UNIT I:

Problem Solving by Search-I: Introduction to AI, Intelligent Agents Problem Solving by Search –II: Problem-
Solving Agents, Searching for Solutions,
Uninformed Search Strategies: Breadth-first search, Uniform cost search, Depth-first search, Iterative
deepening Depth-first search, Bidirectional search, Informed (Heuristic) Search Strategies: Greedy best-first
search, A* search, Heuristic Functions, Beyond Classical Search: Hill-climbing search, Simulated annealing
search,
Local Search in Continuous Spaces, Searching with Non-Deterministic Actions, Searching with Partial
Observations, Online Search Agents and Unknown Environment

Introduction:

 Artificial Intelligence is concerned with the design of intelligence in an artificial device. The term
was coined by John McCarthy in 1956.
 Intelligence is the ability to acquire, understand and apply the knowledge to achieve goals in the
world.
 AI is the study of the mental faculties through the use of computational models
 AI is the study of intellectual/mental processes as computational processes.
 AI program will demonstrate a high level of intelligence to a degree that equals or exceeds the
intelligence required of a human in performing some task.
 AI is unique, sharing borders with Mathematics, Computer Science, Philosophy,
Psychology, Biology, Cognitive Science and many others.
 Although there is no clear definition of AI or even Intelligence, it can be described as an attempt
to build machines that like humans can think and act, able to learn and use knowledge to solve
problems on their own.

History of AI:
Important research that laid the groundwork for AI:

 In 1931, Goedel layed the foundation of Theoretical Computer Science1920-30s:

He published the first universal formal language and showed that math itself is either
flawed or allows for unprovable but true statements.
 In 1936, Turing reformulated Goedel’s result and church’s extension thereof.

Artificial Intelligence Page 1

  In 1956, John McCarthy coined the term "Artificial Intelligence" as the topic of the Dartmouth
Conference, the first conference devoted to the subject.

 In 1957, The General Problem Solver (GPS) demonstrated by Newell, Shaw & Simon
 In 1958, John McCarthy (MIT) invented the Lisp language.
 In 1959, Arthur Samuel (IBM) wrote the first game-playing program, for checkers, to achieve
sufficient skill to challenge a world champion.
 In 1963, Ivan Sutherland's MIT dissertation on Sketchpad introduced the idea of interactive
graphics into computing.
 In 1966, Ross Quillian (PhD dissertation, Carnegie Inst. of Technology; now CMU) demonstrated
semantic nets
 In 1967, Dendral program (Edward Feigenbaum, Joshua Lederberg, Bruce Buchanan, Georgia
Sutherland at Stanford) demonstrated to interpret mass spectra on organic chemical compounds.
First successful knowledge-based program for scientific reasoning.
 In 1967, Doug Engelbart invented the mouse at SRI
 In 1968, Marvin Minsky & Seymour Papert publish Perceptrons, demonstrating limits of simple
neural nets.
 In 1972, Prolog developed by Alain Colmerauer.
 In Mid 80’s, Neural Networks become widely used with the Backpropagation algorithm (first
described by Werbos in 1974).
 1990, Major advances in all areas of AI, with significant demonstrations in machine learning,
intelligent tutoring, case-based reasoning, multi-agent planning, scheduling, uncertain reasoning,
data mining, natural language understanding and translation, vision, virtual reality, games, and
other topics.
 In 1997, Deep Blue beats the World Chess Champion Kasparov
 In 2002,iRobot, founded by researchers at the MIT Artificial Intelligence Lab, introduced Roomba,
a vacuum cleaning robot. By 2006, two million had been sold.

Foundations of Artificial Intelligence:

 Philosophy
e.g., foundational issues (can a machine think?), issues of knowledge and believe, mutual
knowledge

Artificial Intelligence Page 2

  Psychology and Cognitive Science
e.g., problem solving skills
 Neuro-Science
e.g., brain architecture
 Computer Science And Engineering
e.g., complexity theory, algorithms, logic and inference, programming languages, and system
building.
 Mathematics and Physics
e.g., statistical modeling, continuous mathematics,
 Statistical Physics, and Complex Systems.
Sub Areas of AI:

1) Game Playing
Deep Blue Chess program beat world champion Gary Kasparov
2) Speech Recognition
PEGASUS spoken language interface to American Airlines' EAASY SABRE reseration system, which
allows users to obtain flight information and make reservations over the telephone. The 1990s has
seen significant advances in speech recognition so that limited systems are now successful.
3) Computer Vision
Face recognition programs in use by banks, government, etc. The ALVINN system from CMU
autonomously drove a van from Washington, D.C. to San Diego (all but 52 of 2,849 miles), averaging
63 mph day and night, and in all weather conditions. Handwriting recognition, electronics and
manufacturing inspection, photo interpretation, baggage inspection, reverse engineering to
automatically construct a 3D geometric model.
4) Expert Systems
Application-specific systems that rely on obtaining the knowledge of human experts in an area and
programming that knowledge into a system.
a. Diagnostic Systems : MYCIN system for diagnosing bacterial infections of the blood and
suggesting treatments. Intellipath pathology diagnosis system (AMA approved). Pathfinder
medical diagnosis system, which suggests tests and makes diagnoses. Whirlpool customer
assistance center.

Artificial Intelligence Page 3

 b. System Configuration
DEC's XCON system for custom hardware configuration. Radiotherapy treatment planning.
c. Financial Decision Making
Credit card companies, mortgage companies, banks, and the U.S. government employ AI
systems to detect fraud and expedite financial transactions. For example, AMEX credit
check.
d. Classification Systems
Put information into one of a fixed set of categories using several sources of information.
E.g., financial decision making systems. NASA developed a system for classifying very faint
areas in astronomical images into either stars or galaxies with very high accuracy by learning
from human experts' classifications.
5) Mathematical Theorem Proving
Use inference methods to prove new theorems.
6) Natural Language Understanding
AltaVista's translation of web pages. Translation of Catepillar Truck manuals into 20 languages.
7) Scheduling and Planning
Automatic scheduling for manufacturing. DARPA's DART system used in Desert Storm and Desert
Shield operations to plan logistics of people and supplies. American Airlines rerouting contingency
planner. European space agency planning and scheduling of spacecraft assembly, integration and
verification.
8) Artificial Neural Networks:
9) Machine Learning

Application of AI:

AI algorithms have attracted close attention of researchers and have also been applied
successfully to solve problems in engineering. Nevertheless, for large and complex problems, AI
algorithms consume considerable computation time due to stochastic feature of the search
approaches

1) Business; financial strategies

Artificial Intelligence Page 4

 2) Engineering: check design, offer suggestions to create new product, expert systems
for all engineering problems
3) Manufacturing: assembly, inspection and maintenance
4) Medicine: monitoring, diagnosing
5) Education: in teaching
6) Fraud detection
7) Object identification
8) Information retrieval
9) Space shuttle scheduling

Building AI Systems:
1) Perception
Intelligent biological systems are physically embodied in the world and experience the world
through their sensors (senses). For an autonomous vehicle, input might be images from a
camera and range information from a rangefinder. For a medical diagnosis system, perception is
the set of symptoms and test results that have been obtained and input to the system manually.
2) Reasoning
Inference, decision-making, classification from what is sensed and what the internal "model" is
of the world. Might be a neural network, logical deduction system, Hidden Markov Model
induction, heuristic searching a problem space, Bayes Network inference, genetic algorithms,
etc.
Includes areas of knowledge representation, problem solving, decision theory, planning, game
theory, machine learning, uncertainty reasoning, etc.
3) Action
Biological systems interact within their environment by actuation, speech, etc. All behavior is
centered around actions in the world. Examples include controlling the steering of a Mars rover
or autonomous vehicle, or suggesting tests and making diagnoses for a medical diagnosis
system. Includes areas of robot actuation, natural language generation, and speech synthesis.

The definitions of AI:

Artificial Intelligence Page 5

 a) "The exciting new effort to make computers b) "The study of mental faculties through
think . . . machines with minds, in the full and the use of computational models"
literal sense" (Haugeland, 1985) (Charniak and McDermott, 1985)

"The automation of] activities that we "The study of the computations that
associate with human thinking, activities such make it possible to perceive, reason, and
as decision-making, problem solving, act" (Winston, 1992)
learning..."(Bellman, 1978)
c) "The art of creating machines that perform d) "A field of study that seeks to explain and
functions that require intelligence when emulate intelligent behavior in terms of
performed by people" (Kurzweil, 1990) computational processes" (Schalkoff, 1
990)
"The study of how to make computers do
things at which, at the moment, people "The branch of computer science that is
are better" (Rich and Knight, 1 99 1 ) concerned with the automation of
intelligent behavior" (Luger and
Stubblefield, 1993)
The definitions on the top, (a) and (b) are concerned with reasoning, whereas those on the bottom, (c)
and (d) address behavior.The definitions on the left, (a) and (c) measure success in terms of human
performance, and those on the right, (b) and (d) measure the ideal concept of intelligence called
rationality

Intelligent Systems:

In order to design intelligent systems, it is important to categorize them into four categories (Luger and
Stubberfield 1993), (Russell and Norvig, 2003)
1. Systems that think like humans
2. Systems that think rationally
3. Systems that behave like humans
4. Systems that behave rationally
Human- Like Rationally

Cognitive Science Approach Laws of thought Approach

Think:
“Machines that think like humans” “ Machines that think Rationally”

Turing Test Approach Rational Agent Approach

Act:
“Machines that behave like humans” “Machines that behave Rationally”

Artificial Intelligence Page 6

Scientific Goal: To determine which ideas about knowledge representation, learning, rule systems
search, and so on, explain various sorts of real intelligence.
Engineering Goal:To solve real world problems using AI techniques such as Knowledge representation,
learning, rule systems, search, and so on.
Traditionally, computer scientists and engineers have been more interested in the engineering
goal, while psychologists, philosophers and cognitive scientists have been more interested in the
scientific goal.
Cognitive Science: Think Human-Like

a. Requires a model for human cognition. Precise enough models allow simulation by
computers.

b. Focus is not just on behavior and I/O, but looks like reasoning process.

c. Goal is not just to produce human-like behavior but to produce a sequence of steps of the
reasoning process, similar to the steps followed by a human in solving the same task.

Laws of thought: Think Rationally

a. The study of mental faculties through the use of computational models; that it is, the study of
computations that make it possible to perceive reason and act.

b. Focus is on inference mechanisms that are probably correct and guarantee an optimal solution.

c. Goal is to formalize the reasoning process as a system of logical rules and procedures of
inference.

d. Develop systems of representation to allow inferences to be like

“Socrates is a man. All men are mortal. Therefore Socrates is

mortal”

Turing Test: Act Human-Like

a. The art of creating machines that perform functions requiring intelligence when performed by
people; that it is the study of, how to make computers do things which, at the moment, people
do better.

b. Focus is on action, and not intelligent behavior centered around the representation of the world

c. Example: Turing Test

o 3 rooms contain: a person, a computer and an interrogator.

Artificial Intelligence Page 7

 o The interrogator can communicate with the other 2 by teletype (to avoid the
machine imitate the appearance of voice of the person)

Artificial Intelligence Page 8

 o The interrogator tries to determine which the person is and which the machine
is.

o The machine tries to fool the interrogator to believe that it is the human, and
the person also tries to convince the interrogator that it is the human.

o If the machine succeeds in fooling the interrogator, then conclude that the
machine is intelligent.

Rational agent: Act Rationally

a. Tries to explain and emulate intelligent behavior in terms of computational process; that it is
concerned with the automation of the intelligence.

b. Focus is on systems that act sufficiently if not optimally in all situations.

c. Goal is to develop systems that are rational and sufficient

The difference between strong AI and weak AI:

Strong AI makes the bold claim that computers can be made to think on a level (at least) equal to
humans.

Weak AI simply states that some "thinking-like" features can be added to computers to make them
more useful tools... and this has already started to happen (witness expert systems, drive-by-wire cars
and speech recognition software).
AI Problems:

AI problems (speech recognition, NLP, vision, automatic programming, knowledge

representation, etc.) can be paired with techniques (NN, search, Bayesian nets, production systems,
etc.).AI problems can be classified in two types:

1. Common-place tasks(Mundane Tasks)

2. Expert tasks

Common-Place Tasks:
1. Recognizing people, objects.
2. Communicating (through natural language).
3. Navigating around obstacles on the streets.

Artificial Intelligence Page 9

These tasks are done matter of factly and routinely by people and some other animals.
Expert tasks:
1. Medical diagnosis.
2. Mathematical problem solving
3. Playing games like chess
These tasks cannot be done by all people, and can only be performed by skilled specialists.
Clearly tasks of the first type are easy for humans to perform, and almost all are able to
master them. The second range of tasks requires skill development and/or intelligence and only some
specialists can perform them well. However, when we look at what computer systems have been able to
achieve to date, we see that their achievements include performing sophisticated tasks like medical
diagnosis, performing symbolic integration, proving theorems and playing chess.

1. Intelligent Agent’s:
2.1 Agents andenvironments:

Fig 2.1: Agents and Environments

2.1.1 Agent:
An Agent is anything that can be viewed as perceiving its environment through sensors and acting
upon that environment through actuators.

 A human agent has eyes, ears, and other organs for sensors and hands, legs, mouth, and
other body parts foractuators.
 A robotic agent might have cameras and infrared range finders for sensors and various
motors foractuators.
 A software agent receives keystrokes, file contents, and network packets as sensory

Artificial Intelligence Page 10

 inputs and acts on the environment by displaying on the screen, writing files, and sending
network packets.

2.1.2 Percept:
We use the term percept to refer to the agent's perceptual inputs at any given instant.

2.1.3 PerceptSequence:
An agent's percept sequence is the complete history of everything the agent has ever perceived.

2.1.4 Agent function:

Mathematically speaking, we say that an agent's behavior is described by the agent function that
maps any given percept sequence to an action.

2.1.5 Agentprogram
Internally, the agent function for an artificial agent will be implemented by an agent program. It is
important to keep these two ideas distinct. The agent function is an abstract mathematical
description; the agent program is a concrete implementation, running on the agent architecture.

To illustrate these ideas, we will use a very simple example-the vacuum-cleaner world shown in Fig

2.1.5. This particular world has just two locations: squares A and B. The vacuum agent perceives
which square it is in and whether there is dirt in the square. It can choose to move left, move right,
suck up the dirt, or do nothing. One very simple agent function is the following: if the current
square is dirty, then suck, otherwise move to the other square. A partial tabulation of this agent
function is shown in Fig 2.1.6.

Fig 2.1.5: A vacuum-cleaner world with just two locations.

Artificial Intelligence Page 11

 2.1.6 Agentfunction

Percept Sequence Action

[A, Clean] Right

[A, Dirty] Suck

[B, Clean] Left

[B, Dirty] Suck

[A, Clean], [A, Clean] Right

[A, Clean], [A, Dirty] Suck

Fig 2.1.6: Partial tabulation of a simple agent function for the example: vacuum-cleaner
world shown in the Fig 2.1.5

Function REFLEX-VACCUM-AGENT ([location, status]) returns an action If

status=Dirty then return Suck

else if location = A then return Right

else if location = B then return Left

Fig 2.1.6(i): The REFLEX-VACCUM-AGENT program is invoked for each new percept (location, status) and
returns an action each time
Strategies of Solving Tic-Tac-Toe Game Playing

Tic-Tac-Toe Game Playing:

Tic-Tac-Toe is a simple and yet an interesting board game. Researchers have used various approaches to
study the Tic-Tac-Toe game. For example, Fok and Ong and Grim et al. have used artificial neural
network based strategies to play it. Citrenbaum and Yakowitz discuss games like Go-Moku,
Hex and Bridg-It which share some similarities with Tic-Tac-Toe.

Artificial Intelligence Page 12

 Fig 1.

A Formal Definition of the Game:

The board used to play the Tic-Tac-Toe game consists of 9 cells laid out in the form of a 3x3 matrix (Fig.
1). The game is played by 2 players and either of them can start. Each of the two players is assigned a
unique symbol (generally 0 and X). Each player alternately gets a turn to make a move. Making a move is
compulsory and cannot be deferred. In each move a player places the symbol assigned to him/her in a
hitherto blank cell.

Let a track be defined as any row, column or diagonal on the board. Since the board is a square
matrix with 9 cells, all rows, columns and diagonals have exactly 3 cells. It can be easily observed that
there are 3 rows, 3 columns and 2 diagonals, and hence a total of 8 tracks on the board (Fig. 1). The goal
of the game is to fill all the three cells of any track on the board with the symbol assigned to one before
the opponent does the same with the symbol assigned to him/her. At any point of the game, if
there exists a track whose all three cells have been marked by the same symbol, then the player
to whom that symbol have been assigned wins and the game terminates. If there exist no track
whose cells have been marked by the same symbol when there is no more blank cell on the board then
the game is drawn.

Let the priority of a cell be defined as the number of tracks passing through it. The priorities of the
nine cells on the board according to this definition are tabulated in Table 1. Alternatively, let the
priority of a track be defined as the sum of the priorities of its three cells. The priorities of the eight
tracks on the board according to this definition are tabulated in Table 2. The prioritization of the cells
and the tracks lays the foundation of the heuristics to be used in this study. These heuristics are
somewhat similar to those proposed by Rich and Knight.

Artificial Intelligence Page 13

Strategy 1:

Algorithm:

1. View the vector as a ternary number. Convert it to a decimal number.

2. Use the computed number as an index into Move-Table and access the vector stored there.

3. Set the new board to that vector.

Procedure:

1) Elements of vector:

0: Empty

1: X

2: O

→ the vector is a ternary number

2) Store inside the program a move-table (lookuptable):

a) Elements in the table: 19683 (39)

b) Element = A vector which describes the most suitable move from the

Comments:

Artificial Intelligence Page 14

 1. A lot of space to store the Move-Table.

2. A lot of work to specify all the entries in the Move-Table.

3. Difficult to extend

Explanation of Strategy 2 of solving Tic-tac-toe problem

Stratergy 2:

Data Structure:
1) Use vector, called board, as Solution 1
2) However, elements of the
vector: 2: Empty
3: X
5: O
3) Turn of move: indexed by
integer 1,2,3, etc

Function Library:

1. Make2:

a) Return a location on a game-board.

IF (board[5] = 2)
RETURN 5; //the center cell.
ELSE
RETURN any cell that is not at the board’s corner;
// (cell: 2,4,6,8)
b) Let P represent for X or O
c) can_win(P) :
P has filled already at least two cells on a straight line (horizontal, vertical, or
diagonal)
d) cannot_win(P) = NOT(can_win(P))
2. Posswin(P):

Artificial Intelligence Page 15

 IF (cannot_win(P))
RETURN 0;
ELSE
RETURN index to the empty cell on the line of
can_win(P)
Let odd numbers are turns of X
Let even numbers are turns of
O
3. Go(n): make a move

Algorithm:
1. Turn = 1: (X moves)
Go(1) //make a move at the left-top cell
2. Turn = 2: (O moves)
IF board[5] is empty THEN
Go(5)
ELSE
Go(1)
3. Turn = 3: (X moves)
IF board[9] is empty THEN
Go(9)
ELSE
Go(3).
4. Turn = 4: (O moves)
IF Posswin (X) <> 0 THEN
Go (Posswin (X))
//Prevent the opponent to win
ELSE Go (Make2)
5. Turn = 5: (X moves)
Artificial Intelligence Page 16
Artificial Intelligence Page 17
 IF Posswin(X) <> 0 THEN
Go(Posswin(X))
//Win for X.
ELSE IF Posswin(O) <> THEN
Go(Posswin(O))
//Prevent the opponent to win
ELSE IF board[7] is empty THEN
Go(7)
ELSE Go(3).

Comments:
1. Not efficient in time, as it has to check several conditions before making each
move.
2. Easier to understand the program’s strategy.
3. Hard to generalize.

Introduction to Problem Solving, General problem solving

Problem solving is a process of generating solutions from observed data.

−a problem is characterized by a set of goals,
−a set of objects, and
−a set of operations.
These could be ill-defined and may evolve during problem solving.

Searching Solutions:
To build a system to solve a problem:
1. Define the problem precisely
2. Analyze the problem
3. Isolate and represent the task knowledge that is necessary to solve the problem
4. Choose the best problem-solving techniques and apply it to the particular problem.

Artificial Intelligence Page 18

Defining the problem as State Space Search:
The state space representation forms the basis of most of the AI methods.
 Formulate a problem as a state space search by showing the legal problem states, the legal
operators, and the initial and goal states.
 A state is defined by the specification of the values of all attributes of interest in the world
 An operator changes one state into the other; it has a precondition which is the value of certain
attributes prior to the application of the operator, and a set of effects, which are the attributes
altered by the operator
 The initial state is where you start
 The goal state is the partial description of the solution

Formal Description of the problem:

1. Define a state space that contains all the possible configurations of the relevant objects.
2. Specify one or more states within that space that describe possible situations from which the
problem solving process may start ( initial state)
3. Specify one or more states that would be acceptable as solutions to the problem. ( goal states)
Specify a set of rules that describe the actions (operations) available

State-Space Problem Formulation:

Example: A problem is defined by four items:

1. initial state e.g., "at Arad―
2. actions or successor function : S(x) = set of action–state pairs
e.g., S(Arad) = {<Arad  Zerind, Zerind>, … }
3. goal test (or set of goal states)
e.g., x = "at Bucharest‖, Checkmate(x)
4. path cost (additive)
e.g., sum of distances, number of actions executed, etc.
c(x,a,y) is the step cost, assumed to be ≥ 0
A solution is a sequence of actions leading from the initial state to a goal state

Artificial Intelligence Page 19

Example: 8-queens problem

1. Initial State: Any arrangement of 0 to 8 queens on board.

2. Operators: add a queen to any square.
3. Goal Test: 8 queens on board, none attacked.
4. Path cost: not applicable or Zero (because only the final state counts, search cost might
be of interest).

State Spaces versus Search Trees:

 State Space
o Set of valid states for a problem
o Linked by operators
o e.g., 20 valid states (cities) in the Romanian travel problem
 Search Tree
– Root node = initial state
– Child nodes = states that can be visited from parent
– Note that the depth of the tree can be infinite
• E.g., via repeated states
– Partial search tree

Artificial Intelligence Page 20

 • Portion of tree that has been expanded so far
– Fringe
• Leaves of partial search tree, candidates for expansion
Search trees = data structure to search state-space

Properties of Search Algorithms

Which search algorithm one should use will generally depend on the problem domain.
There are four important factors to consider:

1. Completeness – Is a solution guaranteed to be found if at least one solution exists?

2. Optimality – Is the solution found guaranteed to be the best (or lowest cost) solution if there exists
more than one solution?

3. Time Complexity – The upper bound on the time required to find a solution, as a function of the
complexity of the problem.

4. Space Complexity – The upper bound on the storage space (memory) required at any point during the
search, as a function of the complexity of the problem.

General problem solving, Water-jug problem, 8-puzzle problem

General Problem Solver:

The General Problem Solver (GPS) was the first useful AI program, written by Simon, Shaw, and Newell
in 1959. As the name implies, it was intended to solve nearly any problem.

Newell and Simon defined each problem as a space. At one end of the space is the starting point; on the
other side is the goal. The problem-solving procedure itself is conceived as a set of operations to cross
that space, to get from the starting point to the goal state, one step at a time.

The General Problem Solver, the program tests various actions (which Newell and Simon called
operators) to see which will take it closer to the goal state. An operator is any activity that changes the

Artificial Intelligence Page 21

state of the system. The General Problem Solver always chooses the operation that appears to bring it
closer to its goal.

Example: Water Jug Problem

Consider the following problem:

A Water Jug Problem: You are given two jugs, a 4-gallon one and a 3-gallon one, a
pump which has unlimited water which you can use to fill the jug, and the ground on which
water may be poured. Neither jug has any measuring markings on it. How can you get
exactly 2 gallons of water in the 4-gallon jug?

State Representation and Initial State :

We will represent a state of the problem as a tuple (x, y) where x represents the amount of
water in the 4-gallon jug and y represents the amount of water in the 3-gallon jug. Note 0 ≤x≤ 4,
and 0 ≤y ≤3. Our initial state: (0, 0)

Goal Predicate - state = (2, y) where 0≤ y≤ 3.

Operators -we must defi ne a set of operators that will take us from one state to another:

1. Fill 4-gal jug (x,y) → (4,y)

x<4

2. Fill 3-gal jug (x,y) → (x,3)

y<3

3. Empty 4-gal jug on ground (x,y) → (0,y)

x>0

4. Empty 3-gal jug on ground (x,y) → (x,0)

y>0

5. Pour water from 3-gal jug (x,y) →! (4, y - (4 - x))

to ll 4-gal jug 0 < x+y 4 and y > 0
6. Pour water from 4-gal jug (x,y) → (x - (3-y), 3)
to ll 3-gal-jug 0 < x+y 3 and x > 0
7. Pour all of water from 3-gal jug (x,y) → (x+y, 0)

Artificial Intelligence Page 22

 into 4-gal jug 0 < x+y 4 and y 0
8. Pour all of water from 4-gal jug (x,y) → (0, x+y)
into 3-gal jug 0 < x+y 3 and x 0

Through Graph Search, the following solution is found :

Gals in 4-gal jug Gals in 3-gal jug Rule Applied

0 0
1. Fill 4
4 0
6. Pour 4 into 3 to ll
1 3
4. Empty 3
1 0
8. Pour all of 4 into 3
0 1
1. Fill 4
4 1
6. Pour into 3
2 3

Second Solution:

Control strategies
Control Strategies means how to decide which rule to apply next during the process of searching for a
solution to a problem.
Requirement for a good Control Strategy

Artificial Intelligence Page 23

 1. It should cause motion
In water jug problem, if we apply a simple control strategy of starting each time from the top of
rule list and choose the first applicable one, then we will never move towards solution.
2. It should explore the solution space in a systematic manner
If we choose another control strategy, let us say, choose a rule randomly from the applicable
rules then definitely it causes motion and eventually will lead to a solution. But one may arrive
to same state several times. This is because control strategy is not systematic.

Systematic Control Strategies (Blind searches):

Breadth First Search:

Let us discuss these strategies using water jug problem. These may be applied to any search problem.

Construct a tree with the initial state as its root.

Generate all the offspring of the root by applying each of the applicable rules to the initial state.

Now for each leaf node, generate all its successors by applying all the rules that are appropriate.

8 Puzzle Problem.

The 8 puzzle consists of eight numbered, movable tiles set in a 3x3 frame. One cell of the frame is always
empty thus making it possible to move an adjacent numbered tile into the empty cell. Such a puzzle is
illustrated in following diagram.

Artificial Intelligence Page 24

The program is to change the initial configuration into the goal configuration. A solution to the problem
is an appropriate sequence of moves, such as “move tiles 5 to the right, move tile 7 to the left, move tile
6 to the down, etc”.

Solution:

To solve a problem using a production system, we must specify the global database the rules, and the
control strategy. For the 8 puzzle problem that correspond to these three components. These elements
are the problem states, moves and goal. In this problem each tile configuration is a state. The set of all
configuration in the space of problem states or the problem space, there are only 3, 62,880 different
configurations o the 8 tiles and blank space. Once the problem states have been conceptually identified,
we must construct a computer representation, or description of them . this description is then used as
the database of a production system. For the 8-puzzle, a straight forward description is a 3X3 array of
matrix of numbers. The initial global database is this description of the initial problem state. Virtually
any kind of data structure can be used to describe states.

A move transforms one problem state into another state. The 8-puzzle is conveniently interpreted as
having the following for moves. Move empty space (blank) to the left, move blank up, move blank to the
right and move blank down,. These moves are modeled by production rules that operate on the state
descriptions in the appropriate manner.

The rules each have preconditions that must be satisfied by a state description in order for them to be
applicable to that state description. Thus the precondition for the rule associated with “move blank up”
is derived from the requirement that the blank space must not already be in the top row.

The problem goal condition forms the basis for the termination condition of the production system. The
control strategy repeatedly applies rules to state descriptions until a description of a goal state is
produced. It also keeps track of rules that have been applied so that it can compose them into sequence
representing the problem solution. A solution to the 8-puzzle problem is given in the following figure.

Example:- Depth – First – Search traversal and Breadth - First - Search traversal

for 8 – puzzle problem is shown in following diagrams.

Artificial Intelligence Page 25

Exhaustive Searches, BFS and DFS

Search is the systematic examination of states to find path from the start/root state to the goal state.

Many traditional search algorithms are used in AI applications. For complex problems, the traditional
algorithms are unable to find the solution within some practical time and space limits. Consequently,
many special techniques are developed; using heuristic functions. The algorithms that use heuristic

Artificial Intelligence Page 26

functions are called heuristic algorithms. Heuristic algorithms are not really intelligent; they appear to
be intelligent because they achieve better performance.

Heuristic algorithms aremore efficient because they take advantage of feedback from the data to direct
the search path.

Uninformed search

Also called blind, exhaustive or brute-force search, uses no information about the problem to guide the
search and therefore may not be very efficient.

Informed Search:

Also called heuristic or intelligent search, uses information about the problem to guide the search,
usually guesses the distance to a goal state and therefore efficient, but the search may not be always
possible.

Uninformed Search Methods:

Breadth- First -Search:
Consider the state space of a problem that takes the form of a tree. Now, if we search the goal along
each breadth of the tree, starting from the root and continuing up to the largest depth, we call it
breadth first search.

• Algorithm:
1. Create a variable called NODE-LIST and set it to initial state
2. Until a goal state is found or NODE-LIST is empty do
a. Remove the first element from NODE-LIST and call it E. If NODE-LIST was empty,
quit
b. For each way that each rule can match the state described in E do:
i. Apply the rule to generate a new state
ii. If the new state is a goal state, quit and return this state
iii. Otherwise, add the new state to the end of NODE-LIST
BFS illustrated:

Step 1: Initially fringe contains only one node corresponding to the source state A.

Artificial Intelligence Page 27

 Figure 1
FRINGE: A

Step 2: A is removed from fringe. The node is expanded, and its children B and C are generated.
They are placed at the back of fringe.

Figure 2
FRINGE: B C

Step 3: Node B is removed from fringe and is expanded. Its children D, E are generated and put
at the back of fringe.

Figure 3
FRINGE: C D E

Step 4: Node C is removed from fringe and is expanded. Its children D and G are added to the
back of fringe.

Artificial Intelligence Page 28

 Figure 4
FRINGE: D E D G

Step 5: Node D is removed from fringe. Its children C and F are generated and added to the back
of fringe.

Figure 5
FRINGE: E D G C F

Step 6: Node E is removed from fringe. It has no children.

Figure 6
FRINGE: D G C F

Step 7: D is expanded; B and F are put in OPEN.

Figure 7
FRINGE: G C F B F

Artificial Intelligence Page 29

Step 8: G is selected for expansion. It is found to be a goal node. So the algorithm returns
the path A C G by following the parent pointers of the node corresponding to G. The
algorithm terminates.

Breadth first search is:

 One of the simplest search strategies

 Complete. If there is a solution, BFS is guaranteed to find it.
 If there are multiple solutions, then a minimal solution will be found
 The algorithm is optimal (i.e., admissible) if all operators have the same cost. Otherwise,
breadth first search finds a solution with the shortest path length.
 Time complexity : O(bd )
 Space complexity : O(bd )
 Optimality :Yes
b - branching factor(maximum no of successors of any node),
d – Depth of the shallowest goal node
Maximum length of any path (m) in search space
Advantages: Finds the path of minimal length to the goal.
Disadvantages:
 Requires the generation and storage of a tree whose size is exponential the depth of the
shallowest goal node.
 The breadth first search algorithm cannot be effectively used unless the search space is quite
small.

Depth- First- Search.

We may sometimes search the goal along the largest depth of the tree, and move up only when further
traversal along the depth is not possible. We then attempt to find alternative offspring of the parent of
the node (state) last visited. If we visit the nodes of a tree using the above principles to search the goal,
the traversal made is called depth first traversal and consequently the search strategy is called depth
first search.

• Algorithm:
1. Create a variable called NODE-LIST and set it to initial state

Artificial Intelligence Page 30

 2. Until a goal state is found or NODE-LIST is empty do
a. Remove the first element from NODE-LIST and call it E. If NODE-LIST was empty,
quit
b. For each way that each rule can match the state described in E do:
i. Apply the rule to generate a new state
ii. If the new state is a goal state, quit and return this state
iii. Otherwise, add the new state in front of NODE-LIST
DFS illustrated:

A State Space Graph

Step 1: Initially fringe contains only the node for A.

Figure 1
FRINGE: A

Step 2: A is removed from fringe. A is expanded and its children B and C are put in front of
fringe.

Figure 2
FRINGE: B C

Artificial Intelligence Page 31

Artificial Intelligence Page 32
Step 3: Node B is removed from fringe, and its children D and E are pushed in front of fringe.

Figure 3
FRINGE: D E C

Step 4: Node D is removed from fringe. C and F are pushed in front of fringe.

Figure 4
FRINGE: C F E C

Step 5: Node C is removed from fringe. Its child G is pushed in front of fringe.

Figure 5
FRINGE: G F E C
Step 6: Node G is expanded and found to be a goal node.

Artificial Intelligence Page 33

 Figure 6
FRINGE: G F E C

The solution path A-B-D-C-G is returned and the algorithm terminates.

Depth first searchis:

1. The algorithm takes exponential time.

2. If N is the maximum depth of a node in the search space, in the worst case the algorithm will
d
take time O(b ).
3. The space taken is linear in the depth of the search tree, O(bN).

Note that the time taken by the algorithm is related to the maximum depth of the search tree. If the
search tree has infinite depth, the algorithm may not terminate. This can happen if the search space is
infinite. It can also happen if the search space contains cycles. The latter case can be handled by
checking for cycles in the algorithm. Thus Depth First Search is not complete.

Exhaustive searches- Iterative Deeping DFS

Description:

 It is a search strategy resulting when you combine BFS and DFS, thus combining the advantages
of each strategy, taking the completeness and optimality of BFS and the modest memory
requirements of DFS.

 IDS works by looking for the best search depth d, thus starting with depth limit 0 and make a BFS
and if the search failed it increase the depth limit by 1 and try a BFS again with depth 1 and so
on – first d = 0, then 1 then 2 and so on – until a depth d is reached where a goal is found.

Artificial Intelligence Page 34

 Algorithm:

procedure IDDFS(root)
for depth from 0 to ∞
found ← DLS(root, depth)
if found ≠ null
return found

procedure DLS(node, depth)

if depth = 0 and node is a goal
return node
else if depth > 0
foreach child of node
found ← DLS(child, depth−1)
if found ≠ null
return found
return null

Performance Measure:
o Completeness: IDS is like BFS, is complete when the branching factor b is finite.

o Optimality: IDS is also like BFS optimal when the steps are of the same cost.

 Space Complexity:

o IDS is like DFS in its space complexity, taking O(bd) of memory.

Conclusion:

 We can conclude that IDS is a hybrid search strategy between BFS and DFS inheriting their
advantages.

 IDS is faster than BFS and DFS.

 It is said that “IDS is the preferred uniformed search method when there is a large search space
and the depth of the solution is not known”.

Artificial Intelligence Page 35

Heuristic Searches:

A Heuristic technique helps in solving problems, even though there is no guarantee that it will never
lead in the wrong direction. There are heuristics of every general applicability as well as domain specific.
The strategies are general purpose heuristics. In order to use them in a specific domain they are coupler
with some domain specific heuristics. There are two major ways in which domain - specific, heuristic
information can be incorporated into rule-based search procedure.

A heuristic function is a function that maps from problem state description to measures desirability,
usually represented as number weights. The value of a heuristic function at a given node in the search
process gives a good estimate of that node being on the desired path to solution.

Artificial Intelligence Page 36

Greedy Best First Search

Greedy best-first search tries to expand the node that is closest to the goal, on the: grounds that this is
likely to lead to a solution quickly. Thus, it evaluates nodes by using just the heuristic function:
f (n) = h (n).

Taking the example of Route-finding problems in Romania, the goal is to reach Bucharest starting from
the city Arad. We need to know the straight-line distances to Bucharest from various cities as shown in
Figure 8.1. For example, the initial state is In (Arad), and the straight line distance heuristic h SLD (In
(Arad)) is found to be 366. Using the straight-line distance heuristic hSLD, the goal state can be reached
faster.

Arad 366 Mehadia 241 Hirsova 151

Bucharest 0 Neamt 234 Urziceni 80
Craiova 160 Oradea 380 Iasi 226
Drobeta 242 Pitesti 100 Vaslui 199
Eforie 161 Rimnicu Vilcea 193 Lugoj 244
Fagaras 176 Sibiu 253 Zerind 374
Giurgiu 77 Timisoara 329
Figure 8.1: Values of hSLD-straight-line distances to B u c h a r e s t.
The Initial State

After Expanding Arad

After Expanding Sibiu

Artificial Intelligence Page 37

After Expanding Fagaras

Figure 8.2: Stages in a greedy best-first search for Bucharest using the straight-line distance heuristic
hSLD. Nodes are labeled with their h-values.

Figure 8.2 shows the progress of greedy best-first search using h SLD to find a path from Arad to
Bucharest. The first node to be expanded from Arad will be Sibiu, because it is closer to
Bucharest than either Zerind or Timisoara. The next node to be expanded will be Fagaras,
because it is closest.
Fagaras in turn generates Bucharest, which is the goal.

Evaluation Criterion of Greedy Search

 Complete: NO [can get stuck in loops, e.g., Complete in finite space with repeated-
state checking ]
 Time Complexity: O (bm) [but a good heuristic can give dramatic improvement]
 Space Complexity: O (bm) [keeps all nodes in memory]

Artificial Intelligence Page 38

  Optimal: NO

Greedy best-first search is not optimal, and it is incomplete. The worst-case time and space
complexity is O (bm), where m is the maximum depth of the search space.

HILL CLIMBING PROCEDURE:

Hill Climbing Algorithm

We will assume we are trying to maximize a function. That is, we are trying to find a point in the search
space that is better than all the others. And by "better" we mean that the evaluation is higher. We might
also say that the solution is of better quality than all the others.

The idea behind hill climbing is as follows.

1. Pick a random point in the search space.

2. Consider all the neighbors of the current state.
3. Choose the neighbor with the best quality and move to that state.
4. Repeat 2 thru 4 until all the neighboring states are of lower quality.
5. Return the current state as the solution state.
We can also present this algorithm as follows (it is taken from the AIMA book (Russell, 1995) and follows
the conventions we have been using on this course when looking at blind and heuristic searches).

Artificial Intelligence Page 39

Algorithm:
Function HILL-CLIMBING(Problem) returns a solution state
Inputs: Problem, problem
Local variables: Current, a node
Next, a node
Current = MAKE-NODE(INITIAL-STATE[Problem])
Loop do
Next = a highest-valued successor of Current
If VALUE[Next] < VALUE[Current] then
returnCurrent Current = Next
End

Also, if two neighbors have the same evaluation and they are both the best quality, then the algorithm
will choose between them at random.

Problems with Hill Climbing

The main problem with hill climbing (which is also sometimes called gradient descent) is that we are not
guaranteed to find the best solution. In fact, we are not offered any guarantees about the solution. It
could be abysmally bad.

You can see that we will eventually reach a state that has no better neighbours but there are better
solutions elsewhere in the search space. The problem we have just described is called a local maxima.

Simulated annealing search

A hill-climbing algorithm that never makes “downhill” moves towards states with lower value (or higher
cost) is guaranteed to be incomplete, because it can stuck on a local maximum. In contrast, a purely
random walk –that is, moving to a successor chosen uniformly at random from the set of successors – is
complete, but extremely inefficient. Simulated annealing is an algorithm that combines hill-climbing
with a random walk in some way that yields both efficiency and completeness.
Figure 10.7 shows simulated annealing algorithm. It is quite similar to hill climbing. Instead of picking the
best move, however, it picks the random move. If the move improves the situation, it is always
accepted. Otherwise, the algorithm accepts the move with some probability less than 1. The probability
decreases exponentially with the “badness” of the move – the amount E by which the evaluation is

Artificial Intelligence Page 40

worsened. The probability also decreases as the "temperature" T goes down: "bad moves are more
likely to be allowed at the start when temperature is high, and they become more unlikely as T
decreases. One can prove that if the schedule lowers T slowly enough, the algorithm will find a global
optimum with probability approaching 1.
Simulated annealing was first used extensively to solve VLSI layout problems. It has been applied widely
to factory scheduling and other large-scale optimization tasks.
function S I M U L A T E D - A N NEALING( problem, schedule) returns a solution state
inputs: problem, a problem
schedule, a mapping from time to "temperature"
local variables: current, a node
next, a node
T, a "temperature" controlling the probability of downward steps
current MAKE-NODE(INITIAL-STATE[problem])
for tl to ∞ do
T schedule[t]
if T = 0 then return current
next a randomly selected successor of current
EVALUE[next] – VALUE[current]
if E> 0 then current  next
else current  next only with probability eE /T

LOCAL SEARCH IN CONTINUOUS SPACES

 We have considered algorithms that work only in discrete environments, but real-world
environment are continuous.
 Local search amounts to maximizing a continuous objective function in a multi-dimensional
vector space.
 This is hard to do in general.
 Can immediately retreat
 Discretize the space near each state
 Apply a discrete local search strategy (e.g., stochastic hill climbing, simulated annealing)

Artificial Intelligence Page 41

  Often resists a closed-form solution
 Fake up an empirical gradient
 Amounts to greedy hill climbing in discretized state space
 Can employ Newton-Raphson Method to find maxima.
 Continuous problems have similar problems: plateaus, ridges, local maxima, etc.

Best First Search:

 A combination of depth first and breadth first searches.

 Depth first is good because a solution can be found without computing all nodes and breadth
first is good because it does not get trapped in dead ends.
 The best first search allows us to switch between paths thus gaining the benefit of both
approaches. At each step the most promising node is chosen. If one of the nodes chosen
generates nodes that are less promising it is possible to choose another at the same level and in
effect the search changes from depth to breadth. If on analysis these are no better than this
previously unexpanded node and branch is not forgotten and the search method reverts to the

OPEN is a priorityqueue of nodes that have been evaluated by the heuristic function but which have not
yet been expanded into successors. The most promising nodes are at the front.

CLOSED are nodes that have already been generated and these nodes must be stored because a graph is
being used in preference to a tree.

Algorithm:

1. Start with OPEN holding the initial state

2. Until a goal is found or there are no nodes left on open do.

 Pick the best node on OPEN

 Generate its successors
 For each successor Do
• If it has not been generated before ,evaluate it ,add it to OPEN and record its
parent

Artificial Intelligence Page 42

 • If it has been generated before change the parent if this new path is better and
in that case update the cost of getting to any successor nodes.

3. If a goal is found or no more nodes left in OPEN, quit, else return to 2.

Example:

1. It is not optimal.
2. It is incomplete because it can start down an infinite path and never return to try
other possibilities.

Artificial Intelligence Page 43

 3. The worst-case time complexity for greedy search is O (bm), where m is the maximum depth of
the search space.
4. Because greedy search retains all nodes in memory, its space complexity is the same as its
time complexity
A* Algorithm

The Best First algorithm is a simplified form of the A* algorithm.

The A* search algorithm (pronounced "Ay-star") is a tree search algorithm that finds a path from a given
initial node to a given goal node (or one passing a given goal test). It employs a "heuristic estimate"
which ranks each node by an estimate of the best route that goes through that node. It visits the nodes
in order of this heuristic estimate.

Similar to greedy best-first search but is more accurate because A* takes into account the nodes that
have already been traversed.

From A* we note that f = g + h where

g is a measure of the distance/cost to go from the initial node to the current node

his an estimate of the distance/cost to solution from the current node.

Thus fis an estimate of how long it takes to go from the initial node to the solution

Algorithm:

1. Initialize : Set OPEN = (S); CLOSED = ( )

g(s)= 0, f(s)=h(s)

2. Fail : If OPEN = ( ), Terminate and fail.

3. Select : select the minimum cost state, n, from OPEN,

save n in CLOSED

4. Terminate : If n €G, Terminate with success and return f(n)

5. Expand : for each successor, m, of n

Artificial Intelligence Page 44

 a) If m € *OPEN U

CLOSED+ Set g(m) = g(n)

+ c(n , m) Set f(m) =

g(m) + h(m)

Insert m in OPEN

b) If m € *OPEN U CLOSED+

Set g(m) = min { g(m) , g(n) + c(n , m)}

Set f(m) = g(m) + h(m)

If f(m) has decreased and m € CLOSED

Move m to OPEN.

Description:

 A* begins at a selected node. Applied to this node is the "cost" of entering this node (usually
zero for the initial node). A* then estimates the distance to the goal node from the current
node. This estimate and the cost added together are the heuristic which is assigned to the path
leading to this node. The node is then added to a priority queue, often called "open".
 The algorithm then removes the next node from the priority queue (because of the way a
priority queue works, the node removed will have the lowest heuristic). If the queue is empty,
there is no path from the initial node to the goal node and the algorithm stops. If the node is the
goal node, A* constructs and outputs the successful path and stops.
 If the node is not the goal node, new nodes are created for all admissible adjoining nodes; the
exact way of doing this depends on the problem at hand. For each successive node, A*
calculates the "cost" of entering the node and saves it with the node. This cost is calculated from
the cumulative sum of costs stored with its ancestors, plus the cost of the operation which
reached this new node.
 The algorithm also maintains a 'closed' list of nodes whose adjoining nodes have been checked.
If a newly generated node is already in this list with an equal or lower cost, no further
processing is done on that node or with the path associated with it. If a node in the closed list
matches the new one, but has been stored with a higher cost, it is removed from the closed list,
and processing continues on the new node.

Artificial Intelligence Page 45

Artificial Intelligence Page 46
  Next, an estimate of the new node's distance to the goal is added to the cost to form the
heuristic for that node. This is then added to the 'open' priority queue, unless an identical node
is found there.
 Once the above three steps have been repeated for each new adjoining node, the original node
taken from the priority queue is added to the 'closed' list. The next node is then popped from
the priority queue and the process is repeatedThe heuristic costs from each city to Bucharest:

Artificial Intelligence Page 47

Artificial Intelligence Page 48
A* search properties:

 The algorithm A* is admissible. This means that provided a solution exists, the first solution
found by A* is an optimal solution. A* is admissible under the following conditions:

 Heuristic function: for every node n , h(n) ≤ h*(n) .

 A* is also complete.

 A* is optimally efficient for a given heuristic.

 A* is much more efficient that uninformed search.

Iterative Deeping A* Algorithm:

Algorithm:

Let L be the list of visited but not expanded node, and

C the maximum depth
1) Let C=0
2) Initialize Lto the initial state (only)
3) If List empty increase C and goto 2),
else
extract a node n from the front of L
4) If n is a goal node,
SUCCEED and return the path from the initial state to n
5) Remove n from L. If the level is smaller than C, insert at the front of L all the children n' of n
with f(n') ≤ C
6) Goto 3)

Artificial Intelligence Page 49

  IDA* is complete & optimal Space usage is linear in the depth of solution. Each iteration is depth
first search, and thus it does not require a priority queue.

 Iterative deepening A* (IDA*) eliminates the memory constraints of A* search algorithm

without sacrificing solution optimality.

 Each iteration of the algorithm is a depth-first search that keeps track of the cost, f(n) = g(n) +
h(n), of each node generated.

 As soon as a node is generated whose cost exceeds a threshold for that iteration, its path is cut
off, and the search backtracks before continuing.

 The cost threshold is initialized to the heuristic estimate of the initial state, and in each
successive iteration is increased to the total cost of the lowest-cost node that was pruned during
the previous iteration.

 The algorithm terminates when a goal state is reached whose total cost dees not exceed the
current threshold.

Constraint Satisfaction Problems

Sometimes a problem is not embedded in a long set of action sequences but requires picking the best
option from available choices. A good general-purpose problem solving technique is to list the
constraints of a situation (either negative constraints, like limitations, or positive elements that you
want in the final solution). Then pick the choice that satisfies most of the constraints.

Formally speaking, a constraint satisfaction problem (or CSP) is defined by a set of variables, X1;X2; : : :
;Xn, and a set of constraints, C1;C2; : : : ;Cm. Each variable Xi has anonempty domain Di of possible
values. Each constraint Ci involves some subset of tvariables and specifies the allowable combinations of
values for that subset. A state of theproblem is defined by an assignment of values to some or all of the
variables, {Xi = vi;Xj =vj ; : : :} An assignment that does not violate any constraints is called a consistent or

Artificial Intelligence Page 50

legalassignment. A complete assignment is one in which every variable is mentioned, and a solution to a
CSP is a complete assignment that satisfies all the constraints. Some CSPs also require a solution that
maximizes an objectivefunction.

CSP can be given an incremental formulation as a standard search problem as follows:

1. Initial state: the empty assignment fg, in which all variables are unassigned.

2. Successor function: a value can be assigned to any unassigned variable, provided that it does not
conflict with previously assigned variables.

3. Goal test: the current assignment is complete.

4. Path cost: a constant cost for every step

Examples:

1. The best-known category of continuous-domain CSPs is that of linear

programming problems, where constraints must be linear inequalities
forming a convex region.

2. Crypt arithmetic puzzles.

Example: The map coloring problem.

The task of coloring each region red, green or blue in such a way that no neighboring regions
have the same color.

We are given the task of coloring each region red, green, or blue in such a way that the
neighboring regions must not have the same color.

To formulate this as CSP, we define the variable to be the regions: WA, NT, Q, NSW, V, SA, and
T. The domain of each variable is the set {red, green, blue}. The constraints require

Artificial Intelligence Page 51

neighboring regions to have distinct colors: for example, the allowable combinations for WA
and NT are the pairs {(red,green),(red,blue),(green,red),(green,blue),(blue,red),(blue,green)}.
(The constraint can also be represented as the inequality WA ≠ NT). There are many possible
solutions, such as {WA = red, NT = green, Q = red, NSW = green, V = red, SA = blue, T =
red}.Map of Australia showing each of its states and territories

Constraint Graph: A CSP is usually represented as an undirected graph, called constraint graph
where the nodes are the variables and the edges are the binaryconstraints.

The map-coloring problem represented as a constraint graph.

CSP can be viewed as a standard search problem as follows:

> Initial state : the empty assignment {},in which all variables are unassigned.
> Successor function: a value can be assigned to any unassigned variable, provided that it
does not conflict with previously assigned variables.
> Goal test: the current assignment is complete.

Artificial Intelligence Page 52

 > Path cost: a constant cost(E.g.,1) for every step.

Game Playing
Adversarial search, or game-tree search, is a technique for analyzing an adversarial game in order to try
to determine who can win the game and what moves the players should make in order to win.
Adversarial search is one of the oldest topics in Artificial Intelligence. The original ideas for adversarial
search were developed by Shannon in 1950 and independently by Turing in 1951, in the context of the
game of chess—and their ideas still form the basis for the techniques used today.
2- Person Games:

o Players: We call them Max and Min.

o Initial State: Includes board position and whose turn it is.
o Operators: These correspond to legal moves.
o Terminal Test: A test applied to a board position which determines whether the game is
over. In chess, for example, this would be a checkmate or stalemate situation.
o Utility Function: A function which assigns a numeric value to a terminalstate. For
example, in chess the outcome is win (+1), lose (-1) or draw (0). Note that by
convention, we always measure utility relative to Max.

MiniMax Algorithm:
1. Generate the whole game tree.
2. Apply the utility function to leaf nodes to get their values.
3. Use the utility of nodes at level n to derive the utility of nodes at level n-1.
4. Continue backing up values towards the root (one layer at a time).
5. Eventually the backed up values reach the top of the tree, at which point Max chooses the move
that yields the highest value. This is called the minimax decision because it maximises the utility
for Max on the assumption that Min will play perfectly to minimise it.

Artificial Intelligence Page 53

Example:

Properties of minimax:

Artificial Intelligence Page 54

  Complete : Yes (if tree is finite)
 Optimal : Yes (against an optimal opponent)
 Time complexity : O(bm)
 Space complexity : O(bm) (depth-first exploration)
 For chess, b ≈ 35, m ≈100 for "reasonable" games
 exact solution completely infeasible.
Limitations
– Not always feasible to traverse entire tree
– Time limitations

Alpha-beta pruning algorithm:

• Pruning: eliminating a branch of the search tree from consideration without exhaustive
examination of each node
• - Pruning: the basic idea is to prune portions of the search tree that cannot improve the
utility value of the max or min node, by just considering the values of nodes seen so far.
• Alpha-beta pruning is used on top of minimax search to detect paths that do not need to be
explored. The intuition is:
• The MAX player is always trying to maximize the score. Call this .
• The MIN player is always trying to minimize the score. Call this  .
• Alpha cutoff: Given a Max node n, cutoff the search below n (i.e., don't generate or examine any
more of n's children) if alpha(n) >= beta(n)
(alpha increases and passes beta from below)
• Beta cutoff.: Given a Min node n, cutoff the search below n (i.e., don't generate or examine any
more of n's children) if beta(n) <= alpha(n)
(beta decreases and passes alpha from above)
• Carry alpha and beta values down during search Pruning occurs whenever alpha >= beta

Artificial Intelligence Page 55

Algorithm:

Artificial Intelligence Page 56

Example:

1) Setup phase: Assign to each left-most (or right-most) internal node of the tree,
variables: alpha = -infinity, beta = +infinity

2) Look at first computed final configuration value. It’s a 3. Parent is a min node, so
set the beta (min) value to 3.

Artificial Intelligence Page 57

3) Look at next value, 5. Since parent is a min node, we want the minimum of
3 and 5 which is 3. Parent min node is done – fill alpha (max) value of its parent max
node. Always set alpha for max nodes and beta for min nodes. Copy the state of the max
parent node into the second unevaluated min child.

4) Look at next value, 2. Since parent node is min with b=+inf, 2 is smaller, change b.

Artificial Intelligence Page 58

5) Now, the min parent node has a max value of 3 and min value of 2. The value of the
2nd child does not matter. If it is >2, 2 will be selected for min node. If it is <2, it will be
selected for min node, but since it is <3 it will not get selected for the parent max node.
Thus, we prune the right subtree of the min node. Propagate max value up the tree.

6) Max node is now done and we can set the beta value of its parent and propagate
node state to sibling subtree’s left-most path.

Artificial Intelligence Page 59

Artificial Intelligence Page 60
7) The next node is 10. 10 is not smaller than 3, so state of parent does not change. We still
have to look at the 2nd child since alpha is still –inf.

8) The next node is 4. Smallest value goes to the parent min node. Min subtree is done, so
the parent max node gets the alpha (max) value from the child. Note that if the max node
had a 2nd subtree, we can prune it since a>b.

Artificial Intelligence Page 61

9) Continue propagating value up the tree, modifying the corresponding alpha/beta values.
Also propagate the state of root node down the left-most path of the right subtree.

10) Next value is a 2. We set the beta (min) value of the min parent to 2. Since no other
children exist, we propagate the value up the tree.

Artificial Intelligence Page 62

11) We have a value for the 3rd level max node, now we can modify the beta (min) value of
the min parent to 2. Now, we have a situation that a>b and thus the value of the rightmost
subtree of the min node does not matter, so we prune the whole subtree.

12) Finally, no more nodes remain, we propagate values up the tree. The root has a value
of 3 that comes from the left-most child. Thus, the player should choose the left-most
child’s move in order to maximize his/her winnings. As you can see, the result is the same
as with the mini-max example, but we did not visit all nodes of the tree.

Artificial Intelligence Page 63

Artificial Intelligence Page 64