accomplish their work. We might not mind if it failed to handle
every interaction gracefully, so long as it provided an overall productivity gain. On the
other hand, we might set a higher standard for a machine intended to interact with
our friends and associates.

The cybernetic chauffeur

When will computers drive our cars? Without major changes in the methods of traffic
control, the advent of the self-driving car is not likely for a long time. Unlike character
recognition, factory vision, and similar tasks that involve limited visual environments,
driving on existing highways requires the full range of human vision and patten-
recognition skills. Furthermore, because the consequences of failure are so serious,
we would demand a high level of performance from such a system before relying on
it. (On the other hand, with 50,000 traffic deaths each year on American highways,
the human standard of performance is far from perfect.)
Yet there is a simpler form of the cybernetic chauffeur that is easier to build
and could still accomplish a sharp reduction in driving fatalities as well as substantially
reduce the tedium of driving. If specially designed sensors and communication
devices were placed in major thoroughfares, cars could communicate with the road as
well as with other cars. They could then be placed on automatic pilot. Highways
would essentially become electronic railways that our cars could join and leave. The
communication system required would be similar to the complex one already in place
for cellular phones. Algorithms built into each car and into the roads would maintain
safe distances between cars and would handle a variety of situations.'® Although no
specific system has been developed, the underlying technology exists today to deal
with the steady-state situation of cars driving on a highway and maintaining proper
speed and distance. But the situation gets more complicated when several other
contingencies are considered. First, we need to be able to get our car onto a com-
puter-controlled road. Presumably, we would drive to an access road, where our
computer driver would lock onto the road’s communication system, which would take
over (in conjunction with the intelligence of our own car computer). Getting off the
road is perhaps trickier. The system has to consider the possibility that the driver has
fallen asleep or is otherwise not ready to resume manual control. The best protocol
might be for the system to bring the car to a halt at an exit position at which point the
human driver would be expected to take over. Machine intelligence would also have
to deal with possibilities of hardware failure—not just of the computer itself, but also

410
of the engine, tires, and other parts of the vehicle. Furthermore, the designers of such
systems will also have to consider the possibility of people or animals straying onto
the road. .
Even with all of these complications, from a technical standpoint, intelligent
roads represent a substantially easier problem than creating an automatic driver that
can cope with traffic situations as they currently exist. One major nontechnical barrier
to-creating an intelligent road system, however, is that it requires a large measure of
cooperation between car manufacturers and the government agencies that manage
our roads (and drivers). It is not an innovation that can be introduced to a small
number of pioneering users; it needs to be implemented al at once in at least some
roads and in all cars intending to access such roads. Presumably, cars not equipped
with such automatic guidance equipment would not be allowed on intelligent roads
Again, cellular phone technology is similar: it was not feasible until both the portable
phone equipment and the overall computerized communication systems were in
place. Still, | would expect such a system to be introduced gradually. At first it would
be featured on one or a few major highways on an experimental basis. If successful, it
would spread from there
The technology to accomplish this should be available in the first decade of
the next century. But because political decision making is involved, it is difficult to
predict when it will receive priority sufficient to warrant implementation. Though
cellular-phone technology also involved the coordination of a complex system, it has
grown rapidly because it created an entrepreneurial opportunity.
The more advanced scenario of the completely driverless car will take us
well into the first half of the next century. Another approach might be to forget roads
altogether and replace them with computer-controlled flying vehicles that can ascend
and descend vertically. There is, after all, much more space in the three-dimensional
open air than there is on our one-dimensional roads. There are already plans in place
to install a satellite-based collision-avoidance system that will dramatically reduce
airplane collisions by the end of the century. The flying vehicles envisioned here
would be the size of today’s cars and would be even easier to use.'”

Invisible credit cards and keys

During the 1990s we shall see highly reliable person-identification systems that use
pattern-recognition techniques applied to fingerprint scanning and voice patterns
Such systems will typically combine the recognition of a personal attribute (a finger-
print or voice pattern) with a password that the user types in. The password prevents
a misrecognition from causing a problem. The recognition system prevents unauthor-
ized access by a thief who acquires someone's password
Today we typically carry a number of keys and cards to provide us with
access to, and use of, our homes, cars, offices, and a broad variety of financial
services. All of these keys and cards could be eliminated by the widespread adoption
of reliable person-identification technologies. The acceptance of this technology is
likely to be hastened by the loss of confidence in hand-written signatures caused by
the explosion of electronic publishing.
Unfortunately, this type of technology is also capable of helping Big Brother
track and control individual transactions and movements.

ai
Vi

Instant ASICs
One of the remarkable recent innovations in hardware technology is the advent of the
application-specific integrated circuit (ASIC), in which an entire complex electronic
system is placed on a single inexpensive chip. The advent of the ASIC has provided
products that are complex, diverse, customized and highly miniaturized. As Allen
Newell points out in the article following this chapter, one might regard it as an almost
magic technology: once an ASIC is developed, it provides enormous computational
power at very low cost, takes up almost no space, and uses almost no electricity. The
major barrier to greater deployment of this technology is the very long and expensive
engineering cycles required to design such chips. The promise of instant ASICs is the
ability to design an integrated circuit as easily as one writes a high-level computer
program and, once designed, to have the actual chip available immediately.
The development of instant-ASIC technology is a major goal of the U.S.
Department of Defense. Aside from its obvious military applications, it will also greatly
accelerate the availability of innovative consumer products. Just as the difficulty of
programming the early computers was quickly recognized as a major bottleneck, so Is
the difficult design process behind today’s advanced chips. Indeed, the latter is
receiving intense attention from all of the major players in the semiconductor industry.
It is expected that in the early 1990s designers will be able to write chip programs,
(whose output is a piece of silicon) as easily and as quickly as computer programs.'®
The availability of instant-ASIC technology will eliminate for most purposes what little
difference remains today between hardware and software engineering. It will acceler-
ate the trend toward knowledge as the primary component of value in our products
and services.

Artifi ial people and the media of the future

Rather than describe the vision | have in mind, | shall approach the idea of artificial
people by starting with what is feasible today. Consider the artificial creatures that we
interact with in computerized games and educational programs. An early and prime
example is Pac Man, an artificial creature capable of responding and interacting with
us in certain limited ways. One might not consider Pac Man to be much of a creature,
let alone an artificial person. It is kind of a cartoon caricature of a fish with a limited
repertoire of movements. Similarly, our range of emotional responses to it is very
narrow, but it serves as a useful starting point.
Now consider what is feasible today, about a decade after the first introduc-
tion of Pac Man and other computer games. Reasonably lifelike video images of
human faces can be completely synthesized and animated. Experiments at such
advanced media laboratories as the MIT Media Laboratory have created completely
synthetic yet realistic computer-generated images of human faces that can move and
express a wide range of responses and emotions.
Let us imagine the next step: computer games and interactive educational
programs that use synthetically generated human images. Rather than simply
replaying a previously stored animated sequence, such programs would start with
knowledge structures representing the concepts to be expressed and then translate
each concept into spoken language and articulated facial and bodily movements. We
would thus see and hear images—not prestored but created in real time—of people
that are reasonably realistic. The motions and speech sounds would be computed as

a2
needed from an intent to express certain ideas and emotions. These artificial people
would be responding to our actions within the context of the program |
Let us take several more steps. Add speech recognition and natural-
language understanding. Add another several generations of improved image resolu-
tion and computing power for greatly enhanced visual realism. Add a more sophisti-
cated problem-solving capability and more intelligence to provide greater subtlety of
personality. Our artificial person is becoming more like a real person and less like
Pac Man.
Applications would include very realistic games, movies that could include
the viewer as a participant, and educational programs that would engage the student
to learn from direct experience. Benjamin Franklin could take a child on a guided tour
of colonial Philadelphia. Rather than a canned visual tour, this artificial Ben Franklin
could answer questions, engage the child in dialogue, customize the tour on the basis
of the child's own expressed interests, and otherwise provide an engaging experi-
ence. One could debate Abraham Lincoln or take Marilyn Monroe on a date. As with
any creative medium, the possibilities are limited only by our imagination. As another
example, the intelligent assistant could include a persona complete with appearance,
accent, and personality. As time went on, such artificial persons would continue to
grow in sophistication, realism, communicative and interactive ability and of course
intelligence. Ultimately, they would develop a sense of humor.
lt should be noted that personality is not an attribute that can be stuck on an
intelligent machine. A personality is almost certainly a necessary byproduct of any
behavior complex enough to be considered intelligent. People already speak of the
personalities of the software packages they use. Shaping the personality of intelligent
machines will be as important as shaping their intelligence. After all, who wants an
obnoxious machine?
Such artificial persons could eventually use three-dimensional projected
holographic technology (a method for creating three-dimensional images that do not
require the use of special glasses). Currently, most holograms are static three dimen-
sional pictures, although some use multiple images to provide a sense of movement.
The MIT Media Lab has succeeded in creating the world’s first three-dimensional
holographic image generated entirely by computer.'? The ability to project a hologram
entirely from computer data is an important step in imaging technology. If a computer
can project one hologram, then it can be made to project any number. Ultimately, with
sufficient computer power, the images could be generated fast enough to appear
realistically to move. The movements would not be prestored but rather computed in
real time to respond to each situation. Thus, our artificial people can ultimately be
lifelike, life-size, three-dimensional images with sufficiently high resolution and
subtlety of movement to be indistinguishable from real people. These future com-
puter displays will also be able to project entire environments along with the people.
There is concern today regarding the power of television to shape our views
and to engage our emotions and attention. Yet television is essentially noninteractive,
of low resolution, and flat. A medium that provided nearly perfect resolution and
three-dimensional images and interacted with us in an intelligent and natural way
would be far more powerful in its emotional impact, possibly too powerful for many
(real) people. Harnessing and regulating these media of the future will undoubtedly be
an area of much debate and controversy

413
 Adoption of the advanced media technologies described here will begin in
the late 1990s and mature over the first half of the next century. Applications include
entertainment, education, conducting business transactions, even companionship.

Another approach
An entirely different approach to the concept of artificial people lies in the area of
robotics. Robots of the first generation, just like the first generation of computer-
generated creature images (essentially pictures of robots), were neither intelligent nor
realistic. We were as unlikely to mistake an early factory robot for a natural creature,
let alone a person, as we were to mistake Pac Man for an image of a real animal. Here
again, successive generations of technology have provided greater intelligence,
subtlety, and naturalness. The primary drive for robotic technology lies in practical
applications in the manufacturing and service industries. Admittedly, for most of these
applications, resemblance to humans or to any other natural creature is of little
relevance. Yet there will be applications for natural robots (sometimes called androids)
as teachers, entertainers, and companions. Primitive robotic pets have already created
a niche in the toy industry.
Creating a reasonably natural robotic imitation of a person is even more
challenging than creating a convincing media image of a person. Any autonomous
robot, humanoid or otherwise, has to be able to ambulate in a natural environment;
this requires general-purpose vision and a high degree of fine motor coordination. Au-
tonomous robots for exploring hostile environments, such as nuclear reactors and the
surfaces of other planets, exist today. Routine use of autonomous robots in more
conventional settings is likely to begin by the end of the century. Robots that are
reasonably convincing artificial people will not appear until well into the next century.
Marvin Minsky has often said that a good source of insights into the realities
of tomorrow's computer science can be found in today’s science fiction. Isaac Asimov
in his Robots of Dawn describes a society two centuries from now in which people
live alongside a ubiquitous generation of robotic servants, companions, guards, and
teachers. Two of the protagonists are a beautiful female scientist and her lover, a
“male” “humaniform” robot

Passing the Turing test

Scientists from the University of Clear Valley reported today that a computer program
they had created was successful in passing the famous Turing test. Computer
scientists around the world are celebrating the achievement of this long-awaited
milestone. Reached from his retirement home, Marvin Minsky, regarded as one of the
fathers of artificial intelligence (Al), praised the accomplishment and said that the age
of intelligent machines had now been reached. Hubert Dreyfus, a persistent critic of
the Al field, hailed the result, admitting that he had finally been proven wrong.

The advent of computers passing the Turing test will almost certainly not
produce the above sort of coverage. We will more likely read the following:

Scientists from the University of Clear Valley reported today that a computer program
they had created was successful in passing the famous Turing test. Computer
scientists reached at press time expressed considerable skepticism about the accom-
plishment. Reached from his retirement home, Marvin Minsky, regarded as one of the

aia
fathers of artificial intelligence (Al), criticized the experiment, citing a number of
deficiencies in method, including the selection of a human “judge” unfamiliar with the
state of the art in Al. He also said that not enough time had been allowed for the
judge to interview the computer foil and the human. Hubert Dreyfus, a persistent
critic of the Al field, dismissed the report as the usual hype we have come to expect
from the Al world and challenged the researchers to use him as the human judge.

Alan Turing was very precisely imprecise in stating the rules of his widely
accepted test for machine intelligence.”° There is, of course, no reason why a test for
artificial intelligence should be any less ambiguous than our definition of artificial
intelligence. It is clear that the advent of the passing of the Turing test will not come
on a single day. We can distinguish the following milestones:

Level 1 Computers arguably pass narrow versions of the Turing test of believability. A
variety of computer programs are each successful in emulating human ability in some
area: diagnosing illnesses, composing music, drawing original pictures, making
financial judgements, playing chess, and so on.

Level 2 It is well established that computers can achieve human or higher levels of
performance in a wide variety of intelligent tasks, and they are relied upon to diagnose
illnesses, make financial judgements, etc.
Level 3 A single computer system arguably passes the ful/ Turing test, although there
is considerable controversy regarding test methodology.

Level 4 It is well established that computers are capable of passing the Turing test
No reasonable person familiar with the field questions the ability of computers to do
this. Computers can engage in a relatively unrestricted range of intelligent discourse
(and engage in many other intelligent activities) at human or greater levels of
performance.

We are at level 1 today. A wide range of expert systems can meet or

exceed human performance within narrowly defined (yet still intelligent) areas of
expertise. The judgements of expert systems are beginning to be relied upon in a
variety of technical and financial fields, although acceptance in the medical area is
much slower. Also, computer success in a variety of artistic endeavors is beginning to
be at least arguably comparable.
Level 2 is within sight and should be attained around the end of the century.
As expert systems grow in sophistication and achieve more natural human interfaces,
we will begin to rely on their expertise as much as (if not more than) human society
relies on their idiot savant forebears today.
We will probably begin to see reports of level 3, and newspaper articles
similar to the second one given above, during the first decade of the next century,
with continued controversy for at least several decades thereafter. The first reports
will almost certainly involve significant limitations to Turing’s originally proposed
challenge. We are close to having the underlying technology (if not the actual pro-
gram) today if we use sufficiently naive judges and provide them with relatively little
time to make their determinations.
Level 4 is what Turing had in mind when he predicted success by the year
2000. Achieving this level is far more difficult than any of the other three. It requires

ais
 advanced natural-language understanding, vast knowledge bases of commonsense
information, and decision-making algorithms capable of great subtlety and abstraction.
Turing’s prediction, made in 1950, will almost certainly not be fulfilled by the year
2000. | place the achievement of level 4 sometime between 2020 and 2070. If this
turns out to be the case, then Turing will have been off by a factor of between 1.4 and
2.4 (70 to 120 years versus his prediction of 50 years), which actually is not bad for
such a longterm prediction. Of course, there is no assurance that my prediction will be
any more accurate than Turing’s.
As mentioned earlier (see The Debate Goes On), Hubert Dreyfus has
indicated that he will concede that he was wrong (and has been wrong for his entire
professional career) if he can be fooled as the human judge in a Turing test. Will this
happen? If we assume that Dreyfus is in good health and further that continuing
advances in bioengineering technology enable him (and the rest of us) to live longer
than today’s average life expectancy, then it is altogether possible. Personally, | would
be willing to bet on it

Conclusion
The above scenarios provide only a small sampling of the ways in which intelligent
machines of the future can be expected to touch our lives. The computers of today,
dumb as they are, have already infiltrated virtually every area of work and play. The
bureaucracies of our society could hardly function without their extensive computer
networks. If one adds just the sharply focused intelligence of the next phase of the
age of intelligent machines to the already prodigious memory capacity and speed of
today’s computers, the combination will be a formidable one indeed. Our cars,
watches, beds, chairs, walls, floors, desks, books, clothes, phones, homes, appli-
ances, and virtually everything else we come into contact with will be intelligent,
monitoring and servicing our needs and desires. The age of intelligent machines will
not start on a particular day; it is a phenomenon that has started already, with the
breadth and depth of machine intelligence growing each year. Turing predicted a time
when people would talk naturally about machines thinking without expecting anyone
to contradict them or be surprised. We are not there today, but the day will arrive so
gradually that no one (except a few authors) will notice it when it does.

+ Breakthroughs

Ralph Gomory, IBM's chief scientist, predicted in 1987 that within a decade the central processing units of
supercomputers will have to be concentrated within a space of three cubic inches. The supercomputer
core of the 1990s will be suitable for a laptop.
George Gilder

Hardware
As | noted earlier, a continuation of the same rate of exponential improvement in
computer hardware appears likely for the foreseeable future, even if we consider only
improvements using conventional approaches. Progress continues to be made in
manufacturing integrated circuits with ever smaller geometries, which thus allows
ever larger numbers of components to be placed on a single chip. As this book was
being written, designers were passing the mark of several million transistors per chip.

a16
The advent of parallel-processing architectures is also contributing to expanding the
amount of computation we can devote to machine intelligence.”'
At the same time, researchers are experimenting with a number of exotic
materials and techniques that, if perfected, could provide a quantum leap in comput-
ing power. Rather than “just” the orderly doubling of computer power every 18 to 24
months, the possibility exists for a relatively sudden increase by a factor of thousands
or even a million
Probably the most promising is the potential for superconductors to virtually
eliminate the thermal constraints that now govern chip geometries. As mentioned
earlier, the heat generated by resistance in circuits limits how small we can build each
transistor. But superconductors offer no electrical resistance, so superconducting
circuits generate no heat. A superconducting integrated circuit could thus use transis-
tors that are smaller by at least a factor of ten in each of the two dimensions, for a
hundredfold improvement in the number of components. Since these components
would also operate ten times faster, the overall improvement is approximately one
thousand. Furthermore, with the absence of heat, we could build a chip with a
thousand layers of circuitry rather than just one. This use of the third spatial dimension
provides another improvement factor of a thousand, for an overall potential improve-
ment of one million to one.
The only problem so far is that no one yet knows a practical way to build
tiny transistors using superconducting materials. Earlier attempts to harness super-
conducting integrated circuits, using a technology called Josephson junction, never
reached commercial viability.22 One of the problems with the earlier attempts was the
need to cool the circuits to so near absolute zero that very expensive liquid helium
had to be used to cool them. But recent advances in superconducting have provided
materials that can provide superconducting at much higher temperatures, so liquid
nitrogen, at less than one tenth the cost of liquid helium, can be used. There is even
the possibility of developing materials that can provide superconductivity at room
temperature, the Holy Grail of superconductivity. The latest materials exhibit consider-
able brittleness, however, which presents a major difficulty in creating the tiny wires
needed.”? While the challenges are formidable, the potential payoff of supercon-
ducting integrated circuits is so enormous that major efforts are underway on three
continents (North America, Europe, and Asia).7*
Another approach being explored is the development of circuits that use
light instead of electricity. There are two potential advantages of light. First, light is
faster. The speed of light is regarded as a fundamental speed limit that the universe is
obliged to follow. Electricity in wires moves about one-third as fast. More important,
laser light can contain millions of adjacent signals that each carry independent infor-
mation. Thus, the potential exists to provide truly massive parallel processing, with
each light signal providing a separate computational path. Optical techniques are
already revolutionizing communications (with optical fibers) and memory (optical
disks); the promise of optical computing would revolutionize computation itself.2°
There are disadvantages, however, in that the types of computations that
can be performed optically are somewhat limited. A massively parallel optical comput-
ing machine would be well suited for certain pattern-recognition tasks, particularly
those using low-level feature detection, but it could not be easily adapted to
conventional types of computing. For this, superconductivity appears more promising.
A massively parallel computer using superconduction could provide an enormous
number of parallel processors that could each use conventional software techniques.
A third, even more exotic approach is to build circuits using bioengineering
techniques, essentially to grow circuits rather than to make them. Such circuits would
be constructed of the same proteins that form the basis of life on earth. These organic
circuits would be three-dimensional, just as circuits in the brains of natural living
creatures are three-dimensional. They would be cooled in the same way that natural
brains are cooled: by bloodlike liquids circulating through a system of capillaries. We
would not build these circuits; they would grow in much the same way that the brain
of a natural organism grows: their growth and reproduction would be controlled by
genes made up of ordinary DNA. Although still at an early stage, significant progress
has been made in developing such organic circuits. Wires and simple circuits have
already been grown.?6
Still another approach, called molecular computing, has substantial overlap
with all three of the above techniques.?” Molecular computing attempts to employ
light and the finest possible grains of matter (for computing, probably molecules) to
provide techniques for information manipulation and storage. The goal of molecular
computing is massively parallel machines with three-dimensional circuitry grown like
crystals
As | mentioned, the objective of all of these investigations is to provide a
great improvement in the capacity and speed of computation. It should be empha-
sized that the value of doing so is not primarily to speed up existing applications of
computers. Speeding up the computation of a spreadsheet that now takes a few
seconds to a few microseconds is of little value. The real excitement is in being able
to solve important problems that have heretofore proved intractable. | discussed
earlier the massive amount of computation required to emulate human-level vision.
Because the amount of computation provided by the human visual system is not yet
available in a machine, all attempts at machine vision to date have been forced to
make substantial compromises. The enormous improvements in speed and capacity
discussed above would provide the computational power needed to emulate human-
level functionality.

Software
lf three-dimensional superconducting chips or one of the other breakthroughs just
described were perfected and computer hardware were suddenly to expand in
capability by a factor of thousands or even a million, the impact on many problem
areas would not be immediate. While substantial expansion of computing power is
one of the requirements to master problems such as vision, there are other require-
ments as well. We also need continued refinement of our algorithms, greater ability to
represent and manipulate abstractions, more knowledge represented in computer
form, and an enhanced ability to capture and manipulate knowledge.
Two key points are worth making about potential improvements in com-
puter technology. First, breakthroughs and software appear to me to be incompatible,
particularly with regard to issues of machine intelligence. Al applications are so
complex, and the requisite human-interface issues so demanding, that none of the

ais
underlying software technologies is likely to spring suddenly upon us. Inevitably,
solutions to various Al problems—image understanding, speech understanding,
language understanding, problem solving—are solved incrementally and gradually,
with each new generation of a technology providing greater subtlety and depth. Each
new step forward reveals new limitations and issues that are addressed in the next
generation of the technology or product. A point often comes when the level of
performance in a certain area exceeds a critical threshold, and this enables the
technology to move out of a research environment into practical commercial applica-
tions. In a sense, the overall Al movement started to pass that threshold during the
1980s, when it moved from a barely visible industry with a few million dollars per year
of revenue in 1980 to a billion dollars in 1986, with four billion expected in 1990,
according to DM Data. It often appears that a breakthrough must have been made,
because a technology became sufficiently powerful to be commercially viable. But
invariably, the underlying technology was created over an extended period of time. Al

is no exception.
The second point is that a hardware breakthrough of the type described
above is not necessary for significant and sustained progress in each area of Al. There
is no question that more powerful computers are needed for many problems, such as
continuous speech recognition, human-level vision and others, but the exponential
growth in computer power that will occur anyway is sufficient.

419
 Once upon a time, when it was still of some use to wish for what one
wanted, ... there lived a king and queen who had a daughter who
was lovely to behold, but who never laughed.

Or perhaps:

there lived an old fisherman by the side of the sea that had hardly any
| PC a fishes in it.

If you are like me, you are already hooked. You are ready to abandon
Fairy Tales all talk of present matters, of computers and electronic technology,
and settle in to hear a fairy tale. Their attraction reaches almost

Prof. Allen Newell is the U. A. and all of us.

Helen Whitaker University They let us enter into an enchanted world. Magic
Professor of Computer Science at abounds, though always in special ways. Animals talk, and not only
Carnegie-Mellon University, where
animals but trees and bridges as well. Villainy is there, certainly
he has been since 1961. He was
educated in physics (B.S, Stanford danger. There are trials to be overcome—usually three of them. But
University), mathematics (Princeton there is always the happy ending. The spell is broken, and the
University), and industrial princess smiles and marries the youth who made her laugh. The old
administration (Ph.D., 1957,
fisherman gets the Jinni back in the bottle with the top on. And
Carnegie Institute of Technology).
He contributed to the emergence of happiness is ever after, which means at least for a little while.
cognitive psychology and artificial The experts tell us that fairy tales are for childhood. They
intelligence in the mid 1950s and contain lessons for the crises of growing up, and their universal
has continued to make significant
attraction comes because they deal with what is central to this
contributions to them ever since,
many with Herbert Simon. His universal period in life: Like Hansel and Gretel, we have to leave
current research (with John Laird home and find our own way. Like the princess with the frog king, we
and Paul Rosenbloom) is on Soar, a must learn to keep our word and embrace what we find ugly and
general cognitive architecture that
disgusting, to discover that it contains our heart's desire.
embodies a unified theory of human
cognition. He has also made major Or like Jack, in the story of the beanstalk, we can bring
contributions to list processing, home the bacon if we persevere, even if our parents don't think we
expert systems, computer can. But there was more, if you remember your Jack: First, he
structures, and human-computer
escaped back home with a bag of gold. But Jack and his mother used
interaction. He is the recipient of
the Association of Computing up the gold, which shows that one success is not enough. Then he
Machinery’sA. M. Turing Award made a second trip up the beanstalk to the giant's castle. This time he
(with Herbert Simon), and the came home with the magic hen that lays golden eggs. Now Jack had
American Psychological
a technology for satisfying his and his mother's wants. But even so,
Association's Distinguished
Scientific Contribution Award. He
is the author or coauthor of over 250
papers and 10 books
420
Allen Newell (Courtesy of Allen
Newell)

material things are not sufficient for the full life. So on his third trip foolishness, second on disaster, and third on bare recovery. Recall
Jack brought home the golden singing harp, symbolizing the higher the story of the Monkey's Paw, which came to an old English couple.
things of life. Their first wish was for just 200 pounds. That was foolish. They lost a
The experts notwithstanding, fairy tales are for all of us. son, whose accident brought them a 200-pound reward. The second
Indeed, this is true especially in our current times. For we are, all of wish was for the return of their son. That was disaster. He returned
us, children with respect to the future. We do not know what is from the grave, though hardly unscathed. The third wish was to send
coming. The future is as new, and as incomprehensible, as adult life their son back to his opened grave, to try to recover for themselves a
is to children. We find ourselves troubled and fearful at the changes world where life could go on.
taking place in ourselves and our society. We need the hidden
| see it differently. | see the computer as the enchanted
guidance of fairy tales to tell us of the trials we must overcome and
technology. Better, it is the technology of enchantment. | mean that
assure us there will be a happy ending. Whether fairy tales have been
quite literally, so | had best explain.
written that speak to the heart of our own adult crises is not clear.
There are two essential ingredients in computer
How would we, the children, ever know? Perhaps we must get along
technology. First, it is the technology of how to apply knowledge to
with the fairy tales we have.
action to achieve goals. It provides the capability for intelligent
But even more, fairy tales seem to me to have a close
behavior. That is why we process data with computers—to get
connection to technology. The aim of technology, when properly
answers to solve our problems. That is what algorithms and programs
applied, is to build a land of Faerie.
are about—frozen action to be thawed when needed.
Well, that should come as a shock! The intellectual garb
The second ingredient is the miniaturization of the
of the modern academic is cynicism. Like a follower in a great herd,
physical systems that have this ability for intelligent action. This is
as surely as | am an academic, | ama cynic. Yet | have just uttered a
what Angel Jordan, my co-Whitaker professor, has been telling us
sentiment that is, if anything, straight from Pollyanna.
about in his talk. Computers are getting smaller, and cheaper, and
In point of fact, within the small circle of writers who
faster, and more reliable, and less energy demanding. Everything is
manage to put technology and fairy tales between the same covers,
changing together in the right direction. The good things do not trade
the emphasis is always on the negative, on the dark side. The favorite
off against the bad ones. More speed does not mean more dollars.
stories are those that trouble:
Small size does not mean lower reliability. On any given date, the
Like the Sorcerer's Apprentice, who learns only enough magic to start expected painful trade-offs do hold, just as we learned in elementary
the broom of technology hauling water from the River Rhine to the economics. It costs more to buy faster circuits or larger memories.
cistern, but who cannot stop it. But come back next year and everything is better: smaller, cheaper,
faster, more reliable, and for less energy.
Like the Jinni in the bottle, where the story is never permitted to go to
Thus computer technology differs from all other technolo-
the conclusion in the Arabian Nights, with the Jinni snookered back
gies precisely in providing the capability for an enchanted world: for
in the bottle, but is always stopped with the Jinni hanging in air and
little boxes that make out your income-tax forms for you, for brakes
the question along with it—Can we ever put the Jinni back? Or will
that know how to stop on wet pavement, for instruments that can
there only be ink all over the sky till the stars go out?
converse with their users, for bridges that watch out for the safety of
Like the many stories of the three magic wishes, in which, promising
infinite riches just for the asking, they are always spent, irst on
those who cross them, for streetlights that care about those who message of the recent literature of political ecology: Our technolo-
stand under them—who know the way, so no one need get lost. gies inevitably demand that we use up our precious world. There is
In short, computer technology offers the possibility of rather abundant evidence for this view. Here in Western Pennsylva-
incorporating intelligent behavior in all the nooks and crannies of our nia, the price paid in enchantment of our countryside for taking our
world. With it we could build an enchanted land. coal by strip mining is only too evident. Less in our awareness,
All very good, but what about the Sorcerer's Apprentice? because it was so thorough, was what the loggers did to Western
Two half-fallacies feed our fear that his nightmare might be ours. The Pennsylvania. Not once, but thrice, within forty years they swept the
first half-fallacy is that technologies are rigid and unthinking. Start hillsides almost bare. The hot scorching breath of a dragon could
the broom off carrying water, and it does just that and not something hardly have done better for desolation.
else. But every computer scientist recognizes in the Sorcerer's But all is not inevitable. Ecologically, computer technol-
Apprentice simply a program with a bug in it, embedded ina first ogy itself is nearly magic. The better it gets, the less of our environ-
generation operating system with no built-in panic button. Even with ment it consumes. It is clean, unobtrusive, consumes little energy and
our computer systems today, poor things that they are, such blunder- little material. Moreover, as we push it to higher peaks of speed and
bus looping is no longer a specter. memory, it becomes more of all these things. For deep technical
Exactly what the computer provides is the ability motto be reasons this has to be. There is no way to obtain immense amounts of
rigid and unthinking, but rather to behave conditionally. That is what processing power by freezing technology at some cost in dollars,
it means to apply knowledge to action: it means to let the action material, and energy per unit of computation and then just buying
taken reflect knowledge of the situation, to be sometimes this way, more and more of it, consuming our wealth and our environment.
sometimes that, as appropriate. With small amounts of computer Instead, for a long time to come, as we get more and more of it, the
technology—that is, with small amounts of memory and small less it will impact our environment.
amounts of processing per decision—you often can't be conditional Even more, the computer is exactly the technology to
enough. That is certainly the story of the first decades of the computer permit us to cope intelligently with the use of our other resources.
revolution. It was too expensive and involved too much complexity to Again, by providing us with distributed intelligence, it can let us
create systems with enough conditionality. We didn't know how and keep track of the use and abuse of our environment. And not only of
couldn't have afforded it if we did. Consequently, many applications the destruction that we ourselves visit on our world but also that
were rigid and unthinking. It was indeed a Sorcerer's Apprentice who which nature does as well. Mount Vesuvius was hardly bound by any
seemed to run the computerized billing service. antipollution ordinances posted on the walls of ancient Pompeii.
The import of miniaturization is that ultimately we will be In sum, technology can be controlled, especially if it is
able to have the capability for enough conditionality in a small saturated with intelligence to watch over how it goes, to keep
enough space. And the import of our scientific study of computers is accounts, to prevent errors, and to provide wisdom for each decision.
that we shall know how to make all the conditionality work for us. And these guardians of our world, these magic informational dwarfs,
Then the brooms of the world themselves can know enough to stop need not extract too high a price.
when things go wrong. But | said that fear of the plight of the Sorcerer's
The second half-fallacy behind the Sorcerer's Apprentice Apprentice was guided by half-fallacies. | did not dismiss the view
is that technologies by their nature extract too high a price. That is a totally. Because, of course, in fairy tales there are great trials to be
performed before the happy ending. Great dangers must be encoun-

422
tered and overcome. Because, also, in fairy tales, the hero (or the
heroine}—the one who achieves finally the happy ending—must
grow in virtue and mature in understanding. No villains need apply
for the central role. The fairy tale that | am indirectly spinning here
will not come true automatically. We must earn it.
Where are we now? We are not at the end of the story,
though we are surely at the end of my talk. In fact, the fairy tale is
hardly past its “Once upon a time.” Still, | wish to assert that
computer science and technology are the stuff out of which the
future fairy land can be built. My faith is that the trials can be
endured successfully, even by us children who fear that we are not
so wise as we need to be. | might remind you, by the way, that the
hero never has to make it all on his own. Prometheus is not the
central character of any fairy tale but of a tragic myth. In fairy tales,
magic friends sustain our hero and help him overcome the giants and
the witches that beset him.
Finally, | wish to express my feeling of childlike wonder
that my time to be awake on this earth has placed me in the middle of
this particular fairy tale.

423
 The Impact On...
It's hard to predict—especially the future.
Niels Bohr, physicist

+ Employment and the Economy

If every instrument could accomplish its own work, obeying or anticipating the will of others .
shuttle could weave, and the pick touch the lyre, without a hand to guide them, chief workmen would not
need servants, nor masters slaves.
Aristotle

If machines could be so improved and multiplied, then all of our corporeal necessities could be entirely
gratified, without the intervention of human labor, there will be nothing to hinder all mankind from
becoming philosophers and poets.
Timothy Walker, essayist, 1831

Machinery will perform all work—automata will direct all activities and the only tasks of the human race
will be to make love, study and be happy.
The United States Review, 1853

When people consider the impact of computer intelligence, few areas generate as
much controversy as its potential to influence patterns of employment. Other areas—
education, medicine, even warfare—evoke less passion. In education, concern is
sometimes expressed about computers replacing human instructors, but astute
observers realize that computer-assisted instruction is intended to compete with
books, not teachers.’ There is understandable hesitancy to rely on the diagnostic
judgements of medical expert systems even when they have demonstrated superior
skills in some areas, but few expect undue reliance on such systems without strenu-
ous steps to verify reliability. There is certainly controversy surrounding military
applications, but even here there is some recognition that the highly pinpointed
targeting provided by “smart” weapons may cause less indiscriminate destruction
than the shotgun tactics required by older generations of weapon systems (more
about this later). The issue of jobs, on the other hand, strikes at a fundamental and
immediate concern of most people. While machines have been competing with
 human labor for centuries in the physical domain, the more recent competition for
mental work is more threatening, both economically and psychologically.
Views on the impact of this latest wave of automation vary dramatically with
the observer. Some hail the ability of machines to eliminate mental drudgery in the
same way that an earlier generation of machines released us from the bondage of
hard physical labor. Others point to a bleak future in which employment opportunities
remain only for an elite few.
This issue is rarely approached dispassionately. Most views are heavily
influenced by a number of social and ideological assumptions. | do not pretend to be
immune from such influence, but | do feel that an examination of long-term trends of
the past and attempts to project such trends into the future can help to illuminate
these issues. Most of the automation that has taken place in human history has
occurred over the past one hundred years. As | pointed out in the prolog, the macro-
economic trends during that period were quite positive. Jobs grew 10-fold in the
United States (12 million in 1870 to 116 million in 1985) with the percentage of the
U.S. population gainfully employed increasing from 31 percent to 48 percent. More

\ |
significantly, the per-capita gross national product, as well as the average earning
power of jobs, increased 600 percent in constant dollars during the same period.? The
quality of jobs improved as well, a much higher fraction of jobs providing gratification
beyond the paycheck. Nonetheless, the impact of automation on jobs has been a
controversial subject during this entire period. One impediment to sober reflection on
the issue is that the reality of lost jobs and the resulting emotional and social impact
are far easier to see than the possibility of new jobs and new industries. Early in this
century, jobs in factories and in agriculture were disappearing at a rapid rate. Dire
predictions of these trends spiraling to disaster were not uncommon. It was not
possible for leaders at the time to say, Don't worry, millions of jobs will be created in
the electrical industry, and following that, the electronics and computer industries will
create millions more. Indeed at the time it would have been impossible to foresee
even the existence of these industries. The phenomenon continues today, with new
manufacturing technologies rapidly reducing the number of production jobs. Manufac-
turing today provides a quarter of all employment; by the beginning of the next
century it is expected to provide relatively few jobs.° The social and political impact of
these lost jobs is felt far more strongly than the future opportunities which undoubt-
edly will be there
More perspective can be gained by attempting as rigorously as possible to
project these trends into the future. A comprehensive study using a detailed com-
puter model of the U.S. economy was conducted recently by Wassily Leontief, Faye
Duchin, and their colleagues at the Institute for Economic Analysis (IEA).* The study
indicates continued expansion of the per-capita gross national product and average
earning power of the American worker. It projects a rapidly diminishing demand for
clerical workers and other categories of semiskilled and unskilled workers and a
reduction in factory workers, although the latter will be partially offset by the overall
increase in economic activity. It projects a sharp reduction in the need for skilled metal
workers due to increased use of factory automation, including robots and computer-
ized machine tools. Conversely, it projects a sharp increase in the need for profession-
als, particularly engineers (including computer specialists), and for teachers.®

426
 The most significant finding of the study is that the limiting factor on future
economic growth will be the availability of a suitably trained workforce.® There will be
plenty of jobs in the early twenty-first century, but only if society provides a sufficient
number of appropriately skilled workers to fill them. As factories are rebuilt for sub-
stantially fewer personnel, both blue and white collar, as agriculture continues to be
mechanized, as the service industries begin to automate, there will be a correspond-
ing increase in the demand for knowledge workers who can design, build, program,
and run the intelligent machines of the future. At least as important as the knowledge
workers are the teachers to train them. Since power and wealth in the age of intelli-
gent machines will increasingly consist of knowledge and skill, the ability to develop
and foster our human intellectual resources becomes society's most pressing
challenge
The IEA study examines a number of scenarios that differ in assumptions
about the speed with which society can incorporate new computer technologies and
provide the workforce with the necessary skills. Consequently, the scenarios differ
dramatically in their projected employment levels and economic progress. Of course,
the economic model used by the IEA, no matter how detailed, does not by any means
eliminate uncertainty regarding the future. It is difficult to predict how quickly a given
technology will be adopted and integrated into the workplace. Often predictions of
rapid change are premature. For example, the advent of numerically controlled
machine tools in the 1960s was expected to quickly revolutionize metalworking, but
the new technology made little headway until it was integrated with computers in the
late 1970s.’ On the other hand, some technologies move more quickly than any
expert predicted, as was the case with the rapid computerization of offices during the
1980s. A comparison of the different scenarios of the IEA’s study does make clear,
however, that the primary variable that will determine the rate of continued economic
progress and continued growth in the availability of employment is the quality and
availability of appropriate education and training. Interestingly, the conclusion is the
same for both advanced and underdeveloped nations.®

The nature of work

Most factories built today employ substantially fewer workers than the factories they
replace. Even if robotic technology is not employed, the computerization of material
handling and work flow has already substantially reduced the direct-labor content of
most products. There is little question that this trend will continue and intensify over
the next two decades. By early in the next century a relatively small number of
technicians and other professionals will be sufficient to operate the entire production
sector of society.
While employing substantially fewer people, the advent of computer-
controlled manufacturing technologies will permit a degree of individual customization
of products not feasible today. Two centuries ago every item was inevitably a little
different, since it was made by hand. During the first industrial revolution the innova-
tion of mass production substantially reduced the individualization of products. During
the second industrial revolution the innovation of extremely flexible manufacturing will
increase it.? For example, consumers will be able to sit down at their home computers
and design their own clothes to their own precise measurements and style

427
 requirements using friendly, computer-assisted design software. When the user
issues the command “Make clothes,” the design parameters and measurements will
be transmitted to a remote manufacturing facility, where the clothes will be made and
shipped within hours.
As employment in the factory dwindles, employment in the office will be
stable or increase. However, what we do in offices will substantially change.'° Clerical
work will largely disappear. Completing a trend already under way, by early in the next
century, computers will type our letters and reports, intelligently maintain our files and
records, and help to organize our work. The areas likely to continue to require signifi-
cant human involvement, particularly during the first half of the next century, will be
communication, teaching, learning, selling, strategic-decision making, and innovation.
While computers will certainly impact all of these areas, they will continue to be the
primary focus of human efforts in the office. As | pointed out above, the office worker
of the next century will have sustained contact with both human and machine
intelligence
The concept of a document will undergo substantial change. Extremely high

i resolution easy-to-view screens will become as common and as easy to read from as
paper. As a result, we will routinely create, modify, handle, and read documents
without their ever being converted to paper form. Documents will include a variety of
types of information beyond mere text and pictures. They will routinely include voice,
music, and other sound annotations. Even the graphic part of documents will become
more flexible: it may be an animated three-dimensional picture. In addition, docu-
ments will be tailored in that they will include the underlying knowledge and flexibility
to respond intelligently to the inputs and reactions of the reader.'' Finally, documents
will not necessarily be ordered sequentially as they are in this book: they will be
capable of flexible intuitive patterns that reflect the complex web of relationships
among ideas (this is Ted Nelson's “hypertext”).'?
With the ongoing acceleration of the pace of change, the idea of training
someone for a lifelong trade will become even less valid than it is today. Instead, we
shall need to teach our children how to keep learning throughout their adult lives. It is
estimated that the typical worker of the twenty-first century will make changes in the
way they work once or twice each decade, changes that we would now consider
major career changes.'? Thus, the primary skill required for employment in the
workplace of the future will be the ability to adapt and to continue growing
intellectually.
A constructive change in our concept of work will be to think of the process
of learning as part of work, rather than as just a prerequisite for work. The worker of
the future may spend as much as half of his time learning rather than just doing. A
trend toward this concept is already discernible among more enlightened employers.
Employers providing on-the-job education and paying their employees to acquire new
skills is likely to emerge as a major trend by the end of the century.
The trend toward work as a vital component of gratification and personal
satisfaction is also likely to intensify in the decades ahead. It is to be hoped that the
divisions between work on the one hand and learning, recreation, and social relation-
ships on the other will dissolve as work becomes more fully integrated with the other
facets of life

428
+ Education

Common sense is not a simple thing. Instead, it is an immense society of hard-earned practical ideas—of
multitudes of life-learned rules and exceptions, dispositions and tendencies, balances and checks.
Marvin Minsky, The Society of Mind

All instruction given or received by way of argument proceeds from pre-existent knowledge.
Aristotle, Posterior Analytics

The search for the truth is in one way hard and in another easy—for it is evident that no one of us can
master it fully, nor miss it wholly. Each one of us adds a little to our knowledge of nature, and from all the
facts assembled arises a certain grandeur.
Aristotle, 350 8.c.

My conclusion about the impact of intelligent machines on employment rests on

education and its pivotal role in shaping the future world economy. This leads me to
consider the impact of this new technology on education itself.'* Thus far the impact
of computers has been modest. Of much greater effect have been the other media
technologies of television, cinema, and recorded music. While it is true that hundreds
of thousands of computers have found their way into schools, a much larger number
have found their way into homes. In the home videogames have made their presence
felt, digital technology is evident in a broad variety of consumer electronic products,
and calculators are ubiquitous. Yet the content and process of education remains
largely unchanged; there is continued reliance on books that, as Minsky puts it, “do
not talk to each other.”
'® The computer revolution, which is radically transforming the
work of the office and the factory, has not yet made its mark in the schools.
The enormous popularity in American classrooms of the Apple Ile, which is
now several generations behind in hardware technology, is a testament to the conser-
vative nature of the educational establishment (and to the lack of adequate resources)
Seymour Papert, developer of LOGO, the principal educational computer language,
compares the specter of an entire classroom of children sharing one computer to that
of a classroom sharing one pencil. One pencil, according to Papert, is probably better

than none, but it is not likely to lead to a “pencil revolution.”

Nonetheless, computers are infiltrating the schools. The importance of this
has more to do with laying the groundwork for the future than with its current impact
Before any fundamental transformation of the learning process can take place, a
critical mass needs to be reached in the capabilities of personal computers, their
availability to the student population, their portability, the sophistication of educational
software, and their integration into the learning process."* Let us consider the situ-
ation when such a critical mass is reached. This situation will, in my view, include the
following eight developments

Every child has a computer. Computers are as ubiquitous as pencils and books.
They are portable laptop devices about the size of a large book
They include very high resolution screens that are as easy to read as books
They include a variety of devices for entering information, including a keyboard and a
track ball (or possibly a mouse).
They support high quality two-way voice communication, including natural-language
understanding

429
 The LOGO turtle robot draws pictures
under command of LOGO programs
written by the children. (Photo by Bill
Pierce of Rainbow)

Seymour Papert, codirector of the

MIT Artificial Intelligence Laboratory
and inventor of the popular children’s
programming language LOGO,
widely regarded as a pioneer in the
application of computers to
education. (Photo by Bill Pierce of
Rainbow)

430
+ They are extremely easy and intuitive to use.
- A great variety of high-quality interactive intelligent and entertaining courseware is
available.
+ Computers are integrated into wireless networks.

My emphasis on intelligent courseware needs explanation. Much of the

available computer-assisted instruction (CAI) provide little more than repetitive
exercises that could have been provided just as easily by books.'’ The better pro-
grams do provide a measure of interaction, with sequences dependent on the specific
areas of weakness of the student, but they still develop little sense of the student's
true strengths and weaknesses. CAI now under development has a more ambitious
goal. We all have models of the world that we use to understand and respond to
events and to solve problems, whether they be real-world situations or classroom
exercises. If we make a mistake, it may be simply a matter of a few missing or
inaccurate facts. But more often it reflects a structural defect in the organization of
our knowledge. A good teacher attempts to understand the model that the student is
using. Then if the student’s model, and not just his data base, is faulty, the teacher
devises a strategy to modify the model to more accurately reflect the subject matter.
Such researchers as John Seely Brown at Xerox's Palo Alto Research Center are
attempting to develop a CAI technology which will be able to model the relationships
in the knowledge to be taught, diagnose the presumably weaker models that a
student is starting with, develop a strategy to upgrade the student's models to the
desired ones, and provide entertaining and engaging experiences to carry out the
remedial strategy.'® The objective is to incrementally improve the world models of the
student. As Seymour Papert points out, you cannot learn something unless “you
already almost know it.”
Wireless networks will allow easy sharing of courseware, submissions by
students of papers, exams, courseware responses, and other creations, electronic
mail and other communications (e.g., love notes). By being plugged into international
networks of information, children will have immediate access to the great libraries of
the world right from their school bags. In addition to being able quickly to access
virtually all books, magazines, data bases, and other research materials, there will
exist intelligent software assistants to help students quickly find the information they
are looking for.
The above vision of an optimal educational workstation will obviously not
come forth suddenly. Some aspects of it are becoming available now; others will
emerge over the next decade.'? A personal computer with the necessary attributes
should become available around the end of this century. With the historical ten-year
lag of the educational field in adopting new computer technology, we can expect a
critical mass level of ubiquitous utilization of such technology by the end of the first
decade of the next century.”° If society wakes up to the pivotal role of education in
determining our future economic well-being, then the time period for widespread
implementation of this technology may be compressed. Yet it will still be well into the
first decade of the next century before a broad transformation of the educational
Process is complete.
Let us envision education yet several decades farther into the future. By the
second half of the next century, the future media technologies described earlier will

431
 be as widespread as the various video technologies of today. The greatest impact of
the media of the future will be in education. A homework assignment, for example,
might be to participate in the Constitutional Canvention of 1787 and debate the
founding fathers on the separation of powers between the three branches of the U.S.
government. A subsequent assignment might be to negotiate the final language on
behalf of the executive branch: see if you can get a better deal for the presidency on
war powers. Your submission would be the actual debates that you participated in,
and your teacher watches them in the same way that you originally did: in a totally
realistic three-dimensional projected holographic environment with nearly perfect
resolution. The founding fathers that you interact with are extremely lifelike artificial
people that can hear you, understand what you are saying and respond to you just as
the original founding fathers might have. They will be programmed to tolerate a young
student barging in on their constitutional convention and engaging them in debate.
They may also have better comprehension of contemporary language than the real
founding fathers of two hundred years ago might be expected to have had.
For those of us who do not want to wait until these future media technolo-
gies are perfected, we can go back right now and engage the founding fathers in
debate. We just have to use our imaginations.

+ Communications

Of what lasting benefit has been man’s use of science and of the new instruments which his research
brought into existence? First, they have increased his control of his material environment. They have
improved his food, his clothing, his shelter; they have increased his security and released him partly from
the bondage of bare existence. They have given him increased knowledge of his own biological
processes so that he has had a progressive freedom from disease and an increased span of life. They are
illuminating the interactions of his physiological and psychological functions, giving promise of an
improved mental health.
Science has provided the swiftest communication between individuals; it has provided a
record of ideas and has enabled man to manipulate and to make extracts from that record so that
knowledge evolves and endures throughout the life of a race rather than that of an individual.
Vannevar Bush, As We May Think, 1945

By early in the next century, personal computers will be portable laptop devices
containing cellular phone technology for wireless communication with both people
and machines. Our portable computers will be gateways to international networks of
libraries, data bases, and information services.
We can expect the personal computers of 2010 to have considerable
knowledge of where to find knowledge. They will be familiar with the types of
information contained in our own personal data bases, in the data bases of companies
and organizations to which we have access, as well as to all subscription and public-
information services available through (wireless) telecommunications. As described
earlier, we shall be able to ask our personal computers to find, organize, and present
diverse types of information. The computer will have the intelligence to engage us in
dialogue to clarify our requests and needs, to access knowledge from other machines
and people, and to make organized and even entertaining presentations.
Software should be highly standardized by the end of this century. In
general, commercially available software packages will work on any computer with

432
Abrief history of human communica-
tion. (Data from Dataquest) Modern
man

| ee
emerged

oo |-—=

| -=

| ee
|-=
|-—=
= -o -o | =

ew
ee

D
=

es coe eg oe | eo
ow -o -o| —

|=
and

-<o|
<2 — «os |

aw <@ ow <o |
|

|
|

|
began

«<w <<
ew
ee oe
2-2

oe

oe
== = -o

«= -= -= -2 -—@

-w
oo — -o

-o
= © -2 «2 =

me
oe

ee
=<

ce
using
language

eee ee

ees
ws
a em ew ew
oe

ow

ee mee
—- © -—© <2 «ww -@

2 -@ -w -o

<= «<<
=

we =
1,400
generations

es
cm
oe
=

oe -w ow ow
ago

<=
os
«=
Writing

ee
oS oe
©
=

oe
|e e
De

a
a
was

ED

ee
2
<2
=

ee eee | me
invented

- <> |<

© <e|—
|<
200

>)
© «2
«© © generations

«— 2
—« —©
© ©

<2 ae

—- —
OE
2-2.

ee
Se
<<
«©

A a
a
ago

—-
a

82
©g

8&
5
=
first

—-_
te

_—-_>
ee
>

—_

al
o-

=|
=

=
=
=
=

—
printed 20

=
generations
The invention of
ago
the computer
occurred less than
two generations ago

sufficient capability. Standard protocols and formats will be in place for all types of
files: text, spreadsheets, documents, images, and sounds. Paper will be regarded as
just another form of information storage. With highly accurate and rapid technology
available for character and document recognition, paper will be a medium easily
accessed by both people and machines.
Indeed, the advent of the electronic document has not caused the end of
paper. The so-called “paperless office” actually uses more paper than its predeces-
sor. U.S. consumption of paper for printed documents increased from 7 million tons in
1959 to 22 million in 1986.2’ American business use grew from 850 billion pages in
1981 to about 2.5 trillion pages in 1986 and is expected to hit 4 trillion pages in
1990.22 While computers make it possible to handle documents without paper, they
also greatly increase the productivity of producing paper documents.
Telephones will routinely include video capability. Later in the next century
the video images will have improved to become highly realistic, moving, three-
dimensional, projected, holographic images with nearly perfect resolution. A phone
call with a friend or business associate will be very much like visiting that person. It
will appear that they are sitting with you in your living room or office. The only
limitation will be that you cannot touch one another.
Even this limitation will eventually be overcome. Once we create robotic
imitations of people that are extremely lifelike in terms of both look and feel, a robotic
person imitator could be made to move in exactly the same way as the real person
hundreds or thousands of miles away. Thus, if two people who are apart wanted to
spend time together, they would each meet with a robotic imitator of the other.
imitator would move in exactly the same way (and at nearly the same time) as the
remote real person by means of high-speed communication and robotic techniques. In
other words, you lift an arm, and your robotic imitator hundreds of miles away lifts its
arm in exactly the same way. One problem to overcome will be the slight communica-
tions delay if the two humans are a few thousand miles apart. Artificial intelligence
The Impact On

may be able to help out here by anticipating movements. Using this type of communi-
cation service of the late twenty-first century, couples may not necessarily have to be
near each other to maintain their romantic relationships. (| have not yet checked with
my wife on this, however.)
The advent of videophones, even of a conventional two-dimensional type,
will place new demands on telephone etiquette. We may not always want to engage
in a call with video. There will, of course, always be the option of turning the picture
off (in either direction), but doing so may involve an explanation that we currently do
not have to deal with. (In fact, the widespread adoption of cellular technology, even
without pictures, will also put a strain on telephone etiquette. It is now feasible to be
“away” from our telephones when we are busy. But if everyone has a phone in their
wrist watch, it may become harder to avoid answering the phone.)
One major impact of advanced communications technology will be on the
nature of our cities. Cities first developed to facilitate manufacturing and transporta-
tion and thus tended to be located near ports and rivers. With highways and railways
providing greater flexibility in transporting goods and people, a primary purpose of the
city shifted to communication. Congregating people in one place facilitated their ability
to meet and conduct business. But if we can “meet” with anyone regardless of
where we are and our computers can easily share information through wireless tele-
communications networks, the need for cities will diminish. Already our cities are
spreading out, and this trend will accelerate as the communication technologies
described above become available. Ultimately, we will be able to live anywhere and
still work, learn, and play with people anywhere else on the globe. The world will
become one city, the ultimate realization of McLuhan’s vision of the global village.

- Warfare

When all else fails, the future still remains.

Christian Bovee

Knowledge is power and permits the wise to conquer without bloodshed and to accomplish deeds
‘surpassing all others.
Sun Tzu (Chou dynasty philosopher and military strategist), The Art of War (fourth century 8.c.)

Warfare and potential for warfare is taking a paradoxical turn in the last half of the
twentieth century. There is increasing reliance, at least by the more developed
nations, on “smart” weapons and a rapid evolution of such weapons. Missiles can be
launched from air, ground, or sea hundreds and in some cases thousands of miles
from their intended targets. These weapons find their way to their destinations using
a variety of pattern-recognition and other computer technologies. Pilot's Assistants,
for example, are beginning to provide pilots with an electronic copilot that helps fly,
navigate, locate enemy targets, plot weapon trajectories, and other tasks. Recent
military engagements which utilized such technology have resulted in more accurate
destruction of enemy targets and substantially less unintended damage to neighbor-
ing civilian populations and facilities (although there are still a few bugs in these
systems). Among military establishments that can afford routine use of these

asa
Military high tech before the
computer, Leonardo da Vinci's
treadmill-powered cross-bow.
(Supplied by North Wind Picture
Archives)

A device proposed by Leonardo for

foiling the enemy's scaling ladders.
(Supplied by North Wind Picture
Archives)

technologies, a profound reconsideration of military tactics is underway. The primary

thrust is to replace shotgun strategies with the careful pinpointing of targets
Not all nations have access to these new technologies. While Iran and Iraq
do possess small numbers of such advanced weapons, they still primarily used
weapons and battlefield tactics of World War II vintage during their recent war. The
Soviet Union in Afghanistan also used relatively unsophisticated weapons and tactics
This reflects a fundamental reality in the balance of power between East and West
the Warsaw Pact forces are at least a decade behind the NATO forces in Al and
computer technologies. Indeed, our primary strategy in countering the numerically
superior forces of the Soviet Union and its allies is to rely on our superiority in the
intelligence of our conventional (or nonnuclear) weapons as the first line of defense
and on our nuclear threat as the second line.”
In accordance with this second line, the United States and NATO have been
unwilling to make a declaration of no first use of nuclear weapons, stating that we
may use nuclear weapons in the event of a conventional attack on Europe. How Ver,
o

if we can improve our intelligent but conventional weapons to a point where our
confidence in the first line strategy is sufficiently enhanced, then the western allies
would be in a position to issue a no-first-use pledge and forego the nuclear threat in
Europe. Recent political changes in Eastern Europe and the apparent collapse of
communism in many countries may hasten such a development. There are active de-
velopment programs to create a new generation of, for example, ground-to-ground
and air-to-ground antitank missiles that are capable of being launched from hundreds
of miles away, follow irregular trajectories, search intelligently for their targets, locate,
and destroy them.*4 Once perfected, these missiles could be launched without
precise knowledge of the location of the enemy positions. They are being designed to
use a variety of artificial vision, pattern-recognition, and communication technologies
to roam around and reliably locate enemy vehicles. Friendly forces would be avoided
by a combination of electronic communication and pattern-recognition identification.
To the extent that friendly targets are avoided by electronic communication, the
reliability and security of the encoding protocols, another important area of advanced
computer research, will obviously be crucial. Anticipated progress in intelligent
weaponry was a major factor behind the recommendation of four former high ranking
American advisers, including Robert McNamara and McGeorge Bundy, for an Ameri-
can no-first-use pledge in the spring 1982 issue of Foreign Affairs.?>
One result of these changes is the prospect of diminished civilian destruc-
tion from war, but few observers are heralding this development. The reason for this
is, of course, the enormous increases in the destructive capability of weapons that
have also occurred. As terrifying and destructive as the atomic weapons that ended
World War II were, the superpowers now possess more than a million times more
destructive power. Children growing up today belong to the first generation in history
born into an era in which the complete destruction of the human race is at least
plausible. Experts may debate whether or not “nuclear winter” (the catastrophic
global change in climate that some scientists have predicted would follow a large-
scale exchange of nuclear weapons) really has the potential to end all human life. The
end of the human race has never before been seriously debated as a possibility.
Whether an all-out nuclear war would actually destroy all human life or not, the
overwhelming destruction that would certainly ensue has created an unprecedented
level of terror, under which all of the world’s people now live. Ironically, the fear of
nuclear conflict has kept the peace: there has not been a world war for nearly half a
century. It is a peace from which we take limited comfort.
The most evident technologies behind this radical change in the potential
destructiveness of warfare are, of course, atomic fission and fusion. The potential for
worldwide catastrophe would not be possible, however, without weapon-delivery
systems, which rely heavily on computer intelligence to reach their destinations. The
power of conventional munitions has also grown substantially, and political and social
inhibitions against their use are far less than those for nuclear weapons. Thus, the
possibility of eliminating nuclear weapons from the European theater paradoxically
evokes fear that such a development would make Europe “safe” fora conventional
war that would still be far more destructive than World War II. This duality in the
development of military technology—the advent of weapons for fighting weapons
rather than civilian populations and the potential for greatly enhanced destruction—will

continue.

436
 Let us consider military technology and strategy several decades into the
next century, at which time these trends should have fully matured. By that time
flying weapons (missiles, robot planes, and flying munitions) will be highly self-reliant.
They will be capable of being launched from virtually any place on earth or from space
and still finding their targets by using a combination of advanced vision and pattern-
recognition technologies. They will obviously need the ability to avoid or counteract
defensive weapons intended for their destruction. Clearly, of primary strategic impor-
tance will be the sophistication, indeed the intelligence, of both the offensive and
defensive systems of such weapons. Geography is already losing its strategic impor-
tance and should be a relatively minor factor several decades from now. Such slow
moving vehicles as tanks and ships, as well as battle stations, whether land, sea, air,
or space based, will be vulnerable unless defended by arrays of intelligent weapons.
Most weapons today destroy their targets with explosions or, less often,
bullets. Within the next few decades it is likely that laser and particle beam weapons
will be perfected. This will provide such fast-moving weapons as missiles a variety of
means for both offense and defense.
Planes, particularly those closest to combat, will not require pilots. With
sophisticated enough electronic technology, there is no reason why planes cannot be
directed from afar by either human or machine intelligence. Of course, reliable and
secure communications will be essential to prevent an enemy from taking control of
remote-controlled robot aircraft. Indeed, the three Cs—command, control, and
communication—are emerging as the cornerstone of future military strategy.7°
In general, the interactions of future weapons are likely to be so fast that
human reflexes will not be the primary criterion of tactical success. Weapons will
utilize a variety of their tactical offensive and defensive capabilities within seconds or
even milliseconds when meeting comparable enemy systems. In such encounters,
the most capable and reliable electronics and software will clearly prevail
| remember as a child reading a tale about a very advanced civilization that
had outlawed war and replaced it with a more refined form of conflict. Rather than
resort to deadly weapons, two societies challenging each other for supremacy
engaged in a game of chess. Each society could select their best master player or use
a committee. As | recall, no one thought to use machine intelligence for this task.
Whoever won the board conflict won the war and, apparently, the spoils of war. How
this was enforced was not discussed, but one can imagine that warfare in the future
may not be all that dissimilar from this tale. If human reflexes and eventually human
decision making, at least on a tactical level, are replaced with machine intelligence,
then two societies could let their machines fight out the conflict and let them know
who wins (or perhaps it would be obvious who had prevailed). It would be convenient
if the actual conflict took place in some remote place, like outer space. Here the
enforcement of the winner's prerogatives is obvious: the losing society will have lost
its machine defenders, which will render it defenseless. It will have no choice but to
submit to the victor.
This scenario differs in one important respect from the story about conflict
resolution through chess. In the terms | used earlier, chess represents level 2 intelli-
gence and is thus amenable to recursive software techniques combined with massive
 amounts of computer power. Battling weapons, on the other hand, require level 3
intelligence (the ability to abstract) as well as advanced forms of pattern recognition.
They also require reliability. One controversial aspect of this new technology is the
extent to which we can rely on these extremely complex systems, considering the
limited opportunity we will have to test them under realistic wartime conditions. This
issue is particularly salient for the highly centralized communication networks needed
for command and control.?”
Can we take any comfort from this vision? It is entirely possible that military
engagements decades hence may involve relatively few casualties, particularly of a
civilian nature. On the other hand, there is no guarantee that warfare will be con-
strained to weapons fighting weapons. The tactic of holding large civilian populations
hostage will continue to have its adherents among military strategists. What is clear,
however, is that a profound change in military strategy is starting to take place. The
cornerstones of military power from the beginning of recorded history through recent
times—geography, manpower, firepower, and battle-station defenses—are being
replaced by the sophistication of computerized intelligence and communications.
Humans will direct battlefield strategy, but even here computers will play a crucial
role. Yet humans will still be the underlying determinants of military success. Military
strength will be a function of the sophistication of the technology, but a society's
leaders, scientists, engineers, technicians, and other professionals will create and use
the technology. At least, that is likely to remain the case for the next half century.

* Medicine

Aself-balancing 28-jointed adapter-base biped; an electro-chemical reduction plant, integral with

segregated stowages of special energy extract in storage batteries for subsequent actuation of thousands
of hydraulic and pneumatic pumps with motors attached; sixty-two thousand miles of capillaries; ... the
whole extraordinary complex mechanism guided with exquisite precision from a turret in which are
located telescopic and microscopic self-registering and recording range finders, a spectroscope, etc.; the
turret control being closely allied with an air-conditioning intake and exhaust, and a main fuel intake . . .
R. Buckminster Fuller, “A Definition of a Man”

A projection of current trends gives us the following picture of medicine early in the
next century: A variety of pattern-recognition systems will have become a vital part of
diagnosis. Blood tests will be routinely analyzed by cybernetic technicians. Today's
routine blood test generally involves a human technician examining only about 100
cells and distinguishing only a few cell types; the blood test of the early twenty-first
century will involve automatic analysis of thousands or even a million blood cells as
well as a thorough biochemical analysis. With such extensive analysis, precursor cells
and chemicals that indicate the early stages of many diseases will be reliably de-
tected. Most people will have such devices in their homes. A sample of blood will be
painlessly extracted on a routine basis and quickly analyzed
Electrocardiograms will be analyzed entirely by computer; indeed, proto-
types of this technology exist today. Our wristwatches will monitor cardiac functions
and other biological processes that might require immediate attention, in addition to
diagnosing less acute problems. Particularly powerful computerized monitoring will
attend anyone in a special-care situation, such as a hospital, nursery, or old-age home.

ass
 Apart from blood tests there will be almost complete reliance on diagnosis
by noninvasive imaging (like sonic and particle-resonance imaging). The instantly
generated hard-copy images will include the computer's diagnostic findings. The
images themselves will be highly realistic computer-generated views of the interiors
of our bodies and brains, rather than the often confusing, hard-to-interpret pictures
from some of today’s imaging devices. While human diagnosticians will continue for
many years to examine images from X-ray machines, CAT scanners, nuclear-mag-
netic-resonance scanners, and sonic scanners, a high degree of confidence will
ultimately be placed in the ability of computerized systems to detect and diagnose
problems automatically.
Lifetime patient records and histories will be maintained in nationally (or
internationally) coordinated data banks in place of today’s disorganized system of
partial, fragmented, and often illegible records. These records will include all imaging
data, the complete readouts of our home blood tests and wristwatch monitoring

_ Amedical workstation of the future

will instantly connect the doctor to
information from the patient-records Patient Michae
department, images from the
Complaints
radiology department, references .
from the medical library and a ' = ie
suggested diagnosis by the work-
station itself. (Photo by Larry Fagan t J Cough
and Mark Friss) —
5 {

| 111. Radiographic Findingsof Tuberculosis,

ntral to the diagn
he most common
sin » patchy infiltrate [ocared is
apical posterion acca ot Ath

systems. Intelligent software will be available to enable this extensive data bank to be
analyzed and accessed quickly by both human and machine experts
Expert systems will influence virtually all diagnostic and treatment deci-
sions. These expert systems will have access to the output of a patient's most recent
imaging and biochemical analyses, as well as to the entire file of all such past exams
and monitored data. They will also have access to all internationally available research
data and will be updated on a daily basis with the latest research insights. The written
reports of these expert systems will be reviewed by human doctors in critical or
complex cases, but for more routine problems the machine’s opinions will be relied
upon with little or no review.”8
The designing of drugs will be entirely different from present methods.
Most drugs on the market today were discovered accidentally. In addition to their

beneficial effects, they often cause a number of undesirable side effects. Further,
their positive effects are often indirect and not fully effective. In contrast, most drugs

439
of the early twenty-first century will have been specifically designed to accomplish
their missions in the most effective and least harmful ways. Drug designers will work
at powerful computer-assisted design workstations that will have access to relatively
complete mathematical models of all of our biochemical processes. The human drug-
design engineers will specify key design parameters, and the computer will perform
most of the detailed design calculations. Human biochemical simulation software will
allow drugs to be tested with software before any actual drugs are manufactured.”
Human trials will still be required in most cases (at least this will be true during the
first half of the next century), but the simulators will ultimately be sufficiently reliable
that the lengthy multistage process of animal and human testing now required will be
substantially shortened.
One class of drugs will be similar to the smart weapons described in the
previous section. These drugs will be actual living cells with a measure of intelligence.
Like cells of our natural immune systems, they will be smart enough to identify an
enemy pathogen (bacteria, virus, etc.) and destroy or pacify it. Again like immune
cells, once they have completed their missions, they may either self-destruct or
remain on call to defend against a future pathogen invasion. Another class of drugs
will help overcome genetic diseases. Computer-designed genes will be distributed to
our cells by specially designed viruses, which will essentially “infect” the body with
the desired genetic information.°
Surgical operations will make extensive use of robotic assistants. In types of
surgery requiring very precise performance, e.g., in eye surgery, the actual operation
will be carried out by the robot, with human doctors overseeing the operation.
Research will be similar to drug design. Most research will be carried out on
software that simulates human biochemical processes. Experiments that would now
take years will be carried out in minutes. Reporting will be instantaneous, with key
results feeding into data bases that allow access by other humans as well as by
computer expert systems.*"
Will these innovations improve our health and well-being? The answer is
almost certainly yes. Heart disease and cancer are likely to be conquered by early in
the next century. Of course, we have the opportunity right now to dramatically reduce
the incidence of these diseases by applying what is already known about the crucial
role of diet and life style. But that is a subject for a different book.
We have already doubled the average life expectancy in Europe, Japan, and
the United States since the beginning of the first industrial revolution two centuries
ago. The potential exists to substantially increase it again by the end of the twenty-
first century
With machines playing such crucial and diverse roles, what will the doctors
and nurses of the twenty-first century do? Their major role will be in research and in
the organization of medical knowledge. Committees with both human and machine
intelligence will review all research findings with a view to incorporating new rules
and recommendations into our expert systems. Today new research knowledge filters
into the medical community in a slow, inconsistent, and haphazard fashion. The future
dissemination of knowledge and recommendations will be very rapid. Doctors will
continue to be involved in strategic medical decisions and will review diagnostic and
treatment recommendations in complicated cases. Yet some of the new technology

aa0
 will bypass the doctor. There will be a trend toward individuals taking responsibility for
their own health and utilizing computerized diagnostic and remedial methods directly
One area that will still require human attention in the early twenty-first
century will be comfort and caring. Machines will not play a significant role here until
mature versions of the advanced media technologies described earlier become
available later in the century.

+ The Handicapped

A primary interest of mine is the application of computer technology to the needs of

the handicapped. Through the application of computer technology, handicaps associ-
ated with the major sensory and physical disabilities can largely be overcome during
the next decade or two. | am confident of this development because of the fortunate
matching of the strengths of early machine intelligence with the needs of the handi-
capped. The typical disabled person is missing a specific skill or capability but is
otherwise a normally intelligent and capable human being. It is generally possible to
apply the sharply focused intelligence of today’s machines to ameliorate these
handicaps. A reading machine, for example, addresses the inability of a blind or
dyslexic person to read, probably the most significant handicap associated with the
disability of blindness.
In the early twenty-first century lives of disabled persons will be far different
than they are today. For the blind, reading machines will be pocket-sized devices that
can instantly scan not only pages of text but also signs and symbols found in the real
world. These machines will be able to read with essentially perfect intonation and
with a broad variety of voice styles. They will also be able to describe pictures and
graphics, translate from one language to another, and provide access to on-line
knowledge bases and libraries through wireless networks. Most blind and dyslexic
persons will have them, and they may be ubiquitous among the rest of the population.
Blind persons will carry computerized navigational aids that will perform the
functions of seeing-eye dogs, only with greater intelligence than today's canine
navigators. Attempts up to now at electronic navigational assistants for the blind have
not proved useful. Unless such a device incorporates a level of intelligence at least
comparable to a seeing-eye dog, it is not of much value. This is particularly true since
modern mobility training can provide a blind person equipped only with an ordinary
cane with substantial travel skills. | personally know many blind people who can travel
around town and even around the world with ease. With the intelligent navigational
aids of the future, travel skills for the blind will become even easier.
Ultimately, compact devices will be built that combine both reading and
navigational capabilities with the ability to provide intelligent descriptions of real-world
scenes on a real-time basis. At that stage of machine evolution they are probably
more accurately called seeing machines.*? Such a machine would be like a friend that
could describe what is going on in the visible world. The blind user could ask the
device (verbally or in some other way) to elaborate on a description, or he could ask it
questions. The visual sensors of such a device could be built into a pair of eyeglasses,
although it may be just as well to pin it on the user's lapel. In fact, these artificial eyes
need not only look forward; they may as well look in all directions. And they may have
better visual acuity than normal eyes. We may all want to use them.

aa
 The deaf will have hearing machines that can display what people are
saying. The underlying technology required is the Holy Grail of voice recognition:
combining large-vocabulary recognition with speaker independence and continuous
speech. Early versions of speech-to-text aids for the deaf should appear over the next
decade. Artificial hearing should also include the ability to intelligently translate other
forms of auditory information, such as music and natural sounds, into other modali-
ties, such as vision and touch.
Eventually we may find suitable channels of communication directly into the
brain to provide truly artificial sight and hearing. But in any case, there will certainly be
progress in restoring lost hearing and sight.
The physically handicapped (paraplegics and quadriplegics) will have their
ability to walk and climb stairs restored, abilities that will overcome the severe access
limitations wheel chairs impose. Methods to accomplish this will include exoskeletal
robotic devices, or powered orthotic devices, as they are called. These devices will be
as easy to put on as a pair of tights and will be controlled by finger motion, head
motion, speech, and perhaps eventually thoughts. Another option, one that has
shown promise in experiments at a number of research institutes, is direct electrical
stimulation of limb muscles. This technique effectively reconnects the control link that |
was broken by spinal cord damage.
Those without use of their hands will control their environment, create
written text, and interact with computers using voice recognition. This capability
already exists. Artificial hand prostheses controlled by voice, head movement, and
perhaps eventually by direct mental connection, will restore manual functionality.
Substantial progress will be made in courseware to treat dyslexia (difficulty
in reading for neurophysical reasons other than visual impairment) and learning
disabilities. Such systems will also provide richer learning experiences for the
retarded.
Perhaps the greatest handicap associated with sensory and physical
disabilities is a subtle and insidious one: the prejudice and lack of understanding often
exhibited by the general public. Most handicapped persons do not want pity or
charity; instead, they want to be respected for their own individuality and intelligence.
We all have handicaps and limitations; those of a blind or deaf person may be more
obvious, but they are not necessarily more pervasive or limiting. | have worked with
many disabled persons, and | know from personal experience that they are as capable
as other workers and students at most tasks. | cannot ask a blind person to drive a
package across town, but | can ask him to give a speech or conduct a research
project. A sighted worker may be able to drive a car, but he will undoubtedly have
other limitations. The lack of understanding many people have of handicapped
persons is evident in many ways, some obvious, some subtle. By way of example, |
have had the following experience on many occasions while eating a meal with a blind
person in a restaurant. The waiter or waitress will ask me if my blind friend wants
dessert or if he wants cream in his coffee. While the waiter or waitress obviously
intends no harm or disrespect, the message is clear. Since there is no indication that
the blind person is also deaf, the implication is that he must not be intelligent enough
to deal with human language.
A not unimportant side benefit of intelligent technology for the handicapped
should be a substantial alteration of these negative perceptions. If the handicaps

442
 resulting from disabilities are significantly reduced, if blind people can read and
navigate with ease, if deaf persons can hold normal conversations on the phone, then
we can expect public perceptions to change as well. When blind, deaf, and other
disabled persons take their place beside us in schools and the workplace and perform
with the same effectiveness as their nondisabled peers, we shall begin to see these
disabilities as mere inconveniences, as problems no more difficult to overcome than
poor handwriting or fear of public speaking or any of the other minor challenges that
we all face.

+ Music

|... begin to feel an irresistible drive to become a primitive and to create a new world.
August Strindberg, from a letter

Let us step well into the next century for a view of music when current trends have
been fully established. There will still be acoustic instruments around, but they will be
primarily of historical interest, much like harpsichords are today. Even concert pianists
will accept electronic pianos as fully the equivalent of the best acoustic pianos. All
nuances of sound, including interstring resonances, will be captured. Of course, toa
pianist, the feel of a piano is important, not just the sound. Piano actions with time-
varying magnetic actions will faithfully emulate the feel of the finest top-of-the-line
grand pianos. Pianists will prefer the electronic versions because they are more
reliable, have extensive additional capabilities, and are always perfectly in tune. We
are close to this vision today. Controllers emulating the playing techniques of all con-
ventional instruments, such as guitar, violin and drums, will have largely replaced their
acoustic counterparts. Perfect synthesis of acoustic sounds will have long been
established.
With the physical link between the control and generation of musical sound
having long been broken, most musicians will have gravitated to a new generation of
controllers. These new instruments will bear little resemblance to any instrument of
today and will enable musicians optimal expressive control.
While the historically desirable sounds of pianos and violins will continue to
be used, most music will use sounds with no direct acoustic counterpart. Unlike many
synthetic sounds of today, these new musical sounds will have greater complexity
and musical depth than any acoustic instrument we are now familiar with
The concept of a musical sound will include not only sounds that simply
Start and stop but also a class of sounds that changes characteristics according to a
number of continuously controllable parameters. For example, we could use all of our
fingers to control ten continuous parameters of a single sound. Such a sound will exist
as a time-evolving entity that changes according to expression applied in ten different
dimensions. With extremely powerful parallel computing devices, the distinction
between real-time and non-real-time sound modification will largely disappear; virtually
everything will happen in real-time. The ability to modify sound with real-time continu-
ous controls will be regarded as more important than the individual sounds them-
selves.
Computers will almost always respond in real time, although people will not
always create music in real time. There will continue to be a distinction between

443
The ImpactOn
music composition and music performance. Composition will not mean writing down
music notation. It will refer to the creation of music in which the creation takes
substantially longer than the piece created. Sequencers that record all continuous
controls with high resolution will allow the “massaging” of a work of music in the
same way that we now work on a printed document. Music created in this way will
certainly not be subject to the limitations of finger dexterity (this has already been
largely achieved with contemporary sequencers). Once composed, high-quality
notation will be instantly available. It is likely that forms of music notation more
satisfactory than the current five-line staffs will be developed over the next several
decades to keep pace with the added dimensions of control, added sounds, and
added sound modifications.
Live music performance will continue to have the same appeal that it does
today. While much of what is heard may have been previously programmed, musi-
cians’ sharing with their audience musical expression through real-time (and live)
control will continue to be a special form of human communication. The preparation of
a musical performance will involve practice and learning of the musical material as
well as preparation of the knowledge bases of the musical instruments. Cybernetic
musicians generating lines of accompaniment and counterpoint will be commonplace.
The intelligence of these software-based musical accompanists will be partially built
into the instruments and partially programmed by the musicians as they prepare a
performance.**
Intelligent software incorporating extensive knowledge of musical theory
and styles will be extensively used by professional musicians in creating musical com-
Positions and preparing performances, by amateur musicians, who can jam with their
computerized instruments, and by students learning how to play. The advent of
musically knowledgeable software-based accompanists will provide children with
exciting musical experiences at early stages of their knowledge and skill acquisition.
Music will not necessarily take the form of fixed works that we listen to
from beginning to end. One form of musical composition might be a set of rules (or a
modification of a set of rules) and expression structures that together can generate an
essentially unlimited number of actual pieces. This “work” would sound different
every time we listen to it, although each such listening would share certain qualities,
which qualifies it to be considered a single work. Compositions will also allow the
listener to control and modify what he is hearing. We could control the entry and exit
of various instruments and lines of music or control the evolution and emotional
content of a piece without necessarily having the musical knowledge of a musician.
This type of musical work would allow us to explore a musical world that we could
have some influence on. A musical work could respond to our physical movements or
even our emotions, which the computer-based system generating the actual sounds
could detect from the subtleties of our facial expressions or perhaps even our brain-
waves.

In this way music becomes potentially more than entertainment. It can have
powerful effects on our emotional states, influencing our moods and affecting our
learning. There will not be a sharp division between the musician and nonmusician.
Increasingly and regardless of musical talent and training, we shall all be able to
express our feelings through music.

444
+ Politics

Economics, sociology, geopolitics, art, religion all provide powerful tools that have sufficed for centuries
to explain the essential surfaces of life. To many observers, there seems nothing truly new under the
sun—no need for a deep understanding of man’s new tools—no requirement to descend into the
microcosm of modern electronics in order to comprehend the world. The world is all too much with us.
Nonetheless, studying economics and other social sciences, | began to realize that the old
disciplines were breaking down, the familiar categories slipping away. An onslaught of technological
progress was reducing much of economic and social theory to gibberish. For example, such concepts as
land, labor, and capital, nation and society—solemnly discussed in every academic institution as if
nothing had changed—have radically different meanings than before and drastically different values. Yet
the vendors of old expertise continue on as if nothing had happened.
Laws get passed, editorials written, speeches delivered, soldiers dispatched, for all the world
as if we still traveled in clipper ships and communicated chiefly by mail.
Jean Kirkpatrick, for example, gave a speech, quoted respectfully in the Wall Street Journal,
in which she said it was impossible to understand what is going on in the world without a comprehension
of geography, “an idea of where things are.” It is a common notion. . . . Visit the Pentagon, or the New York
Times, and everywhere there are maps, solemnly defining national borders and sovereign territories. No
one shows any signs of knowing that we no longer live in geographic time and space, that the maps of
nations are fully as obsolete as the charts of a flat earth, that geography tells us virtually nothing of
interest about where things are in the real world.
The worldwide network of satellites and fiber optic cables, linked to digital computers,
television terminals, telephones and databases, sustain worldwide markets for information, currency and
capital on line 24 hours a day. Boeing 747s constantly traversing the oceans foster a global community of
commerce. The silicon in sand and glass forms a global ganglion of electronic and photonic media that
leaves all history in its wake. With other new technologies of material science, bioengineering, robotics,
and superconductivity, all also heavily dependent on the microchip, informations systems are radically
reducing the significance of so-called raw materials and natural endowments, nations and ethnic
loyalties, material totems and localities.
Israel, a desert-bound society, uses microelectronic agricultural systems to supply eighty
percent of the cut flowers in Europe and compete in avocado markets in New York. Japan, a set of barren
islands, has used microelectronic devices to become one of the globe's two most important nations. In an
age when men can inscribe worlds on grains of sand, conventional territory no longer matters. . . .
Today the most important products are essentially made of the sand in chips of crystalline
silicon. Their worth derives from the ideas they embody. They are information technologies and their value
is invisible. Yet intellectuals, supposedly masters of ideas, refuse to believe in any value they cannot see
or weigh. ...
While the cost-effectiveness of computer components and related products has risen several
million fold and the price of a transistor has sunk from $9.00 in 1955 to about eight ten-thousandths of a
cent in 1987, the estimates of national productivity have entirely ignored the change. Once again, things
that drop in price are assumed to be dropping in value. Yet it is the astronomical reduction in the price of
computing that has made it the most important force and most valuable industry in the world economy.
George Gilder, The Message of the Microcosm

Where did that knowledge exist? . . . If all records told the same tale, then the lie passed into history and
became truth.
George Orwell, 1984

As | have pointed out throughout this book, wealth and power in the age of intelligent
machines is increasingly becoming a function of innovation and skill. The
of power during the first industrial revolution—geography, natural resources, and
manual labor—are rapidly diminishing in importance and relevance. We are entering 4
world in which wealth can be beamed across the world by satellite, smart weapons

445
 can reach their destinations from thousands of miles away, and some of the most
The

powerful technologies in history require only tiny amounts of material resources and
electricity. We can only conclude that the strategic variables controlling our future are
becoming technology and, in particular, the human intellectual resources to advance
technology.
For thousands of years governments have demonstrated the possibility of
forcing people to perform manual labor (although even here productivity is certainly
diminished by coercion). It is a fortunate truth of human nature that creativity and
innovation cannot be forced. To create knowledge, people need the free exchange of
information and ideas. They need free access to the world’s accumulated knowledge
bases. A society that restricts access to copiers and mimeograph machines for fear of
the dissemination of uncontrolled knowledge will certainly fear the much more
powerful communication technologies of personal computers, local area networks,
telecommunication data bases, electronic bulletin boards, and all of the multifarious
methods of instantaneous electronic communication.
Controlled societies like the Soviet Union are faced with a fundamental
dilemma. If they provide their engineers and professionals in all disciplines with
advanced workstation technology, they are opening the floodgates to free communi-
cation by methods far more powerful than the copiers they have traditionally
banned.*®° On the other hand, if they fail to do so, they will increasingly become an
ineffectual third-rate power. Russia is already on a par with many third-world countries
economically. Russia is a superpower only in the military sphere. If it continues to
stagnate economically and fails to develop advanced computer technologies, this type
of power will dissipate as well.
Innovation requires more than just computer workstations and electronic
communication technologies. It also requires an atmosphere of tolerance for new and
unorthodox ideas, the encouragement of risk taking, and the ability to share ideas and
knowledge. A society run entirely by government bureaucracies is not in a position to
provide the incentives and environment needed for entrepreneurship and the rapid
development of new skills and technologies.
From all appearances, some of the leaders of the Communist world have
had similar thoughts. Mikhail Gorbachev's much-heralded campaigns of glasnost
(openness) and perestroika (restructuring) have taken some initial steps to open com-
munication and provide market incentives. Important steps have been taken in many
of these societies toward achieving individual liberty. But these are only the first steps
in what will need to be a long journey to complete a full transformation. Already the
forces of reaction in China have taken a major step backward. What is not yet clear is
the ability of these societies to succeed in moving deeply entrenched bureaucracies.
What is clear, however, is that the pressures for such change will not go away.
Should these societies opt instead for a continuation of the controlled
society, they will also find computers to be of value. Computers today play an indis-
pensable role in legitimate law enforcement; there is no reason why they would not
be equally useful in enforcing any form of state control. With the advanced vision and

446
 networking technologies of the early twenty-first century, the potential will exist to
realize George Orwell's chilling vision in 1984.
Computer technology may lead to a flowering of individual expression,
creativity, and communication or to an era of efficient and effective totalitarian control.
It will all depend on who controls the technology. A hopeful note is that the nature of
wealth and power in the age of intelligent machines will encourage the open society
Oppressive societies will find it hard to provide the economic incentives needed to
pay for computers and their development.

+ Our Concept of Ourselves

We know what we are, but know not what we may be.

William Shakespeare

What will happen when all these artificially intelligent computers and robots leave us with nothing to do?
What will be the point of living? Granted that human obsolescence is hardly an urgent problem. It will be a
long, long time before computers can master politics, poetry, or any of the other things we really care
about. But a “long time” is not forever; what happens when the computers have mastered politics and
poetry? One can easily envision a future when the world is run quietly and efficiently by a set of exceed-
ingly expert systems, in which machines produce goods, services, and wealth in abundance, and where
everyone lives a life of luxury. It sounds idyllic—and utterly pointless.
But personally, | have to side with the optimists—for two reasons. The first stems from the
simple observation that technology is made by people. Despite the strong impression that we are helpless
in the face of, say, the spread of automobiles or the more mindless clerical applications of computers, the
fact is that technology does not develop according to an immutable genetic code. It embodies human
values and human choices. . .. My second reason for being optimistic stems from a simple question: What
does it mean to be “obsolete”?
M. Mitchell Waldrop

As | discussed earlier, | believe that a computer will be able to defeat all human
players at the game of chess within the next one or two decades. When this hap-
pens, | noted, we shall either think more of computers, less of ourselves, or less of
chess. If history is a guide, we will probably think less of chess. Yet, as | hope this
book has made clear, the world chess championship is but one of many accomplish-
ments that will be attained by future machine intelligence. If our approach to coping
with each achievement of machine intelligence is to downgrade the intellectual value
of the accomplishment, we may have a lot of revision to do over the next half century
Let us review some of the intellectual domains that machines are likely to
master in the near future. A few examples of tasks that computers are now beginning
to accomplish include the following: accompanying musical performances, teaching
us skills and areas of knowledge, diagnosing and recommending remedial treatment
for classes of diseases, designing new bioengineered drugs, performing delicate
medical operations, locating underground resources, and flying planes
Amore difficult task for a computer, one that we shall probably see during
the first half of the next century, is reading a book, magazine, or newspaper and
understanding its contents. This would require the computer to update its own
knowledge bases to reflect the information it read. Such a system would be able to
write a synopsis or a critique of its reading. Of comparable difficulty to this task is

aa7
 passing the Turing test, which requires a mastery of written language as well as
extensive world knowledge.
Of at least comparable difficulty would be to watch a moving scene and
understand what is going on. This task requires human-level vision and the ability to
abstract knowledge from moving images. Add the ability for a robot to imitate humans
with sufficient subtlety, and computers will be able to pass a more difficult form of
the Turing test in which communication is not through the written word transmitted
by terminal but rather by live face-to-face communication. For this achievement we
have to go at least to late in the next century
It is clear that the strengths and weaknesses of machine intelligence today
are quite different from those of human intelligence. The very first computers had
prodigious and virtually unerring memories. In comparison, our memories are quite
limited and of dubious reliability. Yet the early computers’ ability to organize knowl-
edge, recognize patterns, and render expert judgements—all elements of human
intelligence—was essentially nonexistent. If we examine the trends that are already
apparent, we can see that computers have progressed in two ways. They have gained
even greater capacities in their areas of unique strength: today’s computers are a
million times more powerful in terms of both speed and memory than their forebears
At the same time, they have also moved toward the strengths of human intelligence.

(Cartoon by Alan Wallerstein)

They are nowhere near that goal yet, but they are certainly getting closer. Today
computers can organize knowledge incorporating networks of relationships, they are
beginning to recognize patterns contained in visual, auditory, and other modalities,
and they can render judgements that rival those of human experts. They have still not
mastered the vast body of everyday knowledge we call common sense, and ironically,
they are particularly weak in the pattern recognition and fine motor skills that children
and.even animals do so well.
Computer intelligence is not standing still. Radical new massively parallel
computer architectures, together with emerging insights into the algorithms of vision,
hearing, and physical skill acquisition, are propelling computers closer to human
capabilities and also continuing to enhance their historical areas of superiority. While
machine intelligence continues to evolve and move in our direction, human intelli-
gence is moving very slowly, if at all. But since we have computers to serve us,
human intelligence may not need to change.
Thousands of years ago, when the religious and philosophical traditions that
still guide Western civilization were being formed, a human being was regarded as
special. We were different from animals and certainly from material things. The
ultimate intelligence in the universe, God, knew about us, and cared about us. Later
on as we learned that the earth on which we stood was not the only celestial body in
the world, we imagined that all the other entities in the sky revolved around us. In this
world view we were special because of our central location. The sun, the moon, the
stars, the comets, and other celestial objects all paid homage to us. Still later when
we realized that the earth was just the third planet orbiting an unremarkable star
located on an arm of an unremarkable galaxy, our view changed again. Then we were
special because of our unique intelligence: We could derive knowledge from informa-
tion. We could contemplate the relationships among the world’s phenomena. We
could create patterns with aesthetic qualities. We could appreciate those qualities.
True, animals shared in this intelligence, but to a much lesser degree, which only
reinforced the uniqueness of the level of intelligence we possessed.
Now we are entering an era in which this latest concept of our uniqueness
will be challenged once again. To be sure, this challenge will not arrive on a single
day. By the time one can seriously argue that computers possess intellectual capabili-
ties comparable to the human species, it will have been at least a century from the
invention of the electronic computer in the late 1940s. We should have time to adjust.
Perhaps we shall return whence we started, with an appreciation of the inherent value
of being human.
 \s artificial intelligence in human society a utopian dream or a
Faustian nightmare? Will our descendants honor us for making
machines do things that human minds do or berate us for irresponsi-
bility and Aubris? Either of these judgments might be made of us, for
like most human projects this infant technology is ambivalent. Just
which aspects of its potential are realized will depend largely on
| Margaret A. Boden social and political factors. Although these are not wholly subject to
deliberate control, they can be influenced by human choice and
The Social Impact of Artificial Intelligence public opinion. If future generations are to have reason to thank us
rather than to curse us, it’s important that the public (and politicians)
of today should know as much as possible about the potential
Margaret Boden is professor of
philosophy and psychology at effects—for good or ill—of artificial intelligence (Al).
the University of Sussex, What are some of the potential advantages of Al? Clearly,
England, and a fellow of the Al can make knowledge more widely available. We shall certainly
British Academy. Her research
see a wide variety of expert systems: for aiding medical diagnosis
work has focused on the
premise that human thinking, and prescription, for helping scientists, lawyers, welfare advisers,
motivation, and even and other professionals, and for providing people with information
personality might be and suggestions for solving problems in the privacy of their homes.
understood in computational
Educational expert systems include interactive programs that can
terms. She is the author of
Artificial Intelligence and help students (schoolchildren or adults, such as medical students) to
Natural Man (2nd edition, 1987), familiarize themselves with some established domain. This would
Minds and Mechanisms (1981), give us much more than a set of useful tools and educational cribs. In
and most recently, Artificial
virtue of its applications in the communication and exploration of
Intelligence in Psychology:
Interdisciplinary Essays (1983). knowledge, Al could revolutionize our capacity for creativity and
problem solving, much as the invention of printing did.
One advantage of having computers in the schoolroom
and elsewhere is that they are not human. Precisely because they are
not, they will not be bored by their human user's questions, nor scorn
their user's mistakes, as another person might. The user may be
ignorant, stupid, or naive, but the computer will not think so.
Moreover, what looks like ignorance, stupidity, or naiveté is often a
sort of exploratory playing around with ideas that is the essence of
learning and of creativity. Many children have their self-confidence
undermined by their teacher's explicit or implicit rejection of their
attempts at self-directed thinking.
 Similarly, many people—for instance, those who are
female, working class, Jewish, disabled, or black—encounter
unspoken (and often unconscious) prejudice in their dealings with
official or professional bodies. An Al welfare adviser, for example,
would not be prejudiced against such clients unless its data and
inferential rules were biased in the relevant ways. A program could,
of course, be written so as to embody its programmer's prejudices,
but the program can be printed out and examined, whereas social
attitudes cannot.
Artificial intelligence might even lead to a society in
which people have greater freedom and greater incentive to
concentrate on what is most fully human. Too few of us today
(especially men) have time to commit ourselves to developing our
interpersonal relations with family and friends. Increased leisure
time throughout society (on the assumption that appropriate political
and economic structures had been developed to allow for this) would
make room for such conviviality. Partly as a result of this and perhaps
partly as a reaction against the unemotional nature of most Al Margaret A. Boden (Courtesy of
Margaret A. Boden)
programs, the emotional dimension of personality might come to be
more highly valued (again, especially by men) than it is in the West
today. In my view, this would be all to the good. Similarly, the new
technology might make it possible for many more people (yet again,
especially men) to engage in activities, whether paid or unpaid, in the
service sector: education, health, recreation, and welfare. The need
for such activities is pressing, but the current distribution of income
makes these intrinsically satisfying jobs financially unattractive. One
of the most important benefits of all is that Al can rehumanize—yes,
rehumanize—our image of ourselves. How can this be? Most people
assume that Al either has nothing to teach us about the nature of
being human or that it depicts us as “nothing but machines”: poor
deluded folk, we believe ourselves to be purposive, responsible
creatures whereas in reality we are nothing of the kind.
The crucial point is that Al is concerned with
representations, and how they can be constructed, stored, accessed,
compared, and transformed. A computer program is itself a set of

asi
representations, a symbol system that models the world more or less If Al research had developed programs with a capacity for under-
adequately. This is why it is possible for an Al program to reflect the standing text, understanding speech, interpreting images, and
sexist or racist prejudices of its programmer. But representation is updating memory, the amount of information about individuals that
central to psychology as well, for the mind too is a system that was potentially available to government would be enormous. Good
represents the world and possible worlds in various ways. Our hopes, news for Big Brother, perhaps, but not for you and me.
fears, beliefs, memories, perceptions, intentions, and desires all Economic destruction might happen too if changes in the
involve our ideas about (our mental models of) the world and other patterns and/or rates of employment are not accompanied by radical
worlds. This is what humanist philosophers and psychologists have structural changes in industrial society and in the way people think
always said, of course, but until recently they had no support from about work. Economists differ about whether the convivial society
science. Because sciences like physics and chemistry have no place described above is even possible: some argue that no stable
for the concept of representation, their philosophical influence over economic system could exist in which only a small fraction of the
the past four centuries has been insidiously dehumanizing. The people do productive (nonservice) work. Certainly, if anything like
mechanization of our world picture—including our image of man— this is to be achieved, and achieved without horrendous social costs,
was inevitable, for what a science cannot describe it cannot new ways of defining and distributing society's goods will have to be
recognize. Not only can artificial intelligence recognize the mind (as found. At the same time, our notion of work will have to change: the
distinct from the body); it can also help to explain it. It “gives us back Protestant ethic is not appropriate for a high-technology postindus-
to ourselves,” by helping us to understand how it is possible for a trial society.
representational system to be embodied in a physical mechanism Last, what of moral destruction: could we become less
(brain or computer). human—indeed, less than human—as a result of advances in Al? This
So much for the rose-colored spectacles. What of the might happen if people were to come to believe that purpose, choice,
darker implications? Many people fear that in developing Al, we may hope, and responsibility are all sentimental illusions. Those who
be sowing the seeds of our own destruction, our own physical, believe that they have no choice, no autonomy, are unlikely to try to
political, economical, and moral destruction. Physical destruction exercise it. But this need not happen, for our goals and beliefs—in a
could conceivably result from the current plans to use Al within the word, our subjectivity—are not threatened by Al. As we have seen,
U.S. Strategic Defense Initiative (Star Wars). One highly respected the philosophical implications of Al are the reverse of what they are
computer scientist, David Parnas, publicly resigned from the U.S. commonly assumed to be: properly understood, Al is not dehumaniz-
government's top advisory committee on SDI computing on the ing.
grounds that computer technology (and Al in particular) cannot in A practical corollary of this apparently abstract point is
principle achieve the reliability required for a use where even one that we must not abandon our responsibility for evaluating—and, if
failure could be disastrous. Having worked on military applications necessary, rejecting—the “advice” or “conclusions” of computer
throughout his professional life, Parnas had no political ax to grind. programs. Precisely because a program is a symbolic representation
His resignation, like his testimony before the U.S. Senate in December of the world, rather than a part of the world objectively considered, it
1985, was based on purely technical judgment. is in principle open to question. A program functions in virtue of its
Political destruction could result from the exploitation of data, its inferential rules, and its values (decision criteria), each and
Al (and highly centralized telecommunications) by a totalitarian state. every one of which may be inadequate in various ways. (Think of the

a5s2
 example of the racist expert system.) We take it for granted that
human beings, including experts (perhaps especially experts), can
be mistaken or ill advised about any of these three aspects of
thinking. We must equally take it for granted that computer pro-
grams—which in any event are far less subtle and commonsensical
than their programmers and even their users—can be questioned too.
If we ever forget that “It's true because the computer says so” is
never adequate justification, the social impact of Al will be
horrendous indeed.
 Futurists have long seen computers as Big Brother's crucial ally on
the road to 1984, George Orwell's chilling vision of technocracy.
Placed on pedestals in‘the central computing rooms of large
institutions, computers were large, expensive, and arcane. They did
not understand English; to use them, you had to learn what were
called their higher-level languages. As one expert predicted, “There
will be a small, almost separate society of people in rapport with the
advanced machines.” This elite will tend to control the state and
master the commanding heights of the economy.
The year 1984 came and went, and the prophecies of 1984
were fulfilled only in nightmares and totalitarian gulags. One of the

George Gilder (Photo by Jonathan prime reasons for the failure of the prophecy was the success of
Becker) computer technology. Contrary to all the grim predictions, intelligent
machines empower individuals rather than large institutions, spread
power rather than centralize it.
Crucial to the liberating impact of computers was the very

BE cwsss citer | nature of computer technology. In volume, anything on a chip is

cheap. But moving up the hierarchy from the chip to the circuit board
to the network and to the telecommunications system, interconnec-
A Technology of Liberation
tions between components grow exponentially more expensive. So a
George Gilder is the author of first law of the technology is to concentrate components and
eight books on issues of
connections—and thus computing power—on single chips.
technology and society,
Concentrating components on a chip not only enhanced their speed
including Wealth and Poverty
(1981) and The Spirit of and effectiveness but also vastly lowered their price. Finally, in the
Enterprise (1983), both best form of the microprocessor, the computer on a chip costs a few
sellers. His most recent book is
dollars and outperforms the computer on a pedestal. Rather than
Microcosm (1989), a history and
pushing control toward Big Brother at the top, as the pundits
prophesy of the age of VLSI
microchips. Gilder is a regular predicted, the technology, by its very nature, constantly pulled power
contributor to the Wall Street down to the people. The ultimate beneficiary, the individual with a
Journal and lives in Tyringham,
personal computer or workstation, gained powers of creation and
Massachusetts with his
communication far beyond the kings of old.
wife, four children, and four
computers The individual was not only the heir to the throne of the
technology; he also was its leading creator. Although made possible
by hardware innovations from around the world, the move to small
computers was chiefly an American revolution driven by the
invention of new software. As fast as the Japanese and others could
expand the capacity of computer memories, American entrepreneurs
filled them with useful programs. Some 14 thousand new U.S.
companies, many of them led by teenagers and college hackers,
launched a vast array of software packages and changed the
computer from an arcane tool of elites to a popular appliance. As a
result of this software, ranging from spreadsheets and word
processors to data bases and video games, the United States
pioneered and propagated the use of small computers, and the U.S.
share of the world software market rose from under two-thirds to
more than three quarters.

asa
 Analysts focusing on fifth-generation computer projects, converts high-level chip designs and functions into the intersecting

mainframe systems, and supercomputers dismiss these personal polygons of an actual chip layout. This technology allows the

computers as toys. So did the experts at IBM a few years ago. But at complete design of integrated circuits on a computer, from initial

the same time that the United States moved massively into microcom- concept to final silicon. To understand the impact of this develop-
puters, small systems surged far ahead in price performance. In terms ment, imagine that printing firms essentially wrote all the books. This

of cost per MIPS (millions of instructions per second), the new was the situation in the computer industry; in order to author a chip,
personal computers are an amazing ninety times more cost effective you essentially had to own a semiconductor manufacturing plant (a
than mainframes. With the ever growing ability to interconnect these silicon printing press), which cost between $50 and $200 million to
machines in networks and use them in parallel configurations that build on a profitable scale. But with silicon compilers and related
yield mainframe performance, microcomputers are rapidly gaining gear, any computer-literate person with a $20,000 workstation can
market share against the large machines. author a major new integrated circuit precisely adapted to his needs.
Once believed to be a bastion of bureaucratic computing, If mass production is needed, semiconductor companies around the
IBM itself has become a prime source of the redistribution of globe will compete to print your design in the world’s best clean
computer power. As IBM's machines become smaller and cheaper rooms. In a prophetic move a few firms are now even introducing
and more available to the public, they also become more effective forms of silicon desktop publishing. For $3 million, for example,
and more flexible. The trend will continue. According to Ralph Lasarray sells manufacturing modules that do all essential production
Gomory, IBM's chief scientist, the constraints of interconnections steps from etching the design to assembling the chips. Dallas
mean that supercomputers of the future will have to be concentrated Semiconductor acquired an advanced new chip-making facility for
into a space of three cubic inches. The industrial revolution imposed $10 million. Contrary to the analyses of the critics, the industry is not
economies of scale, but the information revolution imposes econo- becoming more capital intensive. Measured in terms of capital costs
mies of microscale. Computer power continually devolves into the per device function (the investment needed to deliver value to the
hands and onto the laps of individuals. customer) the industry is becoming ever cheaper to enter. The silicon
The advance into the microcosm is now accelerating. compiler and related technologies moves power from large corpora-
Propelling it is a convergence of three major developments in the tions to individual designers and entrepreneurs.
industry, developments that once again disperse power rather than The third key breakthrough is the widespread abandon-
centralize it. The first is arti ial intelligence, giving to computers ment of the long cherished von Neumann computer architecture with
rudimentary powers of sight, hearing, and common sense. True, some its single central processing unit, separate memory, and step-in and
of these Al devices still do a pretty limited job. It has been said that fetch-it instruction sets. Replacing this architecture are massively
the computer industry thrived by doing well what human beings do parallel systems with potentially thousands of processors working at
badly. Artificial intelligence often seems to thrive by doing badly once. This change in the architecture of computers resembles the
what human beings do well. But you can understand the significance abandonment of centralized processing in most large companies. In
of Al advances by imagining that you are deaf, dumb, and blind. If the past users had to line up outside the central processing room,
someone gave you a device that allowed you to see and hear even submit their work to the data-processing experts, and then wait hours
poorly, you would hail him as a new Edison. Computer technology has or days for it to be done. Today tasks are dispersed to thousands of
long been essentially deaf, dumb, and blind. Reachable only through desk tops and performed simultaneously. The new architecture of
keyboards and primitive sensors and programmable only in binary parallel processing breaks the similar bottleneck of the central
mathematics, computers remained mostly immured in their digital processing unit at the heart of every computer. It allows the computer
towers. Artificial intelligence promises to vastly enhance the itself to operate like a modern corporate information system, with
accessibility of computers to human language, imagery, and various operations all occurring simultaneously throughout the firm,
expertise. So the computer can continue to leave ivory towers and rather than like an old corporate data processing hierarchy, which
data-processing elites behind and open itself to the needs of forced people to waste time in queues while waiting access to the
untrained and handicapped individuals, even allowing the blind to company mainframe. Promising huge increases in the cost effective-
read and the disabled to write. ness of computing, parallel processing will cheaply bring supercom-
The second major breakthrough is the silicon compiler. puter performance to individuals.
Just as a software compiler converts high-level languages into the Any one of these breakthroughs alone would not bring the
bits and bytes that a computer can understand, the silicon compiler radical advances that are now in prospect. But all together they will

ass
increase computer efficiency by a factor of thousands. Carver Mead cassette, a compact disk, or even a book. All of these devices cost a
of Caltech, a pioneer in all three of these new fields, predicts a few dollars to make; all sell for amounts determined by the value of
10,000-fold advance in the cost effectiveness of information technol- their contents: the concepts and images they bear. What is important
ogy over the next ten years. The use of silicon compilers to create is not the medium but the message.
massively parallel chips to perform feats of artificial intelligence will Microchip entrepreneur Jerry Sanders once declared that
transform the computer industry and the world economy. semiconductors would be “the oil of the eighties.” Some analysts
An exemplary product of these converging inventions is now fear that giant companies will conspire to cartelize chip
speech recognition. Discrete-speech talkwriter technology already production as OPEC once monopolized oil. They predict that by
commands available vocabularies of nearly one hundred thousand dominating advanced manufacturing technology and supplies, a few
words, active vocabularies in the tens of thousands, learning firms will gain the keys to the kingdom of artificial intelligence and
algorithms that adapt to specific accents, and operating speeds of other information technologies. Yet unlike oil which is a substance
over 40 words per minute. To achieve this speed and capacity extracted from sand, semiconductor technologies are written on
combined with the ability to recognize continuous speech on conven- sand, and their substance is ideas. To say that huge conglomerates
tional computer architectures would require some four hundred MIPS will take over the information industry because they have the most
(millions of instructions per second). Yet the new speech-recognition efficient chip factories or the purest silicon is like saying that the
gear will operate through personal computers and will cost only Canadians will dominate world literature because they have the
some $5000. That is just $15.00 per MIPS. IBM mainframes charge tallest trees.
some $150,000 per MIPS, and the most efficient general-purpose small Contrary to all the fears and prophecies, the new
computers charge some $3,000 per MIPS. By using parallel chips technologies allow entrepreneurs to economize on capital and
adapted specifically to process the enigmatic onrush of human enhance its efficiency, m ing sand and ideas to generate new
speech, these machines can increase the cost effectiveness of wealth and opportunity for men and women anywhere in the world.
computing by a factor of hundreds. The chief effect can be summed up in a simple maxim, a hoary cliche:
The talkwriter is only one of hundreds of such products. knowledge is power. Most people agree that this statement conveys
Coming technologies will increase the efficiency of computer an important truth. Today, however, knowledge is not simply a source
simulation by a factor of thousands, radically enhance the effective- of power; is supremely the source of power. The difference is
ness of machine vision, create dramatically improved modes of music crucial. If knowledge is power in this vital sense, it means that other
synthesis, provide new advances in surgical prosthesis, open a world things are not power. The other things that no longer confer power, or
of information to individuals anywhere in the world, all at prices radically less power than before, include all the goals and dreams of
unimaginable as recently as three years ago. As prices decline, the all the tyrants and despots of the centuries: power over natural
new information systems inevitably become personal technologies, resources, territory, military manpower, national taxes, trade
used and extended by individuals with personal computers. surpluses, and national currencies.
With an increasing stress on software and design rather In an age when men can inscribe new worlds on grains of
than on hardware and materials, the computer industry symbolizes sand, particular territories have lost their economic significance. Not
the age of information. Real power and added value in the modern era only are the natural resources under the ground rapidly declining in
lies not in things but in thoughts. The chip is a medium, much like a relative value, but the companies and capital above the ground can
floppy disk, a 35-millimeter film, a phonograph record, a video easily pick up and leave. Capital markets are now global; funds can

456
move around the world, rush down fiber optic cables, and bounce off
satellites at near the speed of light. People—scientists, workers, and
entrepreneurs—can leave at the speed of a 747, or even a Concorde.
Companies can move in weeks. Ambitious men no longer stand still to
be fleeced or exploited by bureaucrats.
The computer age is the epoch of the individual and
family. Governments cannot take power by taking control or raising
taxes, by mobilizing men or heaping up trade surpluses, by seizing
territory or stealing technology. In the modern world even slaves are
useless: they enslave their owners to systems of poverty and decline.
The new source of national wealth is the freedom of creative
individuals in control of information technology. This change is not
merely a gain for a few advanced industrial states. All the world will
benefit from the increasing impotence of imperialism, mercantilism,
and statism. All the world will benefit from the replacement of the
zero-sum game of territorial conflict with the rising spirals of gain
from the diffusion of ideas. Ideas are not used up as they are used;
they spread power as they are shared. Ideas begin as subjective
events and always arise in individual minds and ultimately repose in
them. The movement toward an information economy inevitably
means a movement toward a global economy of individuals and
families. Collective institutions will survive only to the extent that
they can serve the men and women who comprise them.
All the theories of the computer as an instrument of
oppression misunderstand these essential truths of the technology. In
the information age, nations cannot gain strength by coercing and
taxing their citizens. To increase their power, governments must
reduce their powers and emancipate their people on the frontiers of
the age of intelligent machines.

as7
A computer image (Photo by Lou
Jones)
 Postscript

Let us review a few of the fundamental issues underlying the age of intelligent
machines.

) + Will Machines Reach Human Levels of Intelligence?

As | noted in the last section of the previous chapter, the strengths of today’s
machine intelligence are quite different from those of human intelligence and in many
ways complement it. Once we have defined the transformations and methods
underlying intelligent processes, a computer can carry them out tirelessly and at great
speed. It can call upon a huge and extremely reliable memory and keep track of
billions of facts and their relationships. Human intelligence, on the other hand, though
weak at mastering facts, still excels at turning information into knowledge. The ability
to recognize, understand, and manipulate the subtle networks of abstraction inherent
in knowledge continues to set human intelligence apart.
Yet computers are clearly advancing in these skills. Within narrow do-
mains—diagnosing certain classes of disease, performing financial judgements, and
many other specialized tasks—computers already rival human experts. During the
1980s expert systems went from research experiments to commercially viable tools
that are relied upon daily to perform important jobs. Computers have also begun in
recent years to master the pattern-recognition tasks inherent in vision and hearing.
Though not yet up to human standards, pattern-recognition technology is sufficiently
advanced to perform a wide variety of practical tasks. It is difficult to estimate when
these capabilities will reach human levels, but there does not appear to be any
fundamental barrier to achieving such levels. Undoubtedly, computers will achieve
such levels gradually; no bell will ring when it happens.
What is clear is that by the time computers achieve human levels of per-
formance in those areas of traditional human strength, they will also have greatly
enhanced their areas of traditional superiority. (Not all experts agree with this. Doug
Hofstadter, for example, speculates in Gédel, Escher, Bach that a future “actually
intelligent” machine may not be able to do fast accurate arithmetic, because it will get
distracted and confused by the concepts triggered by the numbers—a dubious
 proposition in my view.') Once a computer can read and understand what it is
reading, there is no reason why it should not read everything ever written (encyclope-
dias, reference works, books, journals and magazines, data bases, etc.) and thus
master all knowledge. As Norbert Wiener has pointed out, no human being has had a
complete mastery of human knowledge for the past couple centuries (and it is
doubtful in my view that anyone has ever had such mastery). Even mere human
levels of intelligence combined with a thorough mastery of all knowledge would give
computers unique intellectual skills. Combine these attributes with computers’
traditional strengths of speed, tireless operation, prodigious and unfailing memory,
and extremely rapid communication, and the result will be formidable. We are, of
course, not yet on the threshold of this vision. This early phase of the age of intelli-
gent machines is providing us with obedient servants that are not yet intelligent
enough to question our demands of them.
Minsky points out that we have trouble imagining machines achieving the
capabilities we have because of a deficiency in our concept of a machine.? The human
race first encountered machines (of its own creation) as devices with a few dozen,
and in some cases a few hundred, active parts. Today, our computerized machines
have millions of active components, yet our concept of a machine as a relatively
inflexible device with only a handful of behavioral options has not changed. By the end
of this century chips with over a billion components are anticipated, and we will enter
an era of machines with many billions of components. Clearly, the subtleness and
intelligence of the behavior of machines at those different levels of complexity are
quite different. Emulating human levels of performance will require trillions, perhaps
thousands of trillions, of components. At current rates of progress, we shall achieve
such levels of complexity early in the next century. Human-level intelligence will not
automatically follow, but reasonable extrapolations of the rate of progress of machine
intelligence in a broad variety of skills in pattern recognition, fine motor coordination,
decision making, and knowledge acquisition leads to the conclusion that there is no
fundamental barrier to the Al field’s ultimately achieving this objective.

+ Can a Machine Think?

The question sounds innocuous enough, but our approach to it rests on the meanings
we ascribe to the terms “machine” and “think.” Consider first the question of
whether or not a human being is a machine. A human being is certainly not like the
early human-made machines, with only a handful of parts. Yet are we fundamentally
different from a machine, one with, say, a few trillion parts? After all, our bodies and
brains are presumably subject to the same natural laws as our machines. As | stated
earlier, this is not an easy question, and several thousand years of philosophical
debate have failed to answer it. If we assume that the answer to this question is no
(humans are not fundamentally different from machines), then we have answered the
original question. We presumably think, and if indeed we are machines, then we must
conclude that machines can think. If, on the other hand, we assume that we are in
some way fundamentally different from a machine, then our answer depends on our
definition of the word “think.”

460
 First, let us assume a behavioral definition, that is, a definition of thinking
based on outwardly observable behavior. Under this definition, a machine should be
considered to think if it appears to engage in intelligent behavior. This, incidentally,
appears to be the definition used by the children | interviewed (see the section “Naive
Experts” in chapter 2). Now the answer is simply a matter of the level of performance
we expect. If we accept levels of performance in specific areas that would be consid-
ered intelligent if performed by human beings, then we have achieved intelligent
behavior in our machines already, and thus we can conclude (as did the children |
talked with) that today’s computers are thinking. If, on the other hand, we expect an
overall level of cognitive ability comparable to the full range of human intelligence,
then today’s computers cannot be regarded as thinking. If one accepts my conclusion
above that computers will eventually achieve human levels of intellectual ability, then
we can conclude that it is inherently possible for a machine to think, but that ma-
chines on earth have not yet started to do so.
If one accepts instead an intuitive definition of thinking, that is, an entity is
considered to be thinking if it “seems” to be thinking, then responses will vary widely
with the person assessing the “seeming.” The children | spoke to felt that computers
seemed to think, but many adults would disagree. For myself, | would say that
computers do not yet seem to be thinking most of the time, although occasionally a
clever leap of insight by a computer program | am interacting with will make it seem,
just for a moment, that thinking is taking place.
Now let us consider the most difficult approach. If we define thinking to
involve conscious intentionality, then we may not be in a position to answer the
question at all. | know that | am conscious, so | know that | think (hence Descartes’s
famous dictum “| think, therefore | am”). | assume that other people think (lest | go
mad), but this assumption appears to be built in (what philosophers would call a priori
knowledge), rather than based on my observations of the behavior of other people. |
can imagine machines that can understand and respond to people and situations with
the same apparent intelligence as real people (see some of the scenarios above). The
behavior of such machines would be indistinguishable from that of people; they would
pass any behavioral test of intelligence, including the Turing test. Are these machines
conscious? Do they have genuine intentionality or free will? Or are they just following
their programs? Is there a distinction to be made between conscious free will and just
following a program? Is this a distinction with a difference? Here we arrive once again
at the crux of a philosophical issue that has been debated for several thousand years.
Some observers, such as Minsky and Dennett, maintain that consciousness is indeed
an observable and measurable facet of behavior, that we can imagine a test that could
in theory determine whether or not an entity is conscious. Personally, | prefer a more
subjective concept of consciousness, the idea that consciousness is a reality appreci-
ated only by its possessor. Or perhaps | should say that consciousness is the posses-
sor of the intelligence, rather than the other way around. If this is confusing, then you
are beginning to appreciate why philosophy has always been so difficult.
lf we assume a concept of thinking based on consciousness and hold that
consciousness is detectable in some way, then one has only to carry out the appropri-
ate experiment and the answer will be at hand. (If someone does this, let me know.)
If, on the other hand, one accepts a subjective view of consciousness, then only the

461
machine itself could know if it is conscious and thus thinking (assuming it can truly
know anything). We could, of course, ask the machine if it is conscious, but we would —
not be protected from the possibility of the machine having been programmed to lie.
(The philosopher Michael Serwen once proposed building an intelligent machine that
could not lie and then simply asking it if it was conscious.)
One remaining approach to this question comes to us from quantum
mechanics. In perhaps its most puzzling implication, quantum mechanics actually
ascribes a physical reality to consciousness. Quantum theory states that a particle
cannot have both a precise location and a precise velocity. If we measure its velocity
precisely, then its location becomes inherently imprecise. In other words, its location
becomes a probability cloud of possible locations. The reverse is also true: measuring
its precise location renders its velocity imprecise. It is important to understand exactly
what quantum mechanics is trying to say. It is not saying that there is an underlying
reality of an exact location and velocity and that we are simply unable to measure
them both precisely. It is literally saying that if a conscious being measures the
velocity of a particle, it actually renders the reality of the location of that particle
imprecise. Quantum mechanics is addressing not simply limitations in observation but
the impact of conscious observation on the underlying reality of what is observed.
Thus, conscious observation actually changes a property of a particle. Observation of
the same particle by a machine that was not conscious would not have the same
effect. If this seems strange to you, you are in good company. Einstein found it
absurd and rejected it.* Quantum mechanics is consistent with a philosophical
tradition that ascribes fundamental reality to knowledge, as opposed to knowledge
simply being a reflection of some other fundamental reality.> Quantum mechanics is
more than just a philosophical viewpoint, however: its predictions have been consis-
tently confirmed. Almost any electronic device of the past 20 years demonstrates its
principles, since the transistor is an embodiment of the paradoxical predictions of
quantum mechanics. Quantum mechanics is the only theory in physics to ascribe a
specific role to consciousness beyond simply saying that consciousness is what may
happen to matter that evolves to high levels of intelligence according to physical laws.
If one accepts its notions fully, then quantum mechanics may imply a way
to physically detect consciousness. | would counsel caution, however, to any would-
be builder of a consciousness detector based on these principles. It might be upset-
ting to point a quantum-mechanical consciousness detector at ourselves and discover
that we are not really conscious after all.
As a final note on quantum mechanics let me provide a good illustration of
the central role it ascribes to consciousness. According to quantum mechanics,
observing the velocity of a particle affects not only the preciseness of its location but
also affects the preciseness of the location of certain types of “sister” particles that
may have emerged from the same particle interaction that produced the particle
whose velocity we just observed. For example, if an interaction produces a pair of
particles that emerge in opposite directions and if we subsequently observe the
velocity of one of the particles, we will instantly affect the preciseness of the position
of both that particle and its sister, which may be millions of miles away. This would
appear to contradict a fundamental tenet of relativity: that effects cannot be transmit-
ted faster than the speed of light. This paradox is currently under study.®

462
+ What Impact Will the Age of Intelligent Machines Have on Society?

When computers were first invented in the mid 1940s, they were generally regarded
as curiosities, though possibly of value to mathematics and a few engineering disci-
plines. Their value to science, business, and other disciplines soon became apparent,
and exploration of their practical applications soon began. Today, almost a half-century
later, computers are ubiquitous and highly integrated into virtually all of society's
institutions. If a law were passed banning all computers (and in the doubtful event
that such legislation were adhered to), society would surely collapse. The orderly func-
tioning of both government and business would break down in chaos. We are already
highly dependent on these “amplifiers of human thought,” as Ed Feigenbaum calls
them. As the intelligence of our machines improves and broadens, computer intelli-
gence will become increasingly integrated into our decision making, our economy, our
work, our learning, our ability to communicate, and our life styles. They will be a
driving force in shaping our future world. But the driving force in the growth of
machine intelligence will continue to be human intelligence, at least for the next half
century.

+ AFinal Note

When | was a boy, | had a penchant for collecting magic tricks and was known to give
magic shows for friends and family. | took pleasure in the delight of my audience in
observing apparently impossible phenomena. It became apparent to me that organiz-
ing ordinary methods in just the right sequence could give rise to striking results that
went beyond the methods | started with. | also realized that revealing these methods
would cause the magic to disappear and leave only the ordinary methods.
As | grew older, | discovered a more powerful form of magic: the computer.
Again, by organizing ordinary methods in just the right sequences (that is, with the
right algorithms), | could once again cause delight. Only the delight caused by this
more grown-up magic was more profound. Computerized systems that help over-
come the handicaps of the disabled or provide greater expressiveness and productiv-
ity for all of us provide measures of delight more lasting than the magic tricks of
childhood. The sequences of 7s and Os that capture the designs and algorithms of our
computers embody our future knowledge and wealth. And unlike more ordinary
magic, any revelation of the methods underlying our computer magic does not tarnish
its enchantment.

463
 Chronology
The world has changed less since Jesus Christ than it has in the last thirty years.
Charles Peguy, 1913

Year Event

140-190 in years ago Dinosaurs roam the earth

Less than 100,000 years ago Homo sapiens begin using intelligence to further their goals.

More than 5,000 years ago The abacus, which resembles the arithmetic unit of a modern computer, is developed in the
Orient.

3000-700 e.c. Water clocks are built in China in 3000 s.c., in Egypt c. 1500 8.c., and in Assyria 700 8.c

2500 e.c. Egyptians invent the idea of thinking machines: citizens turn for advice to oracles, which are
statues with priests hidden inside.

b. 469 a.c. Socrates, the mentor of Plato, is the first Western thinker to assert that mental activities occur
in the unconscious.

469-322 a.c. Socrates, Plato, and Aristotle establish the essentially rationalistic philosophy of Western
culture.

427 8.c. In the Phaedo and later works Plato expresses ideas, several millenia before the advent of the
computer, that are relevant to modern dilemmas regarding human thought and its relation to the
mechanics of the machine.

c. 420 8.c. Archytas of Tarentum, a friend of Plato, constructs a wooden pigeon whose movements are
controlled by a jet of steam or compressed air.

b. 415 8.c. Theaetetus, a member of Plato's Academy, creates solid geometry.

387 B.c. Plato founds the Academy for the pursuit of science and philosophy in a grove on the outskirts
of Athens. It results in the fertile development of mathematical theory

343-334 8.c. Aristotle carries on the Platonic tradition by becoming the teacher of Alexander the Great
in 343 8.c. and founding the Lyceum, also known as the peripatetic schoo! of philosophers,
in 334 B.c
Year Event

293 B.c. Euclid, also a member of Plato's Academy, is the expositor of plane geometry. He writes the
Elements, a basic mathematics textbook for the next 2,000 years.

c. 200 8.c. In China artisans develop elaborate automata, including an entire mechanical orchestra.

c. 200 B.c. An Egyptian engineer improves the water clock, making it the most accurate timekeeping
device for nearly 2,000 years.

A.D. 529 Plato's Academy and Aristotle's Lyceum are closed by the emperor Justinian.

c. 600 The earliest works mentioning the game of chess appear in India.

725 A Chinese engineer and a Buddhist monk build the first true mechanical clock, a water-driven
device with an escapement that causes the clock to tick.

c. 1310 The first mechanical clocks appear in Europe, apparently stemming from stories about the
existence of mechanical clocks in China.

1494 Leonardo da Vinci draws a clock with a pendulum.

1530 The spinning wheel is in use in Europe.

1540, 1772 The technology of clock and watch making results in the production of more elaborate automata
during the European Renaissance. Gianello Toriano’s mandolin-playing lady (1540) and P.
Jacquet-Droz’s child (1772) are famous examples.

1543 Nicolaus Copernicus publishes De Revolutionibus, in which he states that the earth and the
other planets revolve around the sun, thereby changing humankind’s relationship with God.

17th—-18th centuries This is the age of the Enlightenment, a philosophical movement to restore the supremacy of
human reason, knowledge, and freedom, with parallel developments in science and theology. It
had its roots in the European Renaissance and the Greek philosophy of twenty centuries earlier
and constitutes the first systematic reconsideration of the nature of human thought and
knowledge since the Platonists.

1617 John Napier invents Napier's Bones, of significance to the future development of calculating
engines.

1637 René Descartes, who formulated the theory of optical refraction and developed the principles of
modern analytic geometry, pushes rational skepticism to its limits in his most comprehensive
work, Discours de la Méthode. His conclusion was “| think, therefore | am."

1642 Blaise Pascal perfects the Pascaline, a machine that can add and subtract. It is the world’s first
automatic calculating machine

c. 1650 Otto von Guericke perfects the air pump and uses it to produce vacuums.

1670 Pensées, by Blaise Pascal, is published posthumously.

1687 lsaac Newton's Philosophiae Naturalis Principia Mathematica, known as Principia, establishes
his three laws of motion and the law of universal gravitation.

1694 Gottfried Wilhelm Liebniz, an inventor of calculus, perfects the Liebniz Computer, a machine
that multiplies by performing repetitive additions, an algorithm still used in modern computers.

466
Year Event

1719 What appears to be the the first factory, an English silk-thread mill, employs 300 workers,
mostly women and children.

1726 Jonathan Swift describes a machine that will automatically write books in Gulliver's Travels

1733 Jofin.Kay paves the way for much faster weaving by patenting his New Engine for Opening and
Dressing Wool, later known as the flying shuttle

1760 Benjamin Franklin, in Philadelphia, erects lightning rods after having found, through his famous
kite experiment in 1752, that lightning is a form of electricity.

c. 1760 Life expectancy at the beginning of the Industrial Revolution is about 37 years in North America
and northwestern Europe.

1764 James Hargreaves invents the spinning jenny, which is able to spin eight threads at once.

1769 Richard Arkwright, the founder of the modern factory system, patents a hydraulic spinning
machine that is too large and expensive to use in family dwellings. He builds a factory for his
machine in 1781, thereby paving the way for many of the economic and social changes that will
characterize the Industrial Revolution

1781 Immanuel Kant publishes his Critique of Pure Reason, which expresses the philosophy of the
Enlightenment while deemphasizing the role of metaphysics. He sets the stage for the
emergence of twentieth-century rationalism.

1792 Edmund Cartwright devises the first machine to comb wool to feed the new mechanized
spinning machines

1792 William Murdock invents coal-gas lighting. The streets of London will be illuminated by 1802

1800 All aspects of the production of cloth are automated

1805 Joseph-Marie Jacquard devises a method for automating weaving with a series of punched
cards. This invention will be used many years later in the development of early computers.

1811 Ned Ludd founds the Luddite movement in Nottingham over the issue of jobs versus
automation

1821 Charles Babbage is awarded the first gold medal by the British Astronomical Society for his
paper "Observations on the Application of Machinery to the Computation of Mathematical
Tables.”

1821 Michael Farraday, widely recognized as the father of electricity, reports his discovery of
electromagnetic rotation and builds the first two motors powered by electricity

1822 Charles Babbage develops the Difference Engine, but its technical complexities exhaust his
financial resources and organizational skills. He eventually abandons it to concentrate his efforts
on a general-purpose computer.

1829 The first electromagnetically driven clock is constructed

1832 Charles Babbage develops the principle of the Analytical Engine, which is the world’s first
computer and can be programmed to solve a wide variety of logical and computational
problems.

467
Chronology
Year Event

1835 Joseph Henry invents the electrical relay, a means of transmitting electrical impulses over long
distances that serves as the basis of the telegraph.

1837 Samuel Finley Breese Morse patents his more practical version of the telegraph, which sends i
letters in codes consisting of dots and dashes.

1843 Ada Lovelace, Lord Byron's only legitimate child and the world’s first computer programmer,
publishes her own notes with her translation of L. P. Menabrea’s paper on Babbage’s Analytical
Machine

1843 Soren Kierkegaard, who will greatly influence the ideas of modern existentialists, publishes
Either-Or, his major work, followed by other writings that denounce the state-organized church
on grounds that religion is a matter for the individual soul.

1846 Alexander Bain uses punched paper tape to send telegraph messages, greatly improving the
speed of transmission

1847 George Boole publishes his first ideas on symbolic logic. He will develop these ideas into his
theory of binary logic and arithmetic that is still the basis of modern computation:

1851 An exhibition in London promotes the application of science to technology and

focuses world attention on British progress in both fields.

1854 An electric telegraph is installed between Paris and London

1855 Heinrich Geissler Igelshieb develops his mercury pump, used to produce the first good vacuum
tubes. These will lead to the development of cathode rays and eventually to the discovery of the
electron

1855 William Thomson develops a successful theory of transmission of electrical signals through
submarine cables

Charles Darwin, in The Origin of Species, explains his principle of natural selection and its
influence on the evolution of various species.

1861 San Francisco and New York are connected by a telegraph line

1864 Ducos de Harron develops a primitive motion-picture device in France.

1866 Cyrus West Field lays a telegraph cable across the Atlantic Ocean

1870 GNP on a per capita basis and in constant 1958 dollars is $530. Twelve million Americans, or 31
percent of the population, have jobs, and only 2 percent of adults have high school diplomas.

1871 Charles Babbage dies, leaving more than 400 square feet of drawings for his Analytic Engine.

1873 Melvil Dewey develops for the Amherst College Library a plan for 999 categories of materials
that becomes known as the Dewey Decimal System. It is refined over time to provide a virtually
unlimited number of subdivisions

1876 Alexander Graham Bell's telephone receives U.S. Patent 174,465, the most lucrative patent
ever granted

1879 G. Frege, one of the founders of modern symbolic language, proposes a notational system for
mechanical reasoning. This work is a forerunner to the predicate calculus, which will be used for
knowledge representation in artificial intelligence.
Year Event

1879 Thomas Alva Edison invents the first incandescent light bulb that can burn for a significant
length of time.

1880 Frederich Nietzsche writes Morgenréte and later works opposing romanticism and holding up
art, philosophy, and religion as illusions. These ideas will strongly influence modern
existentialism.

1882 Thomas Alva Edison’s design for New York City's Pearl Street station on lower Broadway brings
lighting to the United States.

1885 Boston is connected to New York by telephone.

1886 Alexander Graham Bell, with a modified version of Thomas Alva Edison's phonograph, uses wax
discs for recording sound

1887 The first gasoline-engine automobile is sold in Germany.

1888 William S. Burroughs patents an adding machine. This machine is modified, four years later to
include subtraction and printing. It is the world’s first dependable key-driven calculator and will
soon win widespread acceptance.

1888 Heinrich Hertz experiments with the transmission of what are now known as radio waves.

1888 The first commercial roll-film camera is introduced

1890 Herman Hollerith, incorporating ideas from Jacquard’s loom and Babbage’s Analytical Engine,
patents an electromechanical information machine that uses punched cards. It wins the 1890
U.S. Census competition, with the result that electricity is used for the first time in a major data-
processing project

1894 Guglielmo Marconi builds his first radio equipment, which rings a bell from 30 feet away

1894 Niagara Falls is harnessed for electricity

1896 A sound film is first shown before a paying audience in Berlin

1896 Herman Hollerith forms the Tabulating Machine Company, which will become IBM

1897 Joseph John Thomson, with better vacuum pumps than previously available, discovers the
electron, the first known particle smaller than an atom

1897 Alexander Popov, a Russian, uses an antenna to transmit radio waves, and Guglielmo Marconi,
an Italian, receives the first patent ever granted for radio. Marconi helps organize a company to
market his system

1899 The first recording of sound occurs magnetically on wire and on a thin metal strip.

1899 David Hilbert consolidates the accomplishments of nineteenth-century mathematics with such
Publications as The Foundations of Geometry.

1900 Herman Hollerith introduces an automatic card feed into his information machine to process th
1900 census data

469
 Chronology
Year Event

1900 The entire civilized world is connected by telegraph, and in the United States there are more
than 1.4 million telephones, 8,000 registered automobiles, and 24 million electric light bulbs.
Edison's promise of “electric bulbs so cheap that only the rich will be able to afford candles”
is thus realized. In addition, the Gramophone Company is advertising a choice of five thousand
recordings.

1900 More than one-third of all American workers are involved in the production of food.

1900 David Hilbert introduces the “direct method” in the calculus of variations and presents an
agenda for twentieth-century mathematics that includes a list of the 23 most pressing problems
at the International Mathematics Conference in Paris. He predicts that these problems will
occupy the attention of mathematicians for the next century.

1901 Marconi in Newfoundland receives the first transatlantic telegraphic radio transmission.

1901 Sigmund Freud publishes The /nterpretation of Dreams, which, along with his other works,
illuminates the workings of the mind.

I| 1904 John Ambrose Fleming files a patent for the first vacuum tube, a diode.

1906 Reginald Aubrey Fessenden invents AM radio and transmits by radio waves to wireless
operators on U.S. ships off the Atlantic Coast a Christmas carol, a violin trill, and for the first
time the sound of a human voice

1907 Lee De Forest and R. von Lieben invent the amplifier vacuum tube, known as a triode, which
greatly improves radio.

1908 Orville Wright makes his first hour-long airplane flight.

1910-1913 Bertrand Russell and Alfred North Whitehead publish their three-volume Principia Mathematica,
a seminal work on the foundations of mathematics that provides a new methodology for all
mathematics.

1911 Herman Hollerith’s Tabulating Machine Company acquires several other companies and
changes its name to Computing-Tabulating-Recording Company (CTR). In 1914 Thomas J.
Watson is appointed president.

1913 Henry Ford introduces the first true assembly-line method of automated production.

1913 A. Meissner invents a radio transmitter with vacuum tubes. Radio-transmitter triode modulation
is introduced the following year, and in 1915 the radio-tube oscillator is introduced

1915 The first North American transatlantic telephone call is made between Thomas A. Watson in
San Francisco and Alexander Graham Bell in New York.

1915 Albert Einstein completes his theory of gravitation known as the general theory of relativity.

1921 Czech dramatist Karel Capek popularizes the term “robot,” a word he coined in 1917 to
describe the mechanical people in his science fiction drama R.U.A. (Rossum’s Universal
Robots). His intelligent machines, intended as servants for their human creators, end up taking
over the world and destroying all mankind.

1921 Ludwig Wittgenstein, often referred to as the first logical positivist, publishes Tractatus Logico-
Philosophicus, regarded by some as perhaps the most influential philosophical work of the
twentieth century.
Year Event

1923 Vladimir Kosma Zworkin, the father of television, gives the first demonstration of an electronic
television-camera tube, using a mechanical transmitting device. He develops the iconoscope, an
early type of television system, the following year.

1924 Thomas J. Watson becomes the chief executive officer of CTR and renames the company
International Business Machines (IBM). IBM will become the leader of the modern industry and
one of the largest industrial corporations in the world.

1925 Niels Bohr and Werner Heisenberg lay the foundations for quantum mechanics

1925 Vannevar Bush and his coworkers develop the first analog computer, a machine designed to
solve differential equations.

1926 The era of talking motion pictures is introduced by The Jazz Singer, starring Al Jolson.

1927 Charles Lindbergh makes the first solo nonstop flight across the Atlantic Ocean

1927 Martin Heidegger publishes Sein und Zeit, vol. 1, which is rooted in the work of Seren
Kierkegaard and greatly influences the future development of existentialism.

1927 The Powers Accounting Machine Company goes through a series of mergers to become the
Remington Rand Corporation.

1927 Werner Heisenberg postulates his uncertainty principle, which says that electrons have no
precise location but rather probability clouds of possible locations. He wins a Nobel Prize five
years later for his discovery of quantum mechanics.

1928 John von Neumann presents the minimax theorem, which will be widely used in game-playing
programs.

1928 Philo T. Farnsworth demonstrates the world's first all-electronic television, and Vladimir Zworkin
receives a patent for a color television system

1929 FM radio is introduced.

1930 Paul Adrian Maurice Dirac publishes his Principles of Quantum Mechanics, in which he formu-
lates a general mathematical theory.

1930 More than 18 million radios are owned by 60 percent of U.S. households

1930 Vannevar Bush's analog computer, the Differential Analyzer, is built at MIT. It will be used to
calculate artillery trajectories during World War I!

1930s Music has shifted from the romantic style of Brahms and the early Mahler to the atonality of
Schoenberg, art to the cubism and expressionism of Picasso, and poetry to the minimalism of
Ezra Pound, T. S. Eliot, and William Carlos Williams.

1931 Kurt Godel publishes his incompleteness theorem, which has been called the most important in
all mathematics.

1932 RCA demonstrates a television receiver with a cathode-ray picture tube. In 1933 Zworkin
produces a cathode-ray tube. called the iconoscope, that makes high-quality television almost a
reality,

c. 1935 Albert Einstein’s quest for a unified field theory occupies most of the last two decades of his
life.

avi
Year Event

1936 Regular public television transmission begins in Greet Britain.

1937 Building on the work of Bertrand Russell and Charles Babbage, Alan Turing publishes “On
Computable Numbers,” his now celebrated paper introducing the Turing machine, a theoretical
model of a computer.

1937 The Church-Turing thesis, independently developed by Alonzo Church and Alan Turing, states
that all problems solvable by a human being are reducible to a set of algorithms, or more simply,
that machine intelligence and human intelligence are essentially equivalent.

1937 Frank Whittle builds the first working jet engine.

1939 The first regularly scheduled commercial flights begin crossing the Atlantic Ocean.

1940 John V. Atanasoff and Clifford Berry build an electronic computer known as ABC. This is the
first electronic computer, but it is not programmable.

1940 The 10,000-person British computer war effort known as Ultra creates Robinson, the world’s
first operational computer. It is based on electromechanical relays and is powerful enough to
decode messages from Enigma, the Nazi's first-generation enciphering machine.

1941 Konrad Zuse, a German, completes the world’s first fully programmable digital computer, the
Z-3, and hires Arnold Fast, a blind mathematician, to program it. Fast becomes the world’s first
programmer of an operational programmable computer.

1943 Warren McCulloch and Walter Pitts write their influential “Logical Calculus of the Ideas
Immanent in Nervous Activity,” which discusses neural-network architectures for intelligence.

1943 The Ultra team builds Colossus, a computer that uses electronic tubes 100 to 1000 times faster
than the relays used by Robinson. It cracks increasingly complex German codes and contributes
to the Allies’ winning of World War II

1943 Jean-Paul Sartre, a modern existentialist, publishes L’Etre et le Néant and later works that
incorporate the ideas of Soren Kierkegaard and Martin Heidegger while emphasizing the role of
free will in an apparently purposeless world. The spiritual and emotive world, which is meaning
less to logical positivists is to existentialists the seat of true meaning.

1944 Howard Aiken completes the first American programmable computer, the Mark I. It uses
punched paper tape for programming and vacuum tubes to calculate problems.

1945 Konrad Zuse develops Plankalkul, the first high-level language.

1946 John von Neumann publishes the first modern paper on the stored-program concept and starts
computer research at the Institute for Advanced Study in Princeton.

1946 John Presper Eckert and John W. Mauchley develop ENIAC, the world’s first fully electronic,
general-purpose (programmable) digital computer. It is almost 1,000 times faster than the
Mark | and is used for calculating ballistic-firing tables for the Army.

1946 Television enters American life even more rapidly than radio did in the 1920s. The percentage of
American homes having sets jumps from 0.02 percent in 1946 to 72 percent in 1956 and more
than 90 percent by 1983.

1947 William Bradford Schockley, Walter Hauser Brittain, and John Ardeen invent the transistor, a
minute device that functions like a vacuum tube but switches current on and off at much faster
speeds. It launches a revolution in microelectronics, bringing down the cost of computers and
leading to the development of minicomputers and powerful new main frame computers.

472
Year Event

1947 An airplane flies at supersonic speed for the first time, in the United States

1948 Norbert Wiener publishes Cybernetics, a seminal book on information theory.

1949 Maurice Wilkes, influenced by Eckert and Mauchley, builds EDSAC, the world’s first stored-
program computer. Eckert and Mauchley’s new U.S. company brings out BINAC, the first
American stored-program computer, soon after.

1949 George Orwell's novel 1984 envisions a chilling world in which very large bureaucracies employ
computers to enslave the population

1950 The U.S. census is first handled by a programmable computer, UNIVAC, developed by Eckert
and Mauchley. It is the first commercially marketed computer.

1950 Alan Turing’s “Computing Machinery and Intelligence” describes a means for determining
whether a machine is intelligent known as the Turing test.

1950 Commercial color television begins in the U.S. Transcontinental black-and-white television is
inaugurated the following year.

1950 Claude Elwood Shannon writes a proposal for a chess program

1951 EDVAC, Eckert and Mauchley’s first computer that implements the stored-program concept, is
completed at the Moore School at the University of Pennsylvania

1951 A Cybernetics Congress is held in Paris.

1952 The CBS television network uses UNIVAC to correctly predict the election of Dwight D.
Eisenhower as president of the United States.

1952 The pocket-sized transistor radio is introduced

1952 The 701, IBM’s first production-line electronic digital computer, is designed by Nathaniel
Rochester and marketed for scientific use.

1953 James D. Watson and Francis H. C. Crick discover the chemical structure of the DNA molecule.

1953 Two statements of major importance to modern existentialism appear: Philosophical

Investigations by Ludwig Wittgenstein and Waiting for Godot, a play by Samuel Beckett.

1955 The Remington Rand Corporation merges with Sperry Gyroscope to become the Sperry-Rand
Corporation, one of IBM's chief competitors for a time.

1955 IBM introduces its first transistor calculator, with 2,200 transistors instead of the 1,200 vacuum
tubes that would otherwise be required.

1955 The first design is created for a robotlike machine for industrial use in the U.S

1955 Allen Newell, J. C. Shaw, and Herbert Simon develop IPL-II, the first Al language

1955 The beginning space program and the military in the U.S., recognizing the need for computers
powerful enough to steer rockets to the moon and missiles through the stratosphere, fund
major research projects.

1956 Allen Newell, J. C. Shaw, and Herbert Simon create The Logic Theorist, which uses recursive
search techniques to solve mathematical problems.

a73
1956 The first transatlantic telephone cable begins to operate.

1956 Fortran, the first scientific computer programming language, is invented by John Backus and a
team at IBM.

1956 MANIAC |, the first computer program to beat a human being in a chess game, is developed by
Stanislaw Ulam.

1956 Artificial intelligence is named at a computer conference at Dartmouth College.

1957 Allen Newell, J. C. Shaw, and Herbert Simon develop the General Problem Solver, which uses
means-ends analysis to solve problems.

1957 Noam Chomsky writes Syntactic Structures, the first of many important works that will earn
him the title of father of modern linguistics. This work seriously considers the computation
required for natural-language understanding.

1958 Jack St. Clair Kilby invents the first integrated circuit.

1958 John McCarthy and Marvin Minsky found the Artificial Intelligence Laboratory at the Massachu-
setts Institute of Technology.

1958 The first U.S. commercial jet flies from New York to Paris

1958 Allen Newell and Herbert Simon predict that within ten years a digital computer will be the
world’s chess champion.

1958 John McCarthy introduces LISP, an early (and still widely used) Al language.

1958 The Defense Advanced Research Projects Agency is established. It will fund much important
computer-science research in the decades to come.

1958-1959 Jack Kilby and Robert Noyce independently develop the chip, which leads to much cheaper and
smaller computers.

Arthur Samuel's checker-playing program, completed as a study in machine-learning, performs

as well as some of the best players of the time.

1959 Dartmouth’s Thomas Kurtz and John Kemeny find an alternative to batch
processing: time sharing.

1959 The advent of electronic document preparation will increase U.S. paper consumption of printed
documents: the nation now consumes 7 million tons of paper per year; that number will
increase to 22 million in 1986. American businesses will use 850 billion pages in 1981, 2.5
trillion pages in 1986, and 4 trillion in 1990.

1959 Grace Murray Hopper, one of the first programmers of the Mark |, develops COBOL, a
computer language designed for business use.

1960 The Defense Department's Advanced Research Projects Agency substantially increases its
funding of computer research.

1960 About 6,000 computers are in operation in the United States.

1960 Yehoshua Bar-Hillel’s “Demonstration of the Nonfeasibility of Fully Automatic High-Quality

Translation” points out the difficulty of machine translation from one natural language to
another: a program needs to actually understand the world a particular passage refers to.

ava
Year Event

1960s Current neural-net machines incorporate a small number of neurons organized in only one or
two layers. Such simple models are mathematically proved to be limited in what they can do

1961 President John F. Kennedy, addressing a joint session of Congress, says, “| believe we should
go to the moon,” thereby launching Project Apollo, which will provide the impetus for important
research in computer science.

1961 Yuri Gagarin becomes the first human being to orbit the earth.

1962 AU.S. company markets the world’s first industrial robots.

1962 The first Department of Computer Science offering a Ph.D. is established at Purdue University.

1962 Time sharing is introduced on a computer in Philadelphia for inventory control

1962 John Glenn, Jr., in his Mercury 6 space capsule, becomes the first American to orbit the earth
The U.S. space probe Mariner is the first object made by human beings to voyage to another
planet. And America’s Telstar becomes the first active communications satellite, relaying
television pictures around the globe.

1962 D. Murphy and Richard Greenblatt develop the TECO text editor, one of the first word-
processing systems, for use on the PDP1 computer at MIT.

1962 Frank Rosenblatt publishes Principles of Neurodynamics, in which he defines the perceptron, a
simple processing element for neural networks. He first introduced the perceptron at a
conference in 1959.

1962 Thomas Kuhn publishes The Structure of Scientific Revolutions, in which he theorizes about the
nature of the growth of scientific knowledge.

1963 M. Ross Quillian’s work leads to the semantic network as a means of representing knowledge
in terms of concepts and relationships among concepts.

1963 Project MAC is established at MIT for computer-science research

1963 Al researchers of the 1960s, noting the similarity between human and computer languages,
adopt the goal of parsing natural-language sentences. Susumo Kuno’s parsing system reveals
the great extent of syntactic and semantic ambiguity in the English language. It is tested on the
sentence “Time flies like an arrow.”

1963 John McCarthy founds the Artificial Intelligence Laboratory at Stanford University.

1963 Marvin Minsky publishes his influential Steps Towards Artificial Intelligence.

1964 IBM solidifies its leadership of the computer industry with the introduction of its
360 series

1964 Daniel Bobrow completes his doctoral work on Student, a natural-language program that can
solve high-school-level word problems in algebra

1964 Gordon Moore, one of the founders of Fairchild Semiconductor Corporation, predicts that
integrated circuits will double in complexity each year. His statement will become known as
Moore's law and will prove true for decades to come.

1964 Marshall McLuhan's Understanding Media foresees electronic media, especially television, as
creating a “global village” in which “the medium is the message.”

a75
Year Event

1965 Raj Reddy founds the Robotics Institute at CarnegiesMellon University. The institute becomes a
leading research center for Al.

1965 The DENDRAL project begins at Stanford University, headed by Bruce Buchanan, Edward
Feigenbaum, and Nobel laureate Joshua Lederberg. Its purpose is to experiment on knowledge
as the primary means of producing problem-solving behavior. The first expert system,
DENDRAL, embodies extensive knowledge of molecular-structure analysis. Follow-up work,
carried out through the early 1970s, produce Meta-DENDRAL, a learning program that auto-
matically devises new rules for DENDRAL.

1965 Hubert Dreyfus presents a set of philosophical arguments against the possibility of artificial
intelligence in a RAND Corporation memo entitled “Alchemy and Artificial Intelligence.”

1965 Led by Edward Feigenbaum and his associates the Heuristic Programming Project, which will
later become the Knowledge Systems Laboratory, begins at Stanford University.

1965 Herbert Simon predicts that by 1985 “machines will be capable of doing any work a man can
do.”

Mid 1960s Computers are beginning to be widely used in the criminal justice system.

Mid 1960s Scientific and professional knowledge is beginning to be codified in a machine-readable form.

1966 Richard Greenblatt develops a fairly sophisticated chess-playing program, a version of which
defeats Hubert Dreyfus, an Al critic who strongly doubts the ability of computers to play chess.

1967 Seymour Papert and his associates at MIT begin working on LOGO, an education-
oriented programming language that will be widely used by children.

1967 The software business is born when IBM announces it will no longer sell software
and hardware in a single unit.

1968 David Hubel and Torstein Wiesel publish the first of many important papers on the macaque
monkey cortex. They discover edge-detection cells in the outer layers of the visual cortex.

1968 Marvin Minsky publishes Semantic Information Processing, an important collection of papers
describing Al work by himself and his students.

1968 Noam Chomsky and Morris Halle publish The Sound Pattern of English, a landmark study of
English phonetics.

1968 The film 2001: A Space Odyssey, by Arthur C. Clarke and Stanley Kubrick, presents HAL, a
computer that can see, speak, hear, and think like its human colleagues aboard a spaceship.

1969 Neil Armstrong becomes the first human to stand on the moon.

1969 Marvin Minsky and Seymour Papert write Perceptrons, a book that presents limitations of
single-layer neural nets.

1970 The GNP on a per capita basis and in constant 1958 dollars is $3,500, or more than six times as
much as a century ago.

1970 The floppy disc is introduced for storing data in computers.

1970 Harry Pople and Jack Myers of the University of Pittsburgh begin work on Internist, a system
that aids physicians in the diagnosis of a wide range of human diseases.

476
Year Event

1970 Patrick Winston's doctoral work presents a program that learns to recognize an arch, and it also
addresses the problem of machine learning.

1970 Terry Winograd completes his landmark thesis on SHRDLU, a natural- language system that
exhibits diverse intelligent behavior in the small world of children’s blocks. SHRDLU is criticized,
however, for its lack of generality.

1971 Kenneth Colby, Sylvia Weber, and F. D. Hilf present a report on PARRY, a program simulating a
paranoid person, in a paper entitled “Artificial Paranoia.” The program is so convincing that
clinical psychiatrists cannot distinguish its behavior from that of a human paranoid person.

1971 The first microprocessor is introduced in the U.S.

1971 The first pocket calculator is introduced. It can add, subtract, multiply, and divide.

1971 Direct telephone dialing on a regular basis begins between parts of the U.S. and Europe

1972 Hubert Dreyfus publishes What Computers Can't Do, an elaboration of his 1965 criticism of Al.
He argues that symbol manipulation cannot be the basis of human intelligence.

1973 Alain Colmerauer presents an outline of PROLOG, a logic-programming language. The language
will become enormously popular and will be adopted for use in the Japanese Fifth Generation
Program

1973 Roger Shank and Robert Abelson develop scripts, knowledge-representation systems used to
describe familiar everyday situations.

1974 The first computer-controlled industrial robot is developed

1974 Edward Shortliffe completes his doctoral dissertation on MYCIN, an expert system designed to
help medical practitioners prescribe an appropriate antibiotic by determining the precise identity
of a blood infection. Work to augment this program with other important systems, notably
TEIRESIAS and EMYCIN will continue through the early 1980s. TEIRESIAS will be developed in
1976 by Randall Davis to serve as a powerful information-structuring tool for knowledge
engineers. EMYCIN, by William van Melle, will represent the skeletal structure of inferences

1974 Marvin Minsky issues “A Framework for Representing Knowledge” as an MIT Al memo, a
landmark in knowledge representation.

1974 The SUMEX-AIM computer- communications network is established to promote the develop-
ment of applications of artificial intelligence to medicine.

1975 Benoit Mandelbrot writes “Les objet fractals: Forme, hasard, et dimension," his first long essay
on fractal geometry, a branch of mathematics that he developed. Fractal forms will be widely
used to model chaotic phenomena in nature and to generate realistic computer images of
naturally occurring objects.

1975 Medicine is becoming an important area of applications for Al research. Four major medical
expert systems have been developed by now: PIP, CASNET, MYCIN, and Internist.

1975 The Defense Advanced Research Programs Agency launches its Image Understanding Program
to stimulate research in the area of machine vision.

1975 More than 5,000 microcomputers are sold in the U.S., and the first personal computer, with 256
bytes of memory, is introduced

a7?
Chronology
Year Event

1970s The role of knowledge in intelligent behavior is now g major focus of Al research. Bruce
Buchanan and Edward Feigenbaum of Stanford University pioneer knowledge engineering.

1976 Daniel Bell publishes The Post-Industrial Society, which introduces the concept of a society in
which the “axial principle” is the centrality and codification of knowledge.

1976 As a representation of a visual image, David Marr proposes a primal sketch, containing
information that describes brightness changes, blobs, and textures.

1976 Kurzweil Computer Products introduces the Kurzweil Reading Machine, which reads aloud any
printed text that is presented to it. Based on omnifont-character-recognition technology, it is
intended to be a sensory aid for the blind.

1976 Douglas Lenat presents a program called AM (for Automated Mathematician) as part of his
Stanford doctoral dissertation. AM, a precursor to EURISKO, is a knowledge-based system that
makes “discoveries” in number theory and abstract mathematics.

1976 Joseph Weizenbaum, who created the famous ELIZA program, which simulates a Rogerian
psychotherapist, publishes Computer Power and Human Reason. He argues that even if we
could build intelligent machines, it would be unethical to do so.

1976-1977 Lynn Conway and Carver Mead collaborate and put together a collection of principles for VLSI
design. Their classic textbook Introduction to VLS/ Design is published in 1980. VLSI circuits will
form the basis of the fourth generation of computers.

1977 David Marr and Tomaso Poggio point out the salient difference between the human brain and
today’s computer in a paper on computer vision, “From Understanding Computation to.
Understanding Neural Circuitry.” While the connection to components ratio is only 3 in
computers, it is 10,000 in the cortex of a mammal.

1977 Steven Jobs and Stephen Wozniak design and build the Apple Computer.

1977 The first computer camp for children is held in Connecticut.

1977 The film Star Wars features 3CPO and a galaxy of other imaginative true-to-life robots with a
wide spectrum of convincing human emotions.

1977 Voyagers 1 and 2 are launched and radio back billions of bytes of computerized data about new
discoveries as they explore the outer planets of our solar system.

1977 The Apple II, the first personal computer to be sold in assembled form, is successfully
marketed.

1978 David Marr and H. K. Nishihara propose a new representation of visual information. The
2\/2-dimensional sketch, presents the depth and orientation of all visible surfaces.

1978 Digital Equipment Corporation (DEC) and Carnegie-Mellon University begin work on XCON, an
expert system that configures computer systems. By 1980 XCON will come into regular use,
saving millions of dollars at DEC plants.

1979 In a landmark study published in the Journal of the American Medical Association by nine
researchers, the performance of MYCIN is compared with that of doctors on ten test cases of
meningitis. MYCIN does at least as well as the medical experts. The potential of expert systems
in medicine becomes widely recognized.

ams
Year Event

1979 Ada, a computer language developed for use by the armed forces, is named for Ada Lovelace.

1979 Pac Man and other early computerized video games appear.

1980 Al industry revenue is a few million dollars per year.

1980 Douglas R. Hofstadter wins a Pulitzer Prize for his best-selling Godel, Escher, Bach.

1980 David Marr and Ellen Hildreth publish an important study on edge detection

1980 The Propaedia section of the fifteenth edition of the Encyclopaedia Britannica represents an
ambitious attempt to codify an outline of all human knowledge in just 800 pages.

Early 1980s Second-generation robots arrive with the ability to precisely effect movements with five or six
degrees of freedom. They are used for industrial welding and spray painting.

Early 1980s The MYCIN project produces NeoMYCIN and ONCOCIN, expert systems that incorporate
hierarchical knowledge bases. They are more flexible than MYCIN

Early 1980s Expert systems typically have knowledge bases of about a thousand rules.

1980s The neural-network paradigm begins to make a comeback, as neuron models are now
potentially more sophisticated. Multilayered networks are commonly used.

1981 MITI, Japan’s ministry for trade and industry, announces plans to develop by 1990 intelligent
computers that will be at least a thousand times as powerful as the present ones. MITI has a
track record of leading Japanese industry to world dominance in a wide range of fields.

1981 Desktop-publishing takes root when Xerox brings out its Star Computer. However, it will not
become popular until Apple's Laserwriter comes on the market in 1985. Desktop publishing
provides writers and artists an inexpensive and efficient way to compose and print large
documents.

1981 IBM introduces its Personal Computer (PC).

1982 Compact-disc players are marketed for the first time.

1982 A million-dollar advertising campaign introduces Mitch Kapor's Lotus 1-2-3, an enormously
popular spreadsheet program

1982 With over 100,000 associations between symptoms and diseases covering 70 percent of all the
knowledge in the field, CADUCEUS, an improvement on the Internist expert system, is
developed for internal medicine by Harry Pople and Jack Myers at the University of Pittsburgh
Tested against cases from the New England Journal of Medicine, it proves more accurate than
humans in a wide range of categories

1982 Defense robots play a vital role in the Israeli destruction of 29 Russian SAM (surface-to-air
missile) sites in a single hour during the invasion of Lebanon

1982 Japan‘s ICOT, a corporate consortium formed to meet some of the goals of the Fifth Generation
Project, begins active development with funding of $1 billion (half from MITI, half from
Japanese industry) over ten years. A response is initiated by the Americans.

1982 SRI International’s Prospector, a mineralogical expert system initiated in 1976 and updated
annually, pinpoints the location of a major deposit of molybdenum.

a79
Chronology
Year Event

1983 Edward Feigenbaum and Pamela McCorduck publish their influential book The Fifth Generation,
on Japan's computer challenge to the world.

1983 The Defense Advanced Research Projects Agency (DARPA) unveils the Strategic Computing
Initiative, a major program for research in microelectronics, computer architectures, and Al.

1983 Six million personal computers are sold in the U.S.

1983 Isaac Asimov describes in his science fiction novel Robots of Dawn a society two centuries
from now in which a beautiful female scientist and her “humaniform” robot lover live in the
company of a generation of robotic companions, servants, teachers, and guards.

1984 The European Economic Community forms ESPRIT, a five-year program to develop intelligent
computers. It is launched with $1.5 billion in funding.

1984 RACTER, created by William Chamberlain, is the first computer program to author a book.

1984 Ronald Reagan signs legislation to permit the formation of the Microelectronics and Computer
Corp. (MCC), a consortium of 21 companies whose purpose is to develop intelligent computers.
MCC has an annual research budget of $65 million.

1984 Waseda University in Tokyo completes Wabot-2, a 200 pound robot that reads sheet music
through its camera eye and plays the organ with its ten fingers and two feet

1984 Optical disks for the storage of computer data are introduced, and IBM brings out a megabit
RAM memory chip with four times the memory of earlier chips.

1985 Marvin Minsky publishes The Society of Mind, in which he presents a theory of the mind in
which intelligence is seen to be the result of proper organizations of a very large number of
simple mechanisms, each of which is by itself unintelligent.

1985 Jerome Wiesner and Nicholas Negroponte found MIT’s Media Laboratory to do research on
applications of aspects of computer science, sociology, and artificial intelligence to media
technology.

1985 During this year designs for 6,000 application-specific integrated circuits (ASICs) are
produced. These custom-built chips are being recognized as time and money savers.

1985 Jobs have grown tenfold since 1870: from 12 million to 116 million. The percentage of the U.S.
populace gainfully employed has grown from 31 to 48. Per capita GNP in constant dollars has
increased by 600 percent. These trends are expected to continue.

1985 The MIT Media Laboratory creates the first three-dimensional holographic image to be
generated entirely by computer.

c. 1985 Japan leads the world in robotics development, production, and application

Mid 1980s Al research begins to focus seriously on parallel architectures and methodologies for problem
solving

Mid 1980s Third-generation robots arrive with limited intelligence and some vision and tactile sensing.

1986 Al industry revenue is now $1 billion

1986 Albert Lawrence, Alan Schick, and Robert Birge of Carnegie-Mellon University conduct research
focused on the effort to develop a theory of molecular computing

480
Year Event

1986 Dallas Police use a robot to break into an apartment. The fugitive runs out in fright and
surrenders.

1986 Electronic keyboards account for 55.2 percent of the American musical keyboard
market, up from 9.5 percent in 1980. This trend is expected to continue until the
market is almost all electronic.

1986 James McClelland and David Rumelhart edit a set of papers on neural-network models for
intelligence, a collection that will soon become the manifesto of the new connectionists.

1986 Technology for optical character recognition represents a $100 million dollar industry that is
expected to grow to several hundred million by 1990.

1986 New medical imaging systems are creating a mini revolution. Doctors can now make accurate
judgments based on views of areas inside our bodies and brains.

1986 Using image processing and pattern recognition, Lillian Schwartz comes up with an answer to a
500-year-old question: Who was the Mona Lisa? Her conclusion: Leonardo da Vinci himself.

1986 Life expectancy is about 74 years in the U.S. Only 3 percent of the American work force is
involved in the production of food. Fully 76 percent of American adults have high school
diplomas, and 7.3 million U.S. students are enrolled in college.

1986 Russell Andersons doctoral work at the University of Pennsylvania is a robotic ping-pong player
that wins against human beings.

1986 The best computer chess players are now competing successfully at the senior master level,
with HiTech, the leading chess machine, analyzing 200,000 board positions per second

1987 Computerized trading helps push NYSE stocks to their greatest single-day loss

1987 The market for natural-language products (excluding automated speech recognition) is esti-
mated at $80 million and is expected to grow to $300 million by 1990.

1987 Commercial revenue from Al-related technologies in the U.S., excluding robotics, is now $1.4
billion. It is expected to reach $4 billion by 1990.

1987 Current speech systems can provide any one of the following: a large vocabulary, continuous
speech recognition, or speaker independence.

1987 Japan develops the Automated Fingerprint Identification System (AFIS), which enables U.S. law
enforcement agencies to rapidly track and identify suspects

1987 Robotic-vision systems are now a $300 million industry and will grow to $800 million by 1990.

1987 There are now 1,900 working expert systems, 1,200 more than last year. The most popular area
of application is finance, followed by manufacturing control and fault diagnosis

1987 XCON, DEC's expert system for configuring computers, has grown since its introduction in
1980. It now has a knowledge base that incorporates over 10,000 rules and does the work of
300 people more accurately than humans.

1988 Computer memory today costs only 10 of what it did in 1950.

1988 The expert-systems market is now valued at $400 million, up from $4 million in 1981. The
market is projected to grow to $800 million by 1990.
Year Event

1988 Marvin Minsky and Seymour Papert offer their view of recent developments in neural-network
machinery for intelligence in a revised edition of Perceptrons.

1988 The population of industrial robots has increased from a few hundred in 1970 to several hundred
thousand, most of them in Japan.

1988 In the U.S. 4,700,000 microcomputers, 120,000 minicomputers, and 11,500 mainframes are
sold in this year.

1988 W. Daniel Hillis’s Connection Machine is capable of 65,536 computations at the same time.

1988 Warsaw Pact forces are at least a decade behind NATO forces in artificial intelligence and other’
computer technologies.

1989 Computational power per unit of cost has roughly doubled every 18 to 24 months for the past
40 years

1989 The trend from analog to digital will continue to revolutionize a growing number of industries.

1989 Japan, a country very poor in natural resources but rich in expertise, has become the wealthiest
nation on the planet, as measured by the total value of all assets.

Late 1980s The core avionics of a typical fighter aircraft uses 200,000 lines of software. The figure is
expected to grow to about 1 million in the 1990s. The U.S. Military as a whole uses about 100
million lines of software (and is expected to use 200 million by 1993). Software quality becomes
an urgent issue that planners are beginning to address.

Late 1980s The computer is being recognized as a powerful tool for artistic expression.

Early 1990s A profound change in military strategy arrives. The more developed nations increasingly rely on
“smart weapons,” which incorporate electronic copilots, pattern recognitiontechniques, and
advanced technologies for tracking, identification, and destruction.

Early 1990s Continuous speech systems can handle large vocabularies for specific tasks.

Early 1990s Computer processors operate at speeds of 100 MIPS

Early 1990s Application Specific Integrated Circuit (ASIC) technology makes writing chip programs as easy
as writing software programs.

1990s A multi-hundred-billion-dollar computer and information-processing industry is emerging,

together with a generation of ubiquitous machine intelligence that works intimately with its
human creators.

1990s Significant progress is made toward an intelligent assistant, a decision-support system capable
of a wide variety of administrative and information-gathering tasks. The system can, for
example, prepare a feasibility report on a project proposal after accessing several data bases
and talking to human experts.

1990s Reliable person identification, using pattern- recognition techniques applied to visual and speech
patterns, replace locks and keys in many instances.

1990s Accomplished musicians, as well as students learning music, are routinely accompanied by
cybernetic musicians.

482
Year Event

1990s Al technology is of greater strategic importance than manpower, geography, and natural
resources

Late 1990s Documents frequently never exist on paper because they incorporate information in the form of
audio and video pieces.

Late 1990s Media technology is capable of producing computer-generated personalities, intelligent image
systems with some human characteristics.

1999 The several-hundred-billion-dollar computer and information-processing market is largely

intelligent by 1990 standards.

2000 Three-dimensional chips and smaller component geometries contribute to a multithousandfold

improvement in computer power (compared to a decade earlier)

2000 Chips with over a billion components appear.

2000 The world chess champion is a computer.

Early 2000s Translating telephones allow two people across the globe to speak to each other even if they do
not speak the same language.

Early 2000s Speech-to-text machines translate speech into a visual display for the deaf.

Early 2000s Exoskeletal robotic prosthetic aids enable paraplegic persons to walk and climb stairs

Early 2000s Telephones are answered by an intelligent telephone answering machine that converses with
the calling party to determine the nature and priority of the call

Early 2000s The cybernetic chauffeur, installed in one’s car, communicates with other cars and sensors on
roads. In this way it successfully drives and navigates from one point to another.

Early 21st century Computers dominate the educational environment. Courseware is intelligent enough to
understand and correct the inaccuracies in the conceptual model of a student. Media technol-
ogy allows students to interact with simulations of the very systems and personalities they are
studying.

Early 21st century The entire production sector of society is operated by a small number of technicians and
professionals. Individual customization of products is common

Early 21st century Drugs are designed and tested on human biochemical simulators.

Early 21st century Seeing machines for the blind provide both reading and navigation functions.

2010 A personal computer has the ability to answer a large variety of queries because it will know
where to find knowledge. Communication technologies allow it to access many sources of
knowledge by wireless communication

2020-2050 A phone call, which includes highly realistic three-dimensional holographic moving images, is
like visiting with the person called

2020-2070 A computer passes the Turing test, which indicates human-level intelligence

as3
Notes

The Second Industrial Revolution

1. Duncan Bythell, The Handloom Weavers: A Study in the English Cotton Industry during the Industrial
Revolution, p. 70.

2. Bythell, “The Coming of the Powerloom,” The Handloom Weavers, pp. 66-93.

3. Malcolm |. Thomis has written a sound documentation of this important historical movement in The
Luddites: Machine-Breaking in Regency England.
4. See, for example, Sir Percy Snow, speaker, “Scientists and Decision Making,” in Martin Green-
berger, ed., Computers and the World of the Future, p. 5; and Langdon Winner, “Luddism as Episte-
mology,” in Autonomous Technology, pp. 325-395.
5. Ben J. Wattenberg, ed., The Statistical History of the United States from Colonial Times to the
Present.
6. U.S. Department of Commerce, Bureau of the Census, Statistical Abstract of the United States,
1986, 106th ed., p. 390; see also U.S. Bureau of the Census, How We Live: Then and Now.
7. Ben J. Wattenberg, ed., The Statistical History of the United States from Colonial Times to the
Present.

8. U.S. Department of Commerce, Bureau of the Census, Statistical Abstract of the United States,
1986.

9. Ben J. Wattenberg, ed., The Statistical History of the United States from Colonial Times to the
Present, p. 224.
10. U.S. Department of Commerce, Bureau of the Census, Historical Statistics of the U.S.: Colonial
Times to 1970, vol. 1; and National Center for Education Statistics, U.S. Department of Education,
1986.
11. U.S. Department of Commerce, Bureau of the Census, Historical Statistics of the U.S.: Colonial
Times to 1970, vol. 1

12. U.S. Department of Commerce, Bureau of the Census, Historical Statistics of the U.S.: Colonial
Times to 1970, vol. 1

13. Wassily W. Leontief, The Impact of Automation on Employment, 1963-2000

14. Wassily W. Leontief, The Impact of Automation on Employment, 1963-2000
15. This phenomenon is discussed at length by Barry Bluestone and Bennett Harrison, in The Deindus-
trialization of America; also, see Lester Thurow “The Surge in Inequality,” Scientific American, May
1987, 30-38; and Harrison and Bluestone, The Great U-Turn.
16. Tom Forester surveys the cost and power trends in the computer revolution in High-Tech Society.
pp. 21-41

17. Edward Feigenbaum and Pamela McCorduck discuss the impact of expert systems on the field of
molecular biology in The Fifth Generation, p. 66. Sketches of computer-assisted diagnostic programs
presently in use can be found in Katherine Davis Fishman, The Computer Establishment, pp. 361-366;
see also Roger Schank, A Cognitive Computer: On Language, Learning, and Artificial intelligence, pp.
231-234.

18. Ben J. Wattenberg, ed., The Statistical History of the United States from Colonial Times to the
Present, series F, pp. 1-5, 1965.
19. David L. Parnas delivers one perspective on this topic in “Computers in Weapons: The Limits of
Confidence,” in David Bellin and Gary Chapman, eds., Computers in Battle —Will They Work? pp.
209-231; also of interest is a statement on future prospects for Al by Robert Dale, in Allen M. Din, ed.,
Arms and Artificial Intelligence, p. 45.
20. This possibility may be more hypothetical than real because of the close relationship between
manufacturing and services. Loss of manufacturing in key areas, for example, could be perilous to next-
stage prospects for innovation. For an analysis of these and related problems, see S. S. Cohen and J.
Zysman, Manufacturing Matters.
21. Translated from the Russian, “SAM” means surface-to-air missile, or literally, fixed maintenance
depot to air.

22. See Tom Athanasiou, “Artificial Intelligence as Military Technology,” in Bellin and Chapman, eds.,
Computers in Battle.
23. SCI is aimed toward the use of advanced computing to develop weapons and systems “for battle
management in complex environments where human decision-making [is] seen to be inadequate”
(Allan M. Din, ed., Arms and Artificial Intelligence, p. 7, see also pp. 90-91 in the same volume).

What Is Al, Anyway?

1. Similar definitions are found in many standard textbooks on Al.

2. This conference was originally called the Dartmouth Summer Research Project on Artificial Intelli-
gence; for a full account of this landmark event, see Pamela McCorduck, Machines Who Think,
pp. 93 ff
3. Norbert Weiner, the famous mathematician who coined this term (later supplanted by the term
“artificial intelligence”), was clearly fond of the meaning of its Greek root, “kubernetes”: pilot or
governor.
4. These terms were introduced by Edward Feigenbaum; see his "Art of Artificial Intelligence: Themes
and Case Studies in Knowledge Engines,” in AFIPS Conference Proceedings of the 1978 National
Computer Conference 47: 227-240.

5. Roger Schank, The Cognitive Computer, pp. 49-51

6. The layman may also want to see Susan J. Shepard, “Conversing with Tiny ELIZA,” Computer
Language 4 (May 1987). See also notes 61 and 62 to chapter 2.
7. See Hans Berliner, “New Hitech Computer Chess Success,” A/ Magazine9 (Summer 1988): 133.
And, for a brilliant discussion of machine versus human intelligence in chess and of dangers of rigidity
in “learning machines,” see Norbert Weiner, discussant, “Scientists and Decision Making,” in Martin
Greenberger, ed., Computers and the World of the Future, pp. 23-28.

8. See Lofti Zadeh, “Fuzzy Sets,” in Information and Control8: 338-353. See also a fascinating
interview with Zadeh published in Communications of the ACM, April 1984, pp. 304-311, in which he
discusses the inadequacy of precise Al techniques and tools to solve real life ("fuzzy") problems.

9. See Sigmund Freud, The Psychopathology of Everyday Life, in The Basic Writings of Sigmund Freud,
see also his Collected Papers; for another point of view, see Carl Jung et al., Man and His Symbols; and
for a shorter but broad overview on the subject, see William Kessen and Emily D. Cahan, “A Century of
Psychology: From Subject to Object to Agent,” American Scientist, Nov—Dec. 1986, pp. 640-650.

10. Newell's fullest and most current vision can be found in John E. Laird, A. Newell, and Paul S.
Rosenbloom, “SOAR: An Architecture for Intelligence” (University of Michigan Cognitive Science and
Machine Intelligence Laboratory Technical Report no. 2, 1987).

11. See Richard Dawkins's defense of Darwinism in The Blind Watchmaker: Why the Evidence of
Evolution Reveals a Universe without Design; for some classic arguments on design versus necessity,
also see A. Hunter Dupree, in Ada Gray, ed., Darwiniana, pp. 51-71.
12. This subject is eloquently addressed in a slim volume (23 pages) by S. Alexander, Art and Instinct.
13. To some, of course, the concept of God is not applicable to Buddhism; see William James, The
Varieties of Religious Experience, pp. 42-44 and 315.

486
14. Charles Darwin, The Origin of the Species. |n this, his classic work on natural selection and
evolution, Darwin states, “If it could be demonstrated that any complex organlism] existed, which
could not possibly have been formed by numerous, successive, slight modifications, my theory would
absolutely break down” (p. 229)
15. Richard Dawkins, The Blind Watchmaker, pp. 112-113.

16. Note, for example, a compelling argument against this notion (which instead champions the notion
of hierarchy in evolution) in Stephen Jay Gould, “Is a New and General Theory of Evolution Emerging?”
Paleobiology6 (1979): 19-130.

17. See Gould, Paleobiology6 (1979): 119-130. Also, in Stephen Jay Gould, The Mismeasure of Man,
pp. 326-334, mention is made of “human nature” in relation to the concept of natural selection. See
also Richard Dawkins, The Blind Watchmaker, pp. 141-142
18. This idea is supported, at least in theory, by some pioneers of Al; see, for example, Lawrence
Fogel, Alvin Owens, and Michael Walsh, Artificial Intelligence through Simulated Evolution, pp. viii
and 112
19. Edward Fredkin of MIT is credited with saying, “Artificial intelligence is the next step in evolution”
in Sherry Turkle, The Second Self, p. 242

Philosophical Roots
1. The literature on mind as machine is extensive. One provocative work is Daniel C. Dennett's Brain-
storms: Philosophical Essays on Mind and Psychology. Dennett, a philosopher, draws upon the
achievements of Al to formulate a new theory of mind. An important survey of philosophical issues can
be found in Margaret Boden’s Artificial Intelligence and Natural Man, chapter 14; see also Pamela
McCorduck, Machines Who Think. A brief, useful summary of trends is the introduction ("Philosophy
and Al: Some Issues”) to Steve Torrance, ed., The Mind and the Machine: Philosophical Aspects of
Artificial Intelligence. For the philosophy-Al nexus, see the papers in Martin D. Ringle, ed., Philosophical
Perspectives in Artificial Intelligence. Another important source is John Haugeland, ed., Mind Design
The legacy of the mind-body problem is related to contemporary Al debates in succinct, lively fashion
by Paul M. Churchland in Matter and Consciousness.
2. Some theorists who have argued for the mind-beyond-machine approach are J. R. Lucas, Hubert
Dreyfus, and John Searle. Lucas, in 1961, used Gédel’s incompleteness theorem to argue that
computers could never model the human mind successfully; see his “Minds, Machines and Godel,”
Philosophy
36 (1961): 120-124. For a refutation of this position, see Dennett's Brainstorms, chapter 13
Dreyfus’s famous critique of Al is What Computers Can’t Do: The Limits of Artificial Intelligence. Searle
distinguishes between the capacities of “weak Al” and “strong Al” in his 1980 paper “Minds, Brains,
and Programs,” The Behavioral and Brain Sciences 3 (1980): 417-424. (Searle's paper is reprinted as
chapter 10 of Haugeland’s Mind Designs.) Here Searle introduces his famous “Chinese room” example
to criticize what he sees as the “residual behaviorism” of Al. A particularly useful review of criticism of
Alis J. Schwartz, “Limits of Artificial Intelligence,” in Stuart C. Shapiro, ed., Encyclopedia of Artificial
Intelligence, vol. 1. The Encyclopedia is an excellent general source

3. See Boden’s discussion, in Artificial Intelligence, pp. 21-63, of Colby’s attempt to develop a compu-
tational model of human emotions; on pages 440-444 she argues that emotions are not “merely”
feelings; in What Computers Can't Do, Dreyfus argues from a phenomenological standpoint that
computers can never simulate our understanding, in part because of our capacity to experience
emotions. Dennett provides an intriguing discussion of the matter in chapter 11 ("Why You Can't Make
a Computer That Feels Pain”) of Brainstorms.

4. According to Dreyfus in What Computers Can't Do, “The story of artificial intelligence might well
begin around 450 s.c.," when Plato expressed the idea that “all knowledge must be stateable in
explicit definitions which anyone could apply” (p. 67).
5. For an overview see D. A. Rees, “Platonism and the Platonic Tradition,” The Encyclopedia of Philoso-
phy, vol. 6, pp. 333-341 (New York: The Macmillan Company, 1967)

6. See Thomas L. Hankins, Science and the Enlightenment. See also Ernst Cassirer, The Philosophy of
the Enlightenment, the first three chapters provide an important overview of the new studies of mind
and how they reflected methods of the new science.
7. See Reinhardt Grossman, Phenomenology and Existentialism: An Introduction; the critiques of Al
mounted by Hubert and Stuart Dreyfus have their roots in phenomenology. Hubert Dreyfus has
developed the Heideggerian notion that understanding is embedded in a world of social purpose, which
cannot be adequately represented as a set of facts. Stuart Dreyfus emphasizes the importance of skills

487
that elude representations and rules by drawing upon the existential phenomenology of Merleau-Ponty.
See H. Hall, “Phenomenology,” in Shapiro, Encyclopedia of Artificial Intelligence, vol. 2, pp. 730-736.
8. See A. J. Ayer, “Editor's Introduction,” in A. J. Ayer, ed., Logical Positivism, pp. 3-28; Rudolf Carnap,
“The Elimination of Metaphysics through Logical Analysis of Language,” in Ayer’s Logical Positivism,
pp. 3-81; Noam Chomsky, Syntactic Structures (1957). For a review of Chomsky’s achievement and
influence, see Frederick J. Newmeyer's Linguistic Theory in America, 2nd ed., chapter 2, “The
Chomskyan Revolution.”
9. This debate, in its technical and personal dimensions, is described in some detail in McCorduck’s
Machines Who Think.
10. Plato's works are readily available in Greek and English in the Loeb Classical Library editions; some
other English translations of individual works are mentioned below. An excellent place to begin is any of
several reference works: Gilbert Ryle, “Plato,” in The Encyclopedia of Philosophy, vol. 6, pp. 324-333;
D. J. Allan, “Plato,” in The Dictionary of Scientific Biography, vol. 11, pp. 22-31 (New York: Charles
Scribner's Sons, 1975). A more detailed account can be found in J. N. Findlay, Plato and Platonism: An
Introduction

11. In Aristotle: The Growth and Structure of His Thought, chapters 2 and 3, G. E. R. Lloyd describes
Aristotle as both a pupil and a critic of Plato
12. See “The Greek Academy,” in The Encyclopedia of Philosophy, vol. 3, pp. 382-385. The Academy
is also treated in Ryle’s “Plato,” pp. 317-319. In his excellent survey, A History of Greek Philosophy,
vol. 4, p. 19, W. K. C. Guthrie explains that the Academy was by no means like our modern university: it
had religious elements we might more readily associate with a medieval college. Volume 4 of this
survey is devoted to Plato; the Academy is discussed on pp. 8-38. The early years of Plato's Academy
are described in the reprint edition of Eduard Zeller's 1888 classic, Plato and the Older Academy. See
note 17 below.
13. Guthrie (vol. 4, pp. 338-340) points out that Plato was influenced by the mystery religions of his
day, especially in the Phaedo.
14. Plato describes the movements of the planets in important passages in the Republic and the
Timaeus; In Plato’s Timaeus, pp. 33-35, Francis Cornford provides a useful summary of the kinds of
motion Plato describes in the Timaeus. G. E. R. Lloyd has a lucid and concise discussion of Plato's
astronomy in chapter 7 of Early Greek Science: Thales to Aristotle, pp. 80-98. The nature of Plato's
astronomy, long a controversial subject for the history of science, is analyzed in John D. Anton's
Science and the Sciences in Plato.
15. The myth of Er in the Republic (617-618) was Plato's version of a scheme originally developed by
the Pythagorean philosopher Philolaus, who put fire at the extremity and at the center of the universe,
thus displacing the earth from its central position (G. S. Kirk, J. E. Raven, and M. Schofield, The
Presocratic Philosophers, p. 259). The Pythagorean concept of a central fire is described by S. Sam-
bursky in The Physical World of the Greeks, pp. 64-66.
16. The discovery of irrational numbers eventually resulted in the rejection of a Pythagorean "geometric
atomism” and led to the concept of the continuum (S. Sambursky, The Physical World of the Greeks,
pp. 33-35). In Plato’s Theaetetus, the mathematician Theodorus demonstrates the irrationality of
nonsquare numbers up to the root of 17. Plato then claims that the roots of all numbers that are not
squares are irrational. According to G. E. R. Lloyd (Early Greek Science: Thales to Aristotle, pp. 32-34),
the irrationality of the square root of 2 was known even before the time of Plato. The Greeks commonly
expressed the proof in geometrical terms, by showing that the diagonal of a square is not commensu-
rable with its side. (The proof assumes this commensurability, then shows that it leads to an impossibil-
ity because the resulting number is both odd and even.) The discovery that some magnitudes are
incommensurable (c. 450-441 8.c.) is attributed to Hippacos of Mepontum, a member of the Pythago-
rean Brotherhood, in Alexander Helleman and Bryan Bunch, The Timetables of Science, p. 31
17. E. R. Dodds, The Greeks and the Irrational is a classic treatment of this subject. Ananke is described
in detail in F. M. Cornford, Plato’s Cosmology, pp. 159-177. A more recent work is Richard R. Mohr's
Platonic Cosmology.
18. In the Phaedo and The Republic, Plato opposes the activity of intellect to the “brutish” passivity of
desire (Martha Nussbaum, “Rational Animals and the Explanation of Action,” in The Fragility of
Goodness: Luck and Ethics in Greek Tragedy and Philosophy, p. 273). In this book Nussbaum explores
the antithesis in Greek philosophy between the controlling power of reason and events beyond one’s
control, an antithesis central to Plato's dialogues.
19. The first mention of the Forms is in the Phaedo; an excellent discussion can be found in Gilbert
Ryle’s article (pp. 320-324) in The Encyclopedia of Philosophy.

ass
20. Plato's theory of matter in the Timaeus, where the smallest particles are triangles, is a blend of
Pythagorean ideas and Democritan atomism (see S. Sambursky, The Physical World of the Greeks,
p. 31),

21. Cornford, in Plato's Cosmology, pp. 159-177, provides a lucid discussion of this tension between
necessity and reason.

22. On the dialogue as Plato's chosen form, see D. Hyland’s “Why Plato Wrote Dialogues,” Philosophy
and Rhetoric1 (1968): 38-50.
23. Physicist Werner Heisenberg describes how he arrived at his uncertainty principle, which he formu-
lated in 1927 in chapter 6 of his gracefully written and entertaining volume Physics and Beyond: En-
counters and Conversations. Heisenberg was influenced by Plato’s corpuscular physics, and he
explores the relation between Plato's ideas and quantum theory in chapter 20, “Elementary Particles
and Platonic Philosophy (1961-1965).”

24. A refreshing new interpretation of the Phaedrus emphasizing the role of paradox is Martha
Nussbaum’s “'This Story Isn't True’: Madness, Reason, and Recantation in the Phaedrus,” chapter 7 in
The Fragility of Goodness, pp. 200-228
25. D. A. Rees, “Platonism and the Platonic Tradition,” p. 336. It was Xenocrates, who headed the
Academy after the death of Speusippus, Plato’s immediate successor, who identified the Platonic
Ideas with mathematical numbers, not the “ideal” numbers postulated in the Academy under Plato and
discussed in the Phaedo. The fates of the various forms of Platonism are reviewed in several brief
articles in the Dictionary of the History of Ideas (New York: Charles Scribner's Sons, 1973), vol. 3: John
Fisher's “Platonism in Philosophy and Poetry,” pp. 502-508; John Charles Nelson’s “Platonism in the
Renaissance,” pp. 508-515; and Ernst Moritz Manasse’s “Platonism since the Enlightenment,” pp
515-525.

26. D. H. Fowler, in The Mathematics of Plato's Academy, reconstructs in detail the curriculum of the
Academy. A particularly readable account of the work of the geometers can be found in chapter 3 of
Francois Lasserre, The Birth of Mathematics in the Age of Plato. A more technical treatment can be
found in chapter 3 of Wilbur Richard Knorr, The Ancient Tradition of Geometric Problems.

27. For an general overview of Plato's philosophy of numbers, see “Plato,” The New Encyclopedia
Britannica, vol. 14, p. 538. For the text of the Epinomis in Greek and English, see W. R. M. Lamb, ed.,
Plato, Loeb Classical Library, vol. 8. In the Epinomis, 976 D-E, the speaker asks what science is indis-
pensable to wisdom: “it is the science which gave number to the whole race of mortals.” See also R.
S. Brumbaugh, Plato’s Mathematical Imagination.
28. A superb introduction to Enlightenment thought is Peter Gay's two volumes, The Enlightenment:
An Interpretation, vol. 1, The Rise of Modern Paganism and vol. 2, The Science of Freedom.
29. The definitive biography is Richard Westfall’s Never at Rest: A Biography of Isaac Newton. No one
interested in Isaac Newton's scientific achievement should fail to see |. Bernard Cohen's Newtonian
Revolution. Those who wish to tackle Newton in the original should see /saac Newton’s Philosophiae
Naturalis Principia Mathematica, 3rd edition (1726), assembled by Alexander Koyré, |. Bernard Cohen,
and Anne Whitman.
30. Otto Mayr, Authority, Liberty, and Automatic Machinery in Early Modern Europe.
31. A useful overview of Descartes’s life and work can be found in The Dictionary of Scientific
Biography, vol. 4, pp. 55-65. Descartes, by Jonathan Rée, is unsurpassed in giving a unified view of
Descartes's philosophy and its relation to other systems of thought.

32. The brief Discours de la Méthode appeared in 1637 and is written ina lively autobiographical
manner. It is readily available in the Library of the Liberal Arts edition, which includes the appendixes in
which Descartes introduced analytic geometry and his theory of refraction: Discourse on Method,
Optics, Geometry, and Meteorology, trans. by Paul J. Olscamp.

33. Derek J. de Solla Price, “Automata and the Origins of Mechanism and Mechanistic Philosophy,
Technology and Culture 5 (1964): 23.
34. See |. Bernard Cohen on Newton in the Dictionary of Scientific Biography, vol. 10, pp. 42-103, and
Cohen's Newtonian Revolution, mentioned above.
36. Charles Gillispie, The Edge of Objectivity, p. 140. The resulting prestige of science during the
Enlightenment is treated in chapter 5.
36. For a readable and lucid introduction to relativity, see the 1925 classic by Bertrand Russell, The
ABC of Relativity, 4th rev. ed. A more detailed treatment may be found in Albert Einstein, Relativity.
The Special and General Theory, a Popular Exposition, trans. Robert Lawson.

aso
37. Gillispie, The Edge of Objectivity, pp. 145-150.
38. Leibniz’s criticism of the watchmaker God can be found in a letter written in November 1715 to
Samuel Clarke (1675-1729), a renowned disciple of Newton (see pp. 205-206 of Leibniz’s Philosophical
Writings, G. H. R. Parkinson, ed.). For the famous debate this letter initiated, see The Leibniz-Clarke
Correspondence, H. G. Alexander, ed
39. W.T. Jones, Kant and the Nineteenth Century, p. 14. The legacy of Descartes is expressed in
Kant's own definition of the Enlightenment, which is quoted by Ernst Cassirer in The Philosophy of the
Enlightenment, p. 163: “Enlightenment is man’s exodus from his self-incurred tutelage. Tutelage is the
inability to use one’s understanding without the guidance of another person. This tutelage is self-
incurred if its cause lies not in any weakness of the understanding, but in indecision and lack of courage
to use the mind without the guidance of another. ‘Dare to know’ (sapere aude)! Have the courage to
use your Own understanding; this is the motto of the Enlightenment.”

40. Immanuel Kant, Critique of Pure Reason, 1st ed. 1781; Prolegomena to Any Future Metaphysics,
1st ed. 1783. The relations between Kantian philosophy and science are explored in Gordon G. Brittan,
ur., Kant’s Theory of Science.
41. A brief history of logical positivism can be found in A. J. Ayer, Logical Positivism, pp. 3-28. Moritz
Schlick, center of the Vienna Circle in the 1920s, compares the Kantian and positivist treatments of
reality in “Positivism and Realism,” an essay published in 1932 or 1933 and reprinted in Ayer's Logical
Positivism (see p. 97).
42. Ayer, in Logical Positivism, p. 11, points out the positivist nature of Hume's attack on metaphysics
and then claims that he could well have cited Kant instead, “who maintained that human understanding
lost itself in contradictions when it ventured beyond the bounds of possible experience.” Ayer claims
that “the originality of the logical positivists lay in their making the impossibility of metaphysics depend
not upon the nature of what could be known but upon the nature of what could be said” (Logical
Positivism, p. 11).
43. Norman Malcolm, Ludwig Wittgenstein: A Memoir, with a Biographical Sketch by Georg Henrik Von
Wright, p. 10. Whereas Kant distinguished between what can and cannot be known, Wittgenstein
distinguished between what can and cannot be said. See “The Tractatus,” chapter6 of W. T. Jones,
The Twentieth Century to Wittgenstein and Sartre.

44. Ludwig Wittgenstein, Tractatus Logico-Philosophicus, trans. by D. F. Pears and B. F. McGuiness,

first German edition published in 1921
45. Malcolm, Ludwig Wittgenstein, pp. 11-12.
46. Wittgenstein, Tractatus, p. 37
47. Wittgenstein, Tractatus, p. 115.
48. Wittgenstein, Tractatus, p. 115.

49. For a readable discussion of the Church-Turing thesis, see David Harel’s Algorithmics: The Spirit of
Computing, pp. 221-223. The Church-Turing thesis, named after Alonzo Church and Alan Turing, is
based on ideas developed in the following papers: Alan Turing, “On Computable Numbers with an
Application to the Entscheidungsproblem,” Proc. London Math. Soc. 42(1936): 230-265; Alonzo
Church, “An Unsolvable Problem of Elementary Number Theory,” Amer. J. Math. 58 (1936): 345-363.
50. See, for example, statement 4.002 in Wittgenstein’s Tractatus, p. 37.

51. Wittgenstein, Tractatus, pp. 7, 151

52. Ludwig Wittgenstein, Philosophical Investigations, trans. G .E .M. Anscombe.

53. Michael Dummett, in his “Frege and Wittgenstein” (in Irving Block, ed., Perspectives on the Phi-
losophy of Wittgenstein, pp. 31-42), argues that Wittgenstein tried and failed to provide a theory of
language in Philosophical Investigations.
54. In the preface to Philosophical Investigations (p. vi), Wittgenstein claims that he recognized “grave
mistakes” in his earlier work, the Tractatus. The more atomistic approach of the latter is challenged by
a greater emphasis on contexts in Philosophical Investigations. Anthony Kenny compares the two
works in “Wittgenstein’s Early Philosophy of Mind,” Block, ed., Perspectives, pp. 140-147. A. J. Ayer
remarks that Wittgenstein “modified the rigors of his early positivism” as expressed in the Tractatus.
(See Ayer's Logical Positivism, p. 5)
55. In the Preface to his 1936 work Language, Truth and Logic, p. 31, Alfred Ayer asserts that his views
stem from the writings of Russell and Wittgenstein

56. See Reinhardt Grossman, Phenomenology and Existentialism.

57. Tractatus, p. 151

490
58. Hubert L. Dreyfus, “Alchemy and Artificial Intelligence,” The RAND Corporation, December 1965,
publication 3244. For a profile of Dreyfus, see Frank Rose, “The Black Knight of Al,” Science 85, 6
(March 1985): 46-51

59. Pamela McCorduck, Machines Who Think, p. 204. McCorduck devotes chapter 9 ("L’Affair
Dreyfus”) to an engaging history of Dreyfus's critique and the reactions it provoked in the Al community.

60. ELIZA was first announced in Joseph Weizenbaum’s “ELIZA—A Computer Program for the Study
of Natural Language Communication between Man and Machine," Communications of the Association
for Computing Machinery9 (1966): 36-45. Hubert Dreyfus stumped ELIZA by entering the phrase “I’m
feeling happy,” and then correcting it, by adding “No, elated.” ELIZA responded with “Don’t be so
negative,” because it is programmed to respond that way whenever “no” appears anywhere in the
input. See Hubert Dreyfus and Stuart Dreyfus, “Why Computers May Never Think like People,”
Technology Review 89 (1986): 42-61
61. ELIZA mimics a Rogerian psychotherapist, whose technique consists largely of echoing utterances
of the patient; it therefore uses very little memory, and arrives at its “answers” by combining transfor-
mations of the “input” sentences with phrases stored under keywords. Its profound limitations were
acknowledged by its creator. In his 1976 work, Computer Power and Human Reason, Weizenbaum
argues that ELIZA's limitations serve to illustrate the importance of context for natural language under-
standing, a point made in his original paper. He chose this kind of psychotherapeutic dialogue precisely
because the psychotherapist in such a dialogue need know practically nothing about the real world
See Margaret Boden, Artificial Intelligence and Natural Man, p. 108.

62. Dreyfus developed this argument in detail in What Computers Can't Do: The Limits of Artificial
Intelligence. There he sets out objections to “the assumption that man functions like a general-purpose
symbol-manipulating device” (p. 156). Especially drawing his ire was the work of Allen Newell and H
A. Simon, Computer Simulation of Human Thinking, The RAND Corporation, P-2276 (April 1961)
63. PROLOG, a language based upon logic programming, was devised by Colmerauer at Marseille
around 1970 (see W. F. Clocksin and C. S. Mellish, Programming in PROLOG).
64. Fuzzy logic, developed by L. A. Zadeh, guards against the oversimplification of reality by not
assuming all fundamental questions have yes or no answers. See E. H. Mamdani and B. R. Gaines,
Fuzzy Reasoning and Its Applications.
65. See, in particular, the introduction to the revised edition, in Hubert Dreyfus, What Computers Can't
Do.
66. Dreyfus’s predictions about the limitations of chess-playing programs have been proven wrong
time and again. Chess-playing programs have improved their performance through the application of
greater and greater computational power. One of the latest benchmarks occurred when HiTech won
the Pennsylvania State Chess Championship in 1988. See Hans Berliner, “HITECH Becomes First
Computer Senior Master,” A/ Magazine 9 (Fall 1988): 85-87.

67. McCorduck, Machines Who Think, p. 205.

68. In What Computers Can't Do Dreyfus argues, “There is no justification for the assumption that we
first experience isolated facts, or snapshots of facts, or momentary views of snapshots of isolated
facts, and then give them significance. The analytical superfluousness of such a process is what
contemporary philosophers such as Heidegger and Wittgenstein are trying to point out” (p. 270)

69. Hubert L. Dreyfus and Stuart E. Dreyfus, “Making a Mind versus Modeling the Brain: Artificial
Intelligence Back at a Branchpoint,” Daedalus 117 (Winter 1988): 15-43. This issue of Daedalus,
devoted to Al, was subsequently published in book form; see Stephen R. Graubard, ed., The Artificial
Intelligence Debate: False Starts, Real Foundations.
70. Dreyfus and Dreyfus, “Making a Mind,” p. 15
71. Jack Cowan and David H. Sharp review the importance of neural nets for Al in “Neural Nets and
Artificial Intelligence,” Daedalus 117 (Winter 1988): 85-121

72. Dreyfus and Dreyfus, “Making a Mind,” pp. 38-39

73. See “The Role of the Body in Intelligent Behavior,” chapter 7 of Hubert Dreyfus's What Computers
Can't Do.
74. Sherry Turkle also explores children’s responses to computers in her 1984 work, The Second Self.
Computers and the Human Spirit, chapter 1, “Child Philosophers: Are Smart Machines Alive?"

75. Sigmund Freud, Jokes and Their Relation to the Unconscious, \st ed., 1905. Marvin Minsky
provides a new interpretation of jokes, emphasizing the importance of “knowledge about knowledge,”
in his “Jokes and the Logic of the Cognitive Unconscious,” in Lucia Vaina and Jaakko Hintikka, eds
Cognitive Constraints on Communication, pp. 175-200

ant
Mathematical Roots

1. For the relationship between logic and recursion, see Stephen Cole Kleene, “I-Definability and
Recursiveness,” Duke Mathematical Journal 2 (1936): 340-353. See also Stephen Cole Kleene, /ntro-
duction to Metamathematics. For Rosser’s contribution, see J. Barkley Rosser, “Extensions of Some
Theorems of Gédel and Church,” Journal of Symbolic Logic1 (1936): 87-91. Church has made many
important contributions to logic and computation. A coherent presentation of his work appears in
Alonzo Church, Introduction to Mathematical Logic, vol. 1.
2. For the flavor of this theory, see a classic text on numerical analysis and computation: R. W.
Hamming, Introduction to Applied Numerical Analysis.

3. A good example of such thinking is Bertrand Russell, Introduction to Mathematical Philosophy.

4. The paradox was first introduced in Bertrand Russell, Principles of Mathematics, 2nd ed., pp 79-81
Russell's paradox is a subtle variant of the Liar Paradox. See E. W. Beth, Foundations of Mathematics,
p. 485.
5. Gottlob Frege was about to publish a monumental work on arithmetic and set theory when Russell
pointed out the implications of his paradox. Frege could only add a postscript that said, “A scientist can
hardly meet with anything more undesirable than to have the foundations give way just as the work is
finished. In this position | was put by a letter from Mr. Bertrand Russell.” See Bertrand Russell, Letter
to Frege, 1902, published in Jean van Heijenoort, ed., From Frege to Godel.
6. Bertrand Russell, Principles of Mathematics, 2nd ed., 1938, pp. 10-32, 66-81.
7. Bertrand Russell, Principles of Mathematics, 2nd ed., pp. 10-32, 66-81
8, See Bertrand Russell, Principles of Mathematics, 2nd ed., pp. v-xiv.
9. See also Alfred N. Whitehead and Bertrand Russell, Principia Mathematica, 3 vols., 2nd ed., pp.
187-231

10. First introduced in Alan M. Turing, “On Computable Numbers with an Application to the
Entscheidungsproblem,” Proc. London Math. Soc. 42 (1936): 230-265
11. Work on PROLOG began in 1970. A clear presentation of the conceptual foundations of PROLOG
appears in Robert Kowalski, “Predicate Logic as a Programming Language,” University of Edinburgh,
DAI Memo 70, 1973. See also Alain Colmerauer, “Sur les bases théoriques de Prolog,” Groupe de IA,
UER Luminy, Univ. d‘Aix-Marseilles, 1979. This and other aspects of the Japanese program are
discussed in Edward Feigenbaum and Pamela McCorduck, The Fifth Generation, p. 115.

12. These early experiments are described in A. Newell, J. C. Shaw, and H. Simon, “Empirical
Explorations with the Logic Theory Machine,” Proceedings of the Western Joint Computer Conference
15 (1957): 218-239,
13. Turing’s theoretical model was first introduced in Alan M. Turing, “On Computable Numbers with
an Application to the Entscheidungsproblem,” Proc. London Math. Soc." 42 (1936): 230-265
14. An enormously influential paper is Alan M. Turing, “Computing Machinery and Intelligence,” Mind
59 (1950): 433-460, reprinted in E. Feigenbaum and J. Feldman, eds., Computers and Thought.

15. The program is called the “Turochamp” (evidently a contraction of “Turing and Champernowne”).
See Andrew Hodges, Alan Turing: The Enigma, pp. 338-339.
16. Turing researched morphogenesis deeply enough to produce paper on the subject: Alan M.
Turing, “The Chemical Basis of Morphogenesis,” Phil. Trans. Roy. Soc. 1952: B237
17. See Andrew Hodges, Alan Turing: The Enigma, pp. 267-268.

18. For an engineering account of this project, see B. Randell, “The Colossus” (1976), reprinted in N.
Metropolis, J. Howlett, and G. C. Rota, eds., A History of Computing in the Twentieth Century.

19. See David Hilbert, Grundlagen der Geometrie, Liepzig and Berlin, 1899, 7th ed., 1930.

20. Alan M. Turing. “On Computable Numbers with an Application to the Entscheidungsproblem,”
Proc. London Math. Soc. 42 (1936): 230-265.

21. Simpler models that have appeared since have perhaps been ignored. See Marvin Minsky, ,
Computation: Finite and Infinite Machines. f
22. This thesis was independently arrived at by both Church and Turing around 1936
eee

23. For an excellent article on the theory of computation, see John M. Hopcroft, “Turing Machines,”
Scientific American, May 1984, pp. 86-98.
ee

492
24. The busy beaver problem is one example of a large class of noncomputable functions, as one can
see from Tibor Rado, “On Noncomputable Functions,” Bell System Technical Journal 41, no. 3 (1962):
877-884.

25. Church's version of the result appears in Alonzo Church, “An Unsolvable Problem of Elementary
Number Theory,” American Jour. Math 58 (1936): 345-363.
26. We can see Gédel’s concerns about Russell's framework in Kurt Godel, “Russell's Mathematical
Logic” (1944), in P. A. Schilpp, ed., The Philosophy of Bertrand Russell.

27. Gédel's incompleteness theorem first appeared in: Kurt Gédel, “Uber formal unenscheiderbare
Satze der Principia Mathematica und verwandter Systeme |,” Monatsh. Math. Phys. 38 (1931)
173-198.
28. See Alonzo Church, “A Note on the Entscheidungsproblem,” Journal of Symbolic Logic 1 (1936):
40-41, and Kurt Gédel, “On Undecidable Propositions of Formal Mathematical Systems,” mimeo-
graphed report of lectures at the Institute for Advanced Study, Princeton, 1934.

29. Herbert A. Simon, The Shape of Automation for Men and Management (Harper & Row, 1965),
p. 96.
30. A short reflection by Turing on some of the issues behind thinking machines appears as part of
chapter 25 of B. W. Bowden, ed., Faster than Thought.

31. For an introductory account of some of the implications of the Church-Turing thesis, see Douglas
Hofstadter, Godel, Escher, Bach: An Eternal Golden Braid, pp. 559-586.

The Formula for Intelligence

1. See Morris Kline, Mathematics and the Search for Knowledge.

2. See Albert Einstein, Relativity: The Special and the General Theory. A more readable account is
presented in Bertrand Russell, ABC of Relativity. See also A. Einstein, “Zur Elektrodynamic bewegter
Korpen,” Annalen der Physik 17 (1905): 895, 905.

3. For the mathematically mature, an excellent introduction can be found in Enrico Fermi, Thermody-
namics (Englewood Cliffs, N.J.: Prentice-Hall, 1937).

4. Atkins gives an account of thermodynamics and entropy that is fascinating and informal yet scholarly
in P. W. Atkins, The Second Law.

5. A glimpse into the complexity is presented in Allan C. Wilson, “The Molecular Basis of Evolution,”
Scientific American, October 1985, pp. 164-173.

6. See Rudy Rucker, The Five Levels of Mathematical Reality, pp. 14-35.
7. The motivations and quests for anthropomorphic parallels are considered in John D. Barrow and
FrankJ. Tipler, The Anthropic Cosmological Principle, pp. 1-23
8. See Robert P. Crease and Charles C. Mann, The Second Creation, pp. 393-420
9. For one contribution to a “theory of everything,” see Stephen Hawking, A Brief History of Time. A
more popular discussion is given in Heinz R. Pagels, Perfect Symmetry, pp. 269-367
10. In 1666 Gottfried Leibniz contemplated a scientific system of reasoning, the “calculus ratiocinator,”
that could be used to settle arguments formally. George Boole took up this problem and presented his
work in 1854 in An Investigation of the Laws of Thought on Which Are Founded the Mathematical
Theories of Logic and Probabilities, aspects of which are discussed in the next few pages.

11. See Douglas Hofstadter, Gédel, Escher, Bach: An Eternal Golden Braid, pp. 559-586.

12. Meindl, James D. “Chips for Advanced Computing," Scientific American, October 1987, p. 78.
13. Fora detailed treatment, see Carver Mead and Lynn Conway, /ntroduction to VLS/ Systems. Old
but nonetheless broadly relevant is the article lvan E. Sutherland and Carver A. Mead, “Microelectron-
ics and Computer Science,” Scientific American, September 1977, pp. 210-228.
14. For a discussion that is less philosophical than that of Hofstadter, see Rudy Rucker, The Five Levels
of Mathematical Reality, pp. 207-249.
15. See John M. Hopcroft, “Turing Machines,” Scientific American, May 1984, p. 91
16. See A. Newell and H. A. Simon, “GPS: A Program that Simulates Human Thought,” in E. A.
Feigenbaum and J. Feldman, eds., Computers and Thought, pp. 71-105, and Claude Shannon, “A
Chess Playing Machine,” Scientific American, October 1950.
17. For a sketch, see Patrick H. Winston, “The LISP Revolution, " BYTE, April 1985, p. 209.

agg
18. For this reason we've been more successful in building checkers programs. See Arthur L. Samuel,
“Some Studies in Machine Learning Using the Game of Checkers,” (1959), reprinted in E. A. Feigen-
baum and J. Feldman, eds., Computers and Thought, pp. 279-293. An early note is given in Claude
Shannon, “Programming a Computer for Playing Chess,” Philosophical Magazine, series 7, 41 (1950):
256-275.
19. This and some of the other formulations discussed here have been examined in depth by research-
ers in game theory. A seminal work in the area is R. D. Luce and H. Raiffa, Games and Decisions. The
famous Minimax theorem itself was presented in J. von Neumann, “Zur Theorie der
Gesellschaftespiele," Mathematische Annalen 100 (1928): 295-320.
20. This serves to show that in theory a computer can be as good as any human chess player.
21. Researchers have tried various strategies to get around the problems created by this combinatorial
explosion in the number of possible chess moves at each stage. See Peter Frey, “An Introduction to
Computer Chess,” in Peter Frey, ed., Chess Skill in Man and Machine, and also M. M. Botvinnik,
Computers in Chess, pp. 15-21
22. A. K. Dewdney, “The King Is Dead, Long Live the King,” Scientific American, May 1986, p. 13
23. See Gregory Chaitin, “On the Difficulty of Computation,” /EEE Transactions on Information Theory
16 (1970): 5-9, and Gregory Chaitin, “Computing the Busy Beaver Function,” IBM Watson Research
Center Report, RC 10722, 1970. An easy introduction to certain aspects of computability and complex-
ity is in Michael R. Garey and David S. Johnson, Computers and Intractability.
24. See, for instance, the piece by M. A. Tsasfman and B. M. Stilman in M. M. Botvinnik, ed., Comput-
ers in Chess. Also see Carl Ebeling, All the Right Moves, pp. 56-64.
25. A lucid presentation on the two positions can be found in Carl Ebeling, A// the Right Moves,
pp. 1-3
26. Compare the various strategies and systems described in Peter Frey, ed., Chess Skill in Man and
Machine.
27. A recent report on HiTech is Hans Berliner, Al Magazine, Summer 1988.
28. The structure of HiTech is well documented in Carl Ebeling, Ai! the Right Moves.

29. H. A. Simon and Allen Newell, “Heuristic Problem Solving: The Next Advance in Operations
Research,” Operations Research 6 (January-February 1958).
30. A report of the system's performance is given in Danny Kopec and Monty Newborn, “Belle and
Mephisto Dallas Capture Computer Chess Titles at the FICC,” Communications of the ACM, July 1987,
pp. 640-645
31. In 1988 HiTech became the first system to beat a human chess grandmaster, albeit one who has
been out of form. See Harold C. Schonberg, New York Times, September 26, 1988
32. See W. Daniel Hillis, “The Connection Machine,” Scientific American, June 1987

33. Eliot Hearst, "Man and Machine: Chess Achievements and Chess Thinking,” in Peter Frey, ed.,
Chess Skill in Man and Machine.

34. A useful examination of the psychology of chess-playing in the light of the performance of chess
programs is given in Brad Leithauser, “Computer Chess,” New Yorker, May 9, 1987, pp. 41-73. See
also the article by Hearst, cited in note 33.

35. An excellent survey is in Geoffrey C. Fox and Paul C. Messina, “Advanced Computer
Architectures,” Scientific American, October 1987, pp. 66-74. The flurry of research activity is evident
from Richard Miller (project manager), Optical Computers: The Next Frontier in Computing, vols. 1 and
2 (Englewood, N.J.: Technical Insights, 1986)

36. Even the early checkers programs were quite good. See Pamela McCorduck, Machines Who Think,
pp. 152-153
37. H. J. Berliner, “Backgammon Computer Program Beats World Champion,” Artificial Intelligence 14,
no. 1 (1980)
38. See H. J. Berliner, "Computer Backgammon,” Scientific American, June 1980
39. The number of possible moves at each point is estimated at 200 for go. See E. Thorp and W. E
Walden, “A Computer-Assisted Study of Go on M by N Boards,” in R. B. Banerji and M. D. Mesarovic,
eds., Theoretical Approaches to Non-numerical Problem-Solving (Berlin: Springer-Verlag, 1970), pp.
303-343,

aga
40. An early effort on go is described in W. Reitman and B. Wilcox, “Pattern Recognition and Pattern
Directed Inference in a Program for Playing Go,” in D. A. Waterman and F. Hayes-Roth, eds., Pattern-
Directed Inference Systems.
41. As stated by John Laird, the cannibals and missionaries problem is, “Three cannibals and three
missionaries want to cross a river. Though they can all row, they only have available a small boat that
can hold two people. The difficulty is that the cannibals are unreliable: if they ever outnumber the
missionaries on a river bank, they will kill them. How do they manage the boat trips so that all six get
safely to the other side?”

42. A. Newell, J. C. Shaw, and H. A. Simon, “Empirical Explorations with the Logic Theory Machine”
(1957), reprinted in E. A. Feigenbaum and J. Feldman, eds., Computers and Thought, pp. 109-133. The
generalized results can be seen in A. Newell, J. C. Shaw and H. A. Simon, “A Report on a General
Problem Solving Program," Proceedings of the International Conference on Information Processing
(UNESCO, Paris, 1959), pp. 256-264
43. Notably from the Dreyfus brothers. See Hubert Dreyfus, What Computers Can't Do, 2nd ed.

44. A. Newell and H. A. Simon, “GPS: A Program That Simulates Human Thought,” in E. A. Feigen-
baum and J. Feldman, eds., Computers and Thought, pp. 71-105
45. H. A. Simon and Allen Newell, “Heuristic Problem Solving: The Next Advance in Operations
Research,” Operations Research. 6 (January-February 1958)

46. Some problems are described in Patrick H. Winston, Artificial Intelligence, pp. 146-154. The results
and lessons of GPS are detailed in A. Newell and H. A. Simon, Human Problem Solving.
47. See E. Feigenbaum and Avron Barr, The Handbook of Artificial Intelligence, vol. 1, pp. 123-138
48. An excellent paper on intelligence and computer chess is A. Newell, J. C. Shaw, and H. A. Simon,
“Chess Playing Programs and the Problem of Complexity” (1958), reprinted in E. Feigenbaum and J
Feldman, Computers and Thought.
49. Minsky’s views on intelligence serve us well here: Marvin Minsky, “Why People Think Computers
Can't,” Technology Review, November-December 1983, pp. 64-70.
50. Formally defined in Marvin Minsky and Seymour Papert, Perceptrons, p. 12
51. W.S. McCulloch and W. Pitts, “A Logical Calculus of the Ideas Immanent in Neural Nets,” Bulletin
of the Mathematical Biophysics 5 (1943)

52. Marvin Minsky and Seymour Papert, Perceptrons, pp. 136-150.

53. An excellent introductory article on the history and achievements of connectionism is Jerome
Feldman, “Connections,” BYTE, April 1985, pp. 277-284
54. This is reflected in the progress reports issued by the MIT Al Laboratory during that period. See, for
instance, Marvin Minsky, Seymour Papert, “New Progress in Artificial Intelligence,” MIT Artificial
Intelligence Laboratory, Al memo 252, 1972
55. See Douglas Hofstadter, Metamagical Themas, pp. 274-292
56. Widely applicable algorithms are likely to perform weakly in all their domains. See Seymour Papert
“One Al or Many,” Daedalus, Winter 1988
57. A recent survey is in Jack D. Cowan and David. H. Sharp, “Neural Nets and Artificial Intelligence,”
Daedalus, Winter 1988, pp. 85-121
58. Important papers on recent work are put together in the standard reference in the field: D. E
Rumelhart, J. L. McClelland, and the PDP Research Group, Paralle! Distributed Processing
59. See Marvin Minsky, “Connectionist Models and Their Prospects,” in David Waltz, ed., Connection-
ist Models and Their Implications (Norwood, N.J.: Ablex Publishing, 1988)

60. This selection is carried out in the style of “summarizing” in the society theory. See Marvin Minsky,
Society of Mind, p. 95.
61. Decision trees have been used extensively in Management Science. For an enjoyable introduction,
see Howard Raiffa, Decision Analysis: Introductory Lectures (Reading, Mass.: Addison-Wesley)

62. This is a point well brought out in Marvin Minsky, “Why People Think Computers Can't.”
Technology Review, November-December 1983, pp. 64-70.
63. See R. C. Schank and R. Abelson, Scripts, Plans, Goals, and Understanding (Hillsdale, N.J
Erlbaum, Lawrence Associates, 1977)

ag5
64. Marvin Minsky, “Plain Talk about Neurodevelopment Epistemology,” Proceedings of the Fifth
International Joint Conference on Al (Cambridge, Mass., 1977). Minsky's work culminated in a major
book: Marvin Minsky, The Society of Mind. é

65. Minsky, The Society of Mind, p. 17.

66. For a sketch of the society theory, see Marvin Minsky, “Society of Mind,” Artificial Intelligence
Journal 1989.

67. For early related work, see Jerome Lettvin, H. Maturana, W. McCulloch and W. Pitts, “What the
Frog's Eye Tells the Frog's Brain,” Proceedings of the IRE 47 (1959): 1940-1951. This famous paper is.
reprinted with other related papers in Warren S. McCulloch, Embodiments of Mind. Also see: W. S.
McCulloch and W. Pitts, “A Logical Calculus of the Ideas Immanent in Neural Nets,” Bulletin of the
Mathematical Biophysics 5 (1943), reprinted in Warren S. McCulloch, Embodiments of Mind.
68. Jerome Lettvin, H. Maturana, W. McCulloch and W. Pitts, “What the Frog’s Eye Tells the Frog’s
Brain,” Proceedings of the IRE 47 (1959): 1940-1951

69. John McDermott, “R1: A Rule-Based Configurer of Computer Systems,” Artificial Intelligence 19,
no. 1 (1982). Also see John McDermott, “XSEL: A Computer Salesperson’s Assistant,” in J. Hayes, D.
Michie, and Y. H. Pao, Machine Intelligence 10 (New York: Halsted, Wiley, 1982).

70. P. H. Winston and K. A. Prendergast, eds., The A/ Business, pp. 41-49, 92-99.

71. A strong case for the use of computers largely as office environment shapers is in Terry Winograd
and Fernando Flores, Understanding Computers and Cognition: A New Foundation for Design.
72. See Seymour Papert, “One Al or Many?” Daedalus, Winter 1988, p. 7

73. One large-scale effort that takes this problem seriously is described in D. Lenat, M. Shepherd, and
M. Prakash, “CYC: Using Common Sense Knowledge to Overcome Brittleness and Knowledge
Acquisition Bottlenecks,” A/ Magazine, Winter 1986.
74. An enjoyable account of genetics, evolution, and intelligence is in Carl Sagan, The Dragons of Eden.

75. The original reports of Crick and Watson, surprisingly readable, may be found in James A. Peters, j
ed., Classic Papers in Genetics (Englewood Cliffs, N.J.: Prentice-Hall, 1959). An exciting account of the
successes and failures that led to the double helix is given in J. D. Watson, The Double Helix.

76. The structure and behavior of DNA and RNA are described in: Felsenfeld Gary, “DNA,” Scientific 1
American, October 1985. And: James Darnell, “RNA,” Scientific American, October 1985.

77. A fascinating account of the new biology is given in Horace F. Judson, The Eighth Day of Creation.
78. G. L. Stebbins and F. J. Ayala, “The Evolution of Darwinism,” Scientific American, July 1985, p. 73. i

Mechanical Roots |
1. See J. David Bolter, Turing’s Man: Western Culture in the Computer Age, pp. 17-24. Bolter
illustrates the mechanism for astronomical calculation described in detail in Derek J. de Solla Price, “An i
Ancient Greek Computer,” Scientific American, June 1959, pp. 60-67; see also Derek J. de Solla Price, :
Gears from the Greeks: The Antikythera Mechanism—A Calendar Computer from circa 80 B.C. Early i
automata and their relation to Al are discussed in Pamela McCorduck’s popular 1979 history of Al
research, Machines Who Think, chapter 1. Another useful and lively source is John Cohen's Human
Robots in Myth and Science. Perhaps the best detailed sources on automata through the ages are
Derek J. de Solla Price, “Automata and the Origins of Mechanism and Mechanistic Philosophy,”
Technology and Culture 5 (1964): 9-23, and Silvio Bedini, "The Role of Automata in the History of
Technology,” Technology and Culture 5 (1964): 24-42. A classic volume with many illustrations is
Alfred Chapuis and Edmond Droz, Automata: A Historical and Technological Study, trans. Alec Reid.
Otto Mayr describes the significance of automata in European culture in Authority, Liberty, and
Automatic Machinery in Early Modern Europe.

2. For a general history of the mechanical arts see C. Singer, E. J. Holmyard, A. R. Hall, and T. |.
Williams, eds., A History of Technology, and A. P. Usher, A History of Mechanical Inventions, 2nd ed.
Those interested in ancient technologies should consult R. J. Forbes, Studies in Ancient Technology.

3. Price, “Automata and the Origins of Mechanism,” p. 11. Other important works on ancient technolo-
gies are A. G. Drachman, The Mechanical Technology of Greek and Roman Antiquity, A. P. Neuberger,
The Technical Arts and Sciences of the Ancients, and K. D. White, Greek and Roman Technology.

4. Price, “Automata and the Origins of Mechanism,” p. 11. Joseph Needham describes the fascinating
automata in China at the time of the pre-Socratics in his Science and Civilisation in China, vol. 2, pp.
53-54, 516. The Chinese mechanical orchestra, consisting of twelve figures cast in bronze, is also

496
described in Needham’s Science and Civilisation in China., vol. 4, p. 158. Descartes, who was very
interested in automata, described in one of his notebooks how to reproduce the pigeon of Archytas.
See Mayr, Authority, Liberty, and Automatic Machinery, p. 63.

5. For more on androids, see Samuel L. Macey, Clocks and the Cosmos: Time in Western Life and
Thought, see also Carlo M. Cipolla, Clocks and Culture, 1300-1700, and David S. Landes, Revolution in
Time: Clocks and the Making of the Modern World.

6. Bedini describes Torriano’s automaton in “Automata in the History of Technology,” p. 32, where it
appears as figure 5. For more on P. Jacquet-Droz and Ecrivain, see Bedini’s “Automata,” p. 39, and
Macey's Clocks and the Cosmos, pp. 210-211. P. Jacquet-Droz’s son, Henri-Louis, created a mechani-
cal artist that drew flowers and a musician that played a clavecin. He also made a pair of artificial hands
for a general's son, who had lost his own hands in a hunting accident. Henri-Louis’s success in this
venture was praised by the great creator of automata Jacques de Vaucanson (1709-1782). See the
entries for Pierre-Jacquet Droz and Henri-Louis-Jacquet Droz in Nouvelle Biographie Générale, vol. 14
(Paris: Didot, 1868), pp. 812-813. Vaucanson was perhaps best known for his duck automaton, which
ate, drank, chewed, and excreted. See Macey’s Clocks and the Cosmos, p. 210, and Bedini’s
“Automata in the History of Technology,” pp. 36-37, which has a diagram of the duck’s inner
mechanism (figures 11 and 12). Anyone interested in Vaucanson should see Michael Cardy, “Technol-
ogy as Play: The Case of Vaucanson,” Stud. Voltaire 18th Cent. 241 (1986): 109-123. In 1726 Jonathan
Swift described a machine that would automatically write books; see Eric A. Weiss, “Jonathan Swift's
Computing Machine,” Annals of the History of Computing 7 (1985): 164-165.
7. Martin Gardner, “The Abacus: Primitive but Effective Digital Computer,” Scientific American 222
(1970): 124-127; Parry H. Moon, The Abacus: Its History, Its Design, Its Possibilities in the Modern
World; J. M. Pullan, The History of the Abacus (London: Hutchinson, 1968).

8. Napier's bones or rods are described and pictured in Stan Augarten’s Bit by Bit: An Iilustrated History
of Computers, pp. 9-10. A more detailed treatment can be found in M. R. Williams, “From Napier to
Lucas: The Use of Napier's Bones in Calculating Instruments,” Annals of the History of Computing 5
(1983): 279-296
9. An earlier calculating machine was devised by the polymath Wilhelm Shickard (1592-1635).
Shickard’s machine, and Pascal's development of the Pascaline are described in Augarten’s Bit by Bit
An Illustrated History of Computers, pp. 15-30. A more technical account can be found in René
Taton,”Sur |'invention de la machine arithmetique,” Revue a’histoire des sciences et de leurs applica-
tions 16 (1963): 139-160; Jeremy Bernstein, The Analytical Engine: Computers—Past, Present, and
Future, p.40; Herman Goldstine, The Computer from Pascal to von Neumann, p. 7-8.
10. Blaise Pascal, Pensées (New York: E. P. Dutton & Co., 1932), p. 96, no. 340.
11. The Pascaline, of which perhaps ten or fifteen were sold, failed to sell for a variety of reasons. See
Augarten, Bit by Bit, pp. 27-30.
12. The Stepped Reckoner, as Leibniz called his machine, employed a special gear as a mechanical
multiplier. See Augarten, Bit by Bit, pp. 30-35, and Goldstine, The Computer from Pascal to von
Neumann, pp. 7-9. Morland’s career is described in Henry W. Dickinson's biography, Sir Samuel
Morland, Diplomat and Inventor, 1625-1695.

13. Brian Randell, ed., The Origins of Digital Computers: Selected Papers, p. 2
14. Augarten, Bit by Bit, p. 89
15. Babbage’s paper can be found in H. P. Babbage, Babbage’s Calculating Engines, pp. 220-222
16. H. P. Babbage, Babbage’s Calculating Engines, pp. 223-224. On Babbage and the Astronomical
Society, see Anthony Hyman, Charles Babbage: Pioneer of the Computer, pp. 50-53.
17. See chapter 2 of Augarten’s Bit by Bit, which has marvelous illustrations. Babbage’s life and career
are treated in detail in Hyman’s Charles Babbage. Joel Shurkin provides a lively account of Babbage's
work in his Engines of the Mind: A History of the Computer, chapter 2. A biography recently published
almost a century after its completion is H. W. Buxton, Memoirs of the Life and Labours of the Late
Charles Babbage, Esq., F.R.S., ed. A. Hyman.

18. Allen G. Bromley, Introduction to H. P. Babbage, Babbage’s Calculating Engines, pp. xiii-xvi;
Bernstein, The Analytical Engine, pp. 47-57.

19. Augarten, Bit by Bit, pp. 62-63; Bernstein, The Analytical Engine, p. 50; Hyman, Charles Babbage,
p. 166.

20. Augarten, Bit by Bit, pp. 68-64. Babbage describes the features of his machine in “On the
Mathematical Powers of the Calculating Engine,” written in 1837 and reprinted as appendix B in
Hyman’s Charles Babbage.

a9o7
 21. A recent biography is Dorothy Stein, Ada, a Life and a Legacy.

22. Goldstine, The Computer, p. 26

Notes

23. Her translation and notes can be found in H. P. Babbage, Babbage’s Calculating Engines, pp. 1-50.
24. The lonely end of Babbage's life is described in Hyman, Charles Babbage, chapter 16.
25. Joel Shurkin, in Engines of the Mind, p. 104, describes Aiken's machine as “an electromechanical
Analytical Engine with IBM card handling.” For a concise history of the development of the Mark |, see
Augarten’s Bit by Bit, pp. 103-107. |. Bernard Cohen provides a new perspective on Aiken's relation to
Babbage in his article “Babbage and Aiken,” Annals of the History of Computing 10 (1988): 171-193
26. Anyone with a serious interest in the history of calculators should be aware of the following two
classics: D. Baxandall, Calculating Machines and Instruments, and Ellice Martin Horsburgh, ed., Modern
Instruments and Methods of Calculation: A Handbook of the Napier Tercentenary Celebration
Exhibition. Some of the calculators and tabulating machines of the 1940s are described in Charles and
Ray Eames, A Computer Perspective, pp. 128-159. A brief pictorial history of calculating machines can
be found in George C. Chase, “History of Mechanical Computing Machinery,” Annals of the History of
Computing 2 (1980): 198-226. Two important sources in the history of computing, besides the Annals,
are N. Metropolis, J. Howlett, and Gian-Carlo Rota, eds., A History of Computing in the Twentieth
Century, and Brian Randell, The Origins of Digital Computers.
27. See chapter 3 of Augarten’s Bit by Bit, and Eames’s A Computer Perspective, pp. 16-17, 22-30.

28. Augarten, Bit by Bit, pp. 78-83; Randell, Origins, p. 28

29. Shurkin, Engines of the Mind, p. 94; Augarten, Bit by Bit, p. 82; Eames, A Computer Perspective,
p. 39. Burroughs's life and work are described in Molly Gleiser, “William S. Burroughs,” Computer
Decisions, March 1978, pp. 34-36.

30. By 1913 the Burroughs Adding Machine Company had $8 million in sales, according to Augarten’s
Bit by Bit, p. 82
31. The legacy of the census crisis is described in detail in L. E. Truesdell, The Development of Punch
Card Tabulation in the Bureau of the Census, 1890-1940 (Washington D.C.: Government Printing
Office, 1965)

32. See Geoffrey D. Austrian’s biography, Herman Hollerith: Forgotten Giant of Information Processing,
pp. 50-51. Shurkin, in Engines of the Mind, chapter 3, gives a very readable and concise account of
Hollerith and his census work
33. Austrian, Herman Hollerith, pp. 16-17, 51; Augarten, Bit by Bit, p. 75.
34. Austrian, Herman Hollerith, pp. 63-64.
35. Hollerith’s system for the 1890 census is similar to one he described in an 1889 article, “An Electric
Tabulating System,” extracts from which are reprinted in Randell, Origins, pp. 129-139. Also see
Randell’s discussion of Hollerith’s work, pp. 125-126
36. According to Augarten in Bit by Bit, p. 77, the Census Bureau was able to give a preliminary
population total of 62,622,250 just six weeks after all the data arrived in Washington

37. Austrian, Herman Hollerith, p. 153

38. Shurkin, Engines of the Mind, pp. 78-82; Austrian, Herman Hollerith, chapter 13.

39. Austrian, Herman Hollerith, p. 176 ff.

40. See chapters 20 and 21 in Austrian, Herman Hollerith, as well as Shurkin, Engines of the Mind,
p. 86.
41. Austrian, Herman Hollerith, p. 312

42. Shurkin, Engines of the Mind, pp. 91-921; Augarten, Bit by Bit, pp. 177-178; Austrian, Herman
Hollerith, pp. 329. Thomas Watson's career is reviewed in Augarten, Bit by Bit, pp. 168ff.
43. Shurkin, Engines of the Mind, p. 92. See “The Rise of IBM,” chapter 25, in Austrian, Herman
Hollerith, and “The Rise of IBM," chapter 6, in Augarten's Bit by Bit. See also Charles J. Bashe, Lyle R. {
Johnson, John H. Palmer, and Emerson W. Pugh, /BM’s Early Computers.

44. Augarten, Bit by Bit, pp. 217-223. Shurkin examines the relations between IBM and its competitors
in Engines of the Mind, pp. 260-279
45. Aiken is quoted in Bernstein, The Analytical Engine, p. 62
46. Bernstein, The Analytical Engine, p. 73

aos
Electronic Roots
1. The writings of these early thinkers are particularly insightful regarding what it means to compute.
Some representative works are H. P. Babbage, “Babbage’s Analytical Engine.” Monthly Note of the
Royal Astronomical Society 70 (1910): 517-526, 645. George Boole, An Investigation of the Laws of
Thought on Which Are Founded the Mathematical Theories of Logic and Probabilities (Peru, \l .: Open
Court Publishing Co., 1952); Bertrand Russell, Principles of Mathematics, 2nd ed.; and H. Hollerith,
“The Electric Tabulating Machine,” Journal of the Royal Statistical Society 57, no. 4 (1894): 678-682.
For-a.detailed account of Burroughs's contributions, scientific and commercial, see B. Morgan, Total to
Date: The Evolution of the Adding Machine.
2. Zuse’s claim is supported by the patent applications he filed. See, for instance, K. Zuse, “Verfahren
zur Selbst Atigen Durchfurung von Rechnungen mit Hilfe von Rechenmaschinen,” German Patent
Application 223624, April 11, 1936. Translated extracts, titled "Methods for Automatic Execution of
Calculations with the Aid of Computers,” appear in Brian Randell, The Origins of Digital Computers, pp.
159-166

3. From an interview with Computerworld magazine. Published in The History of Computing in 1981 by
CW Communications, Framingham, Mass. The magazine's interviewers were enterprising enough to
locate Zuse in Hunfeld, Germany, (where he now lives) and produce an engaging interview.
4. Jan Lukasiewicz developed two related notations, each intended to ease certain aspects of repre-
sentation and computation in mathematical logic. See Donald Knuth, The Art of Computer
Programming, volume 1, Fundamental Algorithms, 2nd edition (Reading, Mass.: Addison-Wesley,
1973), p. 336.
5. A three page description of a special purpose electromechanical computer used to process flying-
bomb wing data is given in K. Zuse, “Rechengerate fiir Flugelvermessung,” private memorandum,
September 10, 1969.
6. The charge is strongly made by Rex Malik in And Tomorrow . . . the World (London: Millington,
1975)
7. Paul Ceruzzi's 1980 doctoral dissertation gives us a most detailed account of Zuse’s contributions to
computer technology and places them in the proper context. Paul E. Ceruzzi, “The Prehistory of the
Digital Computer, 1935-1945: A Cross-Cultural Study.” Texas Tech University, 1980.
8. Zuse’s own statement on his life and his computers (with many details of construction) appears in
Konrad Zuse, Der Computer—Mein Lebenswerk (Berlin: Verlag Moderne Industrie, 1970). More recent
reminiscences appear in Konrad Zuse, “Some Remarks on the History of Computers in Germany,” in
N. Metropolis, J. Howlett, and G. C Rota, eds., A History of Computing in the Twentieth Century, pp.
611-628.
9. John E. Savage, Susan Magidson, and Alex M. Stein, The Mystical Machine, pp. 25-26
10. See Andrew Hodges, Alan Turing: The Enigma. Hodges’s biography, now a standard reference on
Turing's life, gives an original account of Turing’s war-time computers.

11. For an engineering account of the Colossus project, see B. Randell, “The Colossus,” reprinted in N
Metropolis, J. Howlett, and G. C Rota, eds., A History of Computing in the Twentieth Century.

12. An excellent set of brief biographies of computer pioneers, including one of Aiken, may be found in
Robert Slater, Portraits in Silicon.
13. See Andrew Hodges, Alan Turing: The Enigma
14. See Cuthbert Hurd, “Computer Development at IBM,” in N. Metropolis, J. Howlett, and G. C. Rota,
eds., A History of Computing in the Twentieth Century, pp. 389-418. |BM's role in the development of
these early computers is covered in Charles Bashe et al., /BM’s Early Computers. This detailed book is
successful in showing how exhausting an intellectual and physical effort it was to construct computers

15. The History of Computing (Framingham, Mass.: CW Communications, 1981), p. 52

16. Grace Hopper is brought out as a strong, dedicated, and inspiring intellectual in Slater's biography in
Robert Slater, Portraits in Silicon.

17. John E. Savage, Susan Magidson, and Alex M. Stein, The Mystical Machine, p. 30.
18. For a brief overview of the principles and construction of ENIAC and the lessons learned in the
words of the designers themselves, see J. Presper Eckert, “The ENIAC,” and John W. Mauchly, “The
ENIAC.” Both pieces appear in N. Metropolis, J. Howlett, and G. C Rota, eds., A History of Computing
in the Twentieth Century, pp. 525-540, 541-550

asa
19. The court case brought out thousand of pages of material on early computers, valuable to the
computer historian. Judge Larson’s findings are recorded in E. R. Larson, “Findings of Fact, Conclusion
of Law, and Order for Judgement,” File no. 4-67, Civ. 138, Honeywell Inc. vs. Sperry Rand Corp. and
Illinois Scientific Development, Inc., U.S. District Court, District of Minnesota, Fourth Division,
October 19, 1973.

20. A description of the machine and its applications is given in J. V. Atanasoff, “Computing Machine
for the Solution of Large Systems of Linear Algebraic Equations,” Ames, lowa: lowa State College,
1940. Reprinted in Brian Randell, ed., The Origins of Digital Computers: Selected Papers (Berlin:
Springer-Verlag, 1973), pp. 305-325.
21. The concept of a stored program has proved to be one of the most robust in computer science. For
a history of its development and implementation, and also for a clear analysis of the ENIAC experience,
see Arthur Burks, “From ENIAC to the Stored Program: Two Revolutions in Computers,” in N.
Metropolis, J. Howlett, G. C Rota, eds., A History of Computing in the Twentieth Century, pp. 311-344.
22. For a lucid explanation of the stored-program idea, see John E. Savage, Susan Magidson, and Alex
M. Stein, The Mystical Machine, pp. 31-32, 58-62
23. The excitement of these developments is skillfully captured in Wilkes's autobiography: Maurice
Wilkes, Memoirs ofa Computer Pioneer (Cambridge: MIT Press, 1981).

24. For the role of research and development in the rise of IBM, see Charles Bashe et al., /BM’s Early
Computers.
25. Alan M. Turing, “Computing Machinery and Intelligence,” Mind 59 (1950): 433-460.

26. Pamela McCorduck, Machines Who Think (San Francisco: W. H. Freeman, 1979), pp. 93-102.

27. Von Neumann stressed the differences between the nervous system and the computer in “The
General and Logical Theory of Automata,” in L. A. Jeffress, ed., Cerebral Mechanisms in Behavior (New
York: John Wiley & Sons, 1951). He fails to see how these two can be made to be functionally
equivalent
28. A book was published posthumously, however: J. von Neumann, The Computer and the Brain (New
Haven: Yale University Press, 1958).
29. Norbert Wiener, Cybernetics (Cambridge: MIT Press, 1943)
30. Wiener is a delightful writer: the best biographies of him are perhaps his own. See Norbert Wiener,
Ex-Prodigy (Cambridge: MIT Press, 1963) and Norbert Wiener, | Am a Mathematician (Boston:
Houghton-Mifflin, 1964).

31. Wiener liked to believe that the medium underlying life was not energy but information. For an -
account of how this motivated many of Wiener's projects, see the excellent biography Steve Heims, :
John von Neumann and Norbert Wiener: From Mathematics to the Technologies of Life and Death
(Cambridge: MIT Press, 1980)
32. Many of Fredkin's results come from studying his own model of computation, which explicitly
reflects a number of fundamental principles of physics. See the classic Edward Fredkin and Tommaso
Toffoli, "Conservative Logic,” International Journal of Theoretical Physics 21, nos. 3-4 (1982).

33. A set of concerns about the physics of computation analytically similar to those of Fredkin’s may be
found in Norman Margolus, “Physics and Computation,” Ph.D. thesis, MIT.

34. In his provocative book The Coming of Postindustrial Society, Harvard sociologist Daniel Bell
introduces the idea that the codification of knowledge is becoming central to society. In The Fifth
Generation (Reading, Mass.: Addison-Wesley, 1983), Edward Feigenbaum and Pamela McCorduck
discuss the impending reality of such a society.
35. See Norbert Wiener, Cybernetics.
36. The Differential Analyzer and other such analog computing machines are described in chapter 5 of
Michael Williams, A History of Computing Technology (Englewood Cliffs, N.J.: Prentice-Hall, 1985).
Bush's own account of the computer is presented in “The Differential Analyzer,” Journal of the Franklin
Institute 212, no. 4 (1936): 447-488.
37. The drawbacks of analog computers are considered in chapter 5 of Michael Williams, A History of
Computing Technology (Englewood Cliffs, N.J.: Prentice-Hall, 1985).

38. See Norbert Wiener, Cybernetics.

39. Such trends are fast paced. See Tom Forester, High Tech Society (Cambridge: MIT Press, 1987).
40. A clear account of the technology behind the compact disk appears in John J. Simon, rom Sand
to Circuits’ and other Enquiries,” Harvard University Office of Information Technology, 1986.

500
41. See John J. Simon, “From Sand to Circuits’ and Other Enquiries,” Harvard University Office of
Information Technology, 1986. 6
42. The structure of the transistor is explained in Stephen Senturia and Bruce Wedlock, Electronic
Circuits and Applications (New York: McGraw-Hill, 1983).
43. See Claude Shannon and Warren Weaver, The Mathematical Theory of Communication (Urbana, I I
University of Illinois Press, 1964)
44-\n recent years very interesting work has been done to find out how the brain processes perceptual
inputs. For a sample of current thinking in the area, see Ellen Hildreth and Christof Koch, “The Analysis
of Visual Motion: From Computational Theory to Neural Mechanisms,” MIT Artificial Intelligence
Laboratory, Al memo no. 919, 1986.
45. A clear and technically accurate piece on the revolution in music brought about by the representa-
tion of music in digital forms is presented in Understanding Computers: Input, Output, (Alexandria, Va.
Time-Life Books, 1986)
46. Haugeland clarifies many issues by attempting to formalize our intuitions. What does it mean when
we say that the mind is a computer? asks Haugeland in Artificial Intelligence: The Very Idea.
47. Alan Turing, “Computing Machinery and Intelligence," Mind 59 (1950): 433-460 (reprinted in E.
Feigenbaum and J. Feldman, Computers and Thought). Norbert Wiener, Cybernetics, or Control and
Communication in the Animal and Machine. Warren McCulloch and Walter Pitts, “A Logical Calculus of
the Ideas Immanent in Logical Activity,” Bulletin of Mathematical Biophysics, 5 (1943): 115-137.
Claude Shannon, “Programming a Digital Computer for Playing Chess,” Philosophy Magazine 41
(1950): 356-375. A related paper, more amenable to the layperson, is “Automatic Chess Player,”
Scientific American, October 1950, p. 48.
48. See A. Newell, J. C. Shaw, and H. A. Simon, “Programming the Logic Theory Machine,” Proceed-
ings of the Western Joint Computer Conference, 1957, pp. 230-240.
49. See A. Newell, J. C. Shaw, and H. A. Simon, “Empirical Explorations of the Logic Theory Machine,”
Proceedings of the Western Joint Computer Conference, 1957, pp. 218-239.
50. The broad techniques of the Logic Theory Machine were generalized in GPS. This is described in A
Newell, J. C. Shaw, and H. A. Simon, “Report on a General Problem-Solving Program,” reprinted in E
Feigenbaum and J. Feldman, eds., Computers and Thought. Newell and Simon continued their studies
and summarized their results in Human Problem Solving, which placed less emphasis on the actual
computer implementation of their ideas.

51. A. Newell and H. A. Simon, “Heuristic Problem Solving: The Next Advance in Operations
Research,” Journal of the Operations Research Society of America 6, no. 1 (1958), reprinted in Herbert
Simon, Models of Bounded Rationality, vol. 1, Economic Analysis and Public Policy (Cambridge: MIT
Press, 1982).
52. Notably the Dreyfus brothers. See Hubert Dreyfus, What Computers Can’t Do, 2nd ed.
53. Indeed, the prediction about chess has not yet come true. The Fredkin Prize will go to the first
computer to become world chess champion. Samuel's checker program was not written specifically as
a game-playing program but as an exercise in machine learning. See Arthur L. Samuel, “Some Studies
in Machine Learning Using the Game of Checkers,” reprinted in E. A. Feigenbaum and J. Feldman,
eds., Computers and Thought, pp. 279-293.
54. McCorduck’s delightful book on the history of artificial intelligence, Machines Who Think, contains a
chapter on the now famous Dartmouth Conference.
55. See Edward Feigenbaum’s short reflection on twenty-five years of artificial intelligence: “AAAI
President's Message,” A/ Magazine, Winter 1980-1981
56. The version most referred to is Marvin Minsky, “Steps toward Artificial Intelligence,” in E. A
Feigenbaum and J. Feldman, eds., Computers and Thought, pp. 406-450.
57. LISP was originally introduced in a set of memos at the MIT Artificial Intelligence Laboratory. Much
of this found its way into formal publications. See John McCarthy, “Recursive Functions of Symbolic
Expressions and Their Computation by Machine, Part |,” Communications of the ACM 3, no. 4 (1960)
The language soon became popular enough for McCarthy to publish a manual: Jonn McCarthy, P. W.
Abrahams, D. J. Edwards, T. P. Hart, and M. |. Levin, LISP 1.5 Programmer's Manual (Cambridge: MIT
Press, 1962). See Pamela McCorduck, Machines Who Think, pp. 97-102.

58. Daniel Bobrow, “Natural Language Input fora Computer Problem Solving System," in Marvin
Minsky, Semantic Information Processing, pp. 146-226
59. Thomas Evans, “A Program for the Solution of Geometric-Analogy Intelligence Test Questions,” in
Marvin Minsky, Semantic Information Processing, pp. 271-353.
60. This work is described in R. Greenblatt, D. Eastlake, and S. Crocker, “The Greenblatt Chess
Program,” MIT Artificial Intelligence Laboratory, Al memo 174, 1968. The program defeated Hubert
Dreyfus, who once strongly doubted that a chess program could match even an amateur human player.

61. The lessons of DENDRAL are recorded and analyzed in Robert Lindsay, Bruce Buchanan, Edward
Feigenbaum, and Joshua Lederberg, Applications of Artificial Intelligence for Chemical Inference: The
DENDRAL Project (New York: McGraw-Hill, 1980). A brief and clear explanation of the essential
mechanisms behind DENDRAL is given in Patrick Winston, Artificial Intelligence (1984), pp. 163-164,
195-197,

62. Much has been written about ELIZA, but the clearest account on how ELIZA works is from
Weizenbaum himself: “ELIZA—A Computer Program for the Study of Natural Language Communica-
tion between Man and Machine,” Communications of the ACM 9 (1966): 36-45. ELIZA has, of course,
attracted numerous criticisms, many of which were first voiced by Weizenbaum himself. See Hubert
Dreyfus, What Computers Can't Do.

63. For many years SHRDLU was cited as a prominent accomplishment of artificial intelligence.
Winograd’s thesis has been published in book form: Understanding Natural Language (New York:
Academic Press, 1972). A brief version appears as “A Procedural Model of Thought and Language,” in
Roger Schank and Kenneth Colby, eds., Computer Models of Thought and Language (San Francisco: W.
H. Freeman, 1973).

64. Minsky and Papert point out that these toy examples offer many important abstractions for further
analysis. See Marvin Minsky and Seymour Papert, “Artificial Intelligence Progress Report,” MIT
Artificial Intelligence Laboratory, Al memo 252, 1973.

65. Warren McCulloch and Walter Pitts, “A Logical Calculus of the Ideas Immanent in Logical Activity,”
Bulletin of Mathematical Biophysics 6 (1943): 115-137. Rosenblatt’s classic work is Principles of
Neurodynamics (New York: Spartan Books, 1962)
66. Minsky and Papert trace much of this controversy and history, with technical details, in the prologue
and epilogue of the revised edition of their book, published in 1988.

67. This trend is explained and praised in Edward Feigenbaum, “The Art of Artificial Intelligence: |
Themes and Case Studies in Knowledge Engineering,” Fifth International Joint Conference on Artificial
Intelligence, 1977
68. The approach was compelling in light of what it could do. See the papers in E. Feigenbaum and J.
Feldman, Computer and Thought.
69. Knowledge representation was and continues to be an important area of artificial intelligence

—
research. See R. Brachman and H. Levesque, eds., Readings in Knowledge Representation (Los Altos,
Calif.: Morgan Kaufman, 1986)

70. The restaurant scene is a popular example of scripts as a means of representing knowledge. Scripts
are brought out as a powerful scheme for reasoning in R. Schank and R. Abelson, Scripts, Plans, Goals,
and Understanding (Hillsdale, N.J.: Lawrence Erlbaum Associates, 1977).

71. Minsky’s work on frames is one of the most cited in Al. The most complete written form of the
theory is Marvin Minsky, "A Framework for Representing Knowledge,” MIT Artificial Intelligence
Laboratory, Al memo 304, 1974.
72. See R. Schank and R. Abelson, Scripts, Plans, Goals, and Understanding (Hillsdale, N.J.: Lawrence
Erlbaum Associates, 1977).

73. An excellent introduction to the technology and applications of expert systems is F. Hayes-Roth, D. A.
Waterman, and D. B. Lenat, eds., Building Expert Systems (Reading, Mass.: Addison-Wesley, 1983).
74. Some famous expert systems are described by the creators themselves in F. Hayes-Roth, D. A.
Waterman, and D. B. Lenat, eds., Building Expert Systems (Reading, Mass.: Addison-Wesley, 1983).
75. See Edward Feigenbaum and Pamela McCorduck, The Fifth Generation

76. Artificial intelligence is beginning to have an important effect on the productivity of many organiza-
tions. This phenomena is explored in Edward Feigenbaum, Pamela McCorduck, and Penny Nii, The Rise
of the Expert Company (Reading, Mass.: Addison-Wesley, 1989)

Pattern Recognition: The Search for Order

1. An excellent treatment of the role of imagery and “holistic” representations in cognition may be
found in Ned Block, ed., Imagery (Cambridge: MIT Press, 1981)
2. See Newell and Simon’s analysis of human chess playing in Allen Newell and Herbert Simon, Human
Problem Solving (Englewood Cliffs, N.J.: Prentice-Hall, 1972).

502
3. An essay of special relevance to the discussion here is Zenon Pylyshyn, “Imagery and Artificial
Intelligence,” in C. W. Savage, ed., Perception and Cognition: Issues in the Foundations of Psychology,
Minnesota Studies in the Philosophy of Science, vol. 9 (Minneapolis: University of Minnesota Press,
1978).

4. Imagination is a skill that we develop with age. Piaget's experiments show that to the infant (up toa
certain age), an object that is not visible does not exist. See J. Piaget, Play, Dreams, and Imitation in
Childhood (New York: W. W. Norton, 1951).

5. This technique is simple but surprisingly powerful and has been used extensively in Al programs
See Patrick H. Winston, Artificial intelligence, pp. 159-167.
6. A clear introduction to the essential problems and procedures in machine vision appears in chapter
10 of the classic textbook Patrick H. Winston, Artificial Intelligence.
7. This and other techniques for identifying edges are reviewed in L. Davis, “A Survey of Edge
Detection Techniques,” Computer Graphics and Image Processing 4 (1975): 248-270. A more detailed
review appears in Azriel Rosenfeld and Avinash Kak, Digital Picture Processing (New York: Academic
Press, 1976). A more recent summary of results, including John Canny’s work, is Ellen Hildreth, “Edge
Detection,” MIT Artificial Intelligence Laboratory, Al memo 858, 1985.

8. The use of zero crossings in stereo to isolate edges was introduced in David Marr and Tomaso
Poggio, “A Theory of Human Stereo Vision,” Proceedings of the Royal Society of London 204 (1979).
The use of zero crossings was also addressed in Ellen Hildreth’s work: “The Detection of Intensity
Changes by Computer and Biological Vision Systems,” Computer Vision, Graphics, and Image
Processing 23 (1979). For efficiencies more recently incorporated, see John Canny, “Finding Edges and
Lines in Images,” MIT Artificial Intelligence Laboratory, technical report 720, 1983
9. False hypothesis may be corrected also by some of the techniques detailed in L. S. Davis, “A Survey
of Edge Detection Techniques,” Computer Graphics and Image Processing 4 (1975): 248-270. Also see
Ellen Hildreth, “Edge Detection,” MIT Artificial Intelligence Laboratory, Al memo 858, 1985.

10. Hubel and Wiesel are responsible for many important aspects of our knowledge today about the
biological mechanisms for vision. They conducted many imaginative experiments to reveal the structure
and functional decomposition of the cortex. Notable is their discovery of the presence of edge-
detection neurons. See D. H. Hubei and T. N. Wiesel, “Functional Architecture of Macaque Monkey
Visual Cortex,” Journal of Physiology 195 (1968): 215-242. A truly fascinating book written for the
layperson as an introduction to the brain's vision processing is David Hubel, Eye, Brain, and Vision.
11. For details of the computational aspects of recovering details of surfaces from images by means of
sombrero filtering and other related techniques, see W. Eric L. Grimson, From Images to Surfaces
(Cambridge: MIT Press, 1981)
12. An illuminating article on the eye’s computational capacities for image processing is Tomaso
Poggio, “Vision by Man and Machine,” Scientific American, April 1984
13. See Tomaso Poggio, “Vision by Man and Machine,” Scientific American, April 1984.
14. David Marr is brilliant at fusing studies from biology and machine vision. His highly influential
classic, published posthumously, is Vision. A paper that excellently summarizes and demonstrates his
computational approach to vision is D. Marr and H. K. Nishihara, “Visual Information Processing:
Artificial Intelligence and the Sensorium of Sight,” Technology Review, October 1978.
15. See Tomaso Poggio, “Vision by Man and Machine” Scientific American, April 1984
16. The geometry of stereopsis and stereo vision is discussed well in S. T. Barnard and M. A. Fischler,
“Computational Stereo from an |U Perspective," Proceedings of the Image Understanding Workshop,
1981

17. Edges introduce constraints that greatly reduce the number of ways two images can be fused.
Without such preprocessing, matching would be be extremely difficult. Consider, for example, the
computational complexity of fusing random-dot stereograms. See David Marr, Vision, p. 9.

18. For the details of these techniques, see R. O. Duda and P. E. Hart, Pattern Classification and Scene
Analysis (New York: Wiley, 1973).

19. See W. Eric L. Grimson, From Images to Surfaces. Object recognition and labeling is a hard
problem. For the role of knowledge and preconceived models in this process, see Rodney Brooks.
“Model-Based Three-Dimensional Interpretation of Two-Dimensional Images,” Proceedings of the
Seventh International Joint Conference on Artificial Intelligence, 1981. Generalized cylinders are
frequently used as intermediate representations of objects. See D. Marr and H. K. Nishihara, “Visual
Information Processing: Artificial Intelligence and the Sensorium of Sight,” Technology Review.
October 1978.
20. The parallel nature of the computational processes constituting early vision is examined in an
excellent review article: Tomaso Poggio, Vincent Torre, and Christof Koch, “Computational Vision and
Regularization Theory,” Nature, September 26, 1985. The role of analog computations is also discussed
there.
21. For some lessons that evolution offers for strategies in artificial intelligence, see Rodney Brooks,
“Intelligence without Representation,” Artificial Intelligence, 1989.
22. This point is brought out with particular elegance in Dana Ballard and Christopher Brown, “Vision:
Biology Challenges Technology,” BYTE, April 1985.
23. The structure of the Connection Machine is excellently described, along with some machine vision
applications, in W. Daniel Hillis, “The Connection Machine,” Scientific American, June 1987

24. The role of analog computations in vision is discussed in Tomaso Poggio, Vincent Torre, and
Christof Koch, “Computational Vision and Regularization Theory,” Nature, September 26, 1985.

25. Neural networks and related mechanisms have been applied fairly successfully in vision problems.
For work in early vision, see D. H. Ballard, “Parameter Nets: Toward a Theory of Low-Level Vision,”
Artificial Intelligence Journal 22 (1984): 235-267. For higher-level processes, see D. Sabbah, “Comput-
ing with Connections in Visual Recognition of Origami Objects,” Cognitive Science 9 (1985): 25-50.
26. For instance, early neural networks failed to determine connectedness of drawings. The ability of
more complex neural nets to determine connectedness remains controversial. See Marvin Minsky and
Seymour Papert, Perceptrons, pp. 136-150.

27. The manifesto of the new connectionists is Parallel Distributed Processing, vols. 1 and 2, by David
Rumelhart, James McClelland, and the PDP Research Group. Chapter 2 of this book describes the new
neural-net structures.
28. Marvin Minsky and Seymour Papert, Perceptrons, revised ed., p. vii
29. Distributed systems, whose mechanisms and memory are stored not centrally but over a large
space, are less prone to catastrophic degradation. Neural networks are not only parallel but also
distributed systems. See chapter 1 of D. E. Rumelhart, J. L. McClelland, and the PDP Research Group,
Parallel Distributed Processing: Explorations in the Microstructure of Cognition, vol. 1
30. An excellent study of skill acquisition with some implications for parallel distributed processing is
D. E. Rumelhart and D. A. Norman, “Simulating a Skilled Typist: A Study of Skilled Cognitive-Motor
Performance,” Institute for Cognitive Science, technical report 8102, University of California, San
Diego, 1981
31. This objection is articulated in Hubert Dreyfus and Stuart Dreyfus, Mind Over Machine: The Power
of Human Intuition and Expertise in the Era of the Computer (New York: The Free Press, 1986), pp
101-121 }

32. Daniel Hillis points out that for many physical systems that are inherently parallel, fluid flow, for :
example, it is simply not convenient
to think in terms of sequential processes. Similarly, logic turns out
to be inconvenient for the analysis of, say, early-vision processes. See W. Daniel Hillis, “The Connec-
tion Machine,” Scientific American, June 1987.
33. This multilevel, multiparadigm approach is followed in the society theory of the mind. See Marvin
Minsky, The Society of Mind. i
34. Higher-level descriptions have a smaller volume of information, but they incorporate a larger
number of constraints and require more extensive knowledge about the physical world. See chapter 1
in David Marr, Vision.

35. See the appendix of Marvin Minsky, The Society of Mind.

36. For recent developments in the design and fabrication of chips, see J. D. Meindl, “Chips for
Advanced Computing,” Scientific American, October 1987.
37. David Marr and Tomaso Poggio, “Cooperative Computation of Stereo Disparity,” Science 194
(1976): 283-287.

38. See David Marr and Tomaso Poggio, “From Understanding Computation to Understanding Neural
Circuitry,” Proceedings of the Royal Society of London, 1977, pp. 470-488.

39. Daniel Hillis's thesis suggests areas where parallelism ought to be exploited. See W. Daniel Hillis,
The Connection Machine (Cambridge: MIT Press, 1985).

40. David Marr is responsible for these important representations for vision processing. All three are
clearly explained in D. Marr and H. K. Nishihara, “Visual Information Processing: Artificial Intelligence
and the Sensorium of Sight," Technology Review, October 1978.

soa
 41. Segmentation was one of the chief concerns in the construction of the Hearsay speech-recognition
system. The problem was resolved in part by using multiple knowledge sources and multiple experts
See L. Erman, F. Hayes-Roth, V. Lesser, and D. Raj Reddy, “The HEARSAY-II Speech Understanding
System: Integrating Knowledge to Resolve Uncertainty,” Computing Surveys 12, no. 2 (1980):
213-253.
42. See L. Erman, F. Hayes-Roth, V. Lesser, and D. Raj Reddy, “The HEARSAY-II Speech Understand-
ing System: Integrating Knowledge to Resolve Uncertainty,” Computing Surveys 12, no. 2 (1980):
213-253.
43. The early years of artificial intelligence saw a lot of work on character recognition. But researchers
could not perform extensive experiments on their programs because of a lack of computer power. See
W. W. Bledsoe and |. Browning, “Pattern Recognition and Reading by Machine,” Proceedings of the
Eastern Joint Computer Conference, 1959. A more general article is Oliver Selfridge and U. Neisser,
“Pattern Recognition by Machine,” Scientific American, March 1960, 60-68.

44. The Hearsay system has an interesting implementation of such a manager. See L. Erman, F. Hayes-
Roth, V. Lesser, and D. Raj Reddy, “The HEARSAY-I| Speech Understanding System: Integrating
Knowledge to Resolve Uncertainty,” Computing Surveys 12, no. 2 (1980): 213-253.

45. For some interesting points on the use of multiple experts, see Douglas Lenat, “Computer
Software for Intelligent Systems,” Scientific American, September 1984.
46. For basic techniques for template matching, see R. O. Duda and P. E. Hart, Pattern Ciassification
and Scene Analysis (New York: Wiley, 1973).

47. A playful treatment of the nature of fonts and type styles is presented in chapter 13 of Douglas
Hofstadter, Metamagical Themas.
48. This set of paradigms is successfully applied in the Hearsay system in the context of speech
recognition. See L. Erman, F. Hayes-Roth, V. Lesser, and D. Raj Reddy, “The HEARSAY-II Speech
Understanding System: Integrating Knowledge to Resolve Uncertainty,” Computing Surveys 12, no. 2
(1980): 213-253.

49. For details of the project, see Tomaso Poggio and staff, “MIT Progress in Understanding Images,”
Proceedings of the Image Understanding Workshop (Cambridge, Mass., 1988), pp. 1-16
50. The project tries to incorporate what we know about the nature of vision computation in the brain,
an issue treated in Tomaso Poggio, Vincent Torre, and Christof Koch, “Computational Vision and
Regularization Theory,” Nature, September 26, 1985.

51. See T. Poggio, J. Little, et al., “The MIT Vision Machine,” Proceedings of the Image Understanding
Workshop (Cambridge, Mass., 1988), pp. 177-198.

52. For related work, see Anya Hurlbert and Tomaso Poggio, “Making Machines (and Artificial
Intelligence),” Daedalus, Winter 1988.

53. The Terregator and some other projects of the robotics group at Carnegie-Mellon University are
described in Eric Lerner, “Robotics: The Birth of a New Vision,” Science Digest, July 1985.

54. An informative article on Carver Mead and his specialized chips for vision processing is: Andrew
Pollack, “Chips that Emulate the Function of the Retina,” New York Times, August 26, 1987, p. D6.

55. Harry Newquist, ed., Al Trends ‘87: A Comprehensive Annual Report on the Artificial Intelligence
Industry (Scottsdale, Ariz.: DM Data, 1987).

56. See “Technology Aiding in Fingerprint Identification, U.S. Reports,” New York Times, May 4, 1987,
p. A20.

57. For a description of these new products, see Wesley Iversen, “Fingerprint Reader Restricts Access
to Terminals and PCs,” Electronics, June 11, 1987, p. 104.

58. A very good review paper on Al vision systems and their industrial applications is Michael Brady,
“Intelligent Vision,” in W. Eric Grimson and Ramesh Patil, eds., A/ in the 1980s and Beyond.

59. Harry Newauist, ed., A/ Trends ‘87: A Comprehensive Annual Report on the Artificial Intelligence
Industry (Scottsdale, Ariz.: DM Data, 1987).

60. A substantive article on the role of intelligent systems in modern warfare is J. Franklin, Laura Davis,
Randall Shumaker, and Paul Morawski, “Military Applications,” in Stuart Shapiro ed., Encyclopedia of
Artificial Intelligence, vol. 1 (New York: John Wiley & Sons, 1987).

61. The intelligence of remotely piloted aircraft offers great possibilities, as one can see from Peter
Gwynne, “Remotely Piloted Vehicles Join the Service,” High Technology, January 1987, pp. 38-43

sos
pages

62. Expert systems, pattern recognition and other kinds of medical-information systems will become
increasingly utilized in medicine. See Glenn Rennels and Edward Shortliffe, “Advanced Computing for
Medicine,” Scientific American, October 1987. .
63. A beautifully illustrated article on medical imaging technology is Howard Sochurek, “Medicine's
New Vision,” National Geographic, January 1987, pp. 2-41

64. Schwartz's proposal created excitement in the art world. Her research is described in Lillian
Schwartz, “Leonardo's Mona Lisa,” Arts and Antiques, January 1987. A briefer, more technical
description appears in the following book on computers and art: Cynthia Goodman, Digital Visions (New
York: H. N. Abrams, 1987), pp. 41-43.
65. To J. B. Watson, the founder of behaviorism in America, thinking was like talking to oneself. He
attached great importance to the small movements of the tongue and larynx when one is thinking. See
J.B. Watson, Behaviorism (New York: Norton, 1925).

66. Viewers of the film My Fair Lady will recall that the anatomy of speech production is an important
topic for phoneticians. See M. Kenstowicz and C. Kissebereth, Generative Phonology: Description and
Theory (New York: Academic Press, 1979) and P. Ladefoged, A Course in Phonetics, 2nd ed. (New
York: Harcourt Brace Jovanovich, 1982).

67. The distribution of sound is particular to each language. An important study on English is N.
Chomsky and M. Halle, The Sound Pattern of English (New York: Harper & Row, 1968).

68. Some problems and procedures for early auditory processing are presented in S. Seneff, “Pitch and
Spectral Analysis of Speech Based on an Auditory Perspective,” Ph.D. thesis, MIT Dept. of Electrical
Engineering, 1985.
69. This issue is covered in J. S. Perkell and D. H. Klatt., eds., Variability and Invariance in Speech
Processes (Hillsdale, N.J.: Erlbaum, Lawrence Associates, 1985)

70. H. Sakoe and S. Chita, “A Dynamic-Programming Approach to Continuous Speech Recognition,”

Proceedings of the International Congress of Acoustics, Budapest, Hungary, 1971, pp. 206-213.
71. This approach is faithfully followed in the construction of the Hearsay speech-recognition system.
See L. Erman, F. Hayes-Roth, V. Lesser, and D. Raj Reddy, “The HEARSAY-II Speech Understanding
System: Integrating Knowledge to Resolve Uncertainty,” Computing Surveys 12, no. 2 (1980)
213-253
72. A comprehensive review of ASR is Victor Zue, “Automated Speech Recognition,” in W. Eric L.
Grimson and Ramesh Patil, eds., A/ in the 1980s and Beyond.

73. See Harry Newauist, ed., Al Trends ‘87: A Comprehensive Annual Report on the Artificial Intelli-
gence Industry (Scottsdale, Ariz.: DM Data, 1987)

74. See the fascinating cover stories on computers and music in the June 1986 issue of BYTE.

The Search for Knowledge

1. These and other intriguing aspects of memory are discussed in chapter 8 in Marvin Minsky, Society
of Mind.
2. The most complete written form of the frame theory is Marvin Minsky, “A Framework for Repre-
senting Knowledge,” MIT Artificial Intelligence Laboratory, Al memo 306. Other, less technical versions
have appeared since. See Marvin Minsky, “A Framework for Representing Knowledge,” in John
Haugeland, ed., Mind Design
3. A brief description of the classification systems currently followed is given in Classification:A
Beginner's Guide to Some of the Systems of Biological Classification in Use Today, British Museum
(Natural History), London, 1983

4. The successes and limitations of these systems are discussed in the excellent book Lynn Margulus
and Karlene Schwartz, Five Kingdoms: An Illustrated Guide to the Phyla of Life on Earth, 2nd ed. (New
York: W. H. Freeman, 1988).

5. Dewey first published his classic work anonymously under the title, “A Classification and Subject
Title.” Many editions have appeared since, because the Dewey system has grown to meet every
challenge of the world’s libraries. See Melvil Dewey, Dewey Decimal Classification and Relative Index:
Devised by Melvil Dewey, 19th ed., edited under the direction of Benjamin Custer (Albany, N.Y.: Forest
Press, 1979)

6. Ross Quillian is generally credited with developing semantic networks as a knowledge representa-
tion for Al systems. Although he introduced this representation as early as 1963, the standard
reference for his work in this area is M. Ross Quillian, “Semantic Memory,” in Marvin Minsky,
Semantic Information Processing (1968).
7. See Patrick H. Winston, “Learning Structural Descriptions from Examples,” in Patrick H. Winston,
The Psychology of Computer Vision (New York: McGraw-Hill, 1975).

8. Some of the psychological realities behind semantic networks are discussed in M. Ross Quillian,
“Semantic Memory,” in Marvin Minsky, Semantic Information Processing.
9. Some interesting explanations of cognitive dissonance are given in Henry Gleitman, Psychology, 2nd
ed. (New York: W. W. Norton & Co., 1986), pp. 374-376.

10: An excellent book on the influence of media on political thinking is Edwin Diamond and Stephen
Bates, The Spot: The Rise of Political Advertising on Television (Cambridge: MIT Press, 1984)

11. A revealing book on the psychological aspects of advertising today is William Meyers, The Image
Makers: Power and Persuasion on Medison Avenue (New York: Times Books, 1984).

12. Much research has been done in recent years on the mechanisms for computation and memory in
the human brain, particularly since any new knowledge could contribute significantly to the debate on
connectionism. An introductory account is in Paul M. Churchland, Matter and Consciousness, revised
edition.

13. This last point is strongly brought out by Roger Schank and Peter Childers in The Creative Attitude
(New York: Macmillan, 1988).
14. That computers can never be creative has long been an argument against the possibility of artificial
intelligence. A short rebuttal and an examination of what it means to be creative appears as part of
Marvin Minsky, “Why People Think Computers Can't" A/ Magazine
3, no. 4 (Fall 1982)

15. D. Raj Reddy, Foundations and Grand Challenges of Artificial Intelligence, forthcoming. For a similar
analysis of the brain’s processing capabilities, see J. A. Feldman and D. H. Ballard, “Connectionist
Models and Their Properties,” Cognitive Science 6 (1982): 205-254.
16. Work is being done to allow intelligent systems to exploit past experiences instead of relying solely
on the deep analysis of the current situation. For an example, see Craig Stanfill and David Waltz,
“Toward Memory Based Reasoning,” Communications of the ACM 29, no. 12 (1986).

17. See Craig Stanfill and David Waltz, “Toward Memory Based Reasoning,” Communications of the
ACM 29, no. 12 (1986).
18. Human chess-playing and computer chess are analyzed for similarities and differences in Eliot
Hearst, “Man and Machine: Chess Achievements and Chess Thinking,” in Peter Frey, ed., Chess Skill
in Man and Machine, 2nd ed., 1983.
19. See Eliot Hearst, “Man and Machine.”
20. Newell and his associates maintain that much of learning is reorganization of certain memories into
efficient “chunks.” See John E. Laird, P. Rosenbloom, and Allen Newell, “Towards Chunking as a
General Learning Mechanism,” Proceedings of the National Conference of the American Association
for Artificial Intelligence, Austin, Tex, 1984.

21. To see how chunking fits into the SOAR view of cognition and intelligence, see John E. Laird, P.
Rosenbloom, and Allen Newell, “SOAR: An Architecture for General Intelligence,” Artificial Intelligence
Journal 33 (1987): 1-64.

22. An excellent introductory book on the structure and design of expert systems, with contributions
from many figures notable for their work in this area, is Frederick Hayes-Roth, D. A. Waterman, and
D. B. Lenat, eds., Building Expert Systems.

23. XCON, once called R1, was jointly developed by Carnegie-Mellon University and Digital Equipment
Corporation (DEC). See J. McDermott, “R1: A Rule-Based Configurer of Computer Systems,” Artificial
Intelligence Journal 19, no. 1 (1982).

24. For a DEC view of its experiences with XCON, see Arnold Kraft, “XCON: An Expert Configuration
System at Digital Equipment Corporation,” in Patrick Winston and Karen Prendergast, eds., The A/
Business (Cambridge: MIT Press, 1984).

25. Many techniques were introduced to handle the uncertainty of propositions that an expert system
is asked to deal with. Fuzzy logic is one such system. See Lofti Zadeh, “Fuzzy Logic and Approximate
Reasoning,” Synthese 30 (1975): 407-428. Zadeh’s fuzzy logic has a number of limitations, and other
systems for uncertainty have grown in popularity. See Edward Shortliffe and Bruce Buchanan, “A
Model of inexact Reasoning in Medicine,” Mathematical Biosciences 23 (1975): 350-379

26. The role of expert systems and knowledge-based systems in the economies of the future and the
implications of the Japanese fifth-generation project, are discussed in Edward Feigenbaum and Pamela
McCorduck, The Fifth Generation.

507
27. See Edward Feigenbaum, “The Art of Artificial Intelligence: Themes and Case Studies in Knowl-
edge Engineering,” Fifth International Joint Conference on Artificial Intelligence, Cambridge Mass.,
1977. :
28. The experiences and contributions of the DENDRAL experiments are recorded and analyzed in
detail in R. Lindsay, B. G. Buchanan, E. A. Feigenbaum, and J. Lederberg, DENDRAL: Artificial
Intelligence and Chemistry (New York: McGraw-Hill, 1980).
29. An excellent article reviewing the research on, and lessons from, the two systems is Bruce
Buchanan and Edward Feigenbaum, “DENDRAL and Meta-DENDRAL: Their Applications Dimension”
Artificial Intelligence Journal 11 (1978): 5-24.

30. Victor L. Yu, Lawrence M. Fagan, S. M. Wraith, William Clancey, A. Carlisle Scott, John Hannigan,
Robert Blum, Bruce Buchanan, and Stanley Cohen, “Antimicrobial Selection by Computer: A Blinded
Evaluation by Infectious Disease Experts,” Journal of the American Medical Association 242, no. 12
(1979): 1279-1282
31. The results of the MYCIN project at Stanford have been very influential on current thinking in
artificial intelligence. They are presented and analyzed in Bruce Buchanan and Edward Shortliffe, eds.,
Rule-Based Expert Systems: The MYCIN Experiments of the Stanford Heuristic Programming Project
(Reading, Mass.: Addison-Wesley, 1984).
32. The expert-system industry is a burgeoning one. See Paul Harmon and David King, Expert Systems:
Artificial Intelligence in Business. The diversity of application areas for expert systems is remarkable.
See Terri Walker and Richard Miller, Expert Systems ‘87 (Madison, Ga.: SEAI Technical Publications,
1987),

33. Harry Newquist, ed., Al Trends ‘87: A Comprehensive Annual Report on the Artificial Intelligence
Industry (Scottsdale, Ariz.: DM Data, 1987). Expert systems are changing the way problem solving is
handled in corporations across the world. See Edward Feigenbaum, Pamela McCorduck, and Penny Nii,
The Rise of the Expert Company.

34. See Edward Feigenbaum, “The Art of Artificial Intelligence: Themes and Case Studies in Knowl-
edge Engineering,” Fifth International Joint Conference on Artificial Intelligence, Cambridge Mass.,
1977

35. The MYCIN expert system originally appeared as Edward Shortliffe’s doctoral dissertation in 1974.
Other Stanford dissertations explored further the broad concepts behind MYCIN and produced some
important tools and applications. Randall Davis's TEIRESIAS, an interactive tool to help the knowledge
engineer structure expertise, is presented in Randall Davis, “Applications of Meta-Level Knowledge to
the Construction, Maintenance, and Use of Large Knowledge Bases,” Ph.D. dissertation, Stanford
University, Artificial Intelligence Laboratory, 1976. William van Melle succeeded in showing that in
keeping with the conceptual framework proposed earlier by Feigenbaum et al., the inference engine
and the knowledge base could in fact be separated out. Van Melle’s system, EMYCIN, represented the
structure of inferences and reasoning in MYCIN. See W. van Melle, “A Domain-Independent System
That Aids in Constructing Knowledge-Based Consultation Programs,” Ph.D. dissertation, Stanford
University, Computer Science Department, 1980.
36. EMYCIN was combined with a knowledge-base on pulmonary disorder diagnosis to produce PUFF.
See Janice Aikens, John Kunz, Edward Shortliffe, and Robert Fallat, “PUFF: An Expert System for
Interpretation of Pulmonary Function Data,” in William Clancey and Edward Shortliffe, eds., Readings in
Medical Artificial Intelligence: The First Decade.
37. William Clancey and Reed Letsinger, "NEOMYCIN: Reconfiguring a Rule-Based Expert System for
Application to Teaching,” in William Clancey and Edward Shortliffe, eds., Readings in Medical Artificial
Intelligence: The First Decade; and E. H. Shortliffe, A. C. Scott, M. Bischoff, A. B. Campbell, W. van
Melle, and C. Jacobs, “ONCOCIN: An Expert System for Oncology Protocol Management,” in
Proceedings of the Seventh International Joint Conference on Artificial Intelligence (Menlo Park, Calif.:
American Association for Artificial Intelligence, 1981), pp. 876-881

38. See Ramesh Patil, Peter Szolovits, and William Schwartz, “Causal Understanding of Patient Illness
in Medical Diagnosis,” in William Clancey and Edward Shortliffe, eds., Readings in Medical Artificial
Intelligence: The First Decade.
39. Organization of knowledge is especially difficult when the domains are as broad as that of CADU-
CEUS, the system developed chiefly by Harry Pople and Jack Myers. See Harry Pople, “Heuristic
Methods for Imposing Structure on Ill-Structured Problems: The Structure of Medical Diagnostics,” in
Peter Szolovits, ed., Artificial Intelligence in Medicine (Boulder, Col.: West View Press, 1982).

508
40. A short overview of the performance of CADUCEUS is Harry Pople, “CADUCEUS: An Experimental
Expert System for Medical Diagnosis,” in Patrick Winston and Karen Prendergast, eds., The A/
Business.
41. A recent evaluation from the medical community of the performance and potential of medical
artificial intelligence is William Schwartz, Ramesh Patil, and Peter Szolovits, “Artificial Intelligence in
Medicine: Where Do We Stand?” New England Journal of Medicine 316 (1987): 685-688.
42. See William Schwartz, Ramesh Patil, and Peter Szolovits, “Artificial Intelligence in Medicine: Where
Do We Stand?” New England Journal of Medicine 316 (1987): 685-688.
43. For applications of artificial intelligence in a wide variety of areas, including finance, see Wendy
Rauch-Hindin, Artificial Intelligence in Business, Science, and Industry.

44. The structure of Prospector is explained in R. O. Duda, J. G. Gaschnig, and P. E. Hart, “Model
Design in the PROSPECTOR Consultant System for Mineral Exploration,” in D. Michie, ed., Expert
Systems in the Micro-Electronic Age (Edinburgh: Edinburgh University Press, 1979). A report on
Prospector's role in finding the molybdenum deposit in Washington is in A. N. Campbell, V. F. Hollister,
R. O. Duda, and P. E. Hart, “Recognition of a Hidden Mineral Deposit by an Artificial Intelligence
Program,” Science 217, no. 3 (1982). Prospector is also discussed in Avron Barr, Edward Feigenbaum,
and Paul Cohen, eds., The Handbook of Artificial Intelligence (Los Altos, Calif.: William Kaufman, 1981).

45. Digital Equipment Corporation's Al projects are described in Susan Scown, The Artificial Intelligence
Experience (Maynard, Mass.: Digital Press, 1985)
46. Many of these expert-system products are described in Paul Harmon and David King, Expert
Systems: Artificial Intelligence in Business, pp. 77-133
47. Two overviews of the goals and constituent projects of the Strategic Computing Initiative are
Dwight Davis, “Assessing the Strategic Computing Initiative,” High Technology, April 1985, and Karen
McGraw, “Integrated Systems Development,” DS&E (Defense Science and Electronics), December
1986.
48. The Pilot's Associate is assessed by two Air Force officers in Ronald Morishige and John Retelle,
“Air Combat and Artificial Intelligence,” Air Force Magazine, October 1985.

49. Solutions to some of the limitations of expert systems imposed by current architectures are
discussed in Randall Davis, “Expert Systems: Where Are We? And Where Do We Go From Here?”
MIT Artificial Intelligence Laboratory, Al memo 665, 1982

50. Until recently, machine learning has been a neglected area within artificial intelligence, perhaps
because of the many difficulties underlying the problem. An important collection of papers on machine
learning is Ryszard Michalski, Jaime Carbonell, and Tom Mitchell, eds., Machine Learning—An Artificial
Intelligence Approach (Palo Alto, Calif.: Tioga Publishing Company, 1983)
51. Douglas Lenat wrote AM ( Automated Mathematician) as an experiment in causing machine
learning by discovery, in the area of number theory. EURISKO is an improved discovery program. The
systems are discussed in Douglas Lenat, “Why AM and EURISKO Appear to Work,” Artificial
Intelligence Journal 23 (1984): 269-294.
52. Robert Hink and David Woods, “How Humans Process Uncertain Knowledge,” A/ Magazine, Fall
1987. This paper is written primarily to assist knowledge engineers in structuring domain knowledge in
a statistically accurate manner.
53. The cognitive and behavioral aspects of human decision making under uncertainty are considered in
an important collection of papers: Daniel Kahneman, Paul Slovic, and Amos Tversky, eds., Judgement
under Uncertainty: Heuristics and Biases. The essays in this volume assess intriguing aspects of the
way people process and interpret information.
54. See Samuel Holtzmann, Intelligent Decision Systems (Reading, Mass.: Addison-Wesley, 1989)

55. This is not surprising, since language is a principle means of expressing thought. The entire field of
psycholinguistics is devoted to studying the connection between language and thought. So strong is
the appeal of this connection that some believe the Whorfian hypothesis, which, loosely stated, holds
that there can be no thought without language. Others accept a much weaker form of the Whorfian
hypothesis: that there has to be a language of thought, a language that is not necessarily the same as
one’s spoken language. See J. Fodor, T. Bever, and M. Garrett, The Psychology of Language (New
York: McGraw-Hill, 1975); and Benjamin Whorf, Language, Thought, and Reality: Selected Writings
(Cambridge, Mass.: MIT Press, 1956).
56. These and other theoretical aspects of computational linguistics are covered in Mary D. Harris,
Introduction to Natural Language Processing.
57. Terry Winograd has cogently argued that natural languages assume an enormous quantity of
background knowledge. A computer system that lacks this knowledge will not be able to understand
language in the sense that the speaker would expect a human listener to. See Terry Winograd, “What
Does It Mean to Understand Language,” Cognitive Science4 (1980).
58. Y. Bar-Hillel, “The Present Status of Automatic Translation of Languages,” in F. L. Alt, ed.,
Advances in Computers, vol. 1 (New York: Academic Press, 1960).
59. An account of the impressive performance of Logos appears in Tim Johnson, Natural Language
Computing: The Commercial Applications (London: Ovum, 1985), pp. 160-164.

60. There is more to what our statements mean than what we actually say. We are generally con-
cerned with the practical effects of what we say. Some kinds of speech are actions, and such
expressions are referred to as speech acts. See John Searle, Speech Acts (Cambridge: Cambridge
University Press, 1969).
61. Metaphors and idioms are a powerful way to communicate. Lakoff argues that metaphors are not
merely literary devices but permeate every aspect of everyday thought. See Mark Johnson and George
Lakoff, Metaphors We Live By.
62. Terry Winograd, “What Does It Mean to Understand Language,” Cognitive Science4 (1980)
63. Much has been written about SHRDLU, since it demonstrates deep understanding and reasoning
within its limited area of specialty. Winograd’s 1970 thesis on SHRDLU is slightly modified and
published as Terry Winograd, Understanding Natural Language (New York: Academic Press, 1972). A
brief presentation of the main ideas appears as Terry Winograd, “A Procedural Model of Language
Understanding,” in Roger Schank and Kenneth Colby, eds., Computer Models of Thought and
Language (San Fransisco: W. H. Freeman, 1973).

64. That toy worlds offer abstractions of significant value is argued by Marvin Minsky and Seymour
Papert in “Artificial Intelligence Progress Report,” MIT Artificial Intelligence Laboratory, Al memo 252,
1972
65. A short article about Harris and Intellect is Barbara Buell, “The Professor Getting Straight As on
Route 128,” Business Week, April 15, 1985.

66. Scripts appeared as early as 1973. See Robert Abelson, “The Structure of Belief Systems,” in
Roger Schank and Kenneth Colby, eds., Computer Models of Thought and Language (San Francisco:
W. H. Freeman, 1973). But their use as a powerful mechanism for knowledge representation became
sophisticated only a few years later. The standard reference on scripts is Roger Schank and Robert
Abelson, Scripts, Plans, Goals, and Understanding (Hillsdale, N.J.: Erlbaum, Lawrence Associates,
1977)

67. Schank's efforts at Cognitive Systems are described by him in Frank Kedig, “A Conversation with
Roger Schank,” Psychology Today, April 1983. Roger Schank has since resigned all major roles at
Cognitive Systems.

68. An excellent survey of the natural language business is Tim Johnson, Natural Language Computing:
The Commercial Applications (London: Ovum, 1985)
69. Translating text by computer is a rapidly growing business. See Harry Newquist, ed., Al Trends ‘87:
A Comprehensive Annual Report on the Artificial Intelligence Industry (Scottsdale, Ariz.: DM Data,
1987)
70. See Harry Newquist, ed., Al Trends ’87: A Comprehensive Annual Report on the Artificial Intelli-
gence Industry (Scottsdale, Ariz.: DM Data, 1987)
71. The production of R.U.A. and its implications for robots are discussed in Jasia Reichardt, Robots:
Fact, Fiction, and Prediction, a delightful book on the history and future of robots.

72. Some of these early robots are described in Reichardt, Robots: Fact, Fiction, and Prediction.

73. This generation of robots and their role in factory automation is examined by Isaac Asimov with his
usual scientific clarity in Isaac Asimov and Karen Frenkel, Robots: Machines in Man’s Image.

74. See Harry Newaquist, ed., A/ Trends ‘87: A Comprehensive Annual Report on the Artificial Intelli-
gence Industry (Scottsdale, Ariz.: DM Data, 1987).

75. Today the importance of robot programming is immense, since programming is the primary path to
adaptive robots. See Tomas Lozano-Perez, “Robot Programming,” MIT Artificial Intelligence Labora-
tory, memo 698, 1982.

510
76. |saac Asimov and Karen Frenkel, Robots: Machines in Man's Image.

77. Fully automatic factories are unusual today. More common are plants whose organization and
operation rely significantly on robotic machinery, while human workers handle other important
operations. The structure of such production units is realistically described in Christopher Joyce,
“Factories Will Measure As They Make," New Scientist, September 4, 1986
78. What will the factory of the future be like? Some analyses are put forward in Philippe Villers,
“Intelligent Robots: Moving Toward Megassembly,” and Paul Russo, “Intelligent Robots: Myth or
Reality.“ Both of these essays appear in Patrick Winston and Karen Prendergast eds., The A/ Business.
One writer speculates that fully automated factories will be moved away from earth, and we will soon
be industrializing outer space. See Lelland A. C. Weaver, “Factories in Space,” The Futurist, May-June,
1987
79. \saac Asimov and Karen Frenkel, Robots: Machines in Man's Image.

80. See Gene Bylinsky, “Invasion of the Service Robots,” Fortune, September 14, 1987.
81. Gene Bylinsky, “Invasion of the Service Robots,” Fortune, September 14, 1987
82. For details on Odex and other robots being used to increase safety for human workers in nuclear
plants, see Steve Handel, “Al Assists Nuclear Plant Safety.” Applied Artificial Intelligence Reporter,
June 1986. See “High Tech to the Rescue,” a special report in Business Week, June 16, 1986 for a
description of Allen Bradley's factory. Also see Gene Bylinsky, “Invasion of the Service Robots,”
Fortune, September 14, 1987.
83. Some of these new methodologies are described in the context of artificial legs in Marc H. Raibert
and lvan Sutherland, “Machines That Walk,” Scientific American, January 1983

84. Anderson's Ping-Pong player was an outcome of his doctoral work at the University of Pennsylva-
nia. The design and construction of this robot are detailed in Russell Anderson, “A Robot Ping-Pong
Player” (Cambridge: MIT Press, 1985)
85. The dexterity and versatility of some of today’s robotic hands is certainly encouraging. A report,
accompanied by some excellent photographs, appears in Daniel Edson, “Giving Robot Hands a Human
Touch,” High Technology, September 1985.

86. An informative article on what the voice-activated robots of Leifer and Michalowski could do for the
disabled is Deborah Dakins, “Voice-Activated Robot Brings Independence to Disabled Patients,”
California Physician, August 1986. Studies in robotics are leading to an important industry: the eventual
production of artificial limbs, hearts, and ears. See Sandra Atchison, “Meet the Campus Capitalists of
Bionic Valley,” Business Week, May 5, 1986.
87. The Waseda robotic musician is an interesting synthesis of a variety of technologies. There are two
excellent references on Wabot-2. The performances aspects of the robot are covered in Curtis Roads,
“The Tsukuba Musical Robot,” Computer Music Journal, Summer 1986. The design and engineering
aspects of the robot are covered in a set of articles authored by the Waseda team itself. These articles
appear in a special issue of the university's research bulletin: “Special Issue on WABOT-2,” Bulletin of
Science and Engineering Research Laboratory (Waseda University) no. 112 (1985)
88. Paul McCready’s unconventional experiments in aerodynamics are quite fascinating. One can meet
him and his flying machines in Patrick Cooke, “The Man Who Launched a Dinosaur,” Science 86, April
1986.

89. The Defense Department's Autonomous Land Vehicle project has produced at least two transporta-
tion “robots” that can be used in terrain that is not passable by conventional means. The Adaptive
Suspension Vehicle, which was developed primarily at Ohio State University, is described in Kenneth
Waldron, Vincent Vohnout, Arrie Perry, and Robert McGhee, “Configuration Designing of the Adaptive
Suspension Vehicle,” /nternational Robotics Research Journal, Summer 1984. The Terregator (another
vehicle) and other projects of robotics groups at Carnegie-Mellon University are described in Eric
Lerner, “Robotics: The Birth of a New Vision,” Science Digest, July 1985.

90. The intelligence of remotely piloted aircraft are described in Peter Gwynne, “Remotely Piloted
Vehicles Join the Service,” High Technology, January 1987, pp. 38-43.
91. The contribution of each these disciplines to the technology underlying robots is described in the
important review article Michael Brady, “Artificial Intelligence and Robotics,” MIT Artificial Intelligence
Laboratory, Al memo no. 756, 1983.

92. That most robots today function in only organized or artificial environments has been a major
concern to Rodney Brooks, an MIT roboticist whose mobile robots and artificial insects perform very
simple tasks in the dynamic environments we find ourselves in everyday. See Rodney Brooks,
“Autonomous Mobile Robots,” in W. Eric Grimson and Ramesh Patil, eds., A/ in the 1980s and
Beyond.

93. Noel Perrin, a professor at Dartmouth, argues that even though robots are not yet rampant in
households, research in robotics has been successful enough to warrant a serious look at the impact
robots will ultimately have on society. See Noel Perrin, “We Aren't Ready for the Robots,” Wall Street
Journal, editorial page, February 25, 1986.
94. Japan's Fifth Generation Project and the role of ICOT and MITI are presented in their technological,
personal, and sociopolitical dimensions in the well-written book Edward Feigenbaum and Pamela
McCorduck, The Fifth Generation. This provocative book served as a rallying cry for the American
industry's efforts to respond to ICOT.

95. The first complete description of the the Japanese fifth-generation project is “Outline of Research
and Development Plans for Fifth Generation Computer Systems,” Institute for New Generation
Computer Technology (ICOT), Tokyo, May 1982. Descriptions of work in progress are frequently
released by ICOT through its periodicals, conference proceedings, and research reports. ICOT’s primary
journal is [COT Journal Digest.
96. A brief but complete description of the American and European responses to the Japanese effort
appears as chapter 7 in Susan J. Scown, The Artificial Intelligence Experience: An Introduction
(Maynard, Mass.: Digital Press, 1985). Perhaps the most thorough coverage of these international
efforts appears in Fifth Generation Computers: A Report on Major International Research Projects and
Cooperatives (Madison, Ga.: SEA! Technical Publications, 1985)
97. See Edward Feigenbaum and Pamela McCorduck, The Fifth Generation, 1983, pp. 224-226, and
also Susan J. Scown, The Artificial Intelligence Experience: An Introduction (Maynard, Mass.: Digital
Press, 1985), pp. 150-152
98. Antitrust laws remain a problem in the operation of MCC. See David Fishlock, “The West Picks Up
on the Japanese Challenge: How US Is Rewriting Anti-Trust Laws,” Financial Times (London), January
27, 1986.

99. Susan J. Scown, The Artificial Intelligence Experience: An Introduction (Maynard, Mass.: Digital
Press, 1985), pp. 154-155.

100. Research funded and administered by Alvey is described in its publications. See “Alvey Pro-
gramme Annual Report, 1987,” Alvey Directorate, London, 1987.
101. Susan J. Scown, The Artificial Intelligence Experience: An Introduction (Maynard, Mass.: Digital
Press, 1985), pp. 153-154.
102. See Fifth Generation Computers: A Report on Major International Research Projects and
Cooperatives (Madison, Ga.: SEAI Technical Publications, 1985).

103. Not everyone is concerned about the future of Japan's fifth-generation computer systems. See J.
Marshall Unger, The Fifth Generation Fallacy: Why Japan Is Betting Its Future on Artificial Intelligence
(Oxford: Oxford University Press, 1987).

The Science of Art

1. See the articles published in the Computer Music Journal, where much of this revolution is docu-
mented. A selection of such articles may be found in Curtis Roads, The Music Machine: Selected
Readings from “Computer Music Journal.” Articles on computer music also appear in the journal
Computers and the Humanities.
2. The capacities and limitations of digital technology are reviewed in parts 1 and 2 of Curtis Roads and
John Strawn, eds., Foundations of Computer Music. On digital tone generation, see chapter 13 of Hal
Chamberlin, Musical Applications of Microprocessors.
3. For a general introduction to the principles of music synthesis and a brief history, see chapter 1 of
Chamberlin’s Musical Applications. Chapter 18 describes music-synthesis software, and chapter 19
reviews a number of synthesizers.

4. MIDI is described in Chamberlin, Musical Applications, pp. 312-316. Chamberlin made the prediction
that by 1990 a new music protocol would be developed “as the weaknesses of MIDI become
apparent” (p. 789)
5. In The Technology of Computer Music, a text for composers, Max V. Mathews provides an appendix
on psychoacoustics and music, because, he argued, no intuitions exist for the new sounds possible
with computers.

S12
6. See chapter 16 in Chamberlin's Musical Applications.
7. Hal Chamberlin's “A Sampling of Techniques for Computer Performance of Music,” originally
published in Byte magazine in September 1977, describes how to create four-part melodies on a
personal computer. Stephen K. Roberts described his own polyphonic keyboard system in “Polyphony
Made Easy,” an article originally published in Byte in January 1979. Both articles were reprinted in
Christopher P. Morgan, ed., The “Byte” Book of Computer Music; see pp. 47-64 and pp. 117-120,
respectively
8. In chapter 18 of Musical Applications, Chamberlin describes programming techniques for pro-
grammed performance systems, and claims, “It is immaterial whether the synthesis is performed in
real time or not, since the ‘score’ is definitely prepared outside of real time” (p. 639)

9. On the editing of sequences, see Chamberlin, Musical Applications, chapter 11

10. Bateman discusses the role of the computer in composing in chapters 11 and 12 of Introduction to
Computer Music. He relates the stochastic composition possible with the computer to the fact that
Mozart once composed with the aid of a pair of dice, but he emphasizes the computer's subservience
to the human composer's creativity. For a good review of the various selection techniques involved in
Al composing programs, see C. Ames, “Al in Music,” in Stuart C. Shapiro, ed., Encyclopedia of Artificial
Intelligence, vol. 1, pp. 638-642.
11. See Ames, “Al in Music,” in Shapiro, Encyclopedia of Artificial Intelligence, vol. 1 pp. 638-642. Also
see S. Papert, “Computers in Education: Conceptual Issues,” in Shapiro, Encyclopedia of Artificial
Intelligence, vol. 1, p. 183. Papert points out that the difficulties of performance may be circumvented
by use of the computer, and that students may begin to compose in the same way that they learn to
draw when studying art or to write when studying literature
12. Music systems for personal computers are described in C. Yavelow, “Music Software for the Apple
Macintosh,” Computer Music Journal9 (1985): 52-67. See also C. Yavelow, “Personal Computers and
Music,” Journal of the Audio Engineering Society 35 (1987): 160-193.
13. A comprehensive review of computerized music notation, including a table of important systems
and a useful bibliography, is found in N. P. Carter, R. A. Bacon, and T. Messenger, “The Acquisition,
Representation, and Reconstruction of Printed Music by Computer: A Review,” Computers and the
Humanities 22 (1988): 117-136. Professional Composer is being used by Garland Press to produce
editions of sixteenth-century music (p. 130).
14. Chamberlin discusses the use of synthesizers in music education on p. 710 of Musical Applications.

15. A brief assessment of the role of computers in art can be found in Philip J. Davis and Reuben
Hersh, Descartes’ Dream: The World According to Mathematics, pp. 43-53. Herbert W. Francke
discusses the resistance computer art encountered in “Refractions of Science into Art,” in H.-O.
Peitgen and P. H. Richter, The Beauty of Fractals: Images of Complex Dynamical Systems, pp. 181-187.

16. Neal Weinstock, in Computer Animation, discusses some of the resolution limitations of home
computers (see chapter 1). For an overview of graphics hardware, including output-only and display
hardware, see chapter 2 in Weinstock, and chapter3 in the standard text, J. D. Foley and A. Van Dam,
Fundamentals of Interactive Computer Graphics.

17. Advances in computer graphics are occurring at a rapid rate, and the literature is vast. Several
important sources of up-to-date information are the ACM Transactions on Graphics, Computer
Graphics, Quarterly Report of the ACM Special Interest Group of Graphics, Computer Graphics World,
and articles on computer graphics in Byte. Melvin L. Prueitt’s Art and the Computer has dazzling
pictures produced using a wide variety of computer-graphics techniques and includes a series of
examples of art produced on personal computers (see pp. 29 and 191-194). A review of the early
history of computer graphics, complete with illustrations, is provided by H. W. Francke in Computer
Graphics—Computer Art, pp. 57-105. Examples of practical applications of computer graphics are
found in Donald Greenberg, Aaron Marcus, Allen H. Schmidt, and Vernon Gorter, The Computer Image
Applications of Computer Graphics. The shift from fixed images to the modern transformable computer
image has stimulated a new analytical approach to graphics, exemplified by Jacques Bertin, Semiology
of Graphics, trans. William J. Berg.
18. Recent developments in color-graphics displays are described in H. John Durrett, ed., Color and the
Computer. Chapter 12 reviews the available technology for color hard-copy devices, and there are also
chapters on color in medical images, cartography, and education applications
19. Many of these techniques are described and illustrated in Prueitt, Art and the Computer. Some of
the extraordinary achievements in rendering reflection are illustrated on pp. 144-150, and an example
of the synthesis of natural and artificial scenes can be found on p. 29. An excellent example of

S13
distortion is provided by plates E to J of Foley and Van Dam, Fundamentals of Interactive Computer
Graphics, which illustrate the mapping of images of a mandrill onto a series of different geometric
shapes. Chapter 14 of that work provides a description 9f techniques being used to enhance the
realism of computer imagery.
20. Those interested in the technical aspects of image transformation should consult chapter 8 of Foley
and Van Dam, Fundamentals of Interactive Computer Graphics.

21. Computer artists active during the 1970s describe their relations to their chosen medium in Ruth
Leavitt, Artist and Computer. When asked if his work could be done without the computer, computer
artist Aldo Giorgini said, “Yes, in a fashion analogous to the one of carving marble with a sponge”
(p. 12)
22. The game of life was invented by John Conway in 1970 and is one example of a cellular automaton.
As Tommaso Toffoli and Norman Margolis point out, “A cellular automata machine is a universe
synthesizer.” See their Cellular Automata Machines: A New Environment for Modeling, p. 1. An
entertaining explanation of games of life can be found in chapter 7 of Ivars Peterson, The Mathematical
Tourist: Snapshots of Modern Mathematics. See also A. K. Dewdney, The Armchair Universe, “World
Four: Life in Automata.” Dewdney explores the concepts of one-dimensional computers and three-
dimensional life. Chapter 25 in Elwyn R. Berlekamp, John H. Conway, and Richard K. Guy, Winning
Ways for Your Mathematical Plays, is devoted to games of Life. On page 830 the authors describe how
to make a Life computer: “Many computers have been programmed to play the game of Life. We shall
now return the compliment by showing how to define Life patterns that can imitate computers.” See,
as well, William Poundstone, The Recursive Universe: Cosmic Complexity and the Limits of Scientific
Knowleage, chapters 11 and 12 and the section “Life for Home Computers.”

23. See Richard Dawkins’s 1987 work The Blind Watchmaker. Dawkins devised a simple model for
evolution; starting with a stick figure, his program produces more and more complex figures often
resembling actual natural shapes, such as insects. Sixteen numbers function as “genes” and
determine the resulting forms, or “biomorphs.”
24. In Dawkins’s scheme, the human observer provides the natural selection, choosing “mutations”
that will “survive.” An artist could use aesthetic criteria to determine the direction of “evolution.” A. K.
Dewdney describes Dawkins’s program, which runs on the Mac, in “A Blind Watchmaker Surveys the
Land of Biomorphs,” Scientific American, February 1988, pp. 128-131
25. For an excellent introduction to the concept of recursion, see Poundstone, The Recursive Universe.

26. An account of Mandelbrot’s eclectic career and his discovery of the “geometry of nature” can be
found in James Gleick, Chaos: Making a New Science, pp. 81-118. The bible of fractal geometry is
Benoit Mandelbrot, The Fractal Geometry of Nature (1983). This work superseded Mandelbrot’s earlier
volume Fractals: Form, Chance, and Dimension. \mages produced by fractals are described and
illustrated in the classic work by H-O. Peitgen and P. H. Richter, The Beauty of Fractals: Images of
Complex Dynamical Systems. Peitgen and Richter devoted themselves to the study of the Mandelbrot
set and produced spectacular pictures, which they published and displayed (see Gleick, Chaos,
pp. 229 ff)
27. See pp. 213-240 in Gleick’s Chaos. Chaotic-fractal evolutions are discussed in chapter 20 of
Mandelbrot's Fractal Geometry.
28. Some of Mandelbrot's work on price change and scaling in economics can be found in The Fractal
Geometry of Nature, pp. 334-340. For more insight into Mandelbrot’s application of fractals in
economics and biology, see Gleick, Chaos, pp. 81-118. On the properties of scaling in music, see The
Fractal Geometry of Nature, pp. 374-375
29. Peterson, The Mathematical Tourist, pp. 114-116. Also see Mandelbrot, The Fractal Geometry of
Nature, chapter 5.

30. Peterson, The Mathematical Tourist, pp. 126-127. On modeling clouds with fractals, see
Mandelbrot, The Fractal Geometry of Nature, p. 112.

31. “One should not be surprised that scaling fractals should be limited to providing first approxima-
tions of the natural shapes to be tackled. One must rather marvel that these first approximations are so
strikingly reasonable” (Mandelbrot, The Fractal Geometry of Nature, p. 19.) In “The Computer as
Microscope,” in The Mathematical Tourist, Peterson points out that “natural fractals are often self-
similar in a statistical sense” (p. 119). See also pp. 155-164.
32. “Computer graphics provides a convenient way of picturing and exploring fractal objects, and fractal
geometry is a useful tool for creating computer images” (Peterson, The Mathematical Tourist, p. 123.)
33. Mandelbrot explores the application of fractal geometry to cosmology in The Fractal Geometry of
Nature. See in particular chapter 9.

sia
34. Mandelbrot’s discussion of the etymology is in The Fractal Geometry of Nature, pp. 4-5.

35. The concept of dimensionality is central to an understanding of fractals and to an understanding of

the structure of nature. See chapter 3 in Mandelbrot, The Fractal Geometry of Nature.

36. See Mandelbrot, “Index of Selected Dimensions,” in Fractals: Form, Chance, and Dimension,
p. 365, where the seacoast dimension is given as 1.25. Mandelbrot devotes chapter 6 of The Fractal
Geometry of Nature to “Snowflakes and Other Koch Curves." In “The Koch Curve Tamed” he gives its
fractal dimension as 1.2618 (p. 36). Peterson describes the Koch curve, created in 1904, in The
Mathematical Tourist, pp. 116-119.

37. See chapter 10 of Mandelbrot, The Fractal Geometry of Nature, and the chapter "Strange Attrac-
tors” in Gleick, Chaos.
38. See Mandelbrot on fractal art, pp. 23-24 of The Fractal Geometry of Nature. He points out that the
images created by fractals may be reminiscent of the work of M. C. Escher because Escher was
influenced by hyperbolic tilings, which are related to fractal shapes (p. 23). Mandelbrot also suggests
that the work of certain great artists of the past, when it illustrated nature, exemplified “issues tackled
by fractal geometry”: the examples are the frontispiece of Bible moralisée illustrées, Leonardo's
Deluge, and Hokusai's Great Wave (see pp. C1, 2, 3, 16). Some striking examples of fractal images are
described and illustrated in Prueitt’s Art and the Computer (pp. 119, 121-124, 127, 166, 169). Alan
Norton has produced beautiful and bizarre complex shapes by generating and displaying geometric
fractals in three dimensions (see Prueitt, pp. 123-124). For images that resemble natural landscapes,
see plates C9, C11, C13, and C15 in Mandelbrot's Fractal Geometry of Nature. Mandelbrot himself
claims that these artificial landscapes are the fractal equivalent of the “complete synthesis of hemoglo-
bin from the component atoms and (a great deal of) time and energy” (p. C8).

39. On sophisticated animation systems, see chapter 4, in Nadia Magenat-Thalman and Daniel
Thalman, Computer Animation: Theory and Practice. On fractals and their use in generating images,
see pp. 106-110.

40. Harold Cohen, “How to Draw Three People in a Botanical Garden,” AAAI-88, Proceedings of the
Seventh National Conference on Artificial Intelligence, 1988, pp. 846-855. Some of the implications of
AARON are discussed in Pamela McCorduck, “Artificial Intelligence: An Apercu,” Daedalus, Winter
1988, pp. 65-83. This issue of Daedalus, devoted to Al, has been published in book form as Stephen R.
Graubard, ed., The Artificial Intelligence Debate: False Starts, Real Foundations.

41, Fora comparison between traditional animation procedures and new computer methods, see
chapters 1 and 2 in Magenat-Thalman and Thalman, Computer Animation. See also Weinstock,
Computer Animation.

42. Prueitt, Art and the Computer, p. 30. Indeed, Pruiett suggests that computer art “may be closer to
the human mind and heart than other forms of art. That is, it is an art created by the mind rather than
by the body” (pp. 2-3)
43. The most famous of these editors is EMACS, a real-time display editor that Stallman developed in
1974 from earlier systems, in particular, TECO, developed in 1962 by Richard Greenblatt et al. For a
description of EMACS, see Richard M. Stallman, “EMACS: The Extensible, Customizable, Self-
Documenting Display Editor,” in David R. Barstow, Howard E. Shrobe, and Erik Sandewell, eds.,
Interactive Programming Environments, pp. 300-325
44. For example the electronic Oxford English Dictionary (OED) enables scholars to answer questions
that would have taken a lifetime of work in the recent past. See Cullen Murphy, “Computers: Caught in
the Web of Bytes,” The Atlantic, February 1989, pp. 68-70.

45. New techniques for accessing information (on-line searches) are described in Roy Davies, ed.,
Intelligent Information Systems: Progress and Prospects.

46. For a recent assessment of desktop publishing, see John R. Brockmann, “Desktop Publishing—
Beyond GEE WHIZ: Part 1, A Critical Overview,” and Brockmann, “Desktop Publishing—Beyond GEE
WHIZ: Part 2, A Critical Bibliography of Materials,” both in IEEE Transactions on Professional
Communication, March 1988.
47. New tools for conceptual organization are described in Edward Barrett, Text, Context, and Hy-
pertext.

48. In the Introduction to RACTER, The Policeman’s Beard is Half-Constructed, William Chamberlain
describes the process behind RACTER’s prose: certain rules of English are entered into the computer,
and what the computer produces is based upon the words it finds in its files, which are then combined
according to “syntax directives.” Chamberlain concludes that this process “seems to spin a thread of
what might initially pass for coherent thinking throughout the computer-generated copy so that once
the program is run, its output is not only new and unknowable, it is apparently thoughtful. It is crazy
‘thinking’ | grant you, but ‘thinking’ that is expressed in perfect English.” See the discussion by A. K.
Dewdney, “Conversations with RACTER,” in The Armchair Universe, pp. 77-88. Dewdney points out
that RACTER is not artificially intelligent but “artificially insane” (p. 77).

49. For the history and assessment of various attempts at translation using computers, see Y. Wilks,
“Machine Translation,” in Shapiro, ed. Encyclopedia of Artificial Intelligence, vol. 1, pp. 564-571.

Visions
1. Douglas R. Hofstadter, Metamagical Themas: Questing for the Essence of Mind and Pattern, p.128.

2. David Waltz, “The Prospects for Building Truly Intelligent Machines," Daedelus, Winter 1988, p. 204.
3. The preface in Tom Forester's Information Technology Revolution examines similar issues.

4. In the fall of 1987 an entire issue of Scientific American was devoted to this topic. In particular, see
Abraham Peled, “The Next Computer Revolution,” Scientific American, October 1987, pp. 56-64.

5. See James D. Meindl, “Chips for Advanced Computing,” Scientific American, October 1987, pp.
79-81 and 86-88.

6. See Mark H. Kryder, “Data-Storage Technologies for Advance Computing,” Scientific American,
October 1987, pp. 117-125.
7. Such as massively parallel processors based possibly on superconductors. See Peter J. Denning,
“Massive Parallelism in the Future of Science,” American Scientist, Jan.—Feb. 1989, p. 16
8. Marvin Minsky discusses this problem in “Easy Things Are Hard,” Society of Mind, p. 29.

9. Koji Kobayashi, Computers and Communication: A Vision of C & C, pp. 165-166.

10. See Marshall McLuhan, Understanding Media.

11. For a vision of an office system interfacing with a public communication network, see Koji
Kobayashi, Computers and Communication, chapter 10. See also Roger Shank and Peter G. Childers,
“The World of the Future,” in The Cognitive Computer, pp. 227-230.

12. David N. L. Levy, All about Chess and Computers. See also M. M. Botvinnik, Computers in Chess.
13. An intriguing study of the relevance of comments made by master chess players during play can be
found in Jacques Pitrat, “Evaluating Moves Rather than Positions,” in Barbara Pernici and Mareo
Somalvico, eds., /// Convegno Internazionale L ‘Intelligenza Artificiale ed il Gioco Degli Scacchi (Federazi-
‘one Scacchistico Italiana, Regione Lombardia, Politecnico di Milano, 1981).
14. Evidence of this are early board-game programs modeled on master players’ strategies. The 1959.
checkers program of Arthur Samuels, for example, had 53,000 board positions in memory. See Peter
W. Frey, “Algorithmic Strategies for Improving the Performance of Game-Playing Programs,” in
Evolution, Games, and Learning: Models for Adaptation in Machines and Nature, Proceedings of the
Fifth Annual International Conference of the Center for Nonlinear Studies at Los Alamos, N.M., May
20-24, 1985, p. 355.

15. Recursiveness and massive computational power allow for subtle (and hence enormously varied)
solutions to algorithmic problems. See, for example, Gary Josin, “Neural Net Heuristics” BYTE,
October 1987, pp. 183-192; and Douglas Lenat, “The Role of Heuristics in Learning by Discovery,” in R.
Z. Michalski, J. J. Carbonell, and T. M Mitchell, eds., Machine Learning: An Artificial Intelligence
Approach. Also see Monroe Newborn, Computer Chess, pp. 8-15.
16. See John Hilusha’s article, “Smart Roads Tested to Avoid Traffic Jams,” New York Times,
October 18, 1988.

17. Plans are already in place for the development and use of flying vehicles. See, for instance,
“Simulation of an Air Cushion Vehicle Microform,” final report for period January 1975-December
1976, Charles Stark Draper Laboratory, Cambridge, 1977
18. For recent advances in computer and chip design, see James D. Meindl, “Chips for Advanced
Computing,” Scientific American, October 1987, pp. 78-88. An extensive but less current review of the
technology may be found in Alan Burns, The Microchip.

19. See Stewart Brand, The Media Lab: Inventing the Future at MIT, pp. 83-91
20. Although the Turing test has been discussed at length in chapters 2 and 3, the general reader may
further appreciate a straightforward presentation of this famous test in Isaac Malitz, “The Turing
Machine,” BYTE, November 1987, pp. 348-358.

516
21. Research at the University of Illinois is a case in point. The Center for Superconducting Research
and Development was established there in 1984 for the purpose of demonstrating that high-speed
parallel processing is practical for a wide range of applications.
22. Considerable mention is given to this technology in B. Deaver and John Ruvalds, eds, Advances in
Superconductivity. See especially A. Barone and G. Paterno, “Josephson Effects: Basic Concepts.”

23. See M. A. Lusk, J. A. Lund, A. C. D. Chaklader, M. Burbank, A. A. Fife, S. Lee, B. Taylor, and J.
Vrba, “The Fabrication of a Ceramic Superconducting Wire,” Superconductor Science and Technology
1 (1988): 137-140.

24. For brief but informative article on the subject, see Robert Pool, “New Superconductors Answer
Some Questions,” Science 240 (April 8, 1988): 146-147.
25. See David Chaffee, The Rewiring of America: The Fiber Optics Revolution. Also, a reliable,
technically informative account may be found in Robert G. Seippel, Fiber Optics.
26. A terse and brief account of the new technology that serves as the basis for these advances may
be found in Tom Forester, The Materials Revolution, pp. 362-364.
27. It is interesting to compare the mentions made of molecular computing in Tom Forester’s High
Tech Society, p. 39, with those in his Materials Revolution, pp. 362-364.

The Impact On...

1. Seymour Papert makes a convincing argument for extensive use of computers in the classroom in
“Computers and Computer Cultures,” Mindstorms, pp. 19-37.
2. See notes 5 and 14 to the prolog of this book for sources for these statistics.

3. Wassily Leontief and Faye Duchin, eds., The Future Impact of Automation on Workers, p. 18. See
also note 13 to the prolog of this volume.

4. Wassily Leontief and Faye Duchin, eds., The Future Impact of Automation on Workers, pp. 20-21
5. Wassily Leontief and Faye Duchin, eds., The Future Impact of Automation on Workers, pp. 12-19.
See also James Jacobs, “Training Needs of Small and Medium Size Firms in Advanced Manufacturing
Technologies,” in 1987 IEEE Conference on Management and Technology, Atlanta, Georgia, October
27-30, 1987, pp. 117-123.

6. Wassily Leontief and Faye Duchin, eds., The Future Impact of Automation on Workers. pp.
25-26, 52.

7. Wassily Leontief, “The World Economy to the Year 2000,” Scientific American, Sept. 1980, pp
206-231.

8. Wassily Leontief and Faye Duchin, eds., The Future Impact of Automation on Workers, pp.
25-26, 92

9. Tom Forester, High Tech Society, p. 181.

10. See “The Electronic Office,” in Tom Forester, High Tech Society, pp. 195-217.
11. Indeed, floppy disks and CDs have popularly been called a “new papyrus.”
12. See Ted H. Nelson, “Getting It out of Our System,” in G. Schechler, ed., Information Retrieval, pp.
191-210. For a recent account that reveals changes and advances, see Mark Bernstein, “The
Bookmark and the Compass: Orientation Tools for Hypertext Users,” S/G O/S Bulletin 9 (1988): 34-45.

13. Note, for example, the considerable professional reshuffling in the labor force to accommodate
burgeoning technological advances in the workplace. See S. Norman Feingold and Norma Reno Miller,
Emerging Careers: New Occupations for the Year 2000 and Beyond, vol. 1
14. For a discussion of children’s psychological responses to computers, see “The Question of ‘Really
Alive," in Sherry Turkle, The Second Self, pp. 324-332
15. For the sake of brevity | shall use the term “education” here in the conventional sense of education
during the school years, but the the ideas expressed in this section pertain to education in general
See, for example, Elizabeth Gerver, “Computers and Informal Learning,” in Humanizing Technology:
Computers in Community Use and Adult Education; and Jean-Dominique Warnier, “The Teaching of
Computing,” in Computers and Human Intelligence, p. 113 ff.

16. There is a growing volume of literature on the subject of children and computers. See, for instance
Sherry Turkle, “Child Programmers,” in The Second Self, pp. 93-136; Robert Yin and J. L. White,
“Microcomputer Implementation in Schools,” in Milton Chen and William Paisley, eds., Children and

S17
Microcomputers, pp. 109-128; Seymour Papert, Daniel Watt, Andrea di Dessa, and Sylvia Weir, “Final
Report of the Brookline LOGO Project,” MIT Al memo no. 545, Sept. 1979; and R. D. Pea and D. M.
Kurland, “On the Cognitive and Educational Benefits of Teaching Children Programming: A Critical
Look,” in New Ideas in Psychology., vol. 1
17. Debra Liberman, “Research and Microcomputers,” in Milton Chen and William Paisley, eds.,
Children and Microcomputers, pp. 60-61
18. See John Seely Brown, “Process versus Product,” in Chen and Paisley, eds., Children and
Microcomputers, pp. 248-266.
19. Functioning well right now are networked computerized card catalogs linking various university
libraries. For a skeptical view of networking, see Theodore Roszak, “On-Line Communities: The
Promise of Networking,” in The Cult of Information: The Folklore of Computers and the True Art of
Thinking.
20. Tom Forester swiftly chronicles the obstacles that have contributed to this situation in High Tech
Society, pp. 165-169. These obstacles notwithstanding, from the fall of 1980 to the spring of 1982 the
number of computers more than tripled in American schools. See also Jack Rochester and John Gantz,
The Naked Computer, p. 104.

21. Edward Tenner, Harvard Magazine 90 (1988): 23-29.

22. Tenner, Harvard Magazine 90 (1988): 23-29.

23. For a dispassionate and informative presentation of the Strategic Defense Initiative in which a
“flexible nuclear response” is examined, see Stephen J. Cimbala, “The Strategic Defense Initiative,” in
Stephen J. Andriole and Gerald W. Hople, eds., Defense Applications of Artificial Intelligence, pp.
263-291

24. See Randolf Nikitta, “Artificial Intelligence and the Automated Tactical Battlefield,” in Allan M. Din,
ed., Arms and Artificial Intelligence: Weapons and Arms Control of Applications of Advanced Comput-
ing, pp. 100-134.
25. McGeorge Bundy, George F. Kennan, Robert S. McNamara, and Gerard Smith, “Nuclear Weapons
and the Atlantic Alliance,” Foreign Affairs, Spring 1982, pp. 753-768. Another thoughtful and succinct
appraisal of the subject of nuclear deterrence is Leon Wieseltier’s Nuclear War, Nuclear Peace.
26. See Edward C. Taylor, “Artificial Intelligence and Command and Control—What and When?” in
Andriole and Hople, eds., Defense Applications of Artificial Intelligence, pp. 139-149. See also Alan
Borning, “Computer System Reliability and Nuclear War,” in Communications of the ACM 30 (1987):
124.

27. See Alan Borning, “Computer System Reliability and Nuclear War,” Communications of the
ACM 30 (1987): 112-131

28. See P. R. Cohen, D. Day, J. DeLisio, M. Greenberg, R. Kjeldsen, D. Suthers, and P. Berman,
“Management of Uncertainty in Medicine,” /nternational Journal of Approximate Reasoning, Sept.
1987, pp. 103-116.

29. See Tom Forester’s discussion of recent innovations in biotechnology in his Materials Revolution,
pp. 362-364.

30. For the well initiated, the excellent but highly technical journal Biomaterials, Artificial Cells, and
Artificial Organs: An International Journal frequently publishes articles relevant to the subject.
31. See Glenn D. Rennels and Edward H. Shortliffe, “Advanced Computing for Medicine,” Scientific
American, October 1987, pp. 154-161. See also James S. Bennell, “ROGET: A Knowledge-Based
System for Acquiring the Conceptual Structure of a Diagnostic Expert System,” Journal of Automated
Reasoning| (1985): 41-50.
32. For a clear overview of the vision challenge, see Michael Brady, “Intelligent Vision,” in Grimson and
Patil, eds., Al in the 1980s and Beyond, pp. 201-243
33. While we are still far from this achievement, advances are being made in the field of music
notation. See, for instance, John S. Gourlay, “A Language for Music Printing,” Communication of the
ACM 29 (1986): 388-401

34. Of interest to the reader may be the anonymous publication “Viadimir Ussachevsy: In Celebration
of his Seventy-Fifth Birthday,” University of Utah, 1987, pp. 8-9.
35. See David Dickson, “Soviet Computer Lag,” Science, August 1988, p. 1033. Some other interest-
ing facts about Soviet restrictions are to be found in Rochester and Gantz, “How's Your CPU, Boris?”
The Naked Computer.
Postcript
1. See pp. 677-678 of Douglas Hofstadter’s Gédel, Escher, Bach: An Eternal Golden Braid (New York:
Basic Books, 1979) for a fuller account of his concept of potential computer weaknesses.

2. Marvin Minsky, Society of Mind, pp. 186, 288.

3. The general reader will find pertinent to the topic Paul M. Churchland’s philosophical and scientific
examination throughout Matter and Consciousness.
4. See Einstein's letters of August 9, 1939, and December 22, 1950, to E. Schrodinger, in K. Przibram,
ed., Letters on Wave Mechanics, pp. 35-36 and 39-40.
5. Admittedly, some disavow the applicability of subatomic metaphors to any other aspect of life. See
Paul G. Hewitt, Conceptual Physics, 2nd ed., pp. 486-487
6. Adense and pertinent discussion of the sister particle paradox may be found in Abner Himony,
“Events and Processes in the Quantum World,” in R. Penrose and C. J. Isham, eds., Quantum
Concepts in Space and Time, pp. 182-196
Hibliagraphy and Suggested Readings

The Second Industrial Revolution

Antébi, Elizabeth. Biotechnology: Strategies for Life. Cambridge: MIT Press, 1986.

Bellin, David, and Gary Chapman, eds. Computers in Battle—Will They Work? Boston: Harcourt Brace
Jovanovich, 1987.

Bluestone, Barry, and Bennett Harrison. The Deindustrialization of America. New York: Basic Books,
1982.
Bythell, Duncan. The Handioom Weavers: A Study in the English Cotton Industry during the Industrial
Revolution. Cambridge: Cambridge University Press, 1969.
Cohen, S. S., and J. Zysman. Manufacturing Matters. New York: Basic Books, 1987.
Din, Allan M., ed., Arms and Artificial Intelligence. Oxford: Oxford University Press, 1987.

Feigenbaum, Edward A., and Pamela McCorduck. The Fifth Generation: Artificial Intelligence and
Japan's Computer Challenge to the World. Reading Mass.: Addison-Wesley Publishing Co., 1983.
Forester, Tom. High Tech Society: The Story of the Information Technology Revolution. Cambridge:
MIT Press, 1987.
Forester, Tom. The Information Technology Revolution. Cambridge: MIT Press, 1985.
Fishman, Katherine Davis. The Computer Establishment. New York: McGraw-Hill Book Co., 1981

Greenberger, Martin, ed. Computers and the World of the Future. Cambridge: MIT Press, 1962.

Harrison, Ben, and Barry Bluestone. The Great U-Turn. New York: Basic Books, 1988

Leontief, Wassily W. The Impact of Automation on Employment, 1963-2000. Institute for Economic
Analysis, New York University, 1984.
Schank, Roger, [with Peter G. Childers]. The Cognitive Computer: On Language, Learning, and Artificial
Intelligence. Reading, Mass.: Addison-Wesley, 1984.
Smith, ChristopherA., et al. Discovery ‘84: Technology for Disabled Persons. Materials Development
Center, Stout Vocational Rehabilitation Institute, University of Wisconsin at Stout, 1985.

Taylor, Phillip A., ed. The Industrial Revolution in Britain: Triumph or Disaster? Lexington, Mass.: Heath,
1970.

Thomis, Malcolm|. The Luddites: Machine-Breaking in Regency England. Hamden, Conn.: Archon
Books. 1970
U.S. Department of Commerce, Bureau of the Census. Historical Statistics of the U. S.: Colonial Times
to 1970. Vol. 1. Washington, D.C.: U.S. Government Printing Office, 1975.

U.S. Department of Commerce, Bureau of the Census. How We Live: Then and Now. Washington.
D.C.: Government Printing Office, 1986.
U.S. Department of Commerce, Bureau of the Census. Statistical Abstract of the United States, 1986.
106th ed. Washington, D.C.: Government Printing Office, 1986.

United Nations. Population and Vital Statistics Report. Series A, vol. 38, no. 3. United Nations, 1986.
Wattenberg, Ben J. The Statistical History of the United States from Colonial Times to 1970. New York:
Basic Books, 1976.
Wattenberg, Ben J., ed. The Statistical History of the United States from Colonial Times to the Present.
Stamford, Conn.: Fairfield Publishers, 1965.
Winner, Langdon. Autonomous Technology. Cambridge: MIT Press, 1977.

Zysman, J. Manufacturing Matters. New York: Basic Books, 1987.

What is Al, Anyway?

Alexander, S. Art and Instinct. Oxford: Folcroft Press, 1970. Originally published in 1927.
Baron, Jonathan. Rationality and Intelligence. Cambridge: Cambridge University Press, 1985.

Barrett, Paul H., ed. The Collected Papers of Charles Darwin. Vols. 1 and 2. Chicago: University of
Chicago Press, 1977
Bruner, Jerome S., Jacqueline J. Goodnow, and George A. Austin. A Study of Thinking. New York:
Science Editions, 1965. Originally published in 1956.
Darwin, Charles. The Origin of the Species. New York and London: D. Appleton & Company, 1910.
Originally published in 1859.

Dawkins, Richard. The Blind Watchmaker: Why the Evidence of Evolution Reveals a Universe without
Design. New York: W. W. Norton & Co., 1987.
Dawkins, Richard. The Extended Prototype: The Gene on the Unit of Selection. Oxford: Freeman & Co.,
1982
Dawkins, Richard, The Selfish Gene. New York: Oxford University Press, 1976.

Dobzhansky, Theodosius. Mankind Evolving: The Evolution of the Human Species. New Haven: Yale
University Press, 1962.
Elithorn, Alick, and Ranan Banerji, eds. Artificial and Human Intelligence: Edited Review Papers
Presented at the International NATO Symposium on Artificial and Human Intelligence. Amsterdam:
North-Holland Publishing Co., 1981

Elithorn, A., and D. Jones. Artificial and Human Thinking. San Francisco: Jossey-Bass, 1973.
Eysenck, H. J., and Leon Kamin. The Intelligence Controversy. New York: John Wiley & Sons, 1981.
Feigenbaum, Edward. “The Art of Artificial Intelligence: Themes and Case Studies in Knowledge
Engines,” in AFIPS Conference Proceedings of the 1978 National Computer Conference. Vol. 47.
Anaheim, California, 1978.

Fogel, Lawrence J., Alvin Owens, and Michael Walsh. Artificial Intelligence through Simulated
Evolution. New York: John Wiley & Sons, 1966.
Freud, Sigmund. The Basic Writings of Sigmund Freud. New York: The Modern Library, 1938.
Freud, Sigmund. Collected Papers. London: Hogarth Press and the Institute of Psychoanalysis, 1950.

Furth, Hans. Piaget and Knowledge. Englewood Cliffs, N.J.: Prentice-Hall, 1969.
Gould, Stephen Jay. The Panda‘s Thumb. New York: W. W. Norton. & Co., 1980.

Gould, Stephen Jay. The Mismeasure of Man. New York: W. W. Norton & Co., 1981
Grant, Verne. The Origin of Adaptation. New York: Columbia University Press, 1963.
Gray, Ada, ed. Darwiniana. Cambridge: Harvard University Press, 1963.
Greenberger, Martin, ed. Computers and the World of the Future. Cambridge: MIT Press, 1962.

Hitching, F. The Neck of the Giraffe, or Where Darwin Went Wrong. London: Pan Books, 1982

Hoage, R. J., and Larry Goldman. Animal Intelligence: Insights into the Animal Mind. Washington, D.C.:
Smithsonian Institution Press, 1986
Hookway, Christopher, ed. Minds, Machines and Evolution: Philosophical Studies. Cambridge:
Cambridge University Press, 1984.
Jacob, Francois. The Logic of Life. New York: Pantheon Books, 1973.

James, William. The Varieties of Religious Experience. New York: Collier Books, 1961.
Jung, Carl, et al. Man and His Symbols. Garden City, New York: Doubleday, 1964.

Jung, Carl. Memories Dreams and Reflections. Rev. ed. Ed. Aniela Jaffé and trans. Richard and Clara
Winston. New York: Pantheon Books, 1961

522
Kent, Ernest W. The Brains of Men and Machines. New York: McGraw-Hill Publications, 1981
Klix, Friedhart, ed. Human and Artificial Intelligence. Amsterdam: North-Holland Publishing Co., 1979.

McCorduck, Pamela. Machines Who Think. San Francisco: Freeman Press, 1979.
Mayr, Ernst. Animal Species and Evolution. Cambridge: Harvard University Press, 1963.

Metropolis, N., J. Howlett, and Gian-Carlo Rota, eds. A History of Computing in the Twentieth Century.
New York: Academic Press, 1980.
Minsky, Marvin. Society of Mind. New York: Simon & Schuster, 1985.
Morris, Desmond. The Naked Ape: A Zoologist's Study of the Human Animal. New York: McGraw-Hill,
1967

Piaget, Jean. The Psychology of Intelligence. London: Routledge & Kegan Paul, 1967. Originally
published in 1947
Pratt, Vernon. Thinking Machines: The Evolution of Artificial Intelligence. Oxford: Basil Blackwell, 1987.

Schank, Roger. The Cognitive Computer. Reading, Mass.: Addison-Wesley, 1984.

Shackle, G. L. S. Imagination and the Nature of Choice. Edinburgh: University of Edinburgh Press, 1979.

Simpson, George Gaylord. The Meaning of Evolution. The New American Library of World Literature
New York: A Mentor Book, 1951. Originally published in 1949
Singer, P. Animal Liberation. London: Cape Books, 1976
Stahl, Franklin W. The Mechanics of Inheritance. Englewood Cliffs, N.J.: Prentice-Hall, 1969.
Sternberg, Robert J., ed. Handbook of Human Intelligence. Cambridge: Cambridge University Press,
1982

Turkle, Sherry. The Second Self: Computers and the Human Spirit. New York: Simon and Schuster,
1984.

Von Neumann, John. The Computer and the Brain. New Haven: Yale University Press, 1958.
Warmier, Jean-Dominique. Computers and Human Intelligence. Englewood Cliffs, N.J.: Prentice-Hall,
1986.
Winner, Langdon. Autonomous Technology: Technics-out-of-Control as a Theme in Political Thought
Cambridge: MIT Press, 1977
Zadeh, Lofti. Information and Control. Vol. 8. New York: Academic Press, 1974.

Philosophical Roots
Anton, John D. Science and the Sciences in Plato. New York: EIDOS, 1980
Ayer, Alfred J. The Foundations of Empirical Knowledge. London: Macmillan and Co., 1964.
Ayer, J. Language, Truth, and Logic. New York: Dover Publications, n.d. Originally published in 1936.

Ayer, Alfred J., ed. Logical Positivism. New York: Macmillan Publishing Co., 1959.
Block, Irving, ed. Perspectives on the Philosophy of Wittgenstein. Cambridge: MIT Press, 1981
Boden, Margaret. Artificial Intelligence and Natural Man. New York: Basic Books, 1977.

Brittan, Gordon G., Jr. Kant’s. Theory of Science. Princeton: Princeton University Press, 1978.

Brumbaugh, R. S. Plato’s Mathematical Imagination. Bloomington: Indiana University Press, 1954.

Cassirer, Ernst. The Philosophy of the Enlightenment. Princeton: Princeton University Press, 1951

Cherniak, Christopher. Minimal Rationality. Cambridge: MIT Press, 1986.

Chomsky, Noam. Syntactic Structures. The Hague: Mouton, 1957

Churchland, Paul M. Matter and Consciousness. Cambridge: MIT Press, 1984

Clocksin, W. F., and C. S. Mellish. Programming in PROLOG. Berlin: Springer-Verlag, 1981

Cohen,|. Bernard. The Newtonian Revolution. Cambridge: Cambridge University Press, 1980
Cornford, Francis M. Plato’s Cosmology. London: Routledge and Kegan Paul, 1937

Davis, Philip J., and Reuben Hersh. Descartes’ Dream: The World According to Mathematics. San
Diego: Harcourt Brace Jovanovich, 1986.
Dennett, Daniel C. Brainstorms: Philosophical Essays on Mind and Psychology. Cambridge: MIT Press,
1978.

523
Dennett, Daniel C. Elbow Room: The Varieties of Free Will Worth Wanting. Cambridge: MIT Press,
1984.

Descartes, René. Discourse on Method, Optics, Geometty, and Meteorology. Indianapolis: Bobbs-
Merrill Co., 1956. Original, 1637.

Dodds, E. R. The Greeks and the Irrational. Berkeley: University of California Press, 1951
Dreyfus, Hubert L., ed. Husseri, Intentionality, and Cognitive Science. Cambridge: MIT Press, 1982.
Dreyfus, Hubert L. “Philosophic Issues in Artificial Intelligence.” Department of Humanities, MIT, 1967.

Dreyfus, Hubert L. What Computers Can't Do: The Limits of Artificial Intelligence. Rev. ed. New York:
Harper and Row, 1979. Original, 1972.
Einstein, Albert. Relativity: The Special and General Theory: A Popular Exposition. New York: Crown
Publishers, 1961

Findlay, J. N. Plato and Platonism: An Introduction. New York: Times Books, 1978.

Fodor, Jerry A. Representations: Philosophical Essays on the Foundations of Cognitive Science.

Cambridge: MIT Press, 1981

Fowler, D. H. The Mathematics of Plato’s Academy. Oxford: Clarendon Press, 1987.

Freud, Sigmund. Jokes and Their Relation to the Unconscious. Vol. 8 of Standard Edition of the
Complete Psychological Works of Sigmund Freud. London: Hogarth Press, 1957. Original, 1905.

Gay, Peter. The Enlightenment: An Interpretation. Vol. 1, The Rise of Modern Paganism. New York: W.
W. Norton and Co., 1966
Gay, Peter. The Enlightenment: An Interpretation. Vol. 2, The Science of Freedom. New York: W. W.
Norton and Co., 1969.
Gillispie, Charles. The Edge of Objectivity. Princeton: Princeton University Press, 1960.

Graubard, Stephen R., ed. The Artificial Intelligence Debate: False Starts, Real Foundations. Cambridge:
MIT Press, 1988.
Grossman, Reinhardt. Phenomenology and Existentialism: An Introduction. London: Routledge and K.
Paul, 1984.

Guthrie, W. K. C. A History of Greek Philosophy. 6 vols. Cambridge: Cambridge University Press,

1962-1981

Hankins, Thomas L. Science and the Enlightenment. Cambridge: Cambridge University Press, 1985.

Harel, David. Algorithmics: The Spirit of Computing. Menlo Park: Addison-Wesley Publishing Co., 1987.
Haugeland, John, ed. Mind Design. Cambridge: MIT Press, 1981
Heisenberg, Werner. Physics and Beyond: Encounters and Conversations. New York: Harper and Row,
1971

Helleman, Alexander, and Bryan Bunch. The Timetables of Science. New York: Simon and Schuster,
1988.
Hofstadter, Douglas R. Gédel, Escher, Bach: An Eternal Golden Braid. New York: Basic Books, 1979.

Hume, David. An Inquiry Concerning Human Understanding. Indianapolis: Bobbs-Merrill Co., 1955.
Original, 1748.
Jones, W. T. Kant and the Nineteenth Century. Vol. 4 of A History of Western Philosophy, 2nd ed. New
York: Harcourt Brace Jovanovich, 1975.

Jones, W. T. The Twentieth Century to Wittgenstein and Sartre. Vol. 5 of A History of Western
Philosophy, 2nd ed. New York: Harcourt Brace Jovanovich, 1975.

Kant, Immanuel. Critique of Pure Reason. New York: St. Martin's Press, 1929. Originally published in
1781

Kant, Immanuel. Prolegomena to Any Future Metaphysics. Indianapolis: Bobbs-Merrill Co., 1950.
Original, 1783
Kirk, G. S., J. E. Raven, and M. Schofield. The Presocratic Philosophers. Cambridge: Cambridge
University Press, 1983.
Knorr, Wilbur Richard. The Ancient Tradition of Geometric Problems. Boston: Birkhauser, 1986.
Lasserre, Francois. The Birth of Mathematics in the Age of Plato. New York: World Publishing Co.,
1964

s24
Leibniz, Gottfried Wilhelm. Philosophical Writings. Ed. G. H. R. Parkinson. London and Toronto: J. M.
Dent and Sons, 1973. .

Leibniz, Gottfried Wilhelm, and Samuel Clarke. The Leibniz-Clarke Correspondence. Ed. H. G
Alexander. Manchester: Manchester University Press, 1956.

Lloyd, G. E. R. Aristotle: The Growth and Structure of His Thought. Cambridge: Cambridge University
Press, 1968.
Lloyd, G. E. R. Early Greek Science: Thales to Aristotle. New York: W.W. Norton and Co., 1970.
McCorduck, Pamela. Machines Who Think. San Francisco: W. H. Freeman and Co., 1979.
Malcolm, Norman. Ludwig Wittgenstein: A Memoir, with a Biographical Sketch by Georg Henrik Von
Wright. Oxford: Oxford University Press, 1958
Mamdani, E. H., and B. R. Gaines. Fuzzy Reasoning and Its Applications. London: Academic Press,
1981.

Mayr, Otto. Authority, Liberty, and Automatic Machinery in Early Modern Europe. Baltimore: Johns
Hopkins University Press, 1986.
Mohr, Richard R. The Platonic Cosmology. Leiden: E. J. Brill, 1985.
Mumford, Lewis. The Myth of the Machine: Technics and Human Development. New York: Harcourt
Brace and World, 1966-1967.
Newmeyer, Frederick J. Linguistic Theory in America. 2nd ed. Orlando: Academic Press, 1986.
Newton, Isaac. Philosophiae Naturalis Principia Mathematica. 3rd ed. Ed. Alexander Koyré, |. Bernard
Cohen, and Anne Whitman. 2 vols. Cambridge: Harvard University Press, 1972. Original, 1726

Nussbaum, Martha. The Fragility of Goodness: Luck and Ethics in Greek Tragedy and Philosophy.
Cambridge: Cambridge University Press, 1986.
Plato. Epinomis. The Loeb Classical Library. Ed. W. R. M. Lamb. Vol. 8 New York: G. P. Putnam’s Sons,
1927.
Plato. Timaeus. Indianapolis: Bobbs-Merrill Co., 1959.

Rée, Jonathan. Descartes. New York: Pica Press, 1974.

Ringle, Martin D., ed. Philosophical Perspectives in Artificial Intelligence. Brighton: Harvester Press,
1979
Russell, Bertrand. The ABC of Relativity. 4th ed. London: Allen and Unwin, 1985. Original, 1925.
Russell, Bertrand. The Problems of Philosophy. New York: Oxford University Press, 1959

Sambursky, S. The Physical World of the Greeks. London: Routledge and Kegan Paul, 1963. Original,
1956.

Shapiro, Stuart C., ed. Encyclopedia of Artificial Intelligence. 2 vols. New York: John Wiley and Sons,
1987

Torrance, Steve, ed. The Mind and the Machine: Philosophical Aspects of Artificial Intelligence.
Chichester, England: Ellis Horwood, 1984.

Turkle, Sherry. The Second Self: Computers and the Human Spirit. New York: Simon and Schuster,
1984.

Vaina, Lucia, and Jaakko Hintikka, ed. Cognitive Constraints on Communication. Dordrecht: Reidel,
1985.

Weizenbaum, Joseph. Computer Power and Human Reason. San Francisco: W. H. Freeman and Co.,
1976.

Westfall, Richard. Never at Rest: A Biography of Isaac Newton. Cambridge: Cambridge University
Press, 1980.
Wittgenstein, Ludwig. Philosophical Investigations. New York: Macmillan Co., 1963
Wittgenstein, Ludwig. Tractatus Logico-Philosophicus. New York: Routledge and Kegan Paul, 1961
Original German published in 1921
Zeller, Eduard. Plato and the Older Academy. Reprint ed. New York: Russell and Russell, 1962
Originally published in 1888.

525
Mathematical Roots
Beth, E. W. Foundations of Mathematics. Amsterdam: North-Holland Publishing Co., 1959.

Bolter, J. David. Turing’s Man: Western Culture in the Computer Age. Chapel Hill: University of North
Carolina Press, 1984.
Bowden, B. W., ed. Faster than Thought. London: Pitman, 1953.
Church, Alonzo. /ntroduction to Mathematical Logic. Vol. 1. Princeton: Princeton University Press, 1956.

Dreyfus, Hubert. What Computers Can't Do. New York: Harper and Row, 1971.
Feigenbaum, Edward, and J. Feldman, eds. Computers and Thought. New York: McGraw-Hill, 1963.
Feigenbaum, Edward, and Pamela McCorduck. The Fifth Generation. Reading, Mass.: Addison-Wesley,
1983.
Hamming, R. W. Introduction to Applied Numerical Analysis. New York: McGraw-Hill, 1971.
Hodges, Andrew. Alan Turing: The Enigma. New York: Simon & Schuster, 1983.
Hofstadter, Douglas. Gédel, Escher, Bach: An Eternal Golden Braid. New York: Basic Books, 1979.
Kleene, Stephen Cole. /ntroduction to Metamathematics. New York: D. Van Nostrand, 1952.
Metropolis, N., J. Howlett, G. C. Rota, eds. A History of Computing in the Twentieth Century. New
York: Academic Press, 1980.

Minsky, Marvin. Computation: Finite and Infinite Machines. Englewood Cliffs: Prentice-Hall, 1967.

Russell, Bertrand. The Autobiography of Bertrand Russell: 1872-1914. New York: Bantam Books, 1956.
Russell, Bertrand. The Autobiography of Bertrand Russell: The Middle Years 1914-1944. New York:
Bantam Books, 1956.

Russell, Bertrand. Introduction to Mathematical Philosophy. New York: Macmillan, 1919

Russell, Bertrand. Mysticism and Logic. Garden City, N.Y.: Doubleday, 1957.
Russell, Bertrand. Principles of Mathematics. 2nd ed. Cambridge: Cambridge University Press, 1938.
Ist ed., 1903.

Schilpp, P. A., ed. The Philosophy of Bertrand Russell. Chicago: University of Chicago Press, 1944.
Van Heijenoort, Jean, ed. From Frege to Gédel. Cambridge: Harvard University Press, 1967.
Whitehead, Alfred N., and Bertrand Russell. Principia Mathematica. 3 vols. 2nd ed. Cambridge:
Cambridge University Press, 1925-1927

The Formula for Intelligence

Antébi, Elizabeth, and David Fishlock. Biotechnology: Strategies for Life. Cambridge: MIT Press, 1986.
Atkins, P. W. The Second Law. New York: Scientific American Books, 1984.
Barrow, John D., and Frank J. Tipler. The Anthropic Cosmological Principle. Oxford: Oxford University
Press, 1986.
Boole, George. An Investigation of the Laws of Thought on Which Are Founded the Mathematical
Theories of Logic and Probabilities. Peru, \lI.: Open Court Publishing Co., 1952. Original published in
1854

Botvinnik, M. M. Computers in Chess. New York: Springer-Verlag, 1984

Carver, Mead, and Lynn Conway. /ntroduction to VLS/ Systems. Reading, Mass.: Addison-Wesley,
1980

Chaitin, Gregory J. Algorithmic Information Theory. Cambridge: Cambridge University Press, 1987.
Crease, Robert P., and Charles C. Mann, The Second Creation. New York: Macmillan, 1986.
Dreyfus, Hubert. What Computers Can't Do. 2nd ed. New York: Harper and Row, 1971
Ebeling, Carl, All the Right Moves. Cambridge: MIT Press, 1987
Edelman, G. M. Neural Darwinism: The Theory of Neuronal Group Selection. New York: Basic Books,
1987

Einstein, Albert. Relativity: The Special and the General Theory. 17th ed. New York: Crown Publishers,
1961

526
Fadiman, Clifton, ed. Fantasia Mathematica: Being a Set of Stories, together with a Group of Oddments
and Diversions, All Drawn from the Universe of Mathematics. New York: Simon and Schuster, 1958.
Feigenbaum, E., and Avron Barr, eds. The Handbook of Artificial Intelligence. Vol 1. Los Altos, Calif.
William Kaufman, 1981

Feigenbaum, E. A., and J. Feldman, eds. Computers and Thought. New York: McGraw-Hill, 1963.

Freudenthal, Hans. Mathematics Observed. Trans. Stephen Rudolfer and |. N. Baker. New York:
MeGraw-Hill Book Co., 1967.

Frey, Peter, ed. Chess Skill in Man and Machine. 2nd ed. New York: Springer-Verlag, 1983.

Gamow, George. One, Two, Three, Infinity: Facts and Speculations of Science. New York: Bantam
Books, 1947.
Garey, Michael R., and David S. Johnson. Computers and Intractability. San Francisco: W. H. Freeman
1979.

Greenblatt, R. D., et. al. The Greenblatt Chess Program. Proceedings of the fall joint computer
conference. ACM, 1967.
Guillen, Michael. Bridges to Infinity: The Human Side of Mathematics. Los Angeles: Jeremy P. Tarcher,
1983.
Hawking, Stephen. A Brief History of Time. Toronto: Bantam Books, 1988.
Hildebrandt, Stefan, and Anthony Tromba. Mathematics and Optimal Form. New York: Scientific
American Books, 1985.
Hillis, Daniel. The Connection Machine. Cambridge: MIT Press, 1985.
Hofstadter, Douglas R. Godel, Escher, Bach: An Eternal Golden Braid. New York: Basic Books, 1979.
Hofstadter, Douglas R. Metamagical Themas: Questing for the Essence of Mind and Pattern. New
York: Basic Books, 1985.

Hsu, F. Two Designs of Functional Units for VLSI Based Chess Machines. Technical report. Computer
Science Department, Carnegie Mellon University, 1986.

Judson, Horace F. The Eighth Day of Creation. New York: Simon and Schuster, 1979.
Kasner, Edward, and James Newman. Mathematics and the Imagination. New York: Simon and
Schuster, 1940.
Kline, Morris. Mathematics and the Search for Knowledge. Oxford: Oxford University Press, 1985.

Luce, R. D., and H. Raiffa. Games and Decisions. New York: John Wiley & Sons, 1957
McCorduck, Pamela. Machines Who Think. San Francisco: W. H. Freeman, 1979.
McCulloch, Warren S. Embodiments of Mind. Cambridge: MIT Press, 1965.

Minsky, Marvin. Society of Mind. New York: Simon & Schuster, 1985.
Minsky, Marvin, and Seymour Papert. Perceptrons. Expanded edition. Cambridge: MIT Press, 1988.
Newell, A., and H. A. Simon. Human Problem Solving. Englewood Cliffs, N.J.: Prentice Hall, 1972

Nilsson, Lennart. The Body Victorious: The Illustrated Story of Our Immune System and other
Defences of the Human Body. Trans. Clare James. New York: Delacorte Press, 1985.

Pagels, Heinz R. Perfect Symmetry. New York: Simon & Schuster, 1985.

Rucker, Rudy. The Five Levels of Mathematical Reality. New York: Houghton-Mifflin, 1987.

Rumelhart, D. E., J. L. McClelland, and the PDP Research Group. Parallel Distributed Processing. Vols.
1 and 2. Cambridge: MIT Press, 1986.
Russell, Bertrand. ABC of Relativity. 4th ed. London: George Allen & Unwin, 1958.

Sagan, Carl. The Dragons of Eden. New York: Random House, 1977
Waterman, D. A., and F. Hayes-Roth, eds. Pattern-Directed Inference Systems.

Watson, J. D. The Double Helix. New York: Atheneum Publishers, 1968

Winograd, Terry, and Fernando Flores. Understanding Computers and Cognition: A New Foundation for
Design. Norwood, N.J.: Ablex Publishing, 1985
Winston, Patrick H. Artificial Intelligence. 2nd ed. Reading, Mass.: Addison-Wesley, 1984.

Winston, Patrick H., and K. A. Prendergast, eds. The A/ Business. Cambridge: MIT Press, 1984.

S27
Mechanical Roots
Augarten, Stan. Bit by Bit: An Illustrated History of Computers. New York: Ticknor and Fields, 1984.

Austrian, Geoffrey D. Herman Hollerith: Forgotten Giant of Information Processing. New York:
Columbia University Press, 1982.
Babbage, Henry Prevost. Babbage’s Calculating Engines. Charles Babbage Institute Reprint Series for
the History of Computing, vol. 2. Los Angeles: Tomash Publishers, 1982. Originally published in 1889.
Bashe, Charles J., Lyle R. Johnson, John H. Palmer, and Emerson W. Pugh. /BM’s Early Computers.
Cambridge: MIT Press, 1986.
Baxandall, D. Calculating Machines and Instruments. Rev. ed. London: Science Museum, 1975.
Originally published in 1926.
Bernstein, Jeremy. The Analytical Engine: Computers—Past, Present, and Future. Revised ed. New
York: William Morrow and Co., 1981.
Bolter, J. David. Turing’s Man: Western Culture in the Computer Age. Chapel Hill: University of North
Carolina Press, 1984
Buxton, H. W. Memoirs of the Life and Labours of the Late Charles Babbage, Esq., F.R.S. Ed. A.
Hyman. Los Angeles: Tomash, 1988.

Chapuis, Alfred, and Edmond Droz. Automata: A Historical and Technological Study. Neuchatel and
New York: Griffon, 1958.
Cipolla, Carlo M. Clocks and Culture, 1300-1700. London: Collins, 1967

Cohen, John. Human Robots in Myth and Science. London: Allen and Unwin, 1966
Dickinson, Henry W. Sir Samuel Morland, Diplomat and Inventor, 1625-1695. Cambridge: Heffer, 1970.

Drachman, A. G. The Mechanical Technology of Greek and Roman Antiquity. Madison: University of
Wisconsin Press, 1963.

Eames, Charles, and Ray Eames. A Computer Perspective. Cambridge: Harvard University Press, 1973.
Forbes, R. J. Studies in Ancient Technology. 9 vols. Leiden: E. J. Brill, 1955-1965.

Goldstine, Herman. The Computer from Pascal to von Neumann. Princeton: Princeton University Press,
1972.

Hindle, Brooke, and Steven Lubar. Engines of Change: The American Industrial Revolution, 1790-1860.
Washington, D.C.: Smithsonian Institution Press, 1986.

Horsburgh, Ellice Martin, ed. Modern Instruments and Methods of Calculation: A Handbook of the
Napier Tercentenary Celebration Exhibition. London: G. Bell and Sons, 1914.
Hyman, Anthony. Charles Babbage: Pioneer of the Computer. Oxford: Oxford University Press, 1982.
Landes, David S. Revolution in Time: Clocks and the Making of the Modern World. Cambridge: Harvard
University Press, 1983

McCorduck, Pamela. Machines Who Think. San Francisco: W. H. Freeman and Co., 1979.
Macey, Samuel L. Clocks and the Cosmos: Time in Western Life and Thought. Hamden: Archon Books,
1980.
Mayr, Otto. Authority, Liberty, and Automatic Machinery in Early Modern Europe. Baltimore: Johns
Hopkins University Press, 1986.
Metropolis, N., J. Howlett, and Gian-Carlo Rota, ed. A History of Computing in the Twentieth Century,
New York: Academic Press, 1980
Moon, Parry H. The Abacus: Its History, Its Design, Its Possibilities in the Modern World. New York:
Gordon and Breach, 1971

Needham, Joseph. Science and Civilisation in China. Vols. 2 and 4. Cambridge: Cambridge University
Press, 1956, 1965.

Neuberger, A. P. The Technical Arts and Sciences of the Ancients. London: Methuen, 1930.

Pascal, Blaise. Pensées. New York: E. P. Dutton and Co., 1932. Originally published in 1670

Price, Derek J. de Solla. Gears from the Greeks: The Antikythera Mechanism—A Calendar Computer
from circa 80 B.C. New York: Science History Publications, 1975.

Pullan,J. M. The History of the Abacus. London: Hutchinson, 1968.

Randell, Brian, ed. The Origins of Digital Computers: Selected Papers. New York: Springer-Verlag, 1975

528
Shurkin, Joel. Engines of the Mind: A History of the Computer. New York: W. W. Norton and Co., 1984.
Singer, C., E. J. Holmyard, A. R. Hall, and T. |. Williams, eds. A History of Technology. 5 vols. Oxford:
Oxford University Press, 1954-1958.
Stein, Dorothy. Ada, a Life anda Legacy. Cambridge: MIT Press, 1985.
Truesdell, L. E. The Development of Punch Card Tabulation in the Bureau of the Census, 1890-1940.
Washington, D.C.: Government Printing Office, 1965.
Usher, A. P. A History of Mechanical Inventions. 2nd ed. Cambridge: Harvard University Press, 1958
White, K. D. Greek and Roman Technology. London: Thames and Hudson, 1984

Electronic Roots
Barr, Avron, and Edward A. Feigenbaum, eds. The Handbook of Artificial Intelligence. Vols. 1-2. Los
Altos, Calif.: William Kaufmann, 1981-1982.
Bashe, Charles J., Lyle R. Johnson, John H. Palmer, and Emerson W. Pugh. /BM’s Early Computers.
Cambridge: MIT Press, 1986.

Boole, George. An Investigation of the Laws of Thought on Which Are Founded the Mathematical
Theories of Logic and Probabilities. Peru, \ll.: Open Court Publishing Co., 1952. Originally published in
1854.

Charniak, Eugene, and Drew McDermott. introduction to Artificial Intelligence. Reading, Mass.: Addison-
Wesley Publishing Company, 1985.
Cohen, Paul R., and Edward A. Feigenbaum, eds. The Handbook of Artificial Intelligence. Vol. 3. Los
Altos, Calif.: William Kaufmann, 1982.

Dreyfus, Hubert L. What Computers Can't Do: The Limits of Artificial Intelligence. Rev. ed. New York
Harper and Row, 1979.
Elithorn, A., and D. Jones, eds. Artificial and Human Thinking. San Francisco: Josey Bass, 1973.

Feigenbaum, E. A., and J. Feldman, eds. Computers and Thought. New York: McGraw-Hill, 1963.

Feigenbaum, E. A., and P. McCorduck. The Fifth Generation: Artificial Intelligence and Japan's Computer
Challenge to the World. Reading Mass.: Addison-Wesley Publishing Co., 1983.
Gardner, Howard. The Mind’s New Science: A History of the Cognitive Revolution. New York: Basic
Books, 1985.
Haugeland, John. Artificial Intelligence: The Very Idea. Cambridge: MIT Press, 1985.
Hodges, Andrew. Alan Turing: The Enigma. New York: Simon and Schuster, 1983
Hofstadter, Douglas. Godel, Escher, Bach: An Eternal Golden Braid. New York: Basic Books, 1979
Hopper, Grace Murray, and Steven L. Mandell. Understanding Computers. 2nd ed. St. Paul: West
Publishing Co., 1987.
Lammers, Susan. Programmers at Work. Redmond, Wash.: Microsoft Press, 1986.
Laurie, Peter. The Joy of Computers. Boston: Little, Brown and Co., 1983.
Levy, Steven. Hackers: Heroes of the Computer Revolution. New York: Dell, 1984.

Lundstrom, David E. A Few Good Men from Univac. Cambridge: MIT Press, 1987.

McCorduck, Pamela. Machines Who Think: A Personal Inquiry into the History and Prospects of Artificial
Intelligence. San Francisco: W. H. Freeman and Co., 1979.
Metropolis, N., J. Howlett, and G. C. Rota, eds. A History of Computing in the Twentieth Century. New
York: Academic Press, 1980.

Minsky, Marvin, ed. Semantic Information Processing. Cambridge: MIT Press, 1968.

Minsky, Marvin. The Society of Mind. New York: Simon and Schuster, 1985.
Morgan, B. Total to Date: The Evolution of the Adding Machine. London: Burroughs Machines, 1973.

Newell, Allen. intellectual Issues in the History of Artificial Intelligence. Pittsburgh: Camnegie-Mellon
University, 1982
Newell, A., and H. A. Simon. Human Problem Solving. Englewood Cliffs, N.J.: Prentice Hall, 1972

Nilsson, Nils J. Principles of Artificial Intelligence. Los Altos, Calif. Morgan Kaufman Publishers, 1980.

Peat, F. David. Artificial Intelligence: How Machines Think. New York: Baen Enterprises, 1985

529
 Randell, Brian. The Origins of Digital Computers. Berlin: Springer-Verlag, 1973.
Rich, Elaine. Artificial intelligence. New York: McGraw-Hill Book Co., 1983.
Rucker, Rudy. Mind Tools: The Five Levels of Mathematical Reality. Boston: Houghton Mifflin Co.,
1987.

Russell, Bertrand. Principles of Mathematics. 2nd ed. Cambridge: Cambridge University Press, 1938.
Bi

Savage, John E., Susan Magidson, and Alex M. Stein. The Mystical Machine. Reading, Mass.: Addison-
Wesley, 1986.
Scown, Susan J. The Artificial Intelligence Experience: An Introduction. Maynard, Mass.: Digital
Equipment Corp., 1985.

Siekmann, Jorg, and Graham Wrightson. Automation of Reasoning. Vol. 1, Classical Papers on
Computational Logic, 1957-1966. Berlin: Springer-Verlag, 1983.
Siekmann, Jorg, and Graham Wrightson. Automation of Reasoning Vol. 2, Classical Papers on Compu-
tational Logic, 1967-1970. Berlin: Springer-Verlag, 1983.
Simon, Herbert A. The Sciences of the Artificial. 2nd ed. Cambridge: MIT Press, 1981

Simon, Herbert A., and L. Siklossy, eds. Representation and Meaning: Experiments with Information
Processing Systems. Englewood Cliffs, N.J.: Prentice-Hall, 1972
Slater, Robert. Portraits in Silicon. Cambridge: MIT Press, 1987.

Tanimoto, Steven L. The Elements of Artificial Intelligence: An Introduction Using LISP. Rockville, Md.:
Computer Science Press, 1987.
Torrance, Steve, ed. The Mind and the Machine: Philosophical Aspects of Artificial Intelligence.
Chichester, England: Ellis Horwood Limited, 1984.
Waldrop, M. Mitchell. Man-Made Minds: The Promise of Artificial Intelligence. New York: Walker and
Co., 1987.

Webber, Bonnie Lynn, and Nils J. Nilsson, eds. Readings in Artificial Intelligence. Los Altos, Calif.:
Morgan Kaufmann Publishers, 1981

Wiener, Norbert. Cybernetics, or Control and Communication in the Animal and the Machine. 2nd ed
Cambridge: MIT Press, 1961
Winston, Patrick Henry. Artificial Intelligence. 2nd ed. Reading, Mass.: Addison-Wesley Publishing Co.,
1984.

Winston, Patrick Henry, and Richard Henry Brown. Artificial Intelligence: An MIT Perspective. Vol. 1,
Expert Problem Solving, Natural Language Understanding, Intelligent Computer Coaches, Representa-
tion, and Learning. Cambridge: MIT Press, 1979.

Winston, Patrick Henry, and Richard Henry Brown. Artificial Intelligence: An MIT Perspective. Vol. 2,
Understanding Vision, Manipulation, Computer Design, Symbol Manipulation. Cambridge: MIT Press,
1979.
Winston, Patrick H., and Karen A. Prendergast, eds. The A/ Business: The Commercial Uses of Artificial
Intelligence. Cambridge: MIT Press, 1984.

Pattern Recognition: The Search for Order

Barr, Avron, and Edward A. Feigenbaum, eds. The Handbook of Artificial Intelligence. Vols. 1 and 2. Los
Altos, Calif.: William Kaufmann, 1981 and 1982

Cater, John P. Electronically Hearing: Computer Speech Recognition. Indianapolis: Howard W. Sams
and Co., 1984.
Cater, John P. Electronically Speaking: Computer Speech Generation. Indianapolis: Howard W. Sams
and Co., 1983.
Church, Kenneth W. Phonological Parsing in Speech Recognition. Norwell, Mass.: Kluwer Academic
Publishers, 1987
Cohen, Paul R., and Edward A. Feigenbaum, eds. The Handbook of Artificial Intelligence. Vol. 3. Los
Altos, Calif.: William Kaufmann, 1982.

Denes, Peter B., and Elliot N. Pinson. The Speech Chain: The Physics and Biology of Spoken Language.
Bell Telephone Laboratories, 1963.
Fant, Gunnar. Speech Sounds and Features, Cambridge: MIT Press, 1973.
Fischler, Martin A., and Oscar Firschein. /ntelligence: The Eye, the Brain, and the Computer. Reading,
Mass.: Addison-Wesley Publishing Co., 1987 =

Fischler, Martin A., and Oscar Firschein. Readings in Computer Vision: Issues, Problems, Principles,
and Paradigms. Los Altos, Calif: Morgan Kaufmann Publishers, 1987.

Flanagan, James L. Speech Analysis Synthesis and Perception. 2nd expanded ed. Berlin: Springer-
Verlag, 1972.
Frauenfelder, Uli H., and Lorraine Komisarjevsky Tyler, eds. Spoken Word Recognition. Cambridge: MIT
Press, 1987.

Geissler, Hans-Georg, et al. Modern Issues in Perception Amsterdam: North-Holland Publishing Co.,
1983.
Grimson, W. Eric L., From Images to Surfaces. Cambridge: MIT Press, 1981
Grimson, W. Eric L. and Ramesh Patil, eds. A/ in the 1980s and Beyond. Cambridge: MIT Press, 1987.

Hillis, W. Daniel. The Connection Machine. Cambridge: MIT Press, 1985.

Hoel, Paul G., Sidney C. Port, and Charles J. Stone. Introduction to Stochastic Processes. Boston
Houghton Mifflin Co., 1972
Hofstadter, Douglas. Metamagical Themas. New York: Basic Books, 1985
Hor, Berthold Klaus Paul. Robot Vision. Cambridge: MIT Press, 1986.
Hubel, David. Eye, Brain, and Vision. New York: W. H. Freeman, 1988.

Jamieson, Leah H., Dennis Gannon, and Robert J. Douglass, eds. The Characteristics of Parallel
Algorithms. Cambridge: MIT Press, 1987.
Kohonen, Teuvo. Self-Organization and Associative Memory. Berlin: Springer-Verlag, 1984.
Lea, Wayne A., ed. Trends in Speech Recognition. Englewood Cliffs, N.J.: Prentice-Hall, 1980.
Lim, Jae S., ed. Speech Enhancement. Englewood Cliffs, N.J.: Prentice-Hall, 1983.

McClelland, James, David Rumelhart, and the PDP Research Group. Parallel Distributed Processing.
Vol. 2. Cambridge: MIT Press, 1986.

Markov, A. A Theory of Algorithms. National Academy of Sciences, USSR

Marr, David. Vision. San Francisco: W. H. Freeman, 1982.
Massaro, Dominic W., et al. Letter and Word Perception: Orthographic Structure and Visual Processing
in Reading. Amsterdam: North-Holland Publishing Co., 1980.

Meisel, William S. Computer-Oriented Approaches to Pattern Recognition. New York: Academic Press,
1972.

Miller, Richard K. Parallel Processing: The Technology of Fifth Generation Computers. Madison, Ga.
SEAI Technical Publications, 1985.

Minsky, Marvin, and Seymour Papert. Perceptrons. Revised ed. Cambridge: MIT Press, 1988.

Myers, Terry, John Laver, and John Anderson, eds. The Cognitive Representation of Speech. Amster-
dam: North-Holland Publishing Co., 1981.

Potter, Jerry L., ed. The Massively Parallel Processor. Cambridge: MIT Press, 1985.
Rabiner, Lawrence R., and Ronald W. Schafer. Digital Processing of Speech Signals. Englewood Cliffs,
N.J.: Prentice-Hall, 1978.

Radford, Andrew. Transformational Syntax: A Student’s Guide to Chomsky’s Extended Standard

Theory. Cambridge: Cambridge University Press, 1981
Rock, Irvin. Perception. New York: Scientific American Books, 1984.

Rumelhart, David, James McClelland, and the PDP Research Group. Parallel Distributed Processing.
Vol. 1. Cambridge: MIT Press, 1986.
Sacks, Oliver. The Man Who Mistook His Wife for a Hat and Other Clinical Tales. New York: Harper and
Row, 1985.

Uliman, Shimon. The Interpretation of Visual Motion. Cambridge: MIT Press, 1979.

Watanabe, Satosi. Pattern Recognition: Human and Mechanical. New York: John Wiley and Sons, 1985
Winston, Patrick H. Artificial Intelligence. Reading, Mass.: Addison-Wesley, 1984.
Winston, Patrick H., ed. The Psychology of Computer Vision. New York: McGraw-Hill, 1975.

S31
 Zimmermann, Hans J. Fuzzy Set Theory and Its Applications. Hingham, Mass.: Kluwer-Nijhoff Publish-
ing, 1985
Zue, Victor W., Francine R Chen, and Lori Lamel. Speech"Spectrogram Reading: An Acoustic Study of
English Words and Sentences. MIT special summer course, 1982

The Search for Knowledge

Asimov, Isaac, and Karen Frenkel. Robots: Machines in Man’s Image. New York: Harmony Books,
1985.
Ayres, Robert U., et al. Robotics and Flexible Manufacturing Technologies: Assessment, Impacts, and
Forecast. Park Ridge, N.J.: Noyes Publications, 1985.

Bara, Bruno G., and Giovanni Guida, eds. Computational Models of Natural Language Processing.
Fundamental Studies in Computer Science, vol. 9. Amsterdam, The Netherlands: North-Holland
Publishing Co., 1984.

Barr, Avron, and Edward A. Feigenbaum, eds. The Handbook of Artificial Intelligence. Vols. 1 and 2. Los
Altos, Calif.: William Kaufmann, 1981 and 1982.

Beni, Gerardo, and Susan Hackwood, eds. Recent Advances in Robotics. New York: John Wiley and
Sons, 1985.

Berwick, Robert C. The Acquisition of Syntactic Knowledge. Cambridge: MIT Press, 1985.

Bobrow, Daniel G., and A. Collins, eds. Representation and Understanding. New York: Academic Press,
1975.
Brachman, Ronald J., and Hector J. Levesque, eds. Readings in Knowledge Representation. Los Altos,
Calif: Morgan Kaufmann Publishers, 1985.
Brady, Michael, et al., eds. Robot Motion: Planning and Control. Cambridge: MIT Press, 1982.
Brady, Michael, Lester A. Gerhardt, and Harold F. Davidson, eds. Robotics and Artificial Intelligence.
NATO ASI series. Series F, Computer and Systems Sciences, vol. 2. Berlin: Springer-Verlag, 1984.
Churchland, Paul M. Matter and Consciousness. Revised ed. Cambridge: MIT Press, 1988.
Clancey, William J., and Edward H. Shortliffe, eds. Readings in Medical Artificial Intelligence: The First
Decade. Reading, Mass.: Addison-Wesley Publishing Co., 1984.

Cohen, Paul R., and Edward A. Feigenbaum, eds. The Handbook of Artificial Intelligence. Vol. 3. Los
Altos, Calif.: William Kaufmann, 1982

Critchlow, Arthur J. Introduction to Robotics. New York: Macmillan Publishing Co., 1985.

Davis, R., and D. B. Lenat. Knowledge-Based Systems in Artificial Intelligence. New York: McGraw-Hill,
1980.

DM Data. The Artificial Intelligence Directory: A Comprehensive Listing of the Artificial Intelligence
Companies. Third annual edition. Scottsdale, Ariz.: DM Data, 1987.

Fahiman, Scott E. NETL: A System for Representing and Using Real-World Knowledge. Cambridge:
MIT Press, 1979.

Feigenbaum, Edward, and Pamela McCorduck. The Fifth Generation. Reading, Mass.: Addison-Wesley,
1983

Feigenbaum, Edward, Pamela McCorduck, and Penny Nii. The Rise of the Expert Company. Reading,
Mass.: Addison-Wesley, 1988
Frey, Peter, ed. Chess Skill in Man and Machine. 2nd ed. New York: Springer-Verlag, 1983.

Gardner, Anne von der Lieth. An Artificial Intelligence Approach to Legal Reasoning. Cambridge: MIT
Press, 1987
Genesereth, Michael R., and Nils J. Nilsson, Logical Foundations of Artificial Intelligence. Los Altos,
Calif.: Morgan Kaufmann Publishers, 1987.
Grimson, W. Eric L., and Ramesh Patil, eds, A/ in the 1980s and Beyond. Cambridge: MIT Press, 1987.
Harmon, Paul, and David King. Expert Systems: Artificial Intelligence in Business. New York: John
Wiley & Sons, 1985.
Harris, Mary D. Introduction to Natural Language Processing. Reston, Va.: Reston Publishing Co., 1985.

Haugeland, John, ed. Mind Design. Cambridge: MIT Press, 1981

S32
Hayes-Roth, Frederick, D. A. Waterman, and D. B. Lenat, eds. Building Expert Systems. Reading,
Mass.: Addison-Wesley, 1983. .

Hewett, Julian, and Ron Sasson. Expert Systems 1986. Vol. 1, USA and Canada. London: Ovum, 1986.

Jackendoff, Ray. Semantics and Cognition. Cambridge: MIT Press, 1983.

Johnson, L., and E. T. Keravnou. A Guide to Expert Systems Technology. Abacus Press, 1984.
Johnson, Mark, and George Lakoff. Metaphors We Live By. Chicago: University of Chicago Press, 1980
Johnson, Tim. Natural Language Computing: The Commercial Applications. London: Ovum, 1985.

Kahneman, Daniel, Paul Slovic, and Amos Tversky, eds. Judgement under Uncertainty: Heuristics and
Biases. Cambridge: Cambridge University Press, 1982.

Kuno, Susumu. Functional Syntax: Anaphora, Discourse, and Empathy. Chicago: University of Chicago
Press, 1987.

Kuno, Susumu. The Structure of the Japanese Language. Cambridge: MIT Press, 1973.
Li, Deyi. A PROLOG Database System. Letchworth, England: Research Studies Press, 1984
Lindsay, R. K., et al. Applications of Artificial Intelligence for Organic Chemistry: The Dendral Project.
New York: McGraw-Hill, 1980.

Marsh, Alton K. Guide to Defense and Aerospace Expert Systems. Arlington, Va.: Pasha Publications,
1986.
Martin, James, and Steven Oxman. Building Expert Systems: A Tutorial. Englewood Cliffs, N.J
Prentice-Hall, 1988
Mason, Matthew T., and J. Kenneth Salisbury, Jr. Robot Hands and the Mechanics of Manipulation.
Cambridge: MIT Press, 1985.

Merrett, T. H. Relational Information Systems. Reston, Va.: Reston Publishing Co., 1984.
Michalski, Ryszard S., Jaime G. Carbonell, and Tom M. Mitchell, eds. Machine Learning: An Artificial
Intelligence Approach. Vol. 2. Los Altos, Calif.: Morgan Kaufmann Publishers, 1986.
Miller, Richard K. Computers for Antificial Intelligence: A Technology Assessment and Forecast.
Madison, Ga.: SEAI Technical Publications, 1986
Minsky, Marvin, ed. Semantic Information Processing. Cambridge: MIT Press, 1968.
Minsky, Marvin. Society of Mind. New York: Simon & Schuster, 1986.
Newell, Allen. “Unified Theories of Cognition.” The William James Lectures. Psychology Department,
Harvard University, 1987.

Newauist, Harvey P., Ill, ed. A/ Trends ‘87: A Comprehensive Annual Report on the Artificial Intelli-
gence Industry. Scottsdale, Ariz.: DM Data, 1987
Osherson, Daniel N., Michael Stob, and Scott Weinstein. Systems That Learn: An Introduction to
Learning Theory for Cognitive and Computer Scientists. Cambridge: MIT Press, 1986.
Raibert, Marc H. Legged Robots That Balance. Cambridge: MIT Press, 1986.

Rauch-Hindin, Wendy. Artificial Intelligence in Business, Science, and Industry Englewood Cliffs, N.J.
Prentice-Hall, 1986.
Reichardt, Jasia. Robots: Fact, Fiction, and Prediction. Middlesex, England: Penguin Books, 1978.

Schank, Roger C. Dynamic Memory: A Theory of Reminding and Learning in Computers and People.
Cambridge: Cambridge University Press, 1982
Schank, Roger C., and Kenneth Mark Colby, eds. Computer Models of Thought and Language. San
Francisco: W. H. Freeman and Co., 1973.
SEAI Technical Publications, Fifth Generation Computers: A Report on Major International Research
Projects and Cooperatives. Madison, Ga.: SEAI Technical Publications, 1985
Shortliffe, E. MYCIN: Computer-Based Medical Consultations. New York: American Elsevier, 1976

Sterling, Leon, and Ehud Shapiro. The Art of Prolog: Advanced Programming Techniques. Cambridge
MIT Press, 1986.
Toepperwein, L. L., et al. Robotics Applications for Industry: A Practical Guide. Park Ridge, N.J.: Noyes
Data Corporation, 1983
Tomita, Masaru. Efficient Parsing for Natural Language: A Fast Algorithm for Practical Systems.
Hingham, Mass.: Kluwer Academic Publishers, 1986.

533
Walker, Terri C., and Richard K. Miller. Expert Systems 1986: An Assessment of Technology and
Applications. Madison, Ga.: SEAI Technical Publications, 1986.
Wallace, Mark. Communicating with Databases in Natural Languages. Chichester, England: Ellis
Horwood, 1984.
Weiss, Sholom M., and Casimir A Kulikowski. A Practical Guide to Designing Expert Systems. Totowa,
N.J.: Rowman and Allanheld, 1984.
Winograd, Terry. Understanding Natural Language. New York: Academic Press, 1973.

Winston, Patrick H., and Karen Prendergast, eds. The A/ Business. Cambridge: MIT Press, 1984.

The Science of Art

Barrett, Edward. Text, Context, and Hypertext. Cambridge: MIT Press, 1988.
Barstow, David R., Howard E. Shrobe, and Erik Sandewell, ed. Interactive Programming Environments.
New York: McGraw-Hill Company, 1984.
Bateman, Wayne. Introduction to Computer Music. New York: John Wiley and Sons, 1980.
Berlekamp, Elwyn R., John H. Conway, and Richard K. Guy. Winning Ways for Your Mathematical
Plays. London: Academic Press, 1982
Bertin, Jacques. Semiology of Graphics: Diagrams, Networks, Maps. Madison: University of Wisconsin
Press, 1983

Chamberlin, Hal. Musical Applications of Microprocessors. 2nd ed. Hasbrouck Heights, N.J.: Hatden
Book Co., 1985.
Davies, Roy, ed. Intelligent Information Systems: Progress and Prospects. Chichester: Ellis Horwood,
1986
Davis, Philip J., and Reuben Hersh. Descartes’ Dream: The World According to Mathematics. San
Diego: Harcourt Brace Jovanovich, 1986.
Dawkins, Richard. The Blind Watchmaker: Why the Evidence of Evolution Reveals a Universe without
Design. New York: W. W. Norton & Company, 1987.
Dewdney, A. K. The Armchair Universe: An Exploration of Computer Worlds. New York: W. H.
Freeman and Co., 1988

Durrett, H. John, ed. Color and the Computer. Orlando, Fla.: Academic Press, 1987.
Enderle, G., K. Kansky, and G. Pfaff. Computer Graphics Programming: GKS, the Graphics Standard.
Berlin: Springer-Verlag, 1984
Foley, J. D., and A. Van Dam. Fundamentals of Interactive Computer Graphics. Reprint ed. Reading,
Mass.: Addison-Wesley, 1984.
Franke, Herbert W. Computer Graphics—Computer Art. 1st ed. London: Phaidon Press, 1971. 2nd ed.,
Berlin: Springer-Verlag, 1985
Glassner, Andrew S. Computer Graphics User's Guide. Indianapolis: Howard W. Sams and Co., 1984.
Gleick, James. Chaos: Making a New Science. New York: Viking Penguin, 1987.
Goodman, Cynthia. Digital Visions: Computers and Art. New York: Harry N. Abrams, 1987,
Graubard, Stephen R., ed. The Artificial Intelligence Debate: False Starts, Real Foundations. Cambridge:
MIT Press, 1988.

Greenberg, Donald, Aaron Marcus, Allen H. Schmidt, and Vernon Gorter. The Computer Image:
Applications of Computer Graphics. Reading, Mass.: Addison-Wesley Publishing Co., 1982
Langley, Pat, et al. Scientific Discovery: Computational Explorations of the Creative Processes.
Cambridge: MIT Press, 1987

Leavitt, Ruth, ed. Artist and Computer. New York: Harmony Books, 1976.

Magnenat-Thalmann, Nadia, and Daniel Thalmann. Computer Animation. New York: Springer-Verlag,
1985,
Mandelbrot, Benoit B. The Fractal Geometry of Nature. New York: W. H. Freeman and Co., 1983.

Mandelbrot, Benoit B. Fractals: Form, Chance, and Dimension. San Francisco: W. H. Freeman and Co.,
1977

Markle, Sandra, and William Markle. /n Search of Graphics: Adventures in Computer Art. New York:
Lothrop, Lee and Shepard Books, 1985.

534
Mathews, Max V. The Technology of Computer Music. Cambridge: MIT Press, 1969.
Morgan, Christopher P., ed. The “Byte” Book of Computér Music. Peterborough, N.H.: Byte Books,
1979.

Nallin, Walter E. The Musical Idea: A Consideration of Music and Its Ways. New York: Macmillan Co.,
1968.
Peitgen, H.-O., and P. H. Richter. The Beauty of Fractals: Images of Complex Dynamical Systems.
Berlin: Springer-Verlag, 1986.

Peterson, Dale. Genesis II: Creation and Recreation with Computers. Reston, Va.: Reston Publishing
Co., 1983.

Peterson, Ivars. The Mathematical Tourist: Snapshots of Modern Mathematics. San Francisco: W. H.
Freeman, 1988.
Pierce, John R. The Science of Musical Sound. New York: Scientific American Books, 1983
Poundstone, William. The Recursive Universe: Cosmic Complexity and the Limits of Scientific Knowl-
edge. Chicago: Contemporary Books, 1985.
Prueitt, Melvin L. Art and the Computer. New York: McGraw-Hill Book Co., 1984.

RACTER. The Policeman‘s Beard Is Half Constructed: Computer Prose and Poetry by RACTER. New
York: Warner Books, 1984.

Roads, Curtis, ed. Composers and the Computer. Los Altos, Calif.: William Kaufmann, 1985.
Roads, Curtis, ed. The Music Machine: Selected Readings from “Computer Music Journal.” Cambr-
idge: MIT Press, 1988
Roads, Curtis, and John Strawn, ed. Foundations of Computer Music. Cambridge: MIT Press, 1985.

Schank, Roger. The Creative Attitude: Learning to Ask and Answer the Right Questions. New York
Macmillan Publishing Co., 1988.
Shapiro, Stuart C., ed. Encyclopedia of Artificial Intelligence. 2 vols. New York: John Wiley and Sons,
1987.
Time-Life Books editors. Computer Images. Alexandria, Va.: Time-Life Books, 1986
Tjepkema, Sandra L. A Bibliography of Computer Music: A Reference for Composers. lowa City:
University of lowa Press, 1981

Toffoli, Tommaso, and Norman Margolis. Cellular Automata Machines:A New Environment for
Modeling. Cambridge: MIT Press, 1987
Wawrzynek, John. VLS/ Concurrent Computation for Music Synthesis. California Institute of Technol-
ogy, 1987.

Weinstock, Neal. Computer Animation. Reading, Mass.: Addison-Wesley Publishing Co., 1986.
Wilson, Mark. Drawing with Computers: The Artist’s Guide to Computer Graphics. New York: Perigee
Books, 1985

Visions
Agha, Gul. Actors. Cambridge: MIT Press, 1987.
Asimoy, |saac. Robots of Dawn. New York: Doubleday & Co., 1983.

Botvinnik, M. M. Computers in Chess: Solving Inexact Search Problems. Trans. ArthurA. Brown. New
York: Springer-Verlag, 1984.

Brand, Stewart. The Media Lab: Inventing the Future at MIT. New York: Viking Penguin.

Burns, Alan. The Microchip: Appropriate or Inappropriate Technology? New York: John Wiley & Sons,
1981.
Chaffee, David. The Rewiring of America: The Fiber Optics Revolution. Boston: Boston Academic
Press, 1988.
Churchland, Paul M. Matter and Consciousness: A Contemporary Introduction to the Philosophy of
Mind. Cambridge: MIT Press, 1988.
Deaver, B., and John Ruvalds., eds. Advances in Superconductivity. New York: Plenum Press, 1982
Dreyfus, Hubert L. What Computers Can't Do: A Critique of Artificial Reasoning. New York: Harper &
Row, 1972.

S35
Evolution, Games, and Learning: Models for Adaptation in Machines and Nature. Proceedings of the
Fifth Annual International Conference of the Center for Nonlinear Studies at Los Alamos, N.M., May
20-24, 1985. Amsterdam: North-Holland, 1986.

Fjermedal, Grant. The Tomorrow Makers: A Brave New World of Living-Brain Machines. New York:
Macmillan Publishing Co., 1986.
Forester, Tom. High-Tech Society: The Story of the Information Technology Revolution. Cambridge:
MIT Press, 1987

Forester, Tom. The Materials Revolution. Cambridge: MIT Press, 1988.

Forester, Tom. The Information Technology Revolution. Cambridge: MIT Press, 1985.
Hazen, Robert M. The Breakthrough: The Race for the Superconductor. New York: Summit Books,
1988.
Hofstadter, Douglas R. Metamagical Themas: Questing for the Essence of Mind and Pattern. New
York: Basic Books, 1985.

Hofstadter, Douglas, and Daniel C. Dennett. The Mind's I: Fantasies and Reflections on Self and Soul.
New York: Basic Books, 1981.

Kobayashi, Koji. Computers and Communication:A Vision of C&C. Cambridge: MIT Press, 1986.
Lambert, Steve, and Suzanne Ropiequet, eds. CD ROM, the New Papyrus: The Current and Future
State of the Art. Redmond, Wash.: Microsoft Press, 1986.
Levy, David N. L. All about Chess and Computers. Rockland, Md.: Computer Science Press, 1982
Levy, David N. L. Computer Games. New York: Springer-Verlag, 1988.
Levy, Steven. Hackers: Heroes of the Computer Revolution. Garden City, N.Y.: Doubleday, Anchor
Press, 1984
Lord, Norman W. Advanced Computers: Parallel and Biochip Processors. Ann Arbor, Mich.: Ann Arbor
Science, Butterworth Group, 1983.
McLuhan, Marshall. Understanding Media: The Extension of Man. New York: McGraw-Hill, 1964.

Michalski, R. Z.,J. J. Carbonell, andT. M. Mitchell, eds. Machine Learning: An Artificial Intelligence
Approach. Palo Alto: Tioga Publishing, 1983.
Minsky, Marvin. Society of Mind. New York: Simon & Schuster, 1985.
Newborn, Monroe. Computer Chess. New York: Academic Press, 1975.

Schank, Roger, and Peter G. Childers. The Cognitive Computer: On Language, Learning, and Artificial
Intelligence. Reading, Mass.: Addison-Wesley, 1984.
Seippel, Robert G. Fiber Optics. Reston, Va.: Reston Publishing Co., 1984.
Simon, Randy, and Andrew Smith. Superconductors: Conquering Technology's New Frontier. New
York: Plenum Press, 1988
Smith, Donald N., and Peter Heytler, Jr. Industrial Robots: Forecasts and Trends. Dearborn, Mich.:
Society of Manufacturing Engineers, 1985.
Winkler, Nels, and Iben Browning. Robots on Your Doorstep: A Book about Thinking Machines.
Portland, Oregon: Robotics Press, 1978.

The Impact On...

Abelson, Harold, and Andrea diSessa. Turtle Geometry: The Computer as a Medium for Exploring
Mathematics. Cambridge: MIT Press, 1980.

Andriole, Stephen J., and Gerald W. Hople, eds. Defense Applications of Artificial Intelligence.
Lexington, Mass.: Lexington Books, 1988.
Bellin, David, and Gary Chapman, eds. Computers in Battle—Will They Work? Boston: Harcourt Brace
Jovanovich, 1987.

Bernstein, Jeremy. The Analytical Engine: Computers—Past, Present, and Future. New York: Random
House, 1963.
Brand, Stewart. The Media Lab: Inventing the Future at MIT. New York: Viking, 1987.

Chandler, Daniel. Young Learners and the Microcomputer. New York: Taylor and Francis, 1983.
Chen, Milton, and William Paisley, eds. Children and Microcomputers: Research on the Newest
Medium. Beverly Hills: Sage Publications, 1985.

S36
Cotellessa, Robert G., ed. Identifying Research Areas in the Computer Industry in 1995. Park Ridge,
N.J.: Noyes Publications, 1984.
Deakon, Rose. Women and Computing: The Golden Opportunity. London: Macmillan Publishers, 1984.
Dertouzos, Michael, and Joel Moses, eds. The Computer Age: A Twenty Year View. Cambridge: MIT
Press, 1979.
Din, Allan M., ed. Arms and Artificial Intelligence: Weapons and Arms Control of Applications of
Advanced Computing. Oxford: Oxford University Press, 1987.
Feingold, S. Norman, and Norma Reno Miller. Emerging Careers: New Occupations for the Year 2000
and Beyond. Vol. 1. Garrett Park, Md.: Garrett Park Press, 1983.
Fjermedal, Grant. The Tomorrow Makers: A Brave New World of Living Brain Machines. New York:
Macmillan Publishers, 1986.
Forester, Tom. High-Tech Society: The Story of the Information Technology Revolution. Cambridge:
MIT Press, 1987
Forester, Tom. The Materials Revolution. Cambridge: MIT Press, 1988.
Gerver, Elisabeth. Humanizing Technology: Computers in Community Use and Adult Education. New
York: Plenum Press, 1986
Gilder, George. Microcosm. New York: Simon & Schuster, 1989.

Goldenberg, E. Paul, and Wallace Feurzeig. Exploring Language with Logo. Cambridge: MIT Press,
1987.

Greenberger, Martin, ed. Computers and the World of the Future. Cambridge: MIT Press, 1962
Grimson, W. Eric L., and Ramesh S. Patil, eds. Al in the 1980s and Beyond: An MIT Survey. Cambridge:
MIT Press, 1987
Haugeland, John. Artificial Intelligence: The Very Idea. Cambridge: MIT Press, 1985.

Hawkridge, David. New Information Technology in Education. London: Croom Helm, 1983
Horn, Berthold. Robot Vision. Cambridge: MIT Press, 1986.

Inose, Hiroshi, and John R. Pierce. Information Technology and Civilization. New York: W. H. Freeman
&Co., 1984.

Inose, Hiroshi, et al. Road Traffic Control. Tokyo: University of Tokyo Press, 1975.
Kearsley, Greg, ed. Artificial Intelligence and Instruction: Applications and Methods. Reading: Addison-
Wesley Publishing Co., 1987.
Kidder, Tracy. The Soul of a New Machine. London: Allen Lane, 1982.

Lambert, Steve, and Suzanne Ropiequet, eds. CD ROM, the New Papyrus. Redmond, Wash.
Microsoft Press, 1986.

Langley, Pat, et al. Scientific Discovery: Computational Explorations of the Creative Processes.
Cambridge: MIT Press, 1987.

Lapp, Ralph E. The New Priesthood: The Scientific Elite and the Uses of Power. New York: Harper &
Row, 1985.

Laver, Murray. Computers and Social Change. Cambridge: Cambridge University Press, 1980

Leontief, Wassily W. The Structure of the American Economy, 1919-1939. White Plains, N.Y.
International Arts & Sciences Press, 1941
Leontief, Wassily W., and Faye Duchin, eds. The Future Impact of Automation on Workers. Oxford
Oxford University Press, 1986.
Man-Computer Interactive Research: MACINTER-I. Proceedings of the First Network Seminar of the
International Union of Psychological Science on Man-Computer Interaction Research, Berlin, October
16-19, 1984. New York: Elsevier Science Pub. Co., 1986
Mason, Matthew T., and J. Kenneth Salisbury. Robot Hands and the Mechanics of Manipulation
Cambridge: MIT Press, 1985.
Megarry, Jacquetta. Inside Information: Computers, Communications, and People. London: British
Broadcasting Corp., 1985.
Minsky, Marvin. Society of Mind. New York: Simon & Schuster, 1986.

Moreau, R. The Computer Comes of Age: The People, the Hardware, and the Software. Trans. J.
Howlett. Cambridge: MIT Press, 1986.

S37
 Mowshowitz, Abbe, ed. Inside Information: Computers in Fiction. Reading, Mass.: Addison Wesley
and suggest
Bibliography

Pub. Co. 1977.

Newborn, Monroe. Computer Chess. New York: Academié Press, 1975.

Nova, Simon, and Alain Mince. The Computerization of Society. Cambridge: MIT Press, 1980.

Pea, R. D., and D. M. Kurland. New Ideas in Psychology. Vol. 1. Elmsford, N.Y.: Pergammon, 1983.
Papert, Seymour. Mindstorms: Computers, Children, and Powerful Ideas. New York: Basic Books,
1980.
Piaget, Jean. The Psychology of Intelligence. London: Routledge & Kegan Paul, 1967.
Pool, Ithiel de Sola. Technologies of Freedom: On Free Speech in an Electronic Age. Cambridge:
Harvard University Press, 1983
Rochester, Jack B., and John Gantz. The Naked Computer: A Layperson’s Almanac of Computer Lore,
Wizardry, Personalities, Memorabilia, World Records, Mind Blowers, and Tomfoolery. New York:
William Morrow Co., 1983
Rothchild, Jean, ed. Machina ex Dea: Feminist Perspectives on Technology. New York: Pergammon
Press, 1982.
Roszak, Theodore. The Cult of Information: The Folklore of Computers and the True Art of Thinking.
New York: Pantheon Books, 1986.
Schechler, G., ed. Information Retrieval: A Critical Review. Washington, D.C.: Thompson Books, 1967.
Shackle, G. L. S. Imagination and the Nature of Choice. Edinburgh: Edinburgh University Press, 1979.

Stern, Virginia W., and Martha Ross Redden. Technology for Independent Living: Proceedings of the
1980 Workshops on Science and Technology for the Handicapped. Washington, D.C.: American
Association for the Advancement of Science, 1982.
Torrero, Edward A., ed. Next-Generation Computers. New York: IEEE Press, 1985.
Traub, Joseph F., ed. Cohabiting with Computers. Los Altos, Calif.: William Kaufmann, 1985.
Turkle, Sherry. The Second Self. New York: Simon & Schuster, 1984

U.S. Defense Intelligence Agency. Soviet Military Power, 1986. Washington, D.C.: U.S. Government
Printing Office, 1986.
Vallée, Jacques. The Network Revolution: Confessions of a Computer Scientist. Hammondsworth,
England: Penguin Books, 1984.
Warnier, Jean-Dominique. Computers and Human Intelligence. Englewood Cliffs, N.J.: Prentice-Hall,
1986

Weizenbaum, Joseph. Computer Power and Human Reason: From Judgment to Calculation. San
Francisco: W. H. Freeman and Co., 1976.

Wieseltier, Leon. Nuclear War, Nuclear Peace. New York: Holt, Rinehart and Winston, 1983.
Winner, Langdon. Autonomous Technology: Technics-out-of-Control as a Theme in Political Thought.
Cambridge: MIT Press, 1977
Winston, Patrick H., and Karen A. Prendergast, eds. The A/ Business: Commercial Uses of Artificial
Intelligence. Cambridge: MIT Press, 1984.

Yazdani, Masoud, and Ajit Narayanan, eds. Artificial Intelligence: Human Effects. Chichester, England:
Ellis Horwood, 1984.

Postscript
Belinfante, F. J. Measurements and Time Reversal in Objective Quantum Theory. Oxford: Pergamon
Press, 1975.

Churchland, Paul M. Matter and Consciousness. Cambridge: MIT Press, 1988.

Cohen, |. Bernard. The Birth ofa New Physics. New York: W. W. Norton & Co., 1985.
Dechert, Charles R., ed. The Social Impact of Cybernetics. New York: Simon & Schuster, 1966
Hewitt, Paul G. Conceptual Physics. 2nd ed. Boston: Little Brown & Co., 1974.

Hofstadter, Douglas. Gédel, Escher, Bach: An Eternal Golden Braid. New York: Basic Books, 1979.
Landau, L. D., and A. |. Kitaigorodsky. Physical Bodies. Trans. Martin Greendlinger. Moscow: Mir
Publishers, 1980

S38
Landau, L. D., and A. |. Kitaigorodsky. Physics for Everyone. Trans. Martin Greendlinger. Moscow: Mir
Publishers, 1978. -

March, Robert H. Physics for Poets. New York: McGraw-Hill Book Co., 1970.

Minsky, Marvin. Society of Mind. New York: Simon & Schuster, 1985.
Penrose, R., and C. J. Isham, eds. Quantum Concepts in Space and Time, Oxford: Oxford University
Press, 1986.
Przibram, K., ed. Letters on Wave Mechanics. Trans. Martin J. Klein. New York: Philosophical Library,
1967.

539
hlossary

Boldface words in an explanation cross-refer to an entry for that word

Algorithm
A sequence of well-defined rules and instructions describing a procedure to solve a particular
problem. A computer program expresses one or more algorithms in a manner understandable by
a computer.

Analog
Pertaining to data measurable and representable through continuously variable physical quanti-
ties. An analog computer manipulates physical variables that are analogs of (physically analogous
to) the quantities being computed.

Android
A robot similar to a human being in physical appearance.

Artificial intelligence
Broadly, the study of intelligence as a collection of information-processing tasks. Some other
definitions are the following: (1) The field of research concerned with making machines do things
that people consider to require intelligence. (2) The primary goal of Al is to make machines
smarter; the secondary goals are to understand what intelligence is. (3) The study of the
computational connection between action and perception.

Artificial life
A sequence of outputs produced from a computer program that are presented with an initial
configuration of points (the “organism”) and a set of rules (the “genetic code”) to generate
subsequent generations of the organism. Artificial life is modeled on evolution by natural
selection. Certain initial configurations and rules can produce visually pleasing images. This is
thus one technique for generating computer art.

ASIC (application-specific integrated circuit)

A chip that, unlike a normal computer microprocessor, is designed to execute a specialized
function or procedure. ASICs may be viewed as chip implementations of programs.

ASR (automatic speech recognition)

Refers to the ability of a machine to recognize high-level patterns in human speech. |n general,
the goal of ASR is to identify the words spoken.
Backward chaining
A technique of “reasoning” for an expert system in which the system tries to verify a hypothesis
by verifying all statements that imply (or lead to) that hypothesis.

Binary code
A representation (or encoding) of data that makes use of exactly two distinct characters (say 0
and 1). An encoding is a set of rules that specifies a correspondence between one set of
symbols and another.

Bit
Binary digit. In a binary code, one of the two possible characters, usually 0 and 1. In information
theory, the fundamental unit of information.

Byte
A sequence of eight adjacent bits operated on as a unit for the sake of convenience and fre-
quently used as a measure of memory or information. A byte may correspond, for example, to a
letter of the English alphabet

CAD (computer-assisted design)

Computer technology used in many aspects of the design of a product. Users enter design
criteria and create partial designs. The computer then suggests, tests, and modifies the design in
keeping with industry standards and regulations

CAI (computer-assisted instruction)

Education that uses computers to present tutorials and subsequently to test and monitor the
user's learning.

CAT (computer-assisted translation)

Automation of all or part of the difficult task of accurately translating text from one natural
language to another.

CCD (charge-coupled device)

A device that stores information by representing it as packets of minute electrical charges. Also
used as sensors for scanning visual images

Chip
A (possibly large) collection of related circuits designed to work together on a set of tasks. These
circuits reside on a wafer of semiconductor material (typically silicon).

COBOL (Common Business Oriented Language)

A data-processing computer language, the de facto standard for business programming. It makes
use of statements very similar to business English and was first used in 1960.

Combinatorial explosion
The rapid (exponential) growth in the number of possible ways of choosing distinct combinations
of elements from a set as the number of elements in that set grows. Specifically, in artificial
intelligence, the rapid growth in the number of alternatives to be be explored while performing a
search for a solution to a problem

Compiler
A program that produces a machine code from a source code originally written in a high-level
problem-oriented language.

Complexity theory
The mathematical study of the difficulty of solving any well-stated problem in terms of resources
(time and space) consumed. Complexity theory is used primarily to determine the effect of an
increase in the size of a problem.

542
Computer language
A set of rules and specifications to describe a process-on a computer.

Connection Machine
A parallel-processing computer that makes use of a large number of well-connected low-
computation-power processors (currently up to 65,536)

Connectionism
An approach to studying intelligence based on storing problem-solving knowledge as a pattern of
connections among a very large number of simple processing units operating in parallel. Connec-
tionism is often contrasted with the manipulation of the large symbolic structures traditionally
used to represent knowledge in artificial intelligence. |t was inspired by the structure of synapses
(connections) and neurons (processing units) in the human brain

Constrained search
A search on possible alternatives carried out after discarding those alternatives that seem unlikely
to lead up to a solution

CPU (central processing unit)

The principal operating part of a computer, encompassing the electronic circuits that control the
interpretation and execution of instructions. It includes the arithmetic and logic unit, the control
unit, and possibly the primary memory.

Cybernetics
The comparative study of information handling mechanisms (control and communication) in
animals and machines. Cybernetics is based on the theory that intelligent living beings adapt to
their environments and accomplish objectives primarily by reacting to feedback from their
surroundings.

Demon
A support mechanism that performs simple decision making and low-level information gathering
to aid the main (intelligent) program execute its larger tasks. In an expert system, rules that watch
out for exceptions, special conditions, and dangerous situations and suggest precautionary
action.

Data base
An organized store of data made accessible to a computer, ordinarily designed in connection
with an information-retrieval system. A database management system (DBMS) allows monitor-
ing, updating, and interacting with the database.

Debugging
Discovery and rectification of errors (features leading to unwanted results) in a computer
program

Decision tree
A set of rules written in the form of a tree. At each node (branch point) a rule is examined, and a
decision is made to take a particular branch, which in turn leads to the next node or to the end
result.

DENDRAL
The first expert system. It was designed to determine structures of organic compounds using data
from mass spectrometers and magnetic-resonance imaging (MRI)

Digital
Pertaining to the use of combinations of bits to represent all quantities that occur in a problem or
computation. Compare with analog.

543
 ee
Glossary

Domain specific i]
Applicable only to a particular domain of knowledge and expertise. Chemistry is a very large
domain, but organic structural analysis is narrow enough to be handled by DENDRAL, a domain-
specific expert system

Dynamic programming
A procedure that works backward to solve a multistage decision problem by making an optimal
decision at each stage on the assumption that an optimal decision has been made in the
previous stage. The solution for the final stage thus gives the solution for the entire problem

ENIAC (Electronic Numerical Integrator and Computer) |

Completed in 1946 at the University of Pennsylvania, ENIAC was the first general-purpose
programmable electronic digital computer. It was originally designed to produce ballistic tables
for use in the Second World War.

Entropy
In thermodynamics, a measure of chaos and unavailable energy in a physical system. In other |
contexts (even in the social sciences), a term used by analogy to describe the extent of random-
ness and disorder of a system and consequent lack of knowledge or information about it.

Expert system
A computer program, based on various artificial intelligence techniques, performing a specialized
difficult task at the level of a human expert. An expert system is frequently partitioned into a
knowledge base, an inference engine, and (possibly) an explanation mechanism

Floating-point number
A number expressed as a product of a bounded number and a scale factor (consisting of an
integer power of a number base), for example, 2.3 x 107

Floppy disk
A secondary data-storage device consisting of a thin flexible magnetic disk covered by a
semirigid protective jacket

Fractal geometry
A nascent branch of mathematics named with a Latin word meaning irregular. Fractal geometry
explores a world of crinkly convoluted shapes far removed from the straight lines and smooth
curves of traditional Euclidian geometry. Fractal curves, produced by computer algorithms, bear
striking resemblance to naturally occurring shapes (coastlines, clouds, etc.) and hence are
extensively used to produce realistic computer graphics. The term “fractal” is also an abbrevia-
tion for “fractional dimension”: the dimensions of fractal curves are not limited to integers.
These curves are used to mode! chaotic phenomena appearing in such diverse fields as music,
financial markets, and natural topology.

Fuzzy logic
A branch of logic designed specifically to support human reasoning by allowing such linguistic
labels as “fairly” and “very” so that statements may be made with varying degrees of certainty
and precision. In traditional logic a statement may have one of only two values (true or false)
Fuzzy-logic labels and operators allow statements to have multiple values.

Gaussian filtering
A process by which each pixel is replaced by one whose intensity (brightness) is the weighted
sum of the intensities of adjacent pixels. The weight for a pixel is chosen according to the
Gaussian normal function (the bell-shaped curve) of its distance from the central pixel. This is
often a preliminary step to recognizing the edges of a picture.

General Problem Solver (GPS)

A procedure and program developed by Allen Newell, J. C. Shaw, and Herbert Simon that attains
an objective by using rules to generate many alternatives at any current state and pursuing the ;

saa :
alternatives that appear to be the closest to the objective. It has failed on some classes of
problems due to the lack of sufficient rules and an adequate measure of closeness.

Goal-driven reasoning
Reasoning that focuses attention only on those pieces of knowledge that seem likely to lead to a
goal. A generalization of backward chaining.

GoTo statement
An instruction in a computer program used to explicitly redirect the computer's current operation
to another line of the same program.

Hard disk
A rigid plate with magnetic coating. A hard disk can handle greater amounts of data at a higher
speed than a floppy disk.

Heuristic
A tule of thumb or a technique based on experience and for which our knowledge is incomplete.
A heuristic rule works with useful regularity but not necessarily all the time. More generally, a
heuristic is any knowledge that reduces the amount of search

Heuristic programming
The programming of problem-solving systems to reason and search for solutions by means of
organized collections of heuristics.

Holograph
An interference pattern, often using photographic media, that is encoded by laser beams and
read by means of low-power laser beams. This interference pattern can reconstruct a three-di-
mensional image.

Holy Grail
Any objective of a long and difficult quest. In medieval lore, the Grail refers to the plate used by
Christ at the Last Supper. It subsequently became the object of knights’ quests.

Hypertext
An approach to information management in which data is stored in a network of nodes con-
nected by Jinks. The nodes, which may contain text and even audio or video elements, are meant
to be viewed interactively. Links provide pathways to other nodes for exploring further relevant
information

Idiot savant
A system or person that is highly skilled in a small task area but impaired in other areas of
functioning. The term is taken from psychiatry, where it refers to a person who exhibits brilliance
in one very limited domain but is underdeveloped in common sense, knowledge, and compe-
tence. Idiot savants have been known, for example, to be capable of multiplying very large
numbers in their heads, although they are otherwise mentally disabled.

Image processing
Performing computations on sets of visual signals to recognize and interpret highevel patterns
and to resolve the image into meaningful components.

Image scanner
A device that converts a visual image into sets of electronic signals that may be subjected to
image processing.

Josephson junction
An oxide barrier between two metals across which electrons may be forced to tunnel. When
immersed in liquid helium, a circuit containing such a junction is able to switch at very high

545
Glossary

speeds with low power dissipation. It thus avoids the heat-transfer problems that arise from
silicon VLSI devices and is one way to attain superconductivity.

Knowledge engineering
The art of designing and building expert systems, in particular, collecting knowledge and heuristics
from human experts in their area of specialty and assembling them into a knowledge base or
expert system

Knowledge principle
A “principle” that emphasizes the important role played by sheer knowledge (as opposed to
general reasoning mechanisms) in many forms of intelligent activity. It states that a system |
exhibits intelligence primarily because of the specific knowledge that it contains about its domain
of knowledge

Knowledge representation j
A scheme for organizing human knowledge into a manipulable data structure flexible enough to |
allow one to express facts, rules, relationships, and common sense.

LISP (list processing)

An interpretive computer language developed for manipulating symbolic strings of instructions and
data. The principal data structure is the list, a finite ordered sequence of symbols. Because the
programs themselves are expressed as lists, LISP lends itself to sophisticated recursion, symbol
manipulation, and self-modifying code. It is widely used for Al programming

Logic gate
A device implementing any of the elementary logic (or Boolean) functions (for example, Ano, oR,
NOT, NoR, xoR). A logic gate is characterized by the relationship between its inputs and outputs
and not by its internal mechanisms.

Luddite
One of a group of early nineteenth-century English workmen that destroyed labor-saving
machinery in protest. Today the Luddites are a symbol of opposition to technology.

Machine language
The written representation of machine code, which is the operation code understood directly by
a computer. More generally, it is the language used by a computer for communicating internally
with its own subsystems

Mainframe computer
An expensive, sophisticated, general-purpose computer that can be simultaneously accessed by
many users and that usually has a wide range of peripherals. Mainframes are distinguished from
minicomputers and microcomputers primarily by their computational power.

Microcomputer
A small computer designed for a single user, ordinarily used in homes and by individual business
users. A microcomputer usually uses a single-chip microprocessor.

Microprocessor
The entire CPU (central processing unit) of a computer in the form of a large-scale integrated circuit
built on a single chip

MIDI (Musical Instrument Digital Interface)

Technology that allows communication of digital data (representing musical notes and other
information) between musical instruments and thus allows different electronic musical instru-
ments to control and interact with each other.

546
Mind-body problem
The philosophical question of how a nonphysical entity such as the mind may interact with and
exert control over a physical thing such as the body

Minicomputer
A computer similar to a mainframe computer in that it supports multiple users, but not as powerful.

Minimax procedure or theorem

A basic technique used in game-playing programs. An organized enumeration of possible moves,
together with corresponding possible moves from the opponent, is constructed in the form of a
tree. An evaluation of the final position of the tree's leaves that always chooses the minimum
value on the opponent's levels and the maximum value on the player's levels is then passed
back down the tree

MIPS (millions of instructions per second)

Acconventional measure of the speed of a computer in terms of the number of the steps it can
perform per second.

Molecular computers
Computers based on logic gates that are constructed on principles of molecular mechanics (as
opposed to principles of electronics) by appropriate arrangements of molecules. Since the size of
each logic gate is only a few times that of a molecule, the resultant computer could be micro-
scopic in size. Limitations on molecular computers arise only from the physics of atoms. Thus,
molecular computers are theoretically the fastest possible computers

Multiple experts
A group of intelligent programs, each highly skilled in its own area of specialization, organized
together to achieve a broader and higher level of generality in performance in the hope that at
least one will rise to a problem solution

Music-notation processor
A program or system working on music notation in the functionally equivalent way that word
processors work on text.

MYCIN
A successful expert system developed at Stanford University in the mid 1970s and designed to aid
medical practitioners in prescribing an appropriate antibiotic by determining the exact identity of
a blood infection.

Natural language
Ordinarily spoken or written language (e.g., English) governed by sets of rules and conventions
sufficiently complex and subtle for there to be frequent ambiguity in syntax and meaning.

Neural networks
Machinery implementing the ideas of connectionism and consisting of processing units and their
interconnections patterned after the structure of the human brain.

90-10 rule
Generally, with a given set of resources and objectives, solving 90 percent of a problem will
consume only 10 percent of the resources, and solving the remaining 10 percent of the problem
will consume the remaining 90 percent of the resources

NoR gate
A logic gate whose output is logical 1 (true) only when all its inputs are logical 0 (false) and is
logical 1 otherwise. It is a universal gate, since any logical or Boolean function can be realized
with circuits consisting only of nor gates.

S47
Glossary

Object code
The machine-code output of a compiler after translation of the source code. See compiler.

OCR (optical character recognition)

A process in which a machine scans, recognizes, and encodes printed (and possibly handwritten)
characters into digital form for input into a computer.

Operating system
A large, complex body of programs that control and administer all other programs of a computer.

Optical computers
Computers processing information encoded in patterns of light beams, unlike today’s conven-
tional computers, in which information is represented in electronic circuitry or encoded on
magnetic surfaces. Optical computers present the potential of computing at higher speeds and
with a massive level of parallel processing

Optical disks
A disk, typically of plastic, on whose surface information is etched as a sequence of pits, which
are read by a low-power laser beam. Optical disks can potentially store large amounts of
information (a billion bytes or more).

Organic circuits
Circuitry consisting of processing units that, under favorable conditions, are self-replicating and
self-organizing.

Orthogonal invariances
The distribution of strengths of two expert systems in a way that the chance of solving a particular
problem is greater when using both experts than when using either expert alone. The reason for
the success of multiple experts is that the chance is high that at least one of the experts will rise
to a problem solution

Pandemonium selection
An electionlike decision-making procedure in which a decision is made according to the level of
excitation or response generated from a group of sensory or information-gathering units (or
demons ) when presented with each possible solution.

Parallel processing
Simultaneous operation (rather than sequential operation) of two or more devices to perform
independent tasks within an overall job. More than one particular process is active at any instant.
Compare with serial computer.

Parser
A program resolving a string of characters (representing, for example, an English sentence or a
computer-language statement) into its elemental parts (or parts of speech) as determined by the
particular language.

Pattern recogn
Recognition of patterns with the goal of identifying, classifying, or categorizing inputs.

Perceptron
An early model for the processing units that may be used in neural networks. Perceptrons are
noted for the simplicity of the function they perform on input

Pixel
An abbreviation for picture element. One of the elements in a large array holding information that
represents a picture. Pixels contain data giving brightness and possibly color at particular points
in the picture.

548
 Predicate logic
A system of logical inference based on set theory. Variants are used in artificial intelligence in the
representation of knowledge.

Primal sketch
A representation of images that gives explicit information about brightness changes, textures,
and orientations of surfaces.

Program
A sequence of written statements that conform to the specifications of a computer language, for
use by a computer (or any intelligence system).

PROLOG
A nonprocedural programming language used in artificial intelligence. A PROLOG program
contains descriptions of relevant relationships between entities in the problem and of the rules
governing the solution of a problem but not of the procedures that are to be used to find those
solutions.

Prospector
An expert system designed to aid geologists in interpreting mineral data and predicting locations
of mineral deposits. It is famous for correctly pointing to a previously unknown extension of a
molybdenum deposit.

Punch card
A rectangular card that typically records up to 80 characters of data in a binary coded format as a
pattern of holes punched in it.

Pushdown stack
A list of information in which insertions, and removals of data are made at one end of the stack
(the top), also called a LIFO stack for “last in, first out.” Stacks are convenient structures in
computer science because they facilitate recursive computations.

RAM (random-access memory)

A type of temporary internal memory in which all locations of data can be accessed with equal
speed.

Recursion
The process of defining or expressing a function or procedure in terms of itself. Each level of a
recursive-solution procedure produces a simpler (or possibly smaller) version of the problem than
originally posed. This process continues until a subproblem whose answer is already known is
obtained. A surprisingly large number of symbolic and numerical problems lend themselves to
recursive formulations

Reverse Polish notation

A notation for arithmetic operations on numbers, where each operator follows its operands. For
example, 7 + 5 is expressed as 7 5 + and 1 + (2 x 3) is expressed as 2 3 x 1 +. The order of
evaluation is uniquely defined, and such expressions are readily evaluated on a pushdown stack

Robot
A programmable device consisting of machinery for sensory activity and mechanical manipula-
tion and connected to (or including) a computer. Typically, these machines automatically perform
some task normally done by human beings.

ROM (read only memory)

A permanent internal memory whose contents are built into the device during manufacture
ROM contains data or instructions that can be read quickly but cannot be altered.

Sao
Glossary

Sea-of-logic machine
A machine model proposed in this book constructed only of nor gates. The mechanisms that
account for intelligent behavior can in theory be constructed as a sea-of-logic machine.

Search
A procedure in which an automatic problem solver seeks a solution by iteratively selecting from
various possible alternatives as intermediate steps toward a solution.

Segmentation
Generally, the broad problem of separating a problem into parts. For example, a string of speech
may be analyzed at many levels: in terms of phonemes, in terms of words, or in terms of
complete sentences.

Self-modifying code
A program that causes changes in portions of the program itself. Self-modifying code can thus
selectively store, destroy, and transform information within itself (for example, it can replace
problems with simpler subproblems). This ability is the crux of intelligent adaptive behavior

Semantic networks
A type of knowledge representation that uses nodes to denote concepts and labeled links to
indicate the relationships among these concepts

Semiconductors
A material commonly based on silicon or germanium with a conductivity midway between that of
a good conductor and an insulator. Semiconductors are used to manufacture transistors, which
are used to construct logic gates.

Serial computer
A computer that performs two or more computations one after another, not simultaneously
(even if the computations are independent). Opposed to a parallel processing computer.

SHRDLU
A landmark natural-language program, completed in 1970, that integrates the previously inde-
pendent functions of reasoning, syntax, and semantics. It performed remarkably well in its own
small domain of children’s blocks

Silicon compiler
A system producing a description of a given procedure or algorithm at the level of logic gates so
that this low-level detail may be used to specify the design of a chip to be fabricated.

Simulator
A special-purpose computer or program designed to imitate the behavior of some existing or
intended system in order to study the performance and effects of that system. Simulators are
often used to study complex natural systems, such as chemical interaction or fluid flow. Another
common example is the flight simulator used to train pilots

Smart weapons
Attack systems that exhibit intelligence and skill in identifying, locating, tracking, and destroying
targets

Society of mind
A theory of the mind proposed by Marvin Minsky in which intelligence is seen to be the result of
proper organization of a very large number of simple mechanisms, each of which is by itself unin-
telligent

550
Sombrero filter
A process whereby each pixel is replaced by one whose intensity is the weighted sum of the
intensities of adjacent pixels. The weights are chosen according to a function of the distance of
the adjacent pixel from the center pixel. The function is the Laplacian of a Gaussian convolver
and has the shape of a Mexican sombrero hat.

Speaker independence
Refers to the fact that some features of a speech string are not dependent on who the speaker
is. “Speaker independence” also refers to the ability of a system for automatic speech recognition
to understand any speaker, irrespective of whether or not the system has previously sampled
that speaker's speech.

Subroutine
A program or block of programs organizationally distinct from the main body of the program.

Supercomputer
The fastest and most powerful computers available at any given time. Supercomputers are used
for computations demanding high speed and storage (e.g., analyzing weather data)

Superconductivity
The physical phenomenon whereby some materials exhibit zero electrical resistance at low
temperatures. Superconductivity points to the possibility of great computational power with little
or no heat dissipation (a limiting factor today).

SUR (Speech-Understanding Research) Project

A project funded by DARPA in the early 1970s with the goal of producing an automatic speech-
recognition system with the capacity to understand 90 percent of continuous speech (as opposed
to isolated words) about some limited subject matter and making use of a vocabulary of at least
one thousand words.

Synthesizer
An electronic device (typically for musicians) for the production of a wide range of sounds,
allowing significant control over the nature of these sounds

Stored program
Refers to a computer in which the program is stored in memory along with the data to be
operated on. A stored-program capacity is an important capability. for systems of artificial
intelligence in that recursion and self-modifying code are not possible without it

Template matching
Comparing stored prototype image patterns with shapes derived from digitized input images

Toy world
An artificially simplified system used for studying and testing of ideas and devoid of the com-
plexities and constraints that are ordinarily present in real circumstances.

Turing test
Accriterion proposed by Alan Turing that maintains that a system is intelligent if it can deceive 4
human interrogator into believing that it is human.

2'/2-D sketch
A representation of a visual scene that shows the depth and orientation of all visible surfaces

UNIX
A multiprogramming operating system developed at Bell Laboratories and favored by computer
scientists. One of the goals of UNIX is to provide a uniform environment in which a relatively
small number of users may collaborate on a single system with a considerable degree of
cooperation.
 a
Glossary

Vacuum tube
A device with electrodes in an evacuated glass tube for the the control of current flows in a
electric circuit. Used for the construction of early logic gates.

VLSI (very large scale integration)

Fabrication technology that allows tens of thousands (and possibly up to millions) of logic gates to
reside on a single chip

Von Neumann architecture

The design of conventional computers, based on stored programs and serial processing.

Word processor
A program for creating, altering, viewing, storing, and printing text with maximum flexibility. The
term was coined by IBM in the late 1960s

XCON
An expert system developed to specify how all the components of a computer should be placed
and connected. Developed by Carnegie-Mellon University and Digital Equipment Corporation
(DEC) for use in configuring DEC’s VAX series of computers

Z-series machines
A series of machines designed and constructed by Konrad Zuse in Germany: Z-1 (nonprogram-
mable, mechanical), Z-2 (nonprogrammable, electromechanical), Z-3 (programmable, electrome-
chanical), and Z-4 (an improved version of Z-3). The Z-3 was the first programmable computer.

552
Index

AARON program, 357, 358, 380, 381, 382-383, Art

385 and computer technology, 351
Abacus, 161 literary, 370-379, 390-393
ABC computer, 183, 188 musical, 278-280, 351-355, 373, 378, 386-389,
Abelson, Robert, 211 443-444 (see also Music)
Access control, 270, 411 shift in style of, 116
Active Images, 326 visual, 355-369, 373, 380-385.
Ada programming language, 167 ART (Automated Reasoning Tool), 301
“Nicolai” in, 95 Artificial intelligence (Al). See also Computers;
Agenda program, 373 Future of computer technology
Al. See Artificial intelligence beginnings of, 190
Aiken, Howard, 167, 173, 177, 178, 179, 188 and chess, 68-69, 124-127, 130-133 (see also
Alexander, Sam, 175 Chess playing by computers)
Algorithms, 18, 112, 124, 189 and Church-Turing thesis, 117, 152
in brain simulation, 233 and combinatorial explosion, 54, 59
and Church-Turing thesis, 152 as commodity, 13
and creativity, 394 and consciousness, 63-66, 461-462 (see also
generalized perception, 141 Consciousness)
in intelligent machines, 161, 162, 164 criteria of, 93
minimax, 127, 407, 409 criticism of, 36, 38-39, 62-66, 232-233, 235,
in neural nets, 229 347
and NOR gates, 121 definitions of, 13, 14-15
and stored programs, 185 early development of, 198-205, 212, 325-326
Alice (developmental machine), 347, early optimism over, 117, 135
Allen, Jonathan, 264 expert systems as, 15-16, 56-58, 292-303 (see
Alvey, John, 323 also Expert systems)
Alvey Report and Program, 323, 347-349 and Fifth-Generation Computer System
Ambiguity Project, 340
of characters, 241, 243, 244 growth of, 212-213, 323, 419
of language, 51, 304, 306 and handicapped, 8, 269, 278, 281, 318, 441-
Ames, Charles, 354, 386-390 443, 451, 455
AML language, 315 and human intelligence, 292, 455 (see also
Analog vs. digital computation, 191-196, 227 Human beings and computers/Al)
Analogy as mode of reasoning, 327 and human self-esteem, 60
Analogy problems, 202 and human uniqueness, 69
Analytical Engine of Babbage, 89, 158, 165-167, and intelligence of creature vs. creator, 19, 21
171, 185, 188 as interdisciplinary, 198, 281
Anderson, Russell, 317 and knowledge, 209-212, 284 (see also
Androids, 414. See also Robots and robotics Knowledge)
of Renaissance, 159 limits of, 80
Anzai, Yuichiro, 232 military uses of, 9 (see also Military uses)
Application-specific integrated circuit (ASIC), 412 and mind as machine or beyond machine, 23-
Archytas of Tarentum, 159 24
Aristotle, 24, 62 Minsky on, 214-219
and Plato, 27, 28 naming of, 190, 199, 212
quoted, 425, 429 and natural intelligence, 46-47, 139
Arkwright, Richard, 3 and pattern recognition, 224, 271 (see also
Pattern recognition)

S53
Artificial intelligence (cont.) Blind persons
philosophical roots of (logical positivism), 25, future aids for, 441, 442, 455
27, 34, 35 reading machines for, 249, 272, 274, 275, 441
in popular culture, 68 Bidch, Richard, 179
as Potemkin villages, 56 Block, Ned, 53-54
in robots, 312-322 (see also Robots and Bobrow, Daniel G., 202, 203
robotics) Boden, Margaret A., 450-453
social impact of, 450-453 Bohr, Niels, quoted, 425
and Turing test, 48-60, 62, 116-117, 374 (see Boole, George, 164, 176
also Turing test) “Boris” (program), 93
underestimation/overestimation of problems in, Boucher, David, 371
404 Bourbaki, Nicolas, 97
Artificial life, 361 Bovee, Christian, 434
Artificial people, 412-413 Brain, human, 23
Artificial Stupidity (AS), 14 and Al, 13
Artificial vision, 247-262, 428 (see also Visual analog processes in, 195, 227
perception) capacity of, 289-290
Asada, Hurahiko, 319 and computer, 54, 75-77, 290, 292 (see also
ASICs, instant, 412 Human beings and computers/All)
Asimov, Isaac, 219, 312, 414 and consciousness, 61, 75
ASR (automatic speech recognition), 265-270, edge detector cells in, 226
276-278, 405, 407, 456 image filtering in, 227
Atanasoff, John V., 178, 183, 184, 188 and intentionality, 65
Atkins, Peter, quoted, 149 Minsky/Papert on, 214-215, 223
Authorizer’s Assistant, 299 as neuronal or neural net, 38, 75, 140, 142
Automata, 159, 160 19th-century picture of, 33
of Descartes, 29 and parallel processing, 228-229, 233, 234, 290
Automated Composition Project, 386 perceptron as modeling, 140
Automated Fingerprint Identification System fate of neuron death in, 288
(AFIS), 252 relearning capacity of, 234
Automatic learning, 327-328 simulation of, 233-234
Automatic speech recognition (ASR), 265-270, weakness of, 16
276-278, 405, 407 Britain. See United Kingdom
Automatic story generation, 392-393 Brittain, Skona, 306
Automatic translation, 306, 307-309 Brown, John Seely, 431
Automation, impact of, 4-5, 7, 426. See also Buchanan, Bruce, 294, 325
Future impact of computer technology Bundy, McGeorge, 436
Automatix, 255 Burns and Allen, comedy routine by, 303
Automotive diagnosis, expert system for, 300, Burroughs, William, 169, 171, 176
330-335 Bush, Vannevar, 191, 192, 198
Awareness. See Consciousness quoted, 432
Ayer, Alfred, 25, 35 Busy beaver problem, 113-115
Byrd, Don, 99-100, 354
Babbage, Charles, 158, 165-169, 173, 176, 188
and Lovelace, 117, 167 C (programming language), 176, 208
and machine concept, 89 CADUCEUS system, 211, 296, 299
quoted, 165 Calculation, by computers/cash registers/
and stored program, 185 humans, 84, 89
and Turing, 109 Calculators, 169
Backgammon, 134, 136 vs. computers, 169
Bacon, Francis, 24 early versions of, 161, 162, 164
Baker family (Heather, Janet, Jim), 265 Canny, J. F., 226
Bangert, Charles J., 359 Capek, Karel, 312
Bangert, Colette, 359 Carroll, Lewis, quoted, 124
Battle Management System, 301 Causal reasoning, and diagnosis, 332
Bavel, Zamir, 93, 94, 98 Cellular phone, 409-410, 411, 432
Beckett, Samuel, 24-25 Census Bureau, 171-172, 173, 186
Belle (chess program), 131 Chamberlain, William, 370
Benedetti, Carlo De, 329 Champernowne, David, 109
Benford, Gregory, 219 Chaotic fractals, 361
Bergson, Henri, 196 Character recognition, 238-247
Berliner, Hans, 133, 134 optical, 249, 272-275
Berry, Clifford, 183 Chauffeur, cybernetic, 410-411
Bieri, Peter, 59 Checker Playing Program, 124
BINAC (Binary Automatic Computer), 186, 188, Checkers, 134, 136
205 Chemical analysis systems, for taste and smell,
Binary notation, 164 271
Binford, Tom, 257 Chess
Bioengineering techniques, for circuit-building, 418 as midlevel problem, 136
Bledsoe, Woody, 47 war replaced by, 437

554
Chess playing by computer, 68-69, 124-127, and Russell's paradoxes, 104-108
130-133 and Russell's Principia, 108
and Dartmouth conference, 15 4nd Turing, 109, 112 (see also Turing, Alan)
and definition of Al, 14 Computer-aided design (CAD), 302, 315, 403
development of, 202 Computer art, 355-369
and Dreyfus vs. Al, 36 Computer-assisted instruction (CAI), 431
vs. human, 290-291 Computer-based synthesizers, 271, 278-280,
and human self-concept, 448 352
improvement in, 217, 218 Computerized translation, 306, 307-309
self-model in, 87-88 Computer program, 169. See also Software,
Turing’s program for, 109 computer
and Turing test, 51, 58, 116, 378 and genetic code, 20, 151
Wiener on, 198 and level of reality, 67
world championship, 16, 132-133, 407-409 and predicate logic, 338, 340
and Zuse, 190 recursive, 125
Children stored, 185, 206
and computers, 39, 69-72, 461 and Turing machine, 123
in poetry Turing Test, 377 Computer programming
Chinese Room example, 64-65 by Grace Murray Hopper, 178
Chips, computer by Ada Lovelace, 167, 168
computer-aided design of, 403 of Z machines, 176
and computer cost, 9, 454 Computers. See also Artificial intelligence
improvements in, 412, 416 Al seen as application of, 212
and natural resources, 8 and analytic vs. computational approach, 189
and silicon compiler, 122, 455, 456 Babbage's foreshadowing of, 165-167, 173
with two or three dimensions, 234, 408, 417 children’s view of, 39, 69-72
value of, 445 communication with, 99
Chomsky, Noam, 25 definition of, 169
Chunking of knowledge, 292 fifth-generation, 336-345 (see also under
Church, Alonzo, 103, 115 Japan)
Church-Turing thesis, 112, 117, 121, 123, 152 generations of, 322
and Wittgenstein, 34, 115, 117 and human beings, 71-73, 74-77, 219, 290-292,
Church, Ken, 306 447-449, 459 (see also Human beings and
Clarke, Arthur C., 219 computers/Al)
Classification information vs. knowledge from, 17, 284, 290-
of characters, 240, 242 292, 459
of knowledge, 210-211, 282, 284-288 international competition in, 322-323
COBOL (Common Business-Oriented invention and development of, 175-188, 190
Language), 178 Litvin on, 46-47
Cognex Corporation, 254 mechanical roots of, 173
Cognitive dissonance, 288 mistakes by envisioned, 90-91, 459-460
Cohen, Harold, 357, 358, 366, 380-385 von Neumann architecture in, 185, 205-206,
Colby, Kenneth, 54-56, 346 229, 343
Colossus, 110, 177-178, 183, 188 and NOR gates, 121-122, 123
Combinatorial explosion, 54 personal, 373, 431, 432, 454-455
and expert systems, 293, 301-302 as postindustrial power tools, 329
and prestocking of information, 59 programs as distinct from, 67
Common sense, 216-217, 327 Taw material cost in, 9
from artificial intelligence, 455 searching by, 216
and expert systems, 334 sharing of knowledge among, 292
in intelligent assistant, 407 social uses of, 8, 48, 281 (see also Future
Communication impact of computer technology)
by computer, 99 and Turing machine, 48, 123, 337-338, 343
between human and machine, 276, 309, 326 (see also Turing machine)
impact of future computer technology on, 432- and Turing test, 48-60, 116-117, 377-378, 414-
434 416 (see also Turing test)
between neurons, 147 voice communication with, 268-269, 270
of technical-service bulletins, 332 Computer technology. See a/so Future of
Compiler, silicon, 122, 455, 456 computer technology
Complexity theory, 130 accomplishment level of, 459
Computability, establishment of, 338 and art, 351 (see a/so Art)
Computation as enchanted, 421-423
digital vs. analog, 191-196, 227 as magic, 463
energy required for, 191 improvement in, 8, 218, 275, 401-405, 445
optical, 417 as liberating, 454
Computationalism, 63, 66 Computer vision, 247-262
Computation theory Concepts
foundations of, 103 and data structure, 288
limits of, 115 extension of, 85

555
Conditionality, 422 on intentional stance, 89
Connection Machine, 132, 229, 230 The Mind's |, 80
Consciousness, 42-43, 44, 461. See also and simulated hurricanes, 83
intentionality and intentional stance afd Turing test, 48-60, 98, 116, 378
and Al, 63-66, 461-462 Descartes, René, 24, 29, 50, 62, 461
attributes of, 87 Desktop publishing, 371-373, 455
and brain, 61, 75 Determinism, Platonic-like dialogue on, 44
children’s views of, 71 Dewey, Melvil, and Dewey Decimal system,
and computers, 39, 60-61 285
Freud on, 32 Diagnosticians, 331-332. See also Automotive
knowledge of, in others, 88 diagnosis; Medicine
mysteriousness of, 217 Difference Engine, 165, 167
and Plato, 26-27 Differential Analyzer, 191
and quantum mechanics, 462 Digital vs. analog computation, 191-196, 227
and Turing test, 66, 89 DNA, 149-151, 189
Consistency, of doctors’ vs. expert systems’ Doi, Toshi, 293
judgment, 295 Domino, Michael, 354
Context of interpretation, 51 Doyle, Sir Arthur Conan, quoted, 14
for letters, 244 Dragon systems, 265
Contradiction Drawing. See Visual arts
and brain storage, 288-290 Drexler, Eric, 218
and systematic thought, 40 Dreyfus, Hubert, 25, 36, 38-39, 63, 232
Conversation, and intelligence, 50. See also and ELIZA, 36, 202
Turing test and Turing Test, 38, 414-415, 416
Cooking, robots for, 317 Dreyfus, Stuart, 38
Copernicus, Nicholas, 7, 66 Drugs, therapeutic, future design process for,
Courtier system, 307 439-440
Courts, and technical content of issues, 183 Duchin, Faye, 426
Creativity, 219, 290, 394-397 Duda, R. O., 211
in story telling, 393 Dyer, Michael, 93
Credit cards and keys, invisible, 411
Crick, F.H. C., 149 Ebcioglu, Kemal, 387, 388
Cruise missile, 255, 256 Eckert, J. Presper, Jr., 183, 185, 186, 187, 188
Cybernetic chauffeur, 410-411 Economy, impact of computer intelligence on, 5,
Cybernetic Composer, 354, 386 425-428, 452, 456-457
Cybernetic Poet, Kurzweil, 374-379 Edge detection, 224-227
Cybernetics, 15, 139, 190-191, 199. See also in AARON, 385
Artificial intelligence in artificial system, 247
CYRUS (Computerized Yale Retrieval and Edison, Thomas, 3
Updating System), 56-57 EDSAC (Electronic Delay Storage Automatic
Computer), 175, 186, 188, 205
DARPA (Defense Department Advanced Education
Research Projects Agency), 139, 212, 247, impact of computer technology on, 47, 428,
323 429-432, 450
Darsche, Howard, 98 importance of, 9, 427, 429
Dartmouth College, Al conference at, 15, 190, EDVAC (Electronic Discrete Variable Automatic
199,:325 Computer), 186, 188
Darwin, Charles, 66 Eiho, Shigeru, 258
Data-base systems, accessing information from, Einstein, Albert, 118, 119, 120
407 and Newton, 31
Data structures, 284, 288-290. See also and quantum theory, 462
Knowledge quoted, 119, 401
Dawkins, Richard, quoted, 149 Electronic libraries, 328
Deaf persons, aid for, 269, 278, 442 Electronic music, 271, 278-280
“Debugging” of computer, 178, 179, 183 ELIZA (Computer program), 16, 36, 78, 93, 202,
Debussy, Claude, quoted, 351 204
Decision making under uncertainty, by humans Emerging nations, 8
and expert systems, 302-303 Emotions (feelings)
Decision rules, in expert systems, 292-293 and children on computers, 39
Decision tree, 143, 144 and computers vs. humans, 69, 71-72
Deep Thought (chess machine), 130, 131 and knowledge, 289
Defense Advanced Research Projects Agency and mind-machine relation, 24
(DARPA), 139, 212, 247, 323 physical interpretation of, 76
Defracto vision system, 255 and thinking, 72-73, 87, 88-89, 219
DENDRAL program, 202, 294, 325 Employment, impact of computer technology
Dennett, Daniel C., 48 on, 4-5, 425-428, 452, 456-457
on computationalists vs. holists, 63 EMYCIN (Essential MYCIN), 296, 297
on computer level of reality, 84 Encyclopaedia Britannica, Propaedia section of,
on consciousness, 60-61, 461 286

556
England. See United Kingdom Finance, expert systems in, 298, 299, 301, 302,
ENIAC (Electronic Numerical Integrator and 307
Computer), 179, 182, 183, 185, 188. Fingerprint identification, 252, 411
Enigma code machine, 109, 110, 111, 177 Flynn, Anita, 321
Enlightenment, 24, 29, 31-32 Fodor, Jerry A., quoted, 23
Entropy, 120 Ford, Henry, 3
Entry control, 270, 411 Fractals, 361-366
Erdos, Paul, quoted, 103 Frames, theory or methodology of, 210-211, 284
Essence, and extension of concepts, 85 Fredkin, Edward, 189, 191, 195
Etter, Thomas, 370 quoted, 16, 189
Euclid and Euclidean geometry, 28, 136 Free will
EURISKO system, 302 and children’s view of computers, 72
European Strategic Program for Research in and Church-Turing thesis, 117
Information Technology (ESPRIT), 323, 349 in existentialism, 25
Evans, Thomas G., 202 and Plato, 25, 26
Evolution Platonic-like dialogue on, 44-45
artificial life simulates, 361 Frege, Gottlob, 338
of computers, 91 Freud, Sigmund, 32, 39, 66, 72
and entropy, 120 Fromkin, Victoria, 100
and genetic code, 151 FRUMP program, 56
of horses and Al, 334-335 Fuchi, Kazuhiro, 336-345
and intelligence, 18, 19-21 Fuller, A. Buckminster, quoted, 438
and intentionality, 65 Functional isomorphism, 75-76
Existentialism, 24-25, 26, 31-32, 35 Future of computer technology, 218-219, 403-
Expert manager, 238, 246, 247, 268 405, 416, 449, 455-456, 459-460, 463
Expert systems, 15-16, 56-58, 292-294 artificial people, 412-413
accomplishment level of, 459 chess championship, 132, 407-409
and artificial vs. natural model, 139. cybernetic chauffeur, 410-411
for automotive diagnosis, 300, 330-335 hardware breakthroughs, 416-418, 419
backward chaining in, 334 instant application-specific integrated circuits,
beginning of, 202, 294 412
bottlenecks in, 301-302 intelligent assistant, 406-407
brittleness/fragility of, 292, 299, 326-327 intelligent telephone-answering machine, 409-
educational, 450 410
expansion of, 301, 378 invisible credit cards and keys, 411
and Fifth-Generation Computer System media in, 413-414
Project, 340 robots, 414
in finance, 298, 299, 301, 302, 307 software in, 418-419, 432-433
and intelligent assistant, 407 translating telephone, 405-406
knowledge collection for, 211 Turing Test passed, 414-416
logic employed in, 17, 196 Future impact of computer technology, 8
in medicine, 57, 294-297, 299, 325 on communications, 432-434
military uses of, 301 ‘on education, 427, 428, 429-432, 450
in mining, 299-300 on employment and economy, 5, 425-428, 452,
Minsky on, 217 456-457
and music composition, 353-354 on handicapped, 441-443, 451, 455
number of rules needed in, 292 on medicine, 438-441
and parallel processing, 233, 301 on military, 9, 425, 434-438, 452
as society of intelligences, 148 on music, 443-444
U.K. promotion of, 348 on politics, 445-447, 452, 457
for writers, 373 on self-concept of humans, 447-449
Expert Systems Starter Pack, 348 social, 450-453
Explanation Pattern (XP), 396 Fuzzy logic, 17, 36, 196, 293, 295

Factory robots, 314, 315 Games

Factory vision, 252-255 computer, 412
Facts, and knowledge, 284 (see also role-playing, 392
Information; Knowledge) General Problem Solver (GPS), 124, 135-136
Fairy tales, and technology, 420-423 199, 201
Fast, August, 176 General relativity formula, 119
Feelings. See Emotions Geometry, as midievel problem, 136
Feigenbaum, Edward A., 25, 202, 212, 294, 324- George, Harry, 371
329, 463 Gilder, George, 9, 454-457
quoted, 283 quoted, 402, 416, 445
Fifth-Generation Computer System (FGCS) Go (game), 134, 138
Project in Japan, 212, 294, 322, 336-345 God
impact of announcement of, 322-323 as director of souls, 44
UK. response to, 323, 348 imitating of, 47
Finale system, 354 Godel, Kurt, and Gédel’s theorem, 115, 338

557
Goldhor, Richard, 265 and machine concept, 89, 90, 91-92, 460
Gomory, Ralph, 416, 455 self-concept threatened, 60, 447-449
Good, |. J., 175 ang simulable vs. nonsimulable reason, 69
GOTO statements, 338 in social roles, 450-451
GPS (General Problem Solver), 124, 135-136, and thought vs. feeling, 69, 71-73, 219
199, 201 Human brain. See Brain, human
Greenblatt, Richard D., 202 Humanoid robots, unlikelihood of, 61. See a/so
Griffin, Donald, 100 Androids
Humor, in Turing test, 87
Haiku poems, 370 Hurlbert, Anya, 247
Handicapped persons Hypertext, 428
automatic speech recognition for, 269 Hypothesis and test technique, 224, 228
computer technology to aid, 8, 278, 281
future impact of computer technology on, 441- IBM (International Business Machines), 173,
443, 451, 455 177, 178, 186
robots to aid, 318 and Microelectronics and Computer
Hardware. See also Chips, computer Technology Corporation, 322
application-specific integrated circuits, 412 and personal computers, 455
of early computers, 188 robot of, 315
future breakthroughs in, 416-418, 419 ICOT (Institute for New Generation Computer
logic-gate description of, 157 Technology), Japan, 322, 345, 348
and von Neumann concept, 185, 205-206, 229, IDT (Intelligent Diagnostic Tool), 300
343 IKBS (intelligent knowledge-based systems),
new trend in, 340 346, 347
softwarelike design of, 122, 157 Illiteracy, and computer technology, 8
and Turing machine, 123 Imagination, 224
Harris, Larry, 308 visual, 234
Haugeland, John, 54 Imitation game
Hearing, 263 keystroke behavior in, 99-100
and automatic speech recognition, 265-270 Turing test as, 48-60, 81-83, 198 (see also
Hearing-impaired (deaf), aid for, 269, 278, 442 Turing test)
Heisenberg uncertainty principle, 116, 193. See unit of transmission for, 98-99
also Quantum mechanics Incompleteness theorem of Gédel, 115
Heuristic programming, 389 Industrial Revolution, 3, 6, 8
Hierarchy and image of machines, 89
of intelligence, 145 second industrial revolution, 7-9, 13
of knowledge, 284-286 Inference engine
Hilbert, David, 110, 112 in expert systems, 293-294, 301
quoted, 109 power in, 325
twenty-third problem of, 112, 115 Information
and Wiener, 190 in cybernetic view, 191, 195
Hillis, Daniel, 218, 229 vs. knowledge, 283 (see also Knowledge)
Hink, Robert, 302, 303 selective destruction of, 196-197, 223-224,
HiTech (chess machine), 130, 131, 291, 233-234
408 Information economy, 456-457
Hofstadter, Douglas R., 80-101, 377 Information processing, thinking as, 63
and character recognition, 241 Inheritance, principle of, 284
on Chinese Room example, 65 Inoue, Hirochika, 249, 250
‘on computer speed, 401 Institute for Economic Analysis (IEA), study by,
‘on knowledge, 210 5, 426-427
on machine distraction, 459 Institute for New Generation Computing (ICOT),
The Mind's |, 48 Japan, 322, 345, 348
quoted, 16, 124 Intellect program, 307, 308
“Holism,” 63, 65-66 Intelligence. See also Logic; Reason; Thinking
Hollerith, Herman, 171-172, 173, 176 and Al definition, 14, 15, 334-335
Holographic technology, 231, 413 approaches to modeling of, 139-148
Hopper, Grace Murray, 178-179, 183 artificial vs. natural (human), 46-47, 92, 139,
Hubel, David H., 226 152, 156
Human beings definition of, 16-18, 461
common biological origin of, 86 and embodiment, 38-39
indirect perception of, 84, 85-86 and emotions, 88-89, 219
uniqueness of, 69 and evolution, 18, 19-21, 120
Human beings and computers/Al factors in, 290
and brain, 92, 219 and fifth generation computer, 322
complementary abilities of, 455, 459 hierarchy of, 145
and functional isomorphism, 74-77 and human uniqueness, 7-8
and knowledge vs. search abilities, 17, 284, and incomplete knowledge, 17
290-292, 459, 460 as information selection, 197
and language abilities, 309 and knowledge, 284 (see also Knowledge)

558
 Minsky on, 214 Kanada, Takeo, 249
and NOR gate, 121, 124 Kandinski, Wassily, quoted, 355
and Perceptron, 139-141 Kant, Immanuel, 24, 31
and recursion, 124-131, 133, 205-209 (see also Kapor, Mitch, 372, 373
Recursion and recursive formula) Kay, John, 3
search for unified theory of, 121, 141 KEE (Knowledge Engineering Environment), 301
three levels of, 136, 138-139 KEE Pictures, 326
timeliness in, 130 Kennedy, John F., quoted, 401
Turing test for, 48-60 (see a/so Turing test) Kenner, Hugh, 97
understanding of, 19 Keys, invisible, 411
and universal Turing machine, 123, 124 Kierkegaard, Soren, 24
Intelligence, artificial. See Artificial intelligence Klatt, Dennis, 264
Intelligent assistant, 406-407 Kleene, Stephen, 103
Intelligent machines. See Artificial intelligence; Knowledge, 17, 284
Computers in AARON, 382-383, 385
Intelligent machines, early acquisition of, 381
abacus, 161 and Al, 209-212, 217
Analytical Engine, 89, 158, 165-167, 171, 185, in automatic speech recognition, 268
188 and brain simulation, 233
automata, 29, 159, 160 brain structuring of, 288-290
Difference Engine, 165, 167 commonsense, 327 (see also Common sense)
Leibniz computer, 164, 169 and computer vs. human abilities, 17, 284,
Napier's bones, 161 290-292, 459, 460
Pascaline, 162, 163, 169 in expert systems, 15, 56, 211, 292 (see also
Intelligent missiles, 255, 434-438 Expert systems)
Intelligent telephone-answering machine, 409- hierarchical classification of, 209-210, 282, 284-
410 286
Intentionality and intentional stance. See also vs. information, 283
Representations and language, 304, 306-307
and animals, 89 and library of future, 328
and humans as machines, 91 and PARRY, 55-56.
and human uniqueness, 69 and pattern recognition, 233, 238, 247, 271,
and machines as thinking, 461 289 (see also Pattern recognition)
in nonhumans, 100 as power, 325, 326, 328, 329, 434, 456
and Searle's critique, 64, 65-66 vs. recursive search, 290-292
Interactive computer games, 392 and semantic networks, 286-288, 289
Interface and Turing test, 51
computer-human, 326 (see also. volume of needed, 291-292
Communication) Wittgenstein on, 34
electronic music (MIDI), 352 and writing, 373
Internist program, 211, 296, 325 Knowledge Craft (knowledge-engineering
Intuition environment), 301, 332
in chess, 132, 409 Knowledge engineering, 56, 294. See also
and parallel processing, 232-233 Expert systems
IPL (Information Processing Language II), 198- Kobayashi, Koji, 405
199 Koch, Christof, 247
ISA (Intelligent Scheduling Assistant), 300 Kolodner, Janet, 56, 57
Israel, microelectronic agricultural systems in, Kowalski, Bob, 346
445 Kuipers, Benjamin, 57
IUP (Image Understanding Program), 212 Kuno, Susumu, 205, 304
Kurzweil, Raymond, 40-45, 152-157, 272-281
Jacobsen, Stephen, 318 374-379
Jacquard, Joseph-Marie, 165, 171 Kurzweil Applied Intelligence (KAI), 276-278
Jacquet-Droz, P., 159, 160 Kurzweil Computer Products (KCP), 272-275
Jam Factory system, 354 Kurzweil Cybernetic Poet, 374-379
Japan Kurzweil Foundation Automated Composition
automated animation in, 369 Project, 386
automatic language translation in, 309, 311 Kurzweil Music Systems (KMS), 278-280
fifth-generation computer in, 212, 294, 322,
323, 336-345, 348 Landauer, Rolf, quoted, 124
and innovation, 337 Language, 303-304
lifetime employment in, 4 and logic, 340
and material-resources importance, 8 pattern recognition for, 233 (see
and robotics, 247-248 recognition)
Jelinek, Fred, 264 speaking vs. listening 309, 312
Joral, 2 understanding of, 304-30 7
Jordan, Angel, 421 Wittgenstein on, 34
Judge's dilemma, 104-105, 106 Language, programming. See Programming
language

559
Language translation (LT), 405-406. See also McNamara, Robert, 436
Translation MacPaint program, 358
Law enforcement Macrophage, 147
fingerprint pattern recognition for, 252 MACSYMA program, 216
robots in, 317 Mandelbrot, Benoit, 361, 365, 367
Learning, 17 Mark | computer, 110, 167, 177, 178, 181, 188
automatic, 327-328 Mark Il computer, 178
for expert systems, 302 Mark Ill computer, 178
and prior knowlege, 326 Marr, David, 234
Leban, Roy, 97 Marshall, Wallace, 14
Lebowitz, Michael, 373, 390-393 Marx brothers, quoted, 294
Leibniz, Wilhelm, 31, 159, 164, 189 Masterman, Margaret, 370
Leibniz computer, 164, 169 Mathematics, 103
Leifer, Larry, 318 foundations of, threatened, 105-106
Lenat, Douglas, 327 limits of, 115, 116
Leonardo da Vinci Principia Mathematica, 108, 115, 199, 208
Mona Lisa as self-portrait of, 259, 262 Mathieu, Blaine, 74
war machines designed by, 435 Mauchly, John W., 167, 183, 185, 186, 187,

easementa
Leontief, Wassily, 426 188
Lettvin, Jerome, 146-147 Maze problems, 136
Library of the future, 328 Mazrui, Jamal, 274
Life Mead, Carver A., 249, 456
children’s conceptions of, 70-71 Meaning, and Turing test, 64, 65. See also
and DNA, 149-151 Intentionality and intentional stance
as molecule collection, 67 Media
Lighthill, Sir James, 347 and data structure, 289
Lighthill Study of Artificial Intelligence, 347 of future, 413-414
LISP programming language, 17, 34, 103, 202, Medicine
206 expert systems for, 57, 294-297, 299, 325
Literary arts, and computers, 370-379, 390-393 future impact of computer technology on, 438-
Litvin, Margaret, 46-47 44)
Logic. See also Reason pattern recognition in, 255, 258-261, 271, 438
fuzzy, 17, 36, 196, 293, 295 robots for, 317
limits of, 115, 116 Meehan, Jim, 392
and natural language, 340 Memory, 155-156. See also Knowledge
and parallel processing, 233 capacity of, 292
Platonic-like dialogue on, 45 computer vs. human, 448
predicate, 338, 340, 343 Platonic-like dialogue on, 42, 43, 45
Logical positivism, 25 semantic, 209, 210
of Ayer, 35 Menabrea, L. P., 167
and Kant/Enlightenment, 31-32 Meta-DENDRAL, 294
of Wittgenstein, 32, 34-35 Metaphors, 326, 327
Logical thought Michalowski, Stefan, 318
vs. parallel, 228, 235 Michie, Donald, 346
vs. pattern recognition, 223-229, 231 Microelectronics and Computer Technology
Logical transformation, 106-107 Corporation (MCC), 322
Logic gate, 121-122, 123, 152-157 MIDI (Musical Instrument Digital Interface), 352,
Logic programming, 338, 340, 345 353
Logic Theorist (LT) program, 199, 201 Military uses, 9
LOGO programming language, 78, 429, 430 for artificial vision, 255, 256
Logos system, 306 for auditory pattern recognition, 270
Lovelace, Ada, 117, 167, 168, 178 future impact of computer technology on, 425,
quoted, 165 434-438, 452
LT. See Language translation; Logical Theorist for robots, 317, 320
Lucas, George, 366 and Strategic Computing Initiative, 301
Ludd, Ned, and Luddite movement, 3-5 Mind, and machines, 23-24. See also Artificial
Lukasiewicz, Jan, 176 intelligence; Intelligence; Thinking
LUNAR program, 53 Minimax search, 127, 407, 409
Minsky, Marvin, 63, 78, 103, 139, 200
M (system), 354 on Al, 214-219
McCarthy, John, 100, 103, 199, 202 on books, 429
McCorduck, Pamela, 283, 324 on concept of machine, 460
MacCready, Paul, 320, 321 on consciousness, 461
McCulloch, Warren, 198, 205 and Dartmouth conference, 199
Machines, Minsky on, 214-215 and Dreyfus, 25, 36
Machine intelligence. See Artificial intelligence; on future libraries, 328
Intelligent machines, early and McCarthy, 202
McLuhan, Marshall, 406, 434 moving-frontier definition by, 14
quoted, 351 and neural nets, 141, 142

560
 Perceptrons, 139, 140, 205, 223, 229 Optical computing, 417
quoted, 23, 103, 303, 429 Oracle (robot), 317
on science fiction, 414 Organic circuits, 418
Semantic Information Processing, 209 Orwell, George, 445, 447, 454
and society of mind, 18, 142, 145-146, 217 Ovaltine system, 354
and theory of frames, 210-211, 284 Owens, Christopher, 290, 394-397
and Turing Test, 414-415
Mizoguchi, Hiroshi, 249 Painting. See Visual arts
Molecular computing, 418 Pandemonium selection, 142-143
Mona Lisa, identity of, 259, 262 Panpsychism, 88
Morland, Sir Samuel, 164 Papert, Seymour, 25, 62, 78, 139
MOVE program, 125, 127, 206 and computers in education, 429, 430
Moving-frontier definition of Al, 14 and Dreyfus, 36
Music ‘on learning, 431
analog vs. digital transmission of, 192-193, 195 and neural nets, 141
and computer technology, 271, 278-280, 351- Perceptrons, 139, 140, 205, 223, 229
355, 373, 378, 386-389 Paradoxes
and computer technology of future, 443-444 in reality, 27
robot to play, 318, 320 of Russell, 104-108
shift in style of, 115-116 Parallel processing, 218, 229-233, 235
Music Mouse system, 354 Al improvement through, 36
Muskie, Edmund, and CYRUS program, 57 and analog computing, 196, 227
MYCIN system, 56, 211, 294-297, 325 in automatic speech recognition, 266
Mysticism and brain, 228-229, 233, 234, 290
and consciousness, 88 in character recognition, 246
and Plato, 25-27 and chess playing, 131, 132
in Platonic-like dialogue, 45 and computer improvement, 403
in expert systems, 233, 301
Nagao, Makoto, 251 in fifth-generation computer, 339
NAND gate, 154-156 in inference engines, 294
Napier, John, 161 through laser light, 417
Negroponte, Nicholas, 303 and Microelectronics and Computer
Nelson, Ted, 428 Technology Corporation, 323
NeoMYCIN system, 296 in nervous system, 247
Neumann, John von, 178, 185, 188, 190, 338 new architecture of, 455
von Neumann computer architecture, 185, 205- and pattern recognition, 231-232, 235, 246-
206, 229, 343 247, 249, 271
Neural net or network, 38, 139, 140, 141-142, in robotic vehicle, 301
198, 205, 229 in skill acquisition, 231-232, 235
Newell, Allen and software, 340
and application-specific integrated circuits, 412 U.K. program for, 348
and chess programs, 325 in vision vs. logic, 228-229
creations of, 135, 198-199, 201, 292, 420 Parnas, David, 452
and Dreyfus, 36 “Parrot program,” 93
on fairy tales and technology, 420-423 PARRY, 54-56, 92-93, 346
and intelligence, 18, 54 Pascal, Blaise, 162
and Principia Mathematica, 108 Pascaline, 162, 163, 169
on thinking, 63 Pattern recognition, 223-224, 235-238, 271
Newton, Isaac, 24, 29, 29-31 accomplishment level of, 459
“Nicolai” (program), 93-98, 100 in auditory perception, 270-271, 411
Nietzsche, Friedrich, 24 for automatic speech recognition, 265-270,
1984 (Orwell), 445, 447, 454 276-278
90-10 rule, 212 for characters, 238-247
Nishimoto, Yoshiro, 249 in chess, 291
NOR logic gate, 121-122, 123, 124, 152-156, computer vs. human ability at, 449
157 for computer vision, 247-262
Norman, Donald, 38, 100 of faces, 233
NTC (Network Troubleshooting Consultant), 300 for fingerprint identification, 252, 411
Nuclear weapons, no-first-use policy on, 435- fuzzy logic in, 17
436 and knowledge, 271, 289
Numbers, irrational, 26 in medicine, 255, 258-261, 271, 438
military uses of, 255, 256, 434-438
Oakley, Brian, 346-349 Minsky on, 215
Odex (robot), 317 and music, 271
Ogborn, Rod, 93, 98 optical character recognition, 249,
ONCOCIN system, 296, 299, 326 optical computing for, 417
Operationalism, 52 and parallel processing, 231-232
Optical character recognition (OCR), 249, 272- 249, 271
275 projects on, 212

561
Pattern recognition (cont.) Programming language
in visual perception, 224-228 recursive, 208-209
People. See Human beings Russell's concept of, 108
People, artificial, 412-413 PROLOG language, 34, 36, 108, 301
Pepper, Jeff, 330-335 Prospector system, 56, 211-212, 299-300
Perception Proverbs, and creativity, 396
analysis of, 76-77 Prusinkiewicz, Przemyslaw, 361
and computer functioning, 271 Psychiatry or psychoanalysis
smell, 271 and ELIZA, 16, 36, 78, 93, 202, 204
taste, 271 and PARRY, 54-56, 92-93, 346
touch, 271 and thinking-feeling distinction, 72
and Turing test, 59 Pterodactyl, robot model of, 320, 321
visual, 224-228, 229, 231, 232, 233, 247-262, PUFF system, 296, 299
418 PUMA (robot), 315
Perceptron, 139-141, 205, 229 Pylyshyn, Zenon, 65
Perceptron (company), 255
Personal computers, 373, 431, 432, 454-455 Quantum mechanics, 116, 462
Personal history, and Turing test, 59 Quick-probe assumption, 50, 53. See also Turing
Person-identification systems, 270, 411 test
Persons. See Human beings and ambiguous sentence, 51
Phenomenology, 24, 25 and chess, 51
Philosopher-computer conversation, 74 for Dennett test, 52-53, 58
Philosophers and test for German spies, 98
children as, 70 Quillian, M. Ross, 209, 210
and mind-machine relation, 24-25
on reason and mysticism, 25-32, 34-35 RACTER program, 370, 373
Physical-symbol-system hypothesis, 325 Rado, Tibor, 113
Physics RAIL robots, 315
Paradox in, 27 Random fractals, 361-366
shift in, 116 Rationality. See Reason
Piaget, Jean, 70, 71 Reading, parallel processing in, 232
Pilot's Assistants, 434 Reason, 17. See also Intelligence; Logic;
Pilot's Associate system, 301 Thinking
Ping-Pong playing robots, 317 and children’s view of computer, 72
Pitts, Warren, 198, 205 and computer-human comparison, 69
Pixel, 225-227 and mysticism, 25-32, 34-35
Pixel intensity, 235 and Plato, 25-27
Pixel size, 363 symbolic, 17-18
Plankalkul language, 176, 190 Recursion and recursive formula, 124-131, 138,
Plato, 24, 25-27, 62 143, 205-209
“Platonic” dialogue, on human thought, 40 application of, 134, 135
Platonists, 27-28 in fractals, 361-366
Poetry and knowledge, 290, 291
program-written (haiku), 370 and stored program, 185
shift in style of, 116 Reddy, Raj, 124, 132, 247, 248, 290, 291, 294
Turing test on, 374-379 Redundancy
Poggio, Tomaso, 234, 247, 248 in brain vs. computer, 75, 234
Pohl, Frederick, 219 in brain data structure, 288
Politics, future impact of computer technology Relativity formula, general, 119
on, 445-447, 452, 457 Remington Rand Corporation, 173, 177, 186
POPLOG language, 346 Representations, and Al, 217, 451-452. See also
Postindustrial society, 8, 13, 329, 452 Intentionality and intentional stance
Pragmatic analysis, 306 Return of the Jedi (Lucas film), 366
Predicate logic, 338, 340, 343 Rich, Elaine, quoted, 14
Principia Mathematica (Whitehead and Russell), Richter, Hans, quoted, 32
108, 115, 199, 208 Risk taking, and humans vs. expert systems,
Principle of inheritance, 284 302-303
Probabilistic rules, and humans vs. expert RMS (robot), 317
systems, 302-303 Robinson (machine), 110, 177
Problem-solving. See a/so Expert systems; Robotic hand, 317-318, 320
Intelligence and Al, 15 Robotic person imitator, 433-434
creativity in, 394 Robotics Institute (Rl), Carnegie-Mellon
General Problem Solver, 124, 135-136, 199, 201 University, 247
by intelligent assistant, 407 Robotic vehicle, military, 301
recursion in, 206-208 Robotic-vision systems, 247-249, 253, 315
troubleshooting model for, 332 Robotic Vision Systems (company), 255
Professional Composer system, 354. Robots and robotics, 312-322
Programming, computer. See Computer exoskeletal, 442
programming of future, 414

562
 humanoid, 61 SHRDLU program, 205, 307
Minsky on, 216-217 SIGART (Special Interest Group on Artificial
with sense of touch, 271 * Intelligence), 78
in Star Wars, 90, 309, 312 Sight. See Visual perception
Rochester, Nathaniel, 190, 199 Silicon chips. See chips, computer
Rosenblatt, Frank, 205 Silicon compiler, 122, 455, 456
Rosser, Barkley, 103 Simon, Herbert
Rucker, Rudy, quoted, 283 and chess programs, 325
Rumelhart, David, 38 creations of, 135, 199, 201, 420
R.U.R. (Capek play), 312 and Dreyfus, 36
Russell, Bertrand, 32, 103 on intelligence, 54
and computer, 176 prediction by, 117
paradoxes of, 15, 104-108 and Principia Mathematica, 108
Principia Mathematica, 108, 115, 199 on self-esteem threat, 60
and Turing, 109 on thinking, 63
and Wiener, 190 Simulation, 83, 84. See also Imitation game
of brain, 233-234
SAINT program, 216 of evolution, 361
Salisbury, Kenneth, 318, 320 of hurricane, 83-85
SAM (Script Applier Mechanism), 211 of person, 42
Samuel, Arthur, 124, 199, 325 of reason, 69
Sanders, Jerry, 456 of thinking, 63
Sartre, Jean-Paul, 24-25 of thinking vs. feeling, 72
SBDS (Service Bay Diagnostic System), 330 SIR program (Semantic Information Retrieval),
Schank, Roger, 308 209
and Cognitive Systems, 307 Slagle, James, 215-216
and creativity, 290, 394-397 Sloman, Aaron, 346
and CYRUS, 56 “Smart” products, 81
on intelligence, 15 “Smart” weapons, 9, 255, 425, 434-438
on knowledge, 60, 209 SMPTE (Society of Motion Picture and
and language comprehension, 303 Television Engineers) Time Code, 352
and SAM, 211 Smullyan, Raymond, 93
Schools. See Education Soap opera, 390-391
Schwantz, Lillian, 259, 262 SOAR system, 292
SCI (Strategic Computing Initiative), 9, 301 Social impact of Al, 450-453
Science and Kurzweil industries, 281
compartmentalization in, 189, 198, 281 Social manipulation, 60
as message of despair, 66 Social problem, in overstating cognitive power,
as message of hope, 67 59
Sea of logic, 121-123, 124, 152-157 Society of intelligences, 148
Searle, John, 63-64 Society of mind, 142, 145-146, 217
Second industrial revolution, 7-9, 13 Society of Mind (Minsky), 18, 217
Segmentation Society of neurons, 146-147
in automatic speech recognition, 267-268, 270 Socrates, 25
in pattern recognition, 235 Software, computer. See also Computer
Sejnoha, Viadimir, 276 program
Self-consciousness, and computers, 60-61. See factors in cost of, 8,9
also Consciousness future development of, 418-419, 432-433
Selfreference hardware related to, 122, 157, 340
by “Nicolai,” 94 improvement in, 402, 404
in recursion, 125 (see also Recursion and logic-gate description of, 157
recursive formula) and military needs, 301
Selfridge, Oliver, 199 and personal-computer revolution, 454
Self-similarity, 363, 364 Software crisis, 343
Semantic analysis, 304 Solipsism, 88
Semantic memory, 209, 210 Soul. See also Consciousness
Semantic networks, 286-288, 289 and computer-human distinction, 72
Seneff, Stephanie, 265 and intentionality or intentional stance, 64, 89
Senses. See Perception ‘on computationalist view, 66
Serwen, Michael, 462 Platonic-like dialogue on, 41, 43-44
7535 robot, 315 Speaker identification, 270, 417
7565 robot, 315 Speech, and Dreyfus vs. Al, 36
Shane, Howard, 269 Speech recognition, automatic (ASR), 265-270,
Shannon, Claude, 103, 193, 198, 199 276-278, 405, 407, 456
Shakespeare, William, quote from, 447 Speech synthesis (SS), 405, 406
Shaw, J. C., 108, 135, 199, 201 in Reading Machine, 274
Shiliman, Robert, 254, 255 Spinoza, Benedict de, quoted, 19
Shirai, Yoshiaki, 249, 250 Spirit. See Consciousness; Soul
Shortliffe, Edward H., 211 Star Wars (movie), 90, 309, 312

563
Index

Stereo vision, 228, 234 Translation, 306

Story telling, through Al, 390-393 accuracy of, 373
Strategic Computing Initiative (SCI), 9, 301 automatic, 306, 307-309, 405-406
Strindberg, August, quoted, 443 Tree of possibilities, 127, 130, 143, 382
STUDENT program, 202, 203 Troubleshooting model, 332
Subjectivity, and Al as dehumanizing, 452. See Tsujii, Jun-Ichi, 310
also Consciousness T3 (The Tomorrow Tool), 312
Sun Tzu, 434 Turing, Alan, 48, 80, 100, 109, 346
Supercomputers, Minsky on, 217-218 and Al, 103, 190, 198
Superconductivity and Church-Turing thesis, 117
and chip geometries, 234, 417 and computation theory, 110, 112, 113, 115
and parallel processing, 418 and Enigma code, 109, 110, 175, 177, 188
SUR (Speech Understanding Research) project, on Imitation Game, 98-99
212 quoted, 14
Symbolic reasoning or manipulation, 17-18, 63, Turing machine, 48, 112, 113, 122-123
66. See also Representations and computers, 48, 123, 337-338, 343
Synonyms, and computer-human interaction, Feigenbaum on, 325
326 and Russell, 107, 108
Syntax, 304 universal, 123-124
Synthesizers, computer-based, 271, 278-280, Turing test, 48-60, 62, 100-101, 116-117, 198,
352 374, 377-378
vs. alternative tests, 51-52
Tabulating machines, 171 Chinese Room example of, 64-65
Talkwriter, 456. See also Speech recognition, coffeehouse conversation on, 80-93, 377
automatic computer reading compared to, 448
Tanaka, Hidehiko, 231 domain-specific, 116, 374, 378, 415
Teaching. See Education and Dreyfus vs. Al, 38, 414-415
Technology, and fairy tales, 421-423 and ELIZA conversation, 78
TEIRESIAS system, 296, 298 with face-to-face communication, 448
Telephone future passing of, 414-416
future development of, 433, 434 and operationalism, 52-53
translating, 405-406 on poetry, 374-379
Telephone-answering machine, intelligent, 409- in reverse, 97
410 Turkle, Sherry, 68-73
Tele-Spin program, 392 2001: A Space Odyssey (film), HAL in, 312
Television
emotional presentation in, 289 Ultimation 2000 (robot), 312
of future, 413-414 Ultra project, 177-178
Template matching Uncertainty
in automatic speech recognition, 268 and Al, 322
in character recognition, 246 and decision making by humans and expert
by neural net machines, 229 systems, 302-303
in optical character recognition, 272 Heisenberg uncertainty principle, 116, 193
Terregator (Terrestrial Navigator), 247 Understanding. See Intelligence; Reason;
Theaetetus, 28 Thinking
Theory of everything, 120-121 Underwriting Advisor system, 299
Theory of sets (types), 104, 107-108 Unified field theory, 120-121
Thermodynamics, laws of, 119-120 Unifying formulas, 119-121
Thinking. See also Intelligence; Reason as United Kingdom (England)
beyond automatic, 15-16 Al in, 346-349
computational vs. holistic view of, 63 computer research in, 323
and computers, 62 UNIVAC (Universal Automatic Computer), 186,
and emotions, 72-73, 87, 88-89, 219 187, 188
as logical process vs. pattern recognition, 223- Universal Turing machine, 123-124
224, 228-229, 246 Universe program, 391, 392-393
and machines, 48-61, 460-462 Unsolvable problems, 113, 115
parallel vs. sequential, 231-233, 235 Ussachevsky, Vladimir, quoted, 351
Platonic-like dialogue on, 40-45
as subvocalized speech, 263 VAL language, 315
Third World (emerging nations), 8 Vance, Cyrus, and CYRUS program, 56-57
Thomas Aquinas, Saint, 66, 67 Videophones, 434
Tic-tac-toe, 128, 129, 134 View Engineering, 255
Time Visual arts, and computers, 355-369, 373, 380-
Platonic-like dialogue on, 44-45 385
Wiener on, 196-197 Visual perception, 224-228
Torriano, Giannello, 159 through computers, 229, 247-262, 418
Totalitarianism, and computer technology, 8 degradation in, 231
Tower of Hanoi problem, 206-209, 223 logical and parallel processing in, 246
Translating telephone, 405-406 organization of, 233

564
 parallel processing in, 232
VLSI (very large scale integrated) circuit chip,
122, 131, 227, 322, 323
and fifth-generation computer, 341, 342, 343
Vogt, Eric, 272

Wabot-2 (Waseda Robot), 313, 318, 320

Waldrop, Mitchell, 62-67
quoted, 447
Walker, Timothy, 425
Waltz, David, 401
Wang, An, quoted, 223
Warfare. See Military uses
Watson, J. D., 149
Watson, Thomas J., 172-173
Weapons, “smart,” and pattern recognition, 9,
255, 425, 434-438
Weizenbaum, Joseph, 16, 55, 36, 63, 93, 202,
204
Whitehead, Alfred N.
Principia Mathematica, 108, 115, 199
quoted, 25
Wiener, Norbert, 103, 190-198
‘on human knowledge, 460
quoted, 23, 189, 281
Wiesel, Torsten N., 226
Wilkes, Maurice V., 175, 186, 188
Williams, William Carlos, 116, 374, 378-379
Williamson, Jack, 219
Winograd, Terry, 51, 205, 307
Winston, Patrick, 201
Wittgenstein, Ludwig, 25, 32, 34-35, 35
and Church-Turing thesis, 117
and Dreyfus, 38
on problems as answerable, 115
quoted, 36, 38
Wonder, Stevie, 278
Wood, Robin McKinnon, 370
Wood, William, 53
Woods, David, 302, 303
Word processing, origin of, 370
Work, impact of computer technology on, 4-5,
425-428, 452, 456-457
World knowledge. See Knowledge
Wright, Robert, quoted, 189
Writing, and computers, 371-373

XCON system, 212, 292-293, 300

Yoshikawa, Tsuneo, 318

“Zen holism,” 63, 65-66

Zieber, Fred, quoted, 401
Z machines, 176-177, 188, 190
 “A tour de force which illuminates a rich and diverse subject in an entertaining and
informative style. It is imperative that those of us involved in creating and participating in
TECHNOLOGY

the remarkable decades ahead understand and ponder Mr. Kurzweil’s vision of the
future.”
ROBERT NOYCE
President/CEO, SEMATECH; cofounder
02142

and former Chairman/CEO, Intel

Corporation; coinventor of the integrated
circuit
OF

MASSACHUSETTS
INSTITUTE

“A stunningly comprehensive yet imminently accessible account of human thinking about

thinking machines.”
MITCHELL KAPOR
President, ON Technology, Inc.; founder
and former Chairman/CEO, Lotus
Development Corporation; cocreator of
MASSACHUSETTS
PRESS

Lotus 1-2-3
CAMBRIDGE,

“Ray Kurzweil has written a fascinating, understandable, and compelling book about the
MIT

"second revolution’—the age of intelligent machines. It promises to make the Industrial

Revolution pale by comparison.”
THE

DAVID T. KEARNS
Chairman, Xerox Corporation

“\ superb survey of artificial intelligence, from the theory to the technology, by the
pioneer of the field.”
GEORGE GILDER
Author of Wealth and Poverty,
of The Spirit of Enterprise, and of
Microcosm

“The Age of Intelligent Machines is destined to become the definitive guidebook of the
past, present, and future of artificial intelligence Ray Kurzweil’s omnibus is provocative,
captivating, and highly readable.”
JONATHAN ROTENBERG
Chairman, Boston Computer Society

“Ray Kurzweil’s insightful book fills a major void in the field of artificial intelligence. By
writing a book that a nonprofessional can understand and appreciate, Ray has done
a great service to the community. It represents a comprehensive single source of infor-
mation on the evolution of intelligent machines and the scientists who work on them.”

RAJ REDDY
Founder and director, Robotics Institute,
Carnegie-Mellon University

“A unique and brilliant combination: the inside view of the technology from a leading
a
pioneer of the field, tied to a full view of the history and philosophy of man. This is
must-read book for anyone interested in intelligence, artificial or natural.”

DAN BRICKLIN
Cocreator of Visicalc, the first electronic
spreadsheet

KURAH

O- 262-10 etee