Artificial Intelligence

What is Intelligence?
 Main Entry: in·tel·li·gence
Pronunciation: in-'te-l&-j&n(t)s
Function: noun
Etymology: Middle English, from Middle French, from Latin intelligentia, from
intelligent-, intelligens intelligent

 1 a (1) : the ability to learn or understand or to deal with new or trying situations :
REASON; also : the skilled use of reason (2) : the ability to apply knowledge to
manipulate one's environment or to think abstractly as measured by objective criteria
(as tests) b Christian Science : the basic eternal quality of divine Mind c : mental
acuteness : SHREWDNESS

 2 a : an intelligent entity; especially : ANGEL b : intelligent minds or mind <cosmic

intelligence>

 3 : the act of understanding : COMPREHENSION

 4 a : INFORMATION, NEWS b : information concerning an enemy or possible enemy

or an area; also : an agency engaged in obtaining such information

 5 : the ability to perform computer functions

A Bit of Humor
 What is Artificial Intelligence?

Not just studying intelligent systems, but

building them…

Psychological approach: an intelligent

system is a model of human intelligence

Engineering approach: an intelligent system

solves a sufficiently difficult problem in a
generalizable way
 A Bit of AI History (section 1.3)

 Gestation (1943-1955)
 Early learning theory, first neural network, Turing test
 McCulloch and Pitts artificial neuron, Hebbian learning

 Birth (1956)
 Name coined by McCarthy
 Workshop at Dartmouth

 Early enthusiasm, great expectations (1952-1969)

 GPS, physical symbol system hypothesis
 Geometry Theorem Prover (Gelertner), Checkers (Samuels)
 Lisp (McCarthy), Theorem Proving (McCarthy), Microworlds
(Minsky et. al.)
 “neat” (McCarthy @ Stanford) vs. “scruffy” (Minsky @ MIT)
 A Bit of AI History (section 1.3)

 Dose of Reality (1966-1973)

 Combinatorial explosion

 Knowledge-based systems (1969-1979)

 AI Becomes an Industry (1980-present)

 Boom period 1980-88, then AI Winter

 Return of Neural Networks (1986-present)

 AI Becomes a Science (1987-present)

 SOAR, Internet as a domain
 What is Artificial Intelligence? (again)
 Systems that think like  Systems that think
humans rationally
 Cognitive Modeling  “Laws of Thought”
Approach approach
 “The automation of  “The study of mental
activities that we associate faculties through the use of
with human thinking...” computational models”
 Bellman 1978  Charniak and McDermott

 Systems that act like  Systems that act

humans rationally
 Turing Test Approach  Rational Agent Approach
 “The art of creating  “The branch of CS that is
machines that perform concerned with the
functions that require automation of intelligent
intelligence when behavior”
performed by people”  Lugar and Stubblefield
 Kurzweil 1990
 Acting Humanly
The Turing Test
(1950)
 Can machines think?
 Can machines behave
intelligently? Human
?
Operational test for
Human
intelligent behavior Interrogator
 The Imitation Game

AI System
 Acting Humanly

Turing Test (cont)

 Predicted that by 2000, a machine might have a
30% chance of fooling a lay person for 5 minutes
 Anticipated all major arguments against AI in
following 50 years
 Suggested major components of AI: knowledge,
reasoning, language understanding, learning

Problem!
 The turning test is not reproducible, constructive,
or amenable to mathematical analysis
 Thinking Humanly

1960’s cognitive revolution

Requires scientific theories of internal
activities of the brain
 What level of abstraction? “Knowledge” or
“Circuits”
 How to validate?
 Predicting and testing behavior of human subjects (top-
down)
 Direct identification from neurological data (bottom-up)
Cognitive Science and Cognitive
Neuroscience
 Now distinct from AI
 Thinking Rationally

Normative (or prescriptive) rather than

descriptive
Aristotle: What are correct arguments /
thought processes?
Logic notation and rules for derivation for
thoughts
Problems:
 Not all intelligent behavior is mediated by logical
deliberation
 What is the purpose of thinking? What thoughts
should I have?
 Acting Rationally

Rational behavior
 Doing the right thing

What is the “right thing”

 That which is expected to maximize goal

achievement, given available information

We do many (“right”) things without

thinking
 Thinking should be in the service of rational
action
 Applied Areas of AI

Heuristic Search
Computer Vision
Adversarial Search (Games)
Fuzzy Logic
Natural Language Processing
Knowledge Representation
Planning
Learning
 Examples
Playing chess Recognizing speech
Driving on the Diagnosing diseases
highway Translating
Mowing the lawn languages
Answering questions Data mining
Artificial Intelligence

CHAPTER-2
P R O B L E M S , P R O B L E M S PA C E S A N D
SEARCH
 Problems, Problem Spaces and
Search

Problem:
 A problem, which can be caused for different reasons, and, if
solvable, can usually be solved in a number of different ways,
is defined in a number of different ways.
To build a system or to solve a particular problem we
need to do four things.
 1. Define the problem precisely. This definition must include
precise specification of what the initial situation will be as
well as what final situations constitute acceptable solutions to
the problem
 2. Analyze the problem
 3. Isolate and represent the task knowledge that is necessary
to solve the problem
 4. Choose the best solving technique and apply it to the
particular problem.
Defining the Problem as a State Space
Search
State space is a set of legal positions, starting at the
initial state, using the set of rules to move from one
state to another and attempting to end up in a goal
state.

Methodology of State Space Approach

 1. To represent a problem in structured form using different
states
 2. Identify the initial state
 3. Identify the goal state
 4. Determine the operator for the changing state
 5. Represent the knowledge present in the problem in a
convenient form
 6. Start from the initial state and search a path to goal state
 Production System

The entire procedure for getting a solution for AI problem can be viewed as
“Production System”.

Its main components are:

 A Set of Rules, each consisting of a left side (a pattern) that determines the
applicability of the rule and right side that describes the operation to be
performed if the rule is applied.
 Knowledge Base – It contains whatever information is appropriate for a
particular task.
 Control Strategy – It specifies the order in which the rules will be compared
to the database and the way of resolving the conflicts that arise when several
rules match at one.
o The first requirement of a goal control strategy is that it is cause motion; a
control strategy that does not cause motion will never lead to a solution.
o The second requirement of a good control strategy is that it should be
systematic.
 A rule applier: Production rule is like below
if(condition) then
consequence or action
 contd

Algorithm for Production System:

1. Represent the initial state of the problem
2. If the present state is the goal state then go to
step 5 else go to step 3
3. Choose one of the rules that satisfy the present
state, apply it and change the state to new
state.
4. Go to Step 2
5. Print “Goal is reached ” and indicate the search
path from initial state to goal state
6. Stop
Classification of Production System

Based on the direction they can be

1. Forward Production System
 Moving from Initial State to Goal State
 When there are number of goal states and only
one initial state, it is advantage to use forward
production system.
2. Backward Production System
 Moving from Goal State to Initial State
 If there is only one goal state and many initial
states, it is advantage to use backward
production system.
Production System Characteristics

Two questions we might reasonably ask at this

point are:
 Can production systems, like problems, be
described by a set of characteristics that shed
some light on how they can easily be
implemented?

 If so, what relationships are there between

problem types and the types of production
systems best suited to solving the problems?
Can production systems, like problems, be described by a set
of characteristics that shed some light on how they can easily
be implemented?
The answer can be considered with the following definitions of
classes of production systems:

 A monotonic production system is a production system in

which the applications of a rule never prevents the later
application of another rule that could also have been applied at
the time the first rule was selected.
 A non-monotonic production system is one which this is not
true.
 A partially commutative production system is a production
system with the property that if the application of a particular
sequence of rules transforms state X into state Y, then any
permutation of those rules that is allowable also transforms
state X into state Y.
 A commutative production system is a production system
that is both monotonic and partially commutative.
 The four categories of Production Systems

MONOTONIC NON-MONOTONIC

Partially Theorem proving Robot Navigation

commutative

Not Partially Chemical Synthesis Bridge

commutative
 Problem Characteristics

In order to choose the most appropriate method (or

a combination of methods) for a particular
problem, it is necessary to analyze the problem
along several key dimensions:
 Is the problem decomposable?
 Can solution steps be ignored or undone?
 Is the universe predictable?
 Is a good solution absolute or relative?
 Is the solution a state or a path?
 What is the role of knowledge?
 Does the task require human-interaction?
 Problem Classification
 Heuristic Search

 Very large search space

 Large databases
 Image sequences
 Game playing

 Algorithms
 Guaranteed best answer
 Can be slow – literally years
 Heuristics
 “Rules of thumb”
 Very fast
 Good answer likely, but not guaranteed!
 Searching foreign intelligence for terrorist activity.
 Computer Vision
 Computationally
taxing
 Millions of bytes of data
per frame
 Thirty frames per second
 Computers are scalar /
Images are
multidimensional
 Image Enhancement
vs. Image
Understanding
 Can you find the
terrorist in this
picture?
 Adversarial Search

Game theory...
 Two player, zero sum – checkers, chess, etc.
Minimax
 My side is MAX
 Opponent is MIN
Alpha-Beta
 Evaluation function...”how good is board”
 Not reliable...play game (look ahead) as deep as
possible and use minimax.
 Select “best” backed up value.
Where will Al-Qaeda strike next?
 Adversarial Search
1

X X O

MIN O

...
2 6

X X O X X O

MAX O O O O
X X

3 4 5 7 8 9

X X O X X O X X O X X O X X O X X O

O O X O O O O X O O O O O O

X X X X X X X X X X

1-0=1 1-2=-1 1-1=0 *91* 0 10

 Example: Tic Tac Toe #1

• Precompiled
move table.
move
table

• For each input encode look

up
board, a specific
move (output
board)

• Perfect play, but

is it AI?
 Example: Tic Tac Toe #2

Represent board as a magic square, one

integer per square

If 3 of my pieces sum to 15, I win

Predefined strategy:
 1. Win
 2. Block
 3. Take center
 4. Take corner
 5. Take any open square
 Example: Tic Tac Toe #3

 Given a board, consider all possible moves (future

boards) and pick the best one

 Look ahead (opponent’s best move, your best

move…) until end of game

 Functions needed:
 Next move generator
 Board evaluation function

 Change these 2 functions (only) to play a different

game!
 Fuzzy Logic

Basic logic is binary

 0 or 1, true or false, black or white, on or off, etc...

But in the real world there are of “shades”

 Light red or dark red
 0.64756
Membership functions
 Fuzzy Logic

Linguistic
Appetit
Variable
e

Linguistic
Light Moderat Heav Values
e y
1

Membership
Grade

1000 2000 3000

Calories Eaten Per Day

 Natural Language Processing

 Speech recognition vs. natural language processing

 NLP is after the words are recognized

 Ninety/Ten Rule
 Can do 90% of the translation with 10% time, but 10%
work takes 90% time
 Easy for restricted domains
 Dilation
 Automatic translation
 Control your computer
 Say “Enter” or “one” or “open”
 Associative calculus
 Understand by doing
 Natural Language Processing
Net for Basic Noun Group
adjective

determiner noun
“The big grey dog” S1 S2 S3

Net for Prepositional Group

preposition NOUNG
“by the table in the corner” S1 S2 S3

Net for Basic Noun Group

PREPG
adjective

determiner noun
“The big grey dog by the S1 S2 S3
table in the corner”
 Knowledge Representation

Predicate Logic
 On(table, lamp)
 In(corner, table)
 Near(table, dog)
 Prolog

Graph Based
 Semantic Networks
 Frames

Rule Based
 Expert Systems
 Planning
Robotics
 If a robot enters a
room and sits down,
what is the “route”.

Closed world Table

Rule based systems
Blocks world
Chair
 Planning
 Pickup(x) Robot
 Ontable(x), clear(x), Hand
handempty(),
 Holding(x)
 Putdown(x)
C
 Holding(x) A B
 Ontable(x), clear(x),
Clear(B) On(C, A) OnTable(A)
handempty()
 Stack(x, y) Clear(C) Handempty OnTable(B)

 Holding(x), clear(y)
 Handempty(), on(x, y),
clear(x)
 Unstack(x, y) A
 Handempty(), clear(x), on(x, B
y)
 Holding(x), clear(x) C
Goal: [On(B, C) ^ On(A,
B)]
 Learning

Neural Networks
Evolutionary Computing
Knowledge in Learning
Reinforcement Learning