7

FOUR CHALLENGES TO
FUNCTIONALISM

If functionalism is true, anything that is functionally like us in the

relevant respects is psychologically like us. This chapter is concerned,
first, with three well-known examples o f things that are, in one way or
another, and to one extent or another, functionally like us and yet which
intuitively are very unlike us psychologically. We will consider in turn the
challenge posed to functionalism by the China Brain, the Chinese room,
and Blockhead. In each example we consider (a) whether the example
really is, on reflection, o f a creature that is very unlike us psychologically,
and (b) to the extent that it is, we ask whether the example shows that
functionalism is false, or does it instead teach us something important
about which functional roles are crucial? We will suggest that to the extent
that the examples are indeed o f entities that are psychologically unlike
us, they are examples o f entities that do not have the right functional
similarities to us, and so tell us not that functionalism is false but rather
about the functional roles that functionalists need to include in their story
about psychological nature. We will need to advert at various points to
the fact that functionalism comes in many flavours in our discussion o f
the three examples. Fortunately, however, the examples raise man)' rather
general issues about functionalism, and this means that we will often be
able to think in terms o f a fairly undifferentiated version that captures
general features common to most versions.
In the final section we will discuss a general objection to any kind
o f functionalism or indeed physicalism from the alleged possibility' o f
‘zombies’ .

The China Brain

The China Brain is a putative counter-example to functionalism due origin­
ally to Ned Block. Here is a slighdy updated version o f the example. Imagine
 108 R ivals and O bjections

that artificial intelligence has advanced to the point where a program can
be written which will allow an android with a ‘brain’ consisting o f a com­
puter running the program to behave actually and counterfactually much
as a normal human does. It does not matter for the example how this pro­
gramming is done; to avoid confusion about the nature o f the program
(which we will discuss in a later example), let us suppose that the pro­
gram mimics the operation o f a human brain at a neuron by neuron level.
Neurons are essentially ‘input-output’ devices made from organic matter,
the overall input—output characteristics o f the brain being determined by
how the primitive neuronal devices are assembled. Hence, this supposition
amounts to having the program reflect precisely the input-output nature
o f each neuron and how they are connected one to another.
The next step in the process o f constructing the example is to note
that it won’t matter, or anyway can hardly matter from a functionalist
perspective, if the computer running this program is in fact outside the
android’s body, connected by a two-way radio link to it. The final step
gives us the China Brain. Suppose that instead o f the program being run
on an external computer made o f silicon chips, the entire population of
China is enlisted to run the simulation. As the program mimics the way
the brain operates at the neuronal level, this can be done by assigning each
Chinese citizen the job o f just one neuron. They have, let’s suppose, the
kind o f phones that tell you what number has called you. When certain
numbers, or combinations o f numbers, ring in, they have to dial specified
other numbers. Each citizen is given a precise set o f instructions about
what to do that ensures that what each does exactly models what their
assigned neuron does, and the inputs to and outputs from their phones
are connected up so as to run the program. Also, the initial inputs to the
China brain come from the environment in much the same way as the inputs
to us do, and the final outputs go to the limbs and head o f the android
via the radio link in such a way that its actual and counterfactual beha­
viour is much as ours is. Thus, the android will behave in the various
situations that confront it very much as we do, despite the fact that the
processing o f the environmental inputs into final behavioural outputs goes
via a highly organized set o f Chinese citizens rather than a brain.
This is certainly not a realistic fantasy. The population o f China is not
large enough; the whole process could never take place fast enough; the
citizens would get bored and careless; and anyway the program used
to construct the example does not exist and never will (working at the
neuronal level is ridiculously fine-grained). All the same it does seem clearly
intelligible, and if it is intelligible, it is fair to ask for an answer to the
question whether the system consisting o f the robot plus the population
o f China in the imagined case has mental states like ours. Many have a strong
 Four C hallenges to F unctionalism 109

intuition that it does not. If they are right, functionalism o f just about
any variety must be false. For the system is functionally very like us. Not
only is it like us in all the functional roles seen as crucial by the common-
sense functionalist, it is like us in just about every functional respect.
Functionally, it is us; the difference lies in the dramatic difference in how
die functional roles are realized, and that difference counts for nothing
as far as mental nature is concerned according to functionalists.
We think, however, that the functionalist can rea­
sonably deny the intuition. The source o f the intui­
Denying the intuition
tion that the system consisting o f robot plus China
brain lacks mental states like ours seems to be the fact
that it would be so very much bigger than we are. We cannot imagine
‘seeing’ it as a cohesive parcel o f matter. We cannot see, that is to say,
the forest for the trees. A highly intelligent microbe-sized being moving
through our flesh and blood brains might have the same problem. It
would see a whole mass o f widely spaced entities interacting with each
other in a way that made no sense to it, that formed no intelligible over­
all pattern from its perspective. The philosophers among these tiny beings
might maintain with some vigour that there could be no intelligence
here. All that is happening is an inscrutable sending back and forth o f
simple signals. They would be wrong. We think that the functionalist
can fairly say that those who deny mentality in the China brain example
are making the same mistake.
Before we leave the China brain example, we
Consciousness
should note two important points about its role in the
literature. First, it is sometimes directed simply to the
question o f whether functionalism can account for consciousness. In this
manifestation it is granted that the China brain has beliefs and desires
(after all, the robot will move in various ways in response to the environ­
ment and thereby make changes to it o f just the kind we associate with
purposive, informed behaviour), but it is insisted that it is absurd to hold
that it feels anything. We discuss the difficult question o f feeling and
consciousness in the next chapter. Our concern in this chapter will be
restricted to challenges for functionalism about mental states like belief
and desire, and mental traits like being intelligent.
Secondly, sometimes the example is given in a
Connection to the
version that omits the robot. But then the population
environment
o f China is emulating, in some purely abstract way,
the program in someone’s brain with no obvious right
way to connect the overall inputs and outputs with the environment.
The case becomes essentially the same as the one we discussed when we
considered the charge o f excessive liberalism against certain machine
110 R ivals and O bjections

versions o f functionalism in chapter 5. There we argued that merely

crunching numbers - or more generally inputs and outputs that have no
natural or obvious connection to the environment that our mental states
are about - is mere number (or whatever) crunching. I f this is right,
then the China brain example in the version that omits the appropriate
robotic connection to the environment is an example o f something that
lacks a mental life, or anyway a mental life at all like ours. But in this
version it is not an objection to functionalisms that include, in one form
or another, the right sort o f connection to the environment. These func­
tionalisms might be called ‘arm’s length’ functionalisms. Common-sense
functionalism is an ‘arm’s length’ doctrine in this sense.

The Chinese Room

John Searle’s Chinese room is one o f the most famous examples in the
philosophy o f mind. We will present a variation on the original example.
We suppose that someone called Tex, who understands English but not
Chinese, is locked in a room that has an in-chute, an out-chute and a
book full o f instructions in English concerning the manipulation o f Chinese
characters. Stories in Chinese accompanied by questions in Chinese about
the stories come in through the in-chute. Tex follows the instructions in
the book as applied to the stories and the questions, which in due course
tell him which sentences in Chinese to copy onto pieces o f paper and
place in the out-chute. Tex does not understand the stories, the questions
or the sentences he puts in the out-chute. As far as he is concerned he
is simply operating with squiggles. He is mechanically following some rules
in English for manipulating symbols which are in various ways derived
front the squiggles, and which conclude with his writing down some
more squiggles on the paper that goes into the out-chute. We can make
the example more up to date (and facilitate later variations on it that we
will discuss) by supposing the stories and questions in Chinese are typed
into a computer outside the room and appear on its screen. Tex has a
monitor in his room on which what is typed is also displayed in Chinese.
He types in answers to the questions at a separate keyboard in the room,
following the book’s instructions religiously. His answers appear both
on his monitor, and on the monitor screen outside the room. We will
conduct the discussion in terms o f this version o f the example.
Searle in effect points out that the book might well be such that Tex
will consistently deliver Chinese sentences that, to someone who under­
stands Chinese, count as sensible, intelligent answers to the questions in
Chinese about the stories in Chinese that Tex receives. What appears on
 Four C hallenges to F unctionalism 111

:he screen are good answers in Chinese to questions in Chinese about

¡tories in Chinese. Nevertheless, it would be quite wrong to infer that Tex
understands Chinese. All he is doing is manipulating symbols according
to formal rules without any understanding o f what the various symbols
stand for or mean.
It is clear that Tex does not understand Chinese, for he does not him­
self have the ability to answer the questions. It is Tex together with the
book that has the ability. So the issue that needs to be addressed is whether
the system consisting o f Tex plus the book understands Chinese. What
abilities are distinctively associated with understanding a language? It is
plausible that being able to answer comprehension questions about a range
of stories is part o f what is required, but is very much less than all that is
required. One thing we need to add is the ability to extemporize, embellish,
and generally display the inventiveness and flexibility o f a natural language
speaker. I f we always get back the same answer - accurate and intelligent
though it may seem - in response to a story together with a question,
we might well start to think that we are interacting with an automaton
rather than a thinker and understander.
We can, though, embellish the original example by
The example
supposing that Tex plus room has all these capacities.
embellished
We can suppose that the book Tex is following does
not always deliver the same answer in Chinese to a
given question in Chinese. The book takes account o f whether or not
a question has been asked before. It contains instructions in English
concerning what to do when a given sequence o f squiggles (as Tex thinks
of them) appears on the computer screen that takes into account whether
and how often that sequence o f Chinese characters has appeared before.
Obviously, by making Tex’s book o f instructions sufficiently complex
(it will have to include instructions on how to modify the book itself in
response to input, even if only by leaving different pages open), we can
ensure that the answers Tex generates on the screen are exactly those that
would come back from an intelligent Chinese speaker. That is, Tex plus
the room passes what is known as the T urin g test: to pass the Turing
test is to respond to questions with all the signs o f intelligence and thought
distinctive o f thinkers like us.
But at this point many people’s intuition that we are dealing with
something that does not understand Chinese starts to fade. It is still, o f
course, true that Tex does not understand Chinese, but an awful lot o f
processing is being done by Tex plus the book, and many argue that it
is enough for the system to count as understanding Chinese.
For our part, though, we think that even after this embellishment, the
system does not understand Chinese. In order to count as understanding
 112 R ivals and O bjections

Chinese, we need, amongst other things, the kinds o f abilities distinctive

o f understanding what Chinese sentences and their parts stand for. The
system does not even understand what the word ‘book’ stands for, because
it cannot respond in the appropriate way to a book. It only ever responds
to sentences, not to what they or their parts stand for.
We can put this point in terms o f the distinction between semantics and
syntax. Syntax has to do with questions o f grammatical propriety, sentential
structure, whether a word is a verb or a noun, and so on. Semantics has
to do with the interpretations that attach to words and sentences; it relates
to what words and sentences mean. It is in virtue o f having a semantics
that words and sentences in a language can serve to make claims about
how things are, ask questions, issue commands and so on. Understand­
ing a language is a matter o f mastering its semantics. Tex has mastered
the semantics o f English and in fact uses this mastery to follow the book’s
instructions. But as far as Chinese is concerned, all Tex has are certain
syntactic abilities. He can match up Chinese characters on a screen with
those in the book and type them into a keyboard, but he has no idea what
they stand for or mean. In this terminology, the question we face at this
stage is whether, in the embellished example, the system o f Tex plus the
book has a grasp o f the semantics o f Chinese. And our claim is that the
system does not, for it does not understand what Chinese words stand
for. This is manifested in the fact that it cannot respond appropriately to
what Chinese words stand tor. The system can only respond to words as
they appear on a computer screen, not to what they stand for.
We could, o f course, further embellish the example. Imagine a robot which
sends information about the inputs through its eyes and ears and surfaces
by radio in digital form to the Chinese room. Tex is still
inside, and responds to these by writing them down,
A further
sifting through the book, and following its directions
embellishment
as before. After much ( even quicker than before!) cal­
culation, he goes to the console and types a response
which is relayed to the robot. This makes the robot move through and
respond to its environment much as we do. We can suppose that the book
is detailed enough to constitute a program for the mind o f a Cantonese­
speaking adult woman. Thus, the robot will interact with its environment,
including answering questions, in the very same way she would.
But now the intuition that we are dealing with something that does
not understand Chinese has faded completely, or so it seems to us. It is
still the case that Tex does not understand Chinese,
but there is an entity - call her Lin - that does under­
A system that does
stand Chinese. Lin is composed o f the book that con­
understand Chinese
tains the all-important program, parts o f Tex’s brain,
 Four C hallenges to F unctionalism 113

the robot and the radio. Lin, «'hose robot body is entering Kowloon,
might believe she was entering Kowloon. Tex, in a laboratory in Dallas,
might not even know that Kowloon exists, let alone believe he or anyone
else is entering it.
We should say something quickly about a response a Searle-like figure
might make to replies like this. He might think that it depends crucially on
there being a system bigger than Tex that does much o f the work. This,
he thinks, is what makes it seem plausible that the system is distinct from
Tex, and can thus understand something that Tex doesn’t. He might
ask us instead to imagine that Tex memorizes the book, and stores all
the changing data in his head. We are supposed to think that in this case
there is no plausible distinctness between Tex and Lin, so if one tails to
understand Chinese, so does the other. Tex doesn’t understand Chinese,
so nor does Lin.
We do not think that this variation makes an important difference. What
we would have here is two entities who' share a brain. The idea that dis­
tinct individuals might share a brain is not enormously different from
that which we have grown used to in discussions o f multiple personality
disorder. There is some difference here- Lin relies on Tex to do the cal­
culations that constitute her mental states (so if he gets bored or ill, it’s
very bad news for her indeed).
It may seem puzzling that Tex does all these ~ '
. . . , - , . . . . . . A computer analogy
calculations without knowing what it is he is doing. ________________
But in fact something like this is commonplace in
computer science. When one computer does calculations that emulate
the behaviour o f a different machine, the emulated computer is said to
be a virtual machine. You may have seen an Apple Macintosh computer
which has a window which has the look and feel o f a machine running
Microsoft Windows. In fact, this is done by the Macintosh operating sys­
tem directing that calculations be done at the binary level that emulate
the behaviour o f the Intel chip on which Windows runs. I f you ask the
Macintosh operating system what menus appear in its windows, it will
be able to tell you. But if you ask it what menus appear in the Microsoft
Windows lookalike window that it is supporting by emulating the Intel
chip, it won’t be able to tell you. It doesn’t have information about that
process at that level o f abstraction. I f you interrogate Windows, however,
it can tell you about its windows. This is roughly analogous to asking
Tex about Kowloon directly, and drawing a blank, but getting an answer
when you interrogate Lin.
In sum, the Chinese room example starts out as one where both intui­
tion and any plausible functionalism agree that there is no understanding
of Chinese. We can add to the example to get one where plausible versions
 114 R ivals and O bjections

o f functionalism will have to say that there is understanding o f Chinese

(by the relevant entity, not Tex), but in doing so we turn the example
into one where intuition also says that there is understanding o f Chinese.

Blockhead
We noted in chapter 5 that the most popular version
Input-output
o f empirical functionalism is exposed to the charge
functionalism
____ o f chauvinism. It insists on an excessive degree of
internal similarity to us before something counts as
having a mind; beings might fail to have minds by virtue o f having
internal processors which are better than ours! Would it be right to
take the extra step o f holding that all that matters for having a mind is
being such as to ensure the right connexion between external inputs
and outputs? Something is an amplifier if it is such as to secure the right
relationship between inputs and correspondingly bigger outputs, no
matter how the job is done internally. What is done, not how it is done,
is what counts. Should we say the same about the mind? Such a position
can insist on specifying the inputs and outputs in arm’s length terms,
as is done in common-sense functionalism, and that what goes on inside
matters to the extent that the job o f appropriately mediating between
the environmental inputs and behavioural outputs must be done by what
is inside. But that would be the extent o f its constraints. Such a view
might be called in p u t-o u tp u t or stim ulus-response functionalism . It
takes on board what is right about behaviourism - that behaviour in
situations is crucial - but remedies at least part o f what is wrong with it.
Mental states are internal, causally efficacious states, pace behaviourism,
but internal states that can be characterized fully as far as their psycho­
logical nature is concerned in terms o f the behaviour that they do and
would typically produce, or do or would produce if linked up in some
natural way to the body. Input-output functionalism can be distinguished
from supervenient behaviourism by the fact that the input-output
functionalist insists that (most) o f the states causally responsible for
the behavioural profile must be internal. Suppose that Jane’s normal-
seeming behavioural profile is caused by puppeteers acting at a distance.
The supervenient behaviourist might think she had mental states like ours;
the input-output functionalist would not.
Input-output functionalism is false. A now famous example due to
Ned Block shows that the way the job is done does matter. There are
substantial internal constraints on being a thinker. The remainder of
this chapter is concerned with describing his example - the Blockhead
 Four C hallenges to F unctionalism 115

example - and what is to be learnt from it. We approach the example,

following his lead, via some remarks about chess.

Good chess versus being good a t chess

It is possible to play good chess without being any

Copycat chess
good at chess. Do what expert chess players tell you
to do! Your good chess will then be a sign o f the
ability to follow instructions and perhaps the ability' to identify chess
experts, but not o f ability at chess. Alternatively, we can imagine that instead
o f turning to the experts for advice you turn to a chart or look-up tree
prepared by experts. Here is how the table could be constructed for when
you are playing Black. At the beginning o f a game there are only finitely
many moves allowed by the rules. The chess experts nominate the best
response to each possible move. We obtain a little table like figure 7.1:
the boxes represent the possible opening moves by White, and the circles
the responses to each nominated by the experts.
Now for each response by you as Black to White’s opening move, there
will be only finitely many legal responses by White. So we can extend the
table by asking the experts to nominate their best response to each possible
response by White (figure 7.2). Again, the boxes represent White’s possible
moves and the circles the responses to each as nominated by the experts.
It is easy to see that this sort o f diagram can in principle be extended as
long as you like and that anyone who owned such a diagram - perhaps
prepared by Grand Masters over thousands o f years - would be able to
play very good chess without necessarily being any good at chess.
Anyone playing chess with a diagram like this is very vulnerable to changes
in the rules: change one rule, and the diagram can become completely
useless. But we can imagine tree diagrams with various possible rule changes
allowed for at any stage. Each row o f boxes would be supplemented
with additional boxes representing the possible moves under various rule

Figure 7 .1 Look-up tree for the start of a chess game. The boxes
represent the possible opening moves by White; the circles the
responses to each nominated by experts.
 116 R ivals and O bjections

Figure 7 .2 A more extended look-up tree. Again, the boxes

represent moves by White; the circles represent possible responses
by Black as nominated by experts.

changes, and below each box would be the circle representing the best
response for that move, given that rule change, according to the experts.
This would make an already huge tree even bigger but does not introduce
any new point o f principle. In practice, o f course, there is an insuperable
problem with this plan for playing good chess. At each stage o f a game of
chess there are a large number o f legal moves, and for each o f these legal
moves there are many legal responses. Writing out the look-up tree would
in consequence involve what is known as a combinatorial explosion. Giving
more than a line or two o f the tree would require more distinct states
than there are particles in the universe.

The gam e of life

. , , We are now ready to describe what has come to be

Look-up trees for life . . .
_____ ______________ known as the Blockhead example. At any point m a
game o f chess, there are only finitely many legal moves
and countermoves. It is this fact that makes the chess look-up tree just
described possible in principle. Likewise, at any point in a creature’s life
there are only finitely many discriminably distinct possible inputs and out­
puts at its periphery. Indeed, given quantum theory, there are probably only
finitely many nomologically possible inputs and outputs; but in any case
we know that there is a limit to how finely we distinguish different impacts
on our surfaces, and to how many different movements and responses
our bodies can make. This means that there could be in principle a ‘game
o f life’ look-up tree written for any one o f us - for Jones, say.
 Four C hallenges to Functionalism 117

We list all the different possible combinations o f pressure, light, gravity

and so on impacting at the surface o f Jones’s body at the first moment
o f his life. This would be the first row o f the game o f life look-up tree
modelled on Jones. It would correspond to the row o f boxes in the chess
tree above. The second row would give the behavioural response that
Jones would make (in a wide sense that includes any relevant peripheral
change) to each possible input. It corresponds to the first row o f circles in
the chess tree. The third row would give all the possible peripheral inputs
for each o f the various behavioural responses, and the fourth row would
give Jones’s behavioural responses to each o f these. And so on and so
forth for the whole o f Jones’s life. O f course how long Jones lives depends
in large part on which o f the various possible inputs actually come about,
but we may suppose that a fail-safe strategy is employed - the look-up
tree covers, say, 150 years’ worth o f possible inputs.
Jones’s Blockhead twin is then defined as a creature --------------------
that is superficially like Jones but has inside it a chip Blockhead
on which Jones’s game o f life look-up tree is inscribed,
and this chip controls Jones’s Blockhead twin’s every
response to the environment. In the same way we can define a Blockhead
twin for each and every one o f us. The objection to input-output func­
tionalism can now be stated very simply. It is that (a) Jones and Jones’s
Blockhead twin behave exactly alike, not only in how they respond to
each and every situation, but in how they would respond to each and
every possible situation; (b) Jones’s Blockhead twin is not a puppet - the
connexion between its inputs and outputs is largely a consequence o f how
it is, not o f how some puppeteer is; and yet (c) though Jones is, we may
suppose, intelligent and has a normal psychology, his Blockhead twin is
no more intelligent than a toaster (as Block puts it) and has no mental
life at all. It really is the kind o f automaton that dualists (wrongly) hold
that physicalism would reduce us to.
What is particularly interesting about Blockhead
Blockhead’s challenge
is that it tackles input-output functionalism on its
to us all
favoured ground. Many find input-output function­
alism implausible when applied to mental states in
general, but it is at its most appealing applied to intelligence. There is
no special ‘feel’ associated with intelligence, and it is intelligence that
intuitively connects most closely with behavioural performance: the be-
all and end-all o f intelligence does seem to be certain capacities to deliver
answers to problems set by questioners or by the environment. Yet the
clearest intuition about Blockhead is precisely that it completely lacks
intelligence and understanding. Indeed, given the intuitive appeal of
an essentially behavioural approach to intelligence - while insisting, of
 118 R ivals and O bjections

course, that traditional behaviourism was wrong to refuse to see explana­

tions in terms o f intelligence as genuinely causal ones proceeding by
appeal to internal nature —we should all try and say something sensible
about why Blockhead is not intelligent. It is not enough to say ‘It is
fortunate I am not an input-output functionalist’, or to say ‘I hereby
renounce input-output functionalism’ . We all need to say why Blockhead
is not intelligent.
Before we give our answer to this question, we note what seem to us
to be some wrong turns. You might say that the reason Blockhead is not
intelligent is that everything it does or would do is
determined in advance. It thus lacks the flexibility that
Some wrong turns
is part o f being intelligent and being rational. But o f
course if determinism is true, everything anyone ever
does is determined from the very beginning o f time. Some have inferred
from this that determinism is incompatible with intelligence and especially
rational decision making, and have accordingly taken comfort in the fact
that modern quantum theory is indeterministic. But this seems to us a
pretty desperate position. It is hard to see how throwing in some random
fluctuations makes what would otherwise be irrational, rational.
You might object that the look-up tree could never be written down
because it requires knowing all the possibilities for inputs at any given
time and all the outputs that someone - Jones, as we imagined - would
make to each and every input, and such knowledge is impossible. But
what matters for the argument is that the story' that the look-up tree tells
exists, not whether we could know it. For each and everyone o f us there
is a huge story about what we would do in response to each and every
possible input and sequence o f inputs, and so we can make sense o f the
idea that the story' is written on a chip inside a Blockhead.
You might object that we cannot make sense o f the idea that the story
in the form o f a huge look-up tree is contained inside Blockhead. The
look-up tree could not exist because it would involve a combinatorial
explosion. As we noted above, the look-up tree for a short game o f chess,
let alone the game o f life, would take more particles than there are in
the whole universe. This reply seems to us to misunderstand the role of
thought-experiments.
The fact that Blockhead is practically, and perhaps even nomologically',
impossible seems to us no more to the point than the fact that Twin Earth
is practically, and perhaps also nomologically, impossible. The point o f a
thought experiment is to test a conceptual claim, typically a claim about
the relation between two concepts. In the case o f Twin Earth, we test
the hypothesis that being watery and being water necessarily go together.
We come up with the answer that they do not necessarily go together by
 Four C hallenges to Functionalism 119

making clear sense o f the possibility - Twin Earth - where water is not
water}', and what is watery is not water. In the case o f Blockhead we test
the hypothesis that being behaviourally exactly alike someone intelligent
is sufficient for being intelligent, and come up with the answer that it is
not, by describing a possibility we understand and comprehend (while
realizing that it is in practice quite impossible) - a Blockhead twin o f
an intelligent Jones - where what is behaviourally exactly alike someone
intelligent has no intelligence (and indeed no thoughts) at all.
Finally, you might object that though it is missing the point to
complain that the Blockhead example is impossible either in practice or
perhaps even nomologically, it is right to be suspicious o f intuitions
about cases that far removed from what is possible in any but the most
abstract sense. Perhaps, in particular, we should resist the intuition that
Jones’s Blockhead twin lacks intelligence. The trouble with this objection
is that Blockhead is so like all the cases where we feel that someone lacks
understanding: someone who cannot play chess except by asking an expert
what to do at every stage is someone who does not understand the game,
and someone who cannot give you the square root o f a number other
than by looking up a table o f square roots is someone who does not
fully understand what a square root is. The intuition that Blockhead lacks
intelligence is simply a natural extension o f what we learn from these
simple and familiar cases. Moreover we can give a reason why Blockhead
lacks understanding and intelligence - a reason that, we will argue, makes
sense o f our strong intuition that Blockhead is deficient, and so explains
and justifies the intuition.

Why Blockhead is not a thinker

A message o f much recent philosophy has been the importance o f causal

connections o f the right kind. You do not count as seeing something unless
your perceptual state is caused by that thing. Part o f what justifies think­
ing o f an object - the chair in front o f me or the White House - as a
persisting material thing is the way early states o f the object are causally
responsible for later states o f it. The identity through time o f a person is
in part a matter o f their psychology being causally connected over time.
Likewise, causal connections o f the right kind are central to being intel­
ligent, to rationality and to belief.
It is not irrational per se to believe that the Earth
is flat. It is irrational aiven what else you believe and “ !
.. _ . . . . . c Rationality and
given
c7 jyour history.j Rationality
j is in ipart a matter of causal. h is t o r y
your beliefs evolving in the right way from your earlier ___________ y
 120 R ivals and Objections

beliefs and sensory' data. We all ought to believe that the Earth is round
(or oblate, to be more precise) because that is the right belief to have
caused in us by our pasts. Likewise, being intelligent centrally involves
having trains o f thought that evolve in the right way. Later thoughts have
to be caused in the right way by earlier ones. I f a brain scientist inserts
a probe into your brain that causes the crucial thought that enables you
to announce the proof o f Goldbach’s Conjecture, this is not a sign of
your intelligence or rationality. It is either a fluke or a sign o f the intel­
ligence o f the brain scientist, depending on the causal origins o f her action
in inserting the probe. Moreover, it is part o f being a belief o f a certain
kind that it tends to have certain results. Part o f what makes something
the belief that if P then Q , is that combined with the belief that P, it
tends to cause the belief that Q . (We enlarge on the importance o f tend­
ing to evolve rationally to being a belief when we discuss the intentional
stance in chapter 9.)
Simple input-output devices exhibit massive causal dependencies
between early and late stages. The state o f a sundial or an amplifier or a
carburettor that is responsible for its capacity to generate the appropri­
ate outputs on Monday is typically a major causal factor in its capacity
to do the job on Tuesday. The situation with much more complex struc­
tures like human beings is correspondingly more complex. How we
respond to stimuli on Tuesday depends on all sorts o f factors in addition
to how we are on Monday, including what has impacted on us between
the two days and what we have thought about in the interim. This is part
o f what confers on us the flexibility o f response that makes us intelligent.
Nevertheless, causal dependencies between earlier and later thoughts are
crucial. It is just that how we respond in the future depends on a much
more diverse range o f factors than simply how we are in the past - what
we have thought about and what has happened to us in the interim also
enter the equation.
The trouble with devices that work by look-up tree is that they lack
the appropriate causal dependencies. The state that governs the responses
to inputs early on plays the wrong kind o f role in
causing the state that governs the responses later on.
Blockhead's causal
This is because, for the most part, the Blockhead is
peculiarity
static. It is mostly written down in advance, and the
only thing that varies is which node is active.
We will now explain the idea o f an active node. We will call the various
sets o f pre-recorded possible inputs together with appropriate outputs
nodes. At any given time, a Blockhead can be said to have a certain node
that is active. The active node is the one that will be searched until the
input that has been given to the Blockhead is found, and the pre-recorded
 Four C hallenges to Functionalism 121

Figure 7 .3 The Blockhead example - a look-up tree that simulates

living an intelligent life. W hat actually happens lies within the
shaded area.

output produced. Let’s suppose that the Blockhead partly represented in

figure 7.3 has node y currently active. So if we give it input 1 we will get
output b, and if we give it input 3 we will get output c and so on.
Now thefa c t that it is node y that is currently active does indeed depend
on the past nature o f the Blockhead. Indeed, it depends on the nature o f
node a together with the fact that node a received input 3. But that is the
only thing about node y that depends on the nature o f node a , or indeed
anything else about what has happened to it since its initial creation. Noth­
ing about what possible inputs are encoded in node y or what outputs
are recorded against them depends on the history o f the Blockhead.
So at the time when a is active, the later state y that will govern the
later responses already exists. The input-output profile o f the look-up
tree at any given time - any complete row o f squares followed by circles
- does not generate the profile o f any particular node at any later time.
I f on Monday you make a sundial or Frankenstein makes a person, you
do not need to do anything extra to handle how what you have made
will respond to various inputs on, say, Tuesday. What you made on Monday
plus what happens in the meantime does that for you, and this is crucial
to the way the Tuesday responses depend on the Monday ones.
So the only causal dependency manifested in the look-up tree is in which
node o f the tree’s input-output profile at some time the active input and
output are at that time. That does depend, not on the nature o f the nodes
at earlier times, but on which nodes o f those earlier rows were then active,
and what input was received. It is like a recipe for roast duck that tells
you at the end to go to another recipe for the sauce. Although what is
122 R ivals a n d O bjections

in the first recipe plays a role in what you do subsequently, the content
o f the second recipe is, we may suppose, quite independent o f the con­
tent o f the first. But thinking is not like that; the content o f what we
think at a time typically depends in part on the content o f what we thought
at various earlier times in rich and complex ways, and that is crucial for
it to count as thought and as rational thought.
In sum, Blockhead’s input-output profile at any given time does not
depend in the right way on its input-output profiles at earlier times
for Blockhead to count as a thinker, or even as something displaying
rationality and intelligence. The input-output nature o f the node that
controls Blockhead’s behavioural response at time t is not caused by the
input-output nature o f what controls Blockhead’s behavioural response
at any earlier time t —n. The overall input-output nature at time t depends
on the past states only insofar as it is determined by which pre-existing
node is active. Figure 7.3 helps make the point. In the diagram, suppose
that the shaded area represents what actually happens. The point is that
as you progress through the shaded region you are not progressing
through nodes whose nature depends on the nature o f earlier states in
the shaded region, except in the minimal sense in which being active counts
as part o f their nature.

Common-sense functionalism and Blockhead

The argument to the conclusion that Blockhead is not a thinker rests on

a constraint on belief and intelligence that is supposed to be part o f folk­
lore. The way belief evolves over time, the importance for rationality o f
belief evolving causally in the right way, the fact that what is believed
depends on what was believed and what happens to a subject are plaus­
ibly common knowledge - implicit or explicit - and part o f our ordinary
conception o f belief. This means that Blockhead is not an objection to
common-sense functionalism. Blockhead shows that input-output func­
tionalism is false. How things are inside matters for our mental nature
over and above how our insides manifest themselves in determining our
environmental input-output connexions. But Blockhead does not show
that common-sense functionalism is false. Indeed, we could have seen this
straight off. It is intuition that delivers the answer that Blockhead has
not a thought in its head. We did not describe an experiment that shows
that Blockhead is unintelligent. We followed Block in supposing that once
the case was described, the answer was intuitively clear - and common-
sense functionalism is the version o f functionalism most concerned to
honour clear intuitions about the mind.
 Four C hallenges tg Functionalism 123

The Zombie Objection

A final objection to functionalism - and indeed to any
, ... i. , . Zombies invade the
kind or physicalism - is the so-called zombie obiection. . ^ ..
_7 r ' .. . . , , , , * . physicalist paradise
Versions or this objection have been around tor at least __________________
forty years, but it has become especially prominent in
recent years and calls for separate treatment. The rough thought is this:
we can conceive o f a creature physically just like us, but which lacks those
mental experiences that have distinctive ‘feels’ like pain and hunger, or
qualia in the philosophers’ jargon (‘qualia’ is a technical term for the
phenomenal qualities of conscious experience, the qualities that make it the
case that there is something it is like to have the experience; see the next
chapter for a more detailed explanation). Let’s call them zombies, for they
walk and talk like real people, but lack qualia or conscious experience. From
this we conclude that it is possible that there could be such a creature.
But then there must be something about the actual world other than its
physical make-up which gives us qualia: for this possible world we have
imagined is one which has exactly our physical make-up, but lacks qualia.
The thing that it lacks can’t be physical, since it is exactly like the actual
world in physical respects. So the thing in the actual world that makes it
true that we have qualia must be non-physical - thus physicalism is false.
Let’s be a bit more precise, and lay the argument out:

Premise (1). We can conceive o f a world physically ex­

actly like ours, but which lacks any other
features ours may or may not have (i.e. it
is a minimal physical duplicate o f ours),
in which the people in it lack qualia and
consciousness - they are zombies.
Premise (2). Conceivability is a good guide to possibility.
Premise (3). We are not zombies.
Intermediate conclusion (4): Zombies are possible (from (1) and (2)).
Intermediate conclusion: So there is a minimal physical duplicate o f
the actual world that is mentally different
from the actual world, since it contains only
zombies whereas the actual world does not
(from (3) and (4)).
Conclusion: Physicalism is false, from the definition o f
physicalism in chapter I .
 124 R ivals and O bjections

The argument is logically valid. So if all its premises are true, then so
is its conclusion. Any physicalist must, therefore, deny one or more o f its
premises. Exactly which premise they deny, however, varies according to
the kind o f physicalism involved.
Analytic functionalism needs to deny the first premise. For analytic
functionalism says that it is a matter o f the meaning o f mental state
terms that you have the relevant mental states whenever you have the
right functional roles being played. And if the relevant functional roles
are played actually by physical stuff, then those roles will be played in any
minimal physical duplicate o f the actual world. Suppose that an analytic
functionalist has worked through her theory o f mind, and knows what
roles have to be played for qualia to exist. She ought not be able to con­
ceive o f zombies. It is a priori that zombies are impossible. For knowledge
o f the roles, together with knowledge that a physical set up which plays
them exists, logically entails that qualia exist. T o conceive o f zombies is to
conceive o f things that have what is sufficient for qualia (having the right
roles played), and yet lack qualia. And this is to conceive o f a straight­
forward contradiction. In some good sense o f conceive, one cannot
conceive o f the a priori impossible.
This is a real problem for analytic functionalism, for it seems that we
really can conceive o f zombies - and yet the analytic functionalist says they
are ruled out by our grasp o f the meaning o f mental
state terms. It is intuitively fine to think it might be
Analytic functionalism
true that zombies are impossible, but perhaps only if
and ideal
conceivability
this is a substantial fact that does not simply fall out
o f the meaning o f mental state terms. Many think that
if zombies are impossible it does not seem to be a
merely semantic fact, but rather a metaphysical one. There are, however,
things that analytic functionalism might do to sweeten the pill. Sometimes
we think we can conceive o f something that is impossible. Perhaps you were
asked once, in maths class, to find out at what point a parabola crossed the
y axis. You took very seriously that it was at y = 1 and y = 2. You not only
conceived o f that possibility, but you thought it actually true. But after
some calculation you found, no, it was at y = 2 and y = 3. But o f course,
once we define a parabola by a quadratic equation, it is a logical necessity
that it intersects the y axis (or not) where it does. Your original concep­
tion was incoherent: it was a logical impossibility. Yet you had it none the
less. Exactly what to say about this case is controversial. Everyone agrees,
however, that there is some important distinction between what you can
ideally conceive - that is, when all the logical and semantic truths are
before your mind and you are rational —and whatever is happening when
you ‘conceive’ o f the roots o f a quadratic equation being different from
 Four C hallenges to Functionalism 125

what they, o f necessity, are. The disagreements are about vvliat is going on
in the unideal case, and we can set that aside. For perhaps the analytic
functionalist should say that they doubt that we can ideally conceive o f
zombies. If all the facts about the functional roles were before your mind,
and you could see how the physical states must play those roles, you could
not conceive o f zombies. Our apparent ability to conceiv e o f zombies is
on a par with imagining that mathematical truths are false.
This is a powerful reply, and one o f the authors is very attracted by
it. However, it remains a little mysterious how all the extra clarity and
ideal rationality are supposed to do their work. Certainly i f it follows from
the meaning o f mental state terms that zombies are impossible, then we
ideally can’t conceive o f them. But one might take a strong intuition about
their conceivability to be evidence that we have got the theory o f the
meaning o f mental state terms wrong. So for the reply to work, the extra
clarity that we have ideally will need to make it clearer that the analytic
functionalist theory o f mental state terms is right - which is a punt the
analytic functionalist must take. In addition it is hard to imagine exactly
what form this extra clarity would take. We sort o f know what the extra
clarity would come from in the mathematical case, but what the analogue
is in the qualia case is harder to see.
At first glance, empirical functionalism appears to
Empirical
be in a better position to address the zombie chal­
functionalism and
lenge. For here the discovery that qualia are physical zombies
is in some sense a posteriori. So, the thought runs, we
might be able to conceive o f zombies, for it is not a
priori that they are impossible. They are impossible none the less, but
this is an a posteriori impossibility o f the kind we discuss in chapter 4.
Thus the empirical functionalist might tty to deny the second premise o f
the zombie argument. For they think we can conceive o f the impossible
- even ideally conceive o f the impossible. Thus conceivability is no guide
to impossibility, and the zom bie argum ent fails.
We think, however, that this tempting reply fails. It fails because it ignores
some subtle distinctions within empirical functionalism. On one kind of
empirical functionalism the view is coherent but the reply does not work.
On another kind, the reply seems to work - but the coherence o f the
empirical functionalism itself is problematic.
Many o f the versions o f functionalism that we iden­
Reference-fixing
tify in the table at the end o f chapter 5 use something
- perhaps the folk roles - to pick out some mental
natures, and then rigidify on the internal features o f those entities. Now
it is impossible that something be a physical duplicate o f that internal nature
without possessing that internal nature, on the assumption that the internal
126 R ivals and O bjections

nature we discover is physical. So suppose that whatever it is that we use

to reference-fix, reference fixes on some human brains. When we do some
empirical work on brains we discover Neural Feature X playing the qualia
role, and we conclude that a posteriori we have discovered that qualia
are instances o f Neural Feature X. If we rigidify (see chapter 4 if this is
opaque to you) we will in addition conclude that qualia are necessarily
neural feature X. Thus it is impossible to possess neural feature X with­
out possessing qualia. Thus zombies are impossible - for by definition
they lack qualia, but in virtue o f being physical duplicates o f us they
possess neural feature X, which means they do possess qualia. So on these
versions o f functionalism zombies are impossible, but it would appear
to be a posteriori that they are impossible. It appears not to be a con­
ceptual truth, so the mere conceivability o f zombies poses no threat. After
all it was conceivable that many a posteriori necessities (such as water’s
being H 20 ) were false.
But the empirical functionalist cannot rest too easily. Remember that
on many o f these views the folk roles were used to reference-fix on the
samples whose internal nature was discovered. Grant to the empirical
functionalist that ‘qualia’ is a term that refers to the thing that actually
plays those roles - neural feature X. We still need a term for the thing
we knew about before we knew any neuroscience, and that enabled us
to reference-fix. Let us use ‘qualic’ to mean having the folk roles played.
Before we knew neuroscience, we did not know that neural feature X was
in us, but we knew that we were qualic, and (let us further suppose) we
knew that qualia were the actual qualic things.
Now it is surely conceivable that there are possible physical duplicates
that fail to be qualic (incidentally, if you think that being qualic just is
having qualia, then you agree with us that empirical functionalism is a
bad idea). Call them R-zombies. But if being qualic is just a matter o f
playing the folk roles, then it follows as a matter o f the meaning of
‘qualic’ that nothing could be physically just like us and fail to be qualic.
But this clashes with the intuition that there can be R-zombies - beings
that are physically just like us but lack the feature that we used to
identify neural state X. In a nutshell, while the empirical functionalist can
accept a zombie intuition with respect to the features they call ‘qualia’,
the problem arises again at the level o f the reference-fixing descriptions.
For them, R-zombies must be inconceivable, even though they seem
perfectly conceivable.
You may recall that some versions o f empirical functionalism about qualia
deny that the folk roles have any job to do. They take it that we brutally
discover the empirical nature o f qualia. These versions seem doomed to
run into one o f two major problems. If they say what it is that we use
 Four C hallenges to Functionalism 127

to decide which things have a qualitative nature, then a zombie objection

o f the kind we mention for empirical functionalism with the folk roles
seems near at hand. If, on the other hand, they do not, it is completely
mysterious why they think we should be examining heads rather than rocks
in our search for the empirical nature o f qualia.
‘ Perhaps the best reply proceeds something like
A modification of
this. The analytic functionalist should never have been
analytic functionalism
entirely sure that analytic functionalism is true. It is,
after all, a controversial doctrine. In any case dualism
might be right - no matter how unlikely. So the analytic functionalist
should think at most that if dualism is false, then analytic functionalism
is true. But then the conceivability o f zombies is in part the conceivabil-
ity o f dualism’s being true. Now if the analytic functionalist holds that
the truth o f analytic functionalism is an a priori truth there is still a bit
o f a problem, for zombies should still be ideally inconceivable, since
ideally we will grasp the truth o f analytic functionalism.
But there is a weakening o f analytic functionalism that is available. Perhaps
all that is a priori, and follows from the meaning o f our mental state terms,
is that i f dualism is false then analytic functionalism is true. Perhaps the
analytic truth - the thing that we grasp by grasping the meaning o f our
mental state terms but not knowing empirically how the world is - could
be something like:

I f there are dualistic states then

• In the actual world the qualia are the dualistic states; and all and only
the qualia in counterfactual worlds are the dualistic states.
Else
• In the actual world the qualia are the states that play the functional
roles, in other worlds qualia (if any) are the states that play these roles
in that world.

If this is the right analysis, and if dualism is false, then it will certainly be
right to say ‘zombies are impossible’. But we wouldn’t know that we were
entitled to say that a priori. For we know that we are only entitled to say
that zombies are impossible if dualism is false, and even physicalists should
give some credence to dualism in fact being true, however small. The idea
is that the zombie intuition comes from confusing two things. The first of
these is the thought that there is some chance that dualism is nue - and on
that supposition we would be right to claim that zombies were possible.
The second o f these is the straightforw ard possibility o f zombies.
128 R ivals and Objections

A n n o t a t e d R e a d in g
The Chinese nation example is presented in Ned Block, ‘Troubles with Function­
alism’. John Searle’s Chinese room case has been very widely discussed (at times,
with some heat). Perhaps the best place to start is John Searle, ‘Minds, Brains,
and Programs’. A more informal presentation, combined with replies to the many
objections that have been raised, is his ‘Is the Brain’s Mind a Computer Program?’
Among the many replies he considers are those he christens the systems reply
and the robot reply. The systems reply is the first one we expounded. The reply
we eventually setded on is a combination of the systems and robot replies. A good
recent discussion is in chapter 6 of Jack Copeland, Artificial Intelligence. The
classic source for Blockhead is Ned Block, ‘Psychologism and Behaviourism’. Keith
Campbell’s Body and Mind provides a straightforward description of the zombie
argument (he uses imitation men instead of zombies). The term ‘zombie’ in this
context may have come via Robert Kirk, ‘Zombies versus Materialists’. Recent
interest in the zombie objection has been stimulated by David Chalmers, The
Conscious Mind. A fuller version of the reply we give in the final section can be
found in David Braddon-Mitchell, ‘Qualia and Analytic Conditionals’, and for a
similar approach see John Hawthorne, ‘Advice for Physicalists’.