Artificial Intelligence

What is artificial intelligence?

 Attempt to “mimic” human intelligence?
o One approach: capture human expert intelligence so the machine
evaluates and responds as the human would.
 Make a machine act intelligently
o make rational decisions (within a limited problem domain)
o model human thought processes
o function using mathematical reasoning
 Learning–required for intelligence?
 Perceptions: vision is very difficult. Limited sensors are easy.
 Fuzzy definition: “We know it when we see it.”

Turing Test
 Artificial intelligence (AI)
o Explores techniques for incorporating aspects of intelligence into
computer systems
 Turing test
o A test for intelligent behavior of machines
o Allows a human being to interrogate two entities, both hidden from the
interrogator
 A human being
 A machine (a computer)
o If the interrogator is unable to determine which entity is the human
being and which is the computer, the computer has passed the test
 Artificial intelligence can be thought of as constructing computer models of
human intelligence

The Turing Test

1
Intelligence
 If a human’s behavior can be perfectly modeled with an artificial device does
that mean that
o the device is intelligent? OR people are biomechanical machines after
all?

2
A Division of Labor
 Categories of tasks
o Computational tasks
o Recognition tasks
o Reasoning tasks
 Computational tasks
o Tasks for which algorithmic solutions exist
o Computers are better (faster and more accurate) than human beings
 Recognition tasks
o Sensory/recognition/motor-skills tasks
o Human beings are better than computers
 Reasoning tasks
o Require a large amount of knowledge
o Human beings are far better than computers

Human and Computer Capabilities

3
Problem Solving and Complex Algorithms
Classic Traveling Salesman Problem
 A salesman must travel to visit
clients in many cities, each
separated geographically.
 Objective: to find the best order
of cities to visit that minimizes
the number of miles traveled.
o Very difficult for large
numbers of cities.
o “Exponential Explosion”
 Problem is easy for small
numbers of cities.
 Note how fast the number of
routes grow as cities are added.
(This is complexity analysis.)
 The relationship between the
number of cities and the number
of routes is factorial.
 Review: 4! is 4 × 3 × 2 × 1 = 24

Problems with algorithms that contain factorials

 Any algorithm that needs all combinations need factorials
 This table shows how “bad” factorial problems get even with small numbers
of n.
 Imagine how large 1000! must be, or 1 million factorial.

4
  These problems take too much time to be practical without intelligent
shortcuts.

Combinatorial Explosion

Game Playing
 Game playing is a useful environment for testing and experimenting
o Limited, controlled possibilities
o Finite number of states
o Fixed rules that govern play or actions in the artificial world
 Excellent testing/learning/teaching ground for AI concepts and ideas
 Chess, checkers, other games have been largely explored
 Note that the best computer chess programs still use the searching power of
the computer to find better moves than a human will find using our experience
and insight. They are modified with insights about how humans play.

Harder Games
 In “easy” games like chess and checkers, each player may move only one
piece per turn.
 The computer can “look ahead” many turns in advance to try to find the best
move.
 Imagine a game where:
o Each player can move not one, but two game pieces in each turn.
5
 o Each piece can move to any of three different locations on the board.
o The player’s first piece can move to any of three locations. The second
piece can move to any of three locations as well. For each first-piece
move, there are three second-piece moves: 9 combinations.
 moves_per_piece number_of_pieces
 32 = 9 possible moves in just one turn

An Even Harder Game

 Imagine a game where
o Each player can move 10 pieces in each turn.
o Each piece can move to any of 10 different locations on the board.
 moves_per_piece number_of_pieces
 1010 = possible moves in just one turn
 10,000,000,000 = 10 billion combinations of moves in for just the
first player.
o The second player can make the same number of moves in response to
the first player’s moves.
 1010 ×1010 moves = 1020

Actual “Hard” Games

 There are military/historical games/simulations where:
o Each commander may have 100-200 units to move in each turn.
o Each unit can move to any of 100 (or more) different locations on the
game map.
o There are on the order of 100100 ×100100 moves in one game turn (both
players move).
 Brute force methods “die” in this type of game/simulation.
 Artificial intelligence that at least mimics human intelligence is needed.

Modeling the World

 For many problems, the key insight is the decision about how to represent the
world–the data structure.
 The choice of data structure and the logic of operating on that structure will
determine the performance of the AI routines (measured by how long it takes
to make a decision, and by how effective it is at finding a solution).
6
Machine Learning
 One definition of intelligence: the agent is able to learn
o from experience.
o to extrapolate and generalize rules about how the world works.
 Difficult problem, but limited success in a few areas:
o Expert systems
o Artificial neural networks

Applications of Machine Learning - Expert systems

 Capture human reasoning abstractly
 1970s-80s
 Computer runs a program that computes using captured human intelligence.
 Successful in a variety of applications
o predict jury verdicts

Neural Networks
 A network of nerve cells with many connections among cells
 Artificial networks order the nerve cells into a regular structure.
 When one cell fires (changes state), it sends a signal to other cells (which may
then also change state).
 “Knowledge” is stored in the aggregate connections between the cells and in
the internal requirements for each cell to fire.
 Time to allow cells to fire and feedback and cycle

Fuzzy Logic
 Humans have an innate ability to reason with probabilities.
 In our world, events and timing are uncertain, but have a chance of occurring
within a time window.
 Fuzzy logic allows reasoning with pieces of information that are not facts, but
rather probabilities of some event or situation.
 Logic rules are extended to allow conclusions to be drawn with some
“confidence” or likelihood (or probability).
 Humans are used to this type of thinking and do it naturally.
7
8