Lecture Notes in

Artificial Intelligence

Dr V N Krishnachandran

Department of Computer Applications

(MCA) Vidya academy of Science and
Technology Thrissur – 680 501
This page is intentionally left blank.
 L ECTURE N OTES IN
A RTIFICIAL I NTELLIGENCE

Dr V N Krishnachandran

Department of Computer Applications (MCA)

Vidya Academy of Science & Technology
Thrissur - 680501
Title: L ECTURE NOTES IN ARTIFICIAL I NTELLIGENCE

Author’s name:
DR. V. N. KRISHNACHANDRN

Published by author
Pblisher’s address: Dr. V. N.
Krishnachandran Vidya Academy of Science
& Technology Thalakkottukara, Thrissur -
680501, Kerala

First Edition

The book was typeset by the author using the LATEX document preparation

system. Cover design: Author

Copyright ⃝c Author

Licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0)
License. You may not use this file except in compliance with the License. You may
obtain a copy of the License at
https://creativecommons.org/licenses/by/4.0/.

Price: Rs 0.00.

First printing: April 2021

ISBN: · · · · · · · · ·
Preface

The book is written primarily for the students pursuing the MCA programme of the APJ
Abdul Kalam Technological University. The Curriculum for the programme offers a
course on artificial intelligence as an elective course in the Second Semester with code
and name “20MCA188 Artificial Intelligence”. The selection of topics in the book was
guided by the contents of the syllabus for the course and also by the contents in the books
cited in the syllabus. The book will also be useful to faculty members who teach the
course. The users of these notes are earnestly requested to bring to the notice of the
author any errors they find so that a corrected and revised edition can be published at the
earliest.

VAST Campus Dr V N Krishnachandran

April 2021 Department of Computer
Applications Vidya Academy of Science & Technology,
Thrissur - 680501 (email:
[email protected])

i
This page is intentionally left blank.
Contents

Preface i

Syllabus ix

1 Introduction to artificial intelligence 1

1.1 What is artificial intelligence? . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 History of artificial intelligence . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Applications of artificial intelligence . . . . . . . . . . . . . . . . . . . . . 2
1.3.1 Some well known applications of artificial intelligence . . . . . . . 2
1.3.2 Recent applications . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Sample questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Problem definition and examples 6

2.1 Rational agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.1 Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.2 Rational agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Definition of the concept of a problem in AI . . . . . . . . . . . . . . . . . 7
2.2.1 An illustrative example . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Example problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 The vacuum cleaner world . . . . . . . . . . . . . . . . . . . . . . 11
2.3.2 8-puzzle problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.3 Missionaries and cannibals problem . . . . . . . . . . . . . . . . . 13
2.3.4 Cryptarithmetic puzzle . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.5 Blocks world problem . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.6 Monkey and banana problem . . . . . . . . . . . . . . . . . . . . . 21
2.3.7 A map problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.8 Water jug problem . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Problem characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.1 Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Examples illustrating problem characteristics . . . . . . . . . . . . . . . . 25
2.5.1 Problem is decomposable . . . . . . . . . . . . . . . . . . . . . . 25
2.5.2 Solution steps can be ignored or undone. . . . . . . . . . . . . . . 27
2.5.3 The problem universe is predictable. . . . . . . . . . . . . . . . . . 27
2.5.4 There are obvious good solutions. . . . . . . . . . . . . . . . . . . 28
2.5.5 Desired solution is a state of the universe or a path to a state. . . . . 28
2.5.6 Requires lots of knowledge; or, uses knowledge to constrain solutions. 28

iii
CONTENTS iv

2.5.7 Problem requires periodic interaction between human and computer. 29

2.6 Production systems..............................................................................................29
2.6.1 Definition................................................................................................29
2.6.2 Production system example: 3 × 3 knight’s tour....................................31
2.6.3 Production system example: The water jug problem.............................33
2.6.4 Production system characteristics...........................................................35
2.6.5 Advantages and disadvantages of production systems...........................36
2.7 Sample questions.................................................................................................37

3 Blind search strategies 39

3.1 Preliminaries........................................................................................................39
3.1.1 Graphs....................................................................................................39
3.1.2 Trees.......................................................................................................39
3.2 Search in general.................................................................................................40
3.2.1 Solution as a sequence of actions...........................................................40
3.2.2 Search algorithm.....................................................................................41
3.2.3 Search tree..............................................................................................41
3.2.4 Search strategy........................................................................................42
3.3 Blind search.........................................................................................................42
3.3.1 Example..................................................................................................43
3.4 Breadth-first search.............................................................................................43
3.4.1 Algorithm...............................................................................................43
3.4.2 Examples................................................................................................44
3.5 Depth-first search................................................................................................46
3.5.1 Algorithm...............................................................................................46
3.5.2 Examples................................................................................................46
3.6 Breadth-first search vs. depth-first search...........................................................49
3.7 Depth-first iterative deepening (DFID) search....................................................50
3.7.1 Algorithm...............................................................................................50
3.7.2 Advantages and disadvantages...............................................................51
3.8 Blind search in graphs.........................................................................................51
3.8.1 Graph-search algorithm..........................................................................51
3.8.2 Breadth-first and depth-first search in graphs........................................51
3.8.3 Example..................................................................................................52
3.9 Sample questions.................................................................................................53

4 Heuristic search strategies 56

4.1 Heuristic function................................................................................................56
4.1.1 Definition................................................................................................57
4.1.2 Admissible heuristic functions...............................................................57
4.1.3 Remarks..................................................................................................57
4.1.4 Examples................................................................................................58
4.2 Best first search...................................................................................................62
4.3 Greedy best first search.......................................................................................62
4.3.1 The algorithm.........................................................................................62
4.3.2 Examples................................................................................................63
4.4 A∗ best first search..............................................................................................69
4.4.1 Examples................................................................................................69
 CONTENTS v

4.4.2 A∗ best first search algorithm . . . . . . . . . . . . . . . . . . . . 71

4.5Greedy best first vs A∗ best first . . . . . . . . . . . . . . . . . . . . . . . 72
4.6Simple hill climbing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.6.1 Simple hill climbing algorithm . . . . . . . . . . . . . . . . . . . . 74
4.6.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.7Steepest-ascent hill climbing . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.7.1 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.8Advantages and disadvantages of hill climbing . . . . . . . . . . . . . . . . 77
4.9Simulated annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.9.1 Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.9.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.10 Sample questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5 Game playing 84
5.1Games considered in AI . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.2Game tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.3Solving a game . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4Formulation of a game as a search problem . . . . . . . . . . . . . . . . . 88
5.5The minimax value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.6The minimax algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.6.1 The algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.6.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.7Minimax with alpha-beta pruning . . . . . . . . . . . . . . . . . . . . . . 93
5.7.1 Basic idea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.7.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.7.3 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.7.4 Additional example . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.8Sample questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

6 Knowledge representation 100

6.1Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.1.1 Knowledge in AI . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.1.2 Different types of knowledge . . . . . . . . . . . . . . . . . . . . . 101
6.2Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.3Knowledge representation . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.4Desirable properties of knowledge representation systems . . . . . . . . . . 101
6.5Some knowledge representation systems . . . . . . . . . . . . . . . . . . . 102
6.6Semantic networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.6.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.6.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.6.3 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.6.4 Different kinds of semantic networks . . . . . . . . . . . . . . . . 108
6.6.5 Intersection search . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.6.6 Representing nonbinary predicates . . . . . . . . . . . . . . . . . . 111
6.7Advantages and disadvantages of semantic networks . . . . . . . . . . . . 111
6.7.1 Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.7.2 Disadvantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.8Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
CONTENTS vi

6.8.1 Definition..............................................................................................112
6.8.2 Frame structure.....................................................................................113
6.8.3 Examples..............................................................................................113
6.8.4 Frame languages...................................................................................115
6.9 Conceptual dependency.....................................................................................116
6.9.1 Simple examples of representations of sentences in CD theory...........116
6.9.2 The building blocks of conceptual dependency...................................118
6.9.3 Rules.....................................................................................................121
6.9.4 Advantages and disadvantages of the CD theory.................................123
6.10 Sample questions...............................................................................................123

7 Knowledge representation using logic 127

7.1 Propositional logic............................................................................................127
7.1.1 Propositions..........................................................................................127
7.1.2 Operations on propositions: Logical operators.....................................127
7.1.3 Tautologies and contradictions.............................................................129
7.1.4 Identities...............................................................................................130
7.1.5 Inferences..............................................................................................131
7.1.6 Rules of inference.................................................................................132
7.1.7 Conjuctive normal form........................................................................132
7.1.8 Resolution-refutation in propositional logic.........................................135
7.2 First order predicate logic (FOPL)....................................................................136
7.2.1 Domain of discourse.............................................................................138
7.2.2 Predicates..............................................................................................138
7.2.3 Basic elements......................................................................................138
7.3 Quantifiers.........................................................................................................139
7.3.1 Universal quantification........................................................................139
7.3.2 Existential quantification......................................................................140
7.3.3 Nested quantifiers.................................................................................140
7.3.4 Connections between ∀ and ∃..............................................................140
7.3.5 Examples of quantified statements.......................................................141
7.4 Well-formed formulas in FOPL.........................................................................142
7.4.1 Well-formed formulas...........................................................................142
7.4.2 Clausal forms (also called conjuctive normal forms (CNFs))..............142
7.4.3 Algorithm to convert a wff to the clausal form (or, CNF)...................142
7.4.4 Examples..............................................................................................143
7.5 Inference rules in FOPL....................................................................................145
7.5.1 Rules borrowed from propositional logic.............................................145
7.5.2 Universal instantiation (UI)..................................................................145
7.5.3 Existential instantiation (EI).................................................................146
7.5.4 Generalised modus ponens (GMP).......................................................146
7.5.5 The resolution inference rule................................................................147
7.6 Inference methods in FOPL..............................................................................148
7.6.1 Propositionalisation: Reduction to propositional inference..................148
7.6.2 Resolution-refutation method in FOPL................................................149
7.6.3 Examples..............................................................................................149
7.7 Summary of knowledge representation methods..............................................154
7.8 Sample questions...............................................................................................157
CONTENTS vii
8 Planning 160
8.1 Overview...........................................................................................................160
8.1.1 Decomposition of problems..................................................................160
8.1.2 Nearly decomposable problems............................................................161
8.1.3 Planning................................................................................................161
8.1.4 Impact of predictability on planning....................................................161
8.2 Components of a planning system....................................................................161
8.2.1 Choosing rules to apply........................................................................161
8.2.2 Applying rules......................................................................................162
8.2.3 Detecting a solution..............................................................................162
8.2.4 Detecting dead-ends..............................................................................162
8.2.5 Repairing an almost correct solution....................................................162
8.3 Goal stack planning...........................................................................................163
8.3.1 Stacks....................................................................................................163
8.3.2 Goal stack.............................................................................................163
8.3.3 Basic idea..............................................................................................164
8.3.4 The STRIPS planner.............................................................................164
8.3.5 Algorithm..............................................................................................164
8.3.6 Illustration: Blocks world.....................................................................164
8.4 Hierarchical planning........................................................................................170
8.4.1 The basic approach...............................................................................170
8.4.2 The ABSTRIPS planner system...........................................................170
8.5 Sample questions...............................................................................................171

9 Learning 172
9.1 What is learning?...............................................................................................172
9.2 Forms of learning based on feedback................................................................172
9.3 Forms of learning based on different ways of acquiring knowledge................173
9.3.1 Rote learning.........................................................................................173
9.3.2 Learning by taking advice....................................................................173
9.3.3 Learning in problem solving................................................................174
9.3.4 Inductive learning.................................................................................175
9.3.5 Explanation based learning...................................................................176
9.3.6 Discovery..............................................................................................176
9.3.7 Analogy................................................................................................177
9.4 Neural net learning............................................................................................178
9.4.1 Neurons and perceptrons......................................................................178
9.4.2 Neural networks....................................................................................179
9.4.3 Learning in neural networks.................................................................179
9.4.4 Applications..........................................................................................179
9.5 Genetic algorithms............................................................................................180
9.5.1 Outline..................................................................................................180
9.6 Sample questions...............................................................................................181

10 Expert systems 182

10.1 Introduction.......................................................................................................182
10.2 Architecture of an expert system.......................................................................182
10.2.1 User interface........................................................................................182
10.2.2 Knowledge base....................................................................................183
CONTENTS viii
10.2.3 Inference engine...................................................................................183
10.2.4 Case specific data.................................................................................184
10.2.5 Explanation system...............................................................................184
10.2.6 Knowledge base editor.........................................................................184
10.2.7 Expert shell...........................................................................................185
10.3 Roles of individuals who interact with expert systems.....................................185
10.4 Languages and tools..........................................................................................185
10.4.1 Programming languages.......................................................................185
10.4.2 Knowledge engineering languages.......................................................186
10.4.3 System-building aids.............................................................................187
10.4.4 Tool support environment facilities......................................................187
10.5 Typical expert systems......................................................................................187
10.6 Benefits and limitations.....................................................................................189
10.7 Sample questions...............................................................................................190

11 Fuzzy logic 191

11.1 What is fuzzy logic?..........................................................................................191
11.2 Fuzzy logic as multivalued logic.......................................................................191
11.2.1 Three-valued logic................................................................................191
11.2.2 Logical connectives in three valued logic............................................192
11.2.3 Compound propositions in three-valued logic.....................................193
11.2.4 Other multivalued logics.......................................................................193
11.3 Fuzzy sets..........................................................................................................193
11.3.1 Introduction...........................................................................................193
11.3.2 Example................................................................................................194
11.4 Crisp sets...........................................................................................................194
11.5 Definition and examples....................................................................................195
11.5.1 Definition..............................................................................................195
11.5.2 Examples..............................................................................................196
11.5.3 Some special membership functions....................................................197
11.6 Set-theoretic operations for fuzzy sets..............................................................199
11.6.1 Definitions............................................................................................199
11.6.2 Examples..............................................................................................200
11.7 Some more definitions.......................................................................................202
11.8 Properties of set operations on fuzzy sets.........................................................202
11.9 Fuzzy variables..................................................................................................203
11.9.1 Definition..............................................................................................204
11.9.2 Examples..............................................................................................204
11.10Example for application of fuzzy sets...............................................................204
11.11Sample questions...............................................................................................208

Index 210
 Syllabus

0. Course name and code

Category L T P Credit
20MCA 188 Artificial Intelligence
Elective 3 1 0 4

1. Preamble
This course introduces the techniques of Artificial Intelligence and analyses various meth-
ods of solving problems using it. The concept of expert system architecture and fuzzy
operations are introduced. This course serves as a prerequisite for many advanced
courses in Data Science areas.

2. Prerequisite
Mathematical foundations for computing, advanced data structures

3. Course outcomes
After the completion of the course the student will be able to:

CO 1 Apply the steps needed to provide a formal specification for solv-

ing the problem.
CO 2 Apply and analyze the different types of control and heuristic
search methods to solve problems
CO 3 Understand various Game theory problems and Knowledge struc-
tures
CO 4 Formulate knowledge representation and examine resolution in
predicate and propositional logic
CO 5 Apply feasible planning and learning techniques to solve non-trial
problems
CO 6 Analyze expert systems and fuzzy operations to solve real life
problems.

ix
SYLLABUS x

4. Mapping of course outcomes to programme outcomes

PO PO PO PO PO PO PO PO PO PO PO PO
1 2 3 4 5 6 7 8 9 10 11 12
CO 1 3 3 2
CO 2 3 3 2
CO 3 3 3 2
CO 4 3 3 2
CO 5 3 3 2
CO 6 3 3 3 3 2 2 2

5. Assessment pattern

Continuous Assessment Tests End Semester

Bloom’s Category
1 2 Examination
Remember(K1) 10 10 10
Understand(K2) 20 20 20
Apply(K3) 20 20 30
Analyse(K4)
Evaluate(K5)
Create(K6)

6. Marks distribution

Total Marks Continuous End Semester ESE Duration

Internal Evalua- Examination
tion (CIE) (ESE)
100 40 60 3 hours

7. Continuous internal evaluation pattern

Attendance : 8 marks
Continuous Assessment Test (2 numbers) : 20
marks Assignment/Quiz/Course project : 12 marks

8. Course level assessment questions

Course Outcome 1 (CO 1)

1. Describe the areas of Artificial intelligence. (K1)

2. List the problem formulations & production characteristics. (K1 & K2)

3. Solve the various problems such as 8 puzzle, Crypt arithmetic, etc. (K3)

Course Outcome 2 (CO 2)

1. Describe search strategies in solving problems. (K1 & K2)

2. List the disadvantages of hill climbing algorithm. (K1& K2).

SYLLABUS xi

3. Illustrate A* algorithm for the graph. (K3)

Course Outcome 3 (CO 3)

1. Demonstrate two player Zero sum game. (K3)

2. List and explain the knowledge representation methods in AI. (K1 & K2)

3. Explain how alpha-beta algorithm works in pruning of branches with an example.

(K3)

Course Outcome 4 (CO 4)

1. Translate the following sentence to predicate logic: (K3)

(a) “All pompeians were Romans.”

(b) “All Romans were either loyal to Caesar or hated him.”

2. Explain the algorithm to convert WFF to clause. (K1 & K2)

3. Describe about resolution graph in predicate and propositional logic. (K1 & K2)

Course Outcome 5 (CO 5)

1. Differentiate between Goal stack and Hierarchical planning in AI. (K1 & K2)

2. Discuss about neural net learning. (K1 & K2)

3. List out the steps in genetic learning. (K1 & K2)

Course Outcome 6 (CO 6)

1. Specify the components in expert system. (K1 & K2)

2. Solve various fuzzy operations. (K3)

3. List out and explain various tools and languages in AI. (K1 & K2)

8. Model question paper

Part A

1. List the applications areas in AI.

2. Solve the following cryptarithmetic problem:

SEND + MORE

MONEY

3. Explain iterative deepening search.

4. List the disadvantages of hill climbing.

SYLLABUS xii

5. Solve a simple two player Zero sum game.

6. Explain about conceptual dependency.

7. Explain inference rules in FOPL.

8. List components of a planning system.

9. Give a short note on role of an expert system.

10. List various fuzzy operations.

(10 × 3 = 30 marks)

Part B

11. Consider a water jug problem. You are given two jugs, a 4 gallon and 3 gallons.
Neither has any measuring markers on it. There is a pump that can be used to fill
the jugs with water. How can you get exactly 2 gallons of water into 4-gallon jug.
State the production rule for waterjug problem. (6)

OR

12. Solve missionaries and cannibals problem. (6)

13. Explain blind search strategies in detail. (6)

OR

14. Explain A* Algorithm for the given graph: (6)

S A B C D
6 2 3 0 2

15. List and explain the knowledge representation methods in AI. (6)

OR

16. Explain how alpha-beta algorithm works in pruning of branches with an example.(6)
 SYLLABUS xiii

17. Explain the algorithm to convert WFF to clause with an example. (6)

OR

18. Explain Neural net and Genetic learning methods in AI. (6)
19. Illustrate architecture of an expert system and mention its features. (6)

OR

20. Solve the following using various fuzzy set operations: (6)

A = {0.1/1, 0.3/2, 0.45/3}, B = {0.15/1, 0.34/2}.

(5×6=30 Marks)

(Note: Question 20 appears to be incomplete.)

9. Syllabus
Module 1

Introduction to AI and production systems: AI Problem formulation, Problem

definition, Production systems, Problem characteristics, Production system
characteristics. Example AI problems: 8 puzzle problem, Missionary and cannibals
problem, Cryptarithmetic prob- lems, Block world problems.
Module 2
Search strategies: Blind search strategies: Depth first search, Breadth first search, Best first
search, Iterative deepening search. Heuristic search strategies: Admissible heuristics and
examples, Simple hill climbing and steepest ascending hill climbing, Simulated
annealing, A* algorithm.
Module 3
Game playing: Two-player zero sum games, Modelling two-player zero sum games as
search problems, Min-max algorithm, Optimizing Min-max algorithm using α - β cut off.
Knowledge representation structures: Frames, Semantic networks, Conceptual depen-
dencies.
Module 4
Knowledge representation using logic: First order predicate logic (FOPL), Well formed
formula (WFF) in FOPL, Inference rules for FOPL, Clause form and conversion of WFFs
to clause form, Resolution-refutation. Planning: Overview, components of a planning
system, Goal stack planning, Hierarchical planning. Learning: Forms of learning, Neural
net learning, Genetic learning.
Module 5
Expert systems: Architecture of expert systems, Roles of expert systems, Languages and
tools - Typical expert system examples. Fuzzy logic: Fuzzy variables, Fuzzy sets and
fuzzy set operations, Typical examples using fuzzy sets.
SYLLABUS xiv

10. Text books

1. Kevin Night and Elaine Rich, “Artificial Intelligence (SIE)”, McGraw Hill, 2008.
2. Stuart Russel and Peter Norvig, “AI - A Modern Approach”, 2nd Edition, Pearson
Education, 2007

11. Reference books

1. Peter Jackson, “Introduction to Expert Systems”, 3rd Edition, Pearson Education,
2007.
2. Dan W. Patterson, “Introduction to AI and ES”, Pearson Education, 2007.

12. Course contents and lecture schedule

Number of
No. Topic
lectures
1 Module I: Introduction to AI 9 hours
AI-Problem formulation, Problem Definition -
1.1
Production systems
1.2 Production system characteristics
1.3 AI Problems
2 Module II: Search Strategies 9 hours
2.1 Blind search strategies
2.2 Heuristics search strategies
Simple Hill Climbing and Steepest Ascending Hill
2.3
Climbing,
2.4 Simulated annealing
2.5 A* algorithm
3 Module III: Game playing 9 hours
3.1 Zero sum game
3.2 Minimax algorithm
3.3 Alpha beta pruning
3.4 Knowledge representation structure
Module IV: Knowledge representation using
4 12 hours
Logic
4.1 First Order Predicate Logic (FOPL)
Well Formed Formula(WFF) in FOPL, Inference
4.2
rules for FOPL
The Clause Form and conversion of WFFs to Clause
4.3
Form
4.4 Resolution
4.5 Planning
4.6 Learning
5 Module V: Applications 6 hours
5.1 Expert system Architecture
5.2 Fuzzy logic operations
5.3 Languages and tools
 L ECTURE N OTES IN
A RTIFICIAL I NTELLIGENCE
This page is intentionally left blank.
Chapter 1

Introduction to artificial intelligence

In this introductory chapter, we shall see several definitions of artificial intelligence (the
terminology is commonly abbreviated as AI), shall have a look at the various stages in
the historical development of the discipline of artificial intelligence and shall have a
glimpse of the many fields in which this discipline has been successfully applied.

1.1 What is artificial intelligence?

There is no universally accepted definition of artificial intelligence. The following are
some of the well-known definitions of artificial intelligence:
1. “Artificial intelligence is the art of creating machines that perform functions that
re- quire intelligence when performed by people. (Kurzweil, 1990)”
2. “Artificial intelligence is the study of how to make computers do things at which,
at the moment, people are better.”
3. “An agent is something that acts. A rational agent is an agent that acts so as to
achieve the best outcome or when there is uncertainty, the best expected outcome.
Artificial intelligence is the study of the general principles of rational agents and of
the components for constructing them.”

1.2 History of artificial intelligence

The following are some of the important stages in the development of the study of
artificial intelligence as the term is now generally understood.
1. Work prior to the coining of the term "artificial intelligence" (1943 - 1955)
The first work that is now generally recognized as artificial intelligence was done
by Warren McCulloch and Walter Pitts in around 1943. It was not then referred to
as a work in artificial intelligence.
2. The birth of artificial intelligence (1956)
The terminology “artificial intelligence” was coined by John McCarthy and used
by him in a proposal for organising a two-month workshop in Dartmouth College
in 1956. The year 1956 is generally considered as the year of birth of artificial
intelli- gence.

1
1.3. APPLICATIONS OF ARTIFICIAL INTELLIGENCE 2

3. The period of early successes (1952 - 1969)

There were several successes in the early period. The General Problem Solver (a
pro- gram that imitates human problem-solving protocols) by Allen Newell and
Herbert Simon, Geometry Theorem Prover by Herbert Gelernter (1959), Arthur
Samuel’s pro- grams for playing checkers (draughts), James Slagle’s SAINT
(Symbolic Automatic INTegrator) program (1963) to solve calculus integration
problems, Daniel Bobrow’s STUDENT program (1967) that solved algebra story
problems, etc. were some of the notable achievements of this period.

4. The period of knowledge-based systems (1969 - 1979)

This period saw the use of more powerful, domain-specific knowledge in develop-
ing programs that allow larger reasoning steps and can more easily handle typically
occurring cases in narrow areas of expertise.

5. The period when neural networks became important again(1986 - present)

Although neural networks had been in use in the early years of the development
of artificial intelligence, due to the failure of researchers to meet exaggerated
claims and expectations, funding for research in neural networks was halted till early
1980’s. But due to a combination of several factors, interest in neural networks re-
emerged in around 1986.

6. The period when artificial intelligence began to adopt the scientific method (1987
- present)
This was the period when artificial intelligence came firmly under the scientific
method. To be accepted, hypotheses must be subjected to rigorous empirical
experiments, and the results must be analysed statistically for their importance. It is
now possible to replicate experiments by using shared repositories of test data and
code.

7. The period when very large data sets began to be available (2001 - present)
Throughout the 60-year history of computer science, the emphasis has been on the
algorithm as the main subject of study. For many problems, it makes more sense to
worry about the data and be less concerned about what algorithm to apply. This is
true because of the increasing availability of very large data sources: for example,
trillions of words of English and billions of images from the Web; or billions of
base pairs of genomic sequences.

1.3 Applications of artificial intelligence

Nowadays, the concepts, methods and tools of artificial intelligence are finding
applications in more and more fields of human activity almost on a day by day basis (see
Figure 1.1). So it is difficult to give a concise answer to the question “What are the
applications of artificial intelligence?”.

1.3.1 Some well known applications of artificial intelligence

As a sample, a few of the well known applications of artificial intelligence are listed below.
1.3. APPLICATIONS OF ARTIFICIAL INTELLIGENCE 3

Figure 1.1: Applications of artificial intelligence

1. Robotic vehicles
A self-driving car, also known as an autonomous vehicle (AV), driverless car, or
robo- car is a vehicle that is capable of sensing its environment and moving safely
with little or no human input. Self-driving cars combine a variety of sensors to
perceive their surroundings. Advanced control systems interpret sensory
information to identify appropriate navigation paths, as well as obstacles and
relevant signage.
2. Speech recognition
Speech recognition is concerned with the recognition and translation of spoken lan-
guage into text by computers. It is also known as automatic speech recognition,
computer speech recognition or speech to text. Some speech recognition systems
require “training” where an individual speaker reads text or isolated vocabulary
into the system. The system analyses the person’s specific voice and uses it to fine-
tune the recognition of that person’s speech, resulting in increased accuracy.
3. Autonomous planning and scheduling
NASA’s Remote Agent program was the first on-board autonomous planning
program to control the scheduling of operations for a spacecraft. The Remote Agent
generated plans from high-level goals specified from the ground and monitored the
execution of those plans - detecting, diagnosing, and recovering from problems as
they occurred. Successor program MAPGEN planned the daily operations for
NASA’s Mars Explo- ration Rovers, and MEXAR2 did mission planning for the
European Space Agency’s Mars Express mission in 2008.
4. Game playing
IBM’s DEEP BLUE became the first computer program to defeat the world
champion in a chess match when it defeated Garry Kasparov in an exhibition match
in 1997.
5. Spam fighting
Each day, learning algorithms classify over a billion messages as spam, saving the
recipient from having to waste time deleting them.
6. Logistics planning
During the Persian Gulf crisis of 1991, U.S. forces deployed a Dynamic Analysis
and Replanning Tool (DART) to do automated logistics planning and scheduling
for
1.3. APPLICATIONS OF ARTIFICIAL INTELLIGENCE 4

transportation. This involved up to 50,000 vehicles, cargo, and people at a time,

and had to account for starting points, destinations, routes, and conflict resolution
among all parameters. The artificial intelligence planning techniques generated in
hours a plan that would have taken weeks with older methods.

7. Robotics
The iRobot Corporation has sold over two million Roomba robotic vacuum clean-
ers for home use. The company also deploys the more rugged PackBot to Iraq and
Afghanistan, where it is used to handle hazardous materials, clear explosives, and
identify the location of snipers.

8. Machine translation
There are computer programs that can automatically translate from, say, Arabic to
English. The program uses a statistical model built from examples of Arabic-to-
English translations and from examples of English text totalling two trillion words.

1.3.2 Recent applications

The following are some of the recent applications of artificial intelligence.

1. Agriculture
In agriculture new artificial intelligence advancements show improvements in
gaining yield and to increase the research and development of growing crops. New
artificial intelligence methods now predict the time it takes for a crop like a tomato
to be ripe and ready for picking thus increasing efficiency of farming.

2. Cybersecurity
The more advanced of these solutions use artificial intelligence and natural
language processing (NLP) methods to automatically sort the data in networks into
high risk and low-risk information. This enables security teams to focus on the
attacks that have the potential to do real harm to the organization, and not become
victims of attacks such as denial-of-service (DoS), malware and others.

3. Finance
Banks use artificial intelligence systems today to organize operations, maintain
book- keeping, invest in stocks, and manage properties. Artificial intelligence can
react to changes overnight or when business is not taking place.

4. Healthcare
The primary aim of health-related artificial intelligence applications is to analyse
rela- tionships between prevention or treatment techniques and patient outcomes.
Artificial intelligence programs are applied to practices such as diagnosis
processes, treatment protocol development, drug development, personalized
medicine, and patient moni- toring and care. Artificial intelligence algorithms can
also be used to analyse large amounts of data through electronic health records for
disease prevention and diagno- sis.
1.4. SAMPLE QUESTIONS 5

5 and many more

The tools of artificial intelligence are used in many, many more fields like virtual
assistant or chatbots, retail, sports analytics, manufacturing and production,
inventory management, etc.

1.4 Sample questions

(a) Short answer questions
1. Define artificial intelligence.

2. Why a particular year is considered as the year of birth of artificial intelligence?

3. What is “DEEP BLUE”?

4. Why artificial intelligence is now considered as a branch of science?

5. Briefly describe some recent application of artificial intelligence.

6. How artificial intelligence is used in healthcare?

7. List application areas of artificial intelligence.

(b) Long answer questions

1. Describe some of the important periods in the history of the development of
artificial intelligence methods.

2. Describe some of the well known applications of artificial intelligence (at least five
applications).
 Chapter 2

Problem definition and examples

In this chapter we consider the formal definition of the concept of a “problem” and illustrate
the concept with several well known examples. We also discuss different characteristics
of a problem. The concept of a “production system” is also introduced in this chapter.
We begin the chapter with a very brief discussion of a “rational agent” because the
chapter is concerned with the behaviour of a special kind of rational agent known as
“problem solving agent”.

2.1 Rational agents

2.1.1 Agents

Figure 2.1: Schematic diagram of an agent

An agent is anything that can be viewed as perceiving (becoming aware of) its envi-
ronment through sensors and acting upon that environment through actuators (or
effectors) (see Fig 2.1).

Examples
• A human agent has eyes, ears, and other organs for sensors and hands, legs, vocal
tract, and so on for actuators.

6
2.2. DEFINITION OF THE CONCEPT OF A PROBLEM IN AI 7

Figure 2.2: Schematic diagram of a human agent

• A robotic agent might have cameras and infrared range finders for sensors and
vari- ous motors for actuators.

• A software agent receives keystrokes, file contents, and network packets as

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

An agent’s behaviour is described by the agent function that maps any given input
sequence to an action. The agent function for an artificial agent is implemented by an
agent program.

2.1.2 Rational agents

A rational agent is defined as follows:

“A rational agent is an agent which, for each possible input sequence sensed
by the sensors, selects an action that maximises its expected performance
mea- sure.”

Throughout the rest of these notes, by an “agent” we shall always mean a “rational
agent”. Artificial intelligence is sometimes defined as a study of rational agents. A
rational agent could be anything which makes decisions such as a person, firm, machine,
or software.
This chapter describes the behaviour of a special kind of rational agent called a
problem- solving agent. Our discussion of problem solving begins with precise
definitions of prob- lems and their solutions and give several examples to illustrate these
definitions.

2.2 Definition of the concept of a problem in AI

In this section, we first consider an illustrative example to familiarise ourselves with the
terminology associated with the formal definition of a problem. This is followed by the
formal definition of a problem.

2.2.1 An illustrative example

We illustrate the the various components that constitute a problem by considering a well
known problem called the “3–puzzle”. The 3–puzzle is a special case of a more
general puzzle known as the “(n2 − 1)–puzzle”, namely, the case where n = 2. The
“8–puzzle”, the (n2 − 1)–puzzle with n = 3, is discussed in Section 2.3.2. Such
problems are also referred to as “sliding-tile puzzles” or “sliding-tile problems”.
2.2. DEFINITION OF THE CONCEPT OF A PROBLEM IN AI 8

Problem statement

Figure 2.3: 3-puzzle board

“The problem consists of a 2 × 2 board with three numbered tiles and a blank space
(see Figure 2.3). A tile adjacent to the blank space can slide into the space. The objective
is to reach the goal position shown in Figure 2.4 by sliding the tiles one at a time from
start position as in Figure 2.4.”

1 2 3 1
3 2

Start Goal

Figure 2.4: Start position and goal position in a 3-puzzle

States of the problem

The “Start” and “Goal” shown in Figure 2.4 represent different states of the problem. As
we move tiles to reach the goal from the start, the problem passes through several states.
In fact, there are 24 different possible states for the problem as shown in Figure 2.5.

1 2 1 2 1 3 1 3 1 1
3 3 2 2 2 3 3 2
s1 s2 s3 s4 s5 s6

2 1 2 1 2 3 2 3 2 2
3 3 1 1 1 3 3 1
s7 s8 s9 s10 s11 s12

3 1 3 1 3 2 3 2 3 3
2 2 1 1 1 2 2 1
s13 s14 s15 s16 s17 s18

1 1 2 2 3 3
2 3 3 2 1 3 3 1 1 2 2 1
s19 s20 s21 s22 s23 s24

Figure 2.5: The different states of the 3-puzzle

2.2. DEFINITION OF THE CONCEPT OF A PROBLEM IN AI 9

Actions
To move from one state to another state we apply some action. In this problem an action
can be thought of as moving the blank tile. Hence there are four possible actions that can
be applied to any state: “move blank tile up”, “move blank tile down”, “move blank tile
left” and “move blank tile right”. We may abbreviate these actions as “Up”, “Down”,
“Left” and “Right”. Note that not all actions are applicable to all states.
The set of actions applicable to a state s is denoted by ACTIONS(s). Thus, from
Figure 2.5, we can see that

ACTIONS(s1) = {Up, Left}

ACTIONS(s19) = {Right, Down}

Results of actions
The result of applying an action a to a state s is denoted by RESULT(s, a). For example,
we have

RESULT(s1, Up) = s6
RESULT(s19, Down) = s8

A successor of a state is a state reachable by a single action. For example, s8 is a

successor of s19. Also, s5 is a successor of s19.

State space
The initial state, the actions applicable to the initial states, the resulting states, the actions
applicable to the resulting states and the resulting successors, and so on, all together con-
stitute the state space for the problem. The state space can be represented in the form of a
tree1. Figure 2.6 shows part of the state space assuming that the start state is as shown in
Figure 2.4.

1 2
3
Left Up
1 2 1
3 3 2
Right Up Left Down
1 2 2 1 1 2
3 1 3 3 2 3
. . . .

Figure 2.6: A part of the state space for the 3-puzzle

1
For a precise definition of a tree, see Section 3.1.
2.2. DEFINITION OF THE CONCEPT OF A PROBLEM IN AI 10

Solution
A solution is an action sequence that leads from the initial state to a goal state. It can be
easily verified that the following sequence of actions is a solution to the problem that we
are talking about:
Up → Left → Down → Right
The resulting states are shown in Figure 2.7.

Start 1 2 Up 1 Left 1 Down 3 1 Right 3 1

2 Goal
3 3 2 3 2 2

Figure 2.7: Solution to the 3-puzzle

Optimal solutions
We may assign a cost for executing each action in a problem. The costs depend on the
problem. In the present problem, let us assign a cost of 1 unit for each action. Then, in the
solution specified in Figure 2.7, the the total cost of arriving at the goal state is 4 units. In
other words, the cost of the solution is 4 units. There may be solutions with lower costs.
The solution with the least cost is referred to as the optimal solution.
The reader may verify that the solution given in Figure 2.7 is indeed an optimal
solution for the problem under discussion.

2.2.2 Definition
A problem is defined formally by the following five components:

1. Initial state
The initial state that the agent starts in.

2. Actions
A description of the possible actions available to the agent. Given a particular state
s, ACTIONS(s) returns the set of actions that can be executed in s. We say that
each of these actions are applicable in s.

3. Transition model
A description of what each action does. The formal name for this is the transition
model, specified by a function RESULT(s, a) that returns the state that results from
doing action a in state s. We also use the term successor to refer to any state
reachable from a given state by a single action.
The initial state, the actions, and the transition model all taken together implicitly
define the state space of the problem – the set of all states reachable from the
initial state by any sequence of actions. The state space forms a directed network or
graph in which the nodes are states and the links between nodes are actions. A
path in the state space is a sequence of states connected by a sequence of actions.
2.3. EXAMPLE PROBLEMS 11

4. Goal test
The goal test, which determines whether a given state is a goal state. Sometimes
there is an explicit set of possible goal states, and the test simply checks whether
the given state is one of them.

5. Path cost
A path cost function that assigns a numeric cost to each path. The problem-solving
agent chooses a cost function that reflects its own performance measure. We
assume that the cost of a path can be described as the sum of the costs of the
individual actions along the path.

A solution to a problem is an action sequence that leads from the initial state to a goal
state. The optimal solution is the solution having the lowest path cost among all solutions.

2.3 Example problems

2.3.1 The vacuum cleaner world
Description
This 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.

Figure 2.8: The Vacuum cleaner world

Standard formulation
• States: The state is determined by both the agent location and the dirt locations.
The agent is in one of two locations, each of which might or might not contain dirt.
Thus, there are 2 × 22 = 8 possible world states.

• Initial state: Any state can be designated as the initial state.

• Actions: In this simple environment, each state has just three actions: Left, Right,
and Suck.

• Transition model: The actions have their expected effects, except that moving
Left in the leftmost square, moving Right in the rightmost square, and Sucking in a
clean square have no effect. The complete state space is shown in Figure 2.9.

• Goal test: This checks whether all the squares are clean.
 2.3. EXAMPLE PROBLEMS 12

• Path cost: Each step costs 1, so the path cost is the number of steps in the path.

Figure 2.9: The state space for the vacuum world. Links denote actions: L = Left, R =
Right, S = Suck.

2.3.2 8-puzzle problem

Recall the detailed discussion of the 3-puzzle in Section 2.2.1.

Description
The 8-puzzle consists of a 3 × 3 board with eight numbered tiles and a blank space. A tile
adjacent to the blank space can slide into the space. The objective is to reach a specified
goal state such as the one shown in Figure 2.10 by sliding the tiles one at a time, given
some initial or start state as in Figure 2.10

Figure 2.10: 8-puzzle

Standard formulation
• States: A state description specifies the location of each of the eight tiles and the
blank in one of the nine squares. (The total number of states is 9! = 362880.)

• Initial state: Any state can be designated as the initial state.

• Actions: Actions may be defined as movements of the blank space Left, Right, Up,
or Down. Different subsets of these are possible depending on where the blank is.
 2.3. EXAMPLE PROBLEMS 13

• Transition model: Given a state and action, this returns the resulting state; for ex-
ample, if we apply Left to the start state in Figure 2.10, the resulting state has the 5
and the blank cell switched.

• Goal test: This checks whether the state matches the goal configuration shown in
Figure 2.10.

• Path cost: Each step costs 1, so the path cost is the number of steps in the path.

2.3.3 Missionaries and cannibals problem

This problem is famous in AI because it was the subject of the first paper that approached
problem formulation from an analytical viewpoint.

Description
Three missionaries and three cannibals2 are on one side of a river, along with a boat that can
hold one or two people. Find a way to get everyone to the other side without ever leaving
a group of missionaries in one place outnumbered by the cannibals in that place. (See
Figure 2.11.)

Figure 2.11: The missionaries and cannibals problem

Standard formulation
Ignore all irrelevant parts of the problem (e.g., weather conditions, possible presence of
crocodiles in the river, etc.).

• States: Each state is represented by an ordered sequence of 3 numbers (x, y, z) where

x : number of missionaries on initial river bank
y : number of cannibals on initial river bank
z : number of boats on initial river bank
For example, if triangles represent missionaries, circles represent cannibals and the
left bank is the initial bank then the state shown in Figure 2.12 is specified by the
ordered triple (0, 1, 0). Note that the fact that the boat is at the destination bank is
indicated by z = 0. Also it may be noted that not all such ordered triples represent
2
A cannibal is a person who eats the flesh of other humans.
2.3. EXAMPLE PROBLEMS 14

legal states. For example, the triple (2, 3, 1) is not allowed because it represents a
state where the number of cannibals outnumber the missionaries.

Figure 2.12: The state (0, 1, 0) of the missionaries and cannibals problem

• Initial state: The ordered sequence = (3, 3, 1) (see Figure 2.11).

• Actions: Take two missionaries, two cannibals, or one of each across in the boat.
We have to take care to avoid illegal states.

• Transition model: Given a state and action, this returns the resulting state; for
exam- ple, if we apply the action ”Take 1 cannibal” to the state (0, 2, 0) the
resulting state is (0, 3, 1).

• Goal test: We have reached state (0, 0, 0).

• Path cost: Number of crossings.

Solution
Figure 2.13 shows the possible legal states reachable from the initial state applying only
legal actions. In the figure, the notation “−2c ” indicates the operation of taking 2 cannibals
from the bank where the boat is currently→ located to the other bank and the notation “ −2m

” indicates taking two missionaries. From the figure we can easily get the solution to the→
problem as follows:
2c 1c
2c 1c 1m 1m, 1c
(3, 3, 1) −→ (3, 1, 0) −→ (3, 2, 1) −→ (3, 0, 0) −→ (3, 1, 1) −→ (1, 1, 0) −−−→
2m 1c 2c 1c 2c
(2, 2, 1) −→ (0, 2, 0) −→ (0, 3, 1) −→ (0, 1, 0) −→ (0, 2, 1) −→ (0, 0, 0)

Figure 2.13: The state space of the missionaries and cannibals problem (with triangles
representing missionaries and circles representing cannibals
2.3. EXAMPLE PROBLEMS 15

2.3.4 Cryptarithmetic puzzle

This is an example for a special type of problem known as “constraint satisfaction
problem”. Constraint satisfaction problems are extensively studied and are the subject of
intense re- search in artificial intelligence.

Description
A cryptarithmetic puzzle is a mathematical exercise where the digits of some numbers are
represented by letters. Each letter represents a unique digit. It will be assumed that the
leading digits of numbers are not zero. The problem is to find the digits such that a given
mathematical equation is satisfied.

Illustrative example
Find the digits to be assigned to the letters such that the following equation is true (the
digits assigned to T and F should not be 0):

T W O +
T W O

F O U R

Solution
With some trial and error we can get the following

solution: F = 1, O = 4, R = 8, T = 7, U = 6, W = 3.

That this is indeed a solution can be verified easily as follows.

7 3 4 +
7 3 4

1 6 8
4

Here is another solution.

8 2 6 +
8 2 6

1 5 2
6

There may still be other solutions.

2.3. EXAMPLE PROBLEMS 16

Problem formulation
Using the idea of states the problem can be formulated as follows:

• States: Let there be n different letters where n ≤ 10. A problem state is an ordered
n-tuple of digits (d1, d2, . . . , dn) representing the digits to be assigned to the integers.

• Initial state: The initial state can be considered as the ordered n-tuple all of whose
elements are 0’s.

• Actions: Increase the value assigned to a letter by 1.

• Transition model: Given a state and action, this returns the resulting state.

• Goal test: We have reached state (d1, d2, . . . , dn) which satisfies the constraints
of the problem.

• Path cost: Number of actions.

Solution
There are algorithms for solving the cryptarithmetic puzzle. Implementations of these al-
gorithms by hand computations is difficult. The method of trial and error keeping in mind
the constraints to be satisfied may be applied to obtain solutions. We illustrate the
method by solving some simple cryptarithmetic problems.

Example problem 1
Solve the following cryptarithmetic puzzle:

T O
+ G O

O U T

Solution by hand computation

• Since leading digits are not 0, we have T /= 0, G /= 0, O /= 0.

• Since O is a non-zero carry digit when the two digits T an G are added, we must
have O = 1.

• Since O = 1 and O + O = T, we must have T = 2.

• Since T = 2 and T +G ≥ 10, T + G must be either 10 or 11.

• If T + G = 10, then G = 8 and U = 0.

• This gives a solution to the problem, namely,

 2.3. EXAMPLE PROBLEMS 17

T = 2, O = 1, G = 8, U = 0.

2 1
+ 8 1

1 0 2

• If T + G = 11, then U = 1 and in that case U = O = 1 which is not acceptable.

• Thus the problem has a unique solution as given above.

Example problem 2
Solve the following cryptarithmetic puzzle:
S E N D
+ M O R E

M O N E Y

Solution
1. We number the columns as follows:

Col:5 4 3 21

S E ND
+ M O RE

M O N EY

2. From column 5, we must have M = 1 since it is the only carry possible from the
sum of two single digit numbers.

Col:5 4 3 21

S E ND
+ 1 O RE

1 O N EY

3. To produce a carry from column 4 to column 5, S + 1 must be at least 9. We

have S + 1 = 9 (if there is a carry over from column 3) or 10 (if there is no carry
from column 3). So, we must have S = 8 or 9. Then O = 0 or 1. But we have
already seen that M = 1. The same value cannot be assigned to O. Therefore we
must have O = 0.
2.3. EXAMPLE PROBLEMS 18

Col:5 4 3 21

S E ND
+ 1 0 RE

1 0 N EY

4. If there is carry from column 3 to 4 then E = 9 and so N = 0. But O = 0 so there

is no carry and S = 9 and the carry-over from column 3 is 0.

Col:5 4 3 21

9 E ND
+ 1 0 RE

1 0 N EY

5. If there is no carry from column 2 to 3 then E = N which is impossible,

therefore there is carry and
N =E+1
and the carry from column 2 to 3 is 1.

6. (a) If there is no carry from column 1 to 2 then

N + R = 10 + E

and since N = E + 1, we have

E + 1 + R = 10 + E.

Therefore R = 9. But S = 9 and so R cannot be 9.

(b) So there must be carry from column 1 to 2 and hence R = 8.

Col:5 4 3 21

9 E ND
+ 1 0 8E

1 0 N EY

7. (a) To produce a carry of 1 from column 1 to 2, we must have

D + E = 10 + Y.

(b) Since 0 and 1 has already been assigned, Y cannot be 0 or 1. Hence D + E

is at least 12.
 2.3. EXAMPLE PROBLEMS 19

(c) As 8 and 9 have already been assigned, D is at most 7 and so E is at least 5.

(d) Also N is at most 7 and N = E + 1 so we must have E = 5 or 6.

8. If E were 6 and D + E at least 12 then D would be 7, but N = E + 1 and N

would also be 7 which is impossible. Therefore E = 5 and N = 6.

Col:5 4 3 21

9 5 6D
+ 1 0 85

1 0 6 5Y

9. D + E is at least 12 for that we must have D = 7 and Y = 2.

Col:5 4 3 21

9 5 67
+ 1 0 85

1 0 6 52

10. The problem has unique solution, namely,

D = 7, E = 5, M = 1, N = 6, O = 0, R = 8, S = 9, Y = 2.

2.3.5 Blocks world problem

Blocks world (or, the world of blocks) is a model domain used in artificial intelligence to
explore different approaches to automated reasoning. This model is used to illustrate that
a given algorithm can perform planning, or that it is efficient in terms of the number of
calculations required to find a solution or in terms of the length of that solution.

Description
There is a table on which some uniform blocks (cubes) are placed. Some blocks may or may
not be stacked on other blocks. We have a robot arm to pick up or put down the blocks.
The robot arm can move only one block at a time, and no other block should be stacked
on top of the block which is to be moved by the robot arm. The problem is to change the
configuration of the blocks from a given initial state to a given goal state (see, for
example, Figure 2.14).

Problem formulation
The problem can be formulated as follows:
 2.3. EXAMPLE PROBLEMS 20

Figure 2.14: A blocks world problem

• States: Configurations of the blocks satisfying the conditions. The configurations

can be specified using the following predicates where A and B denote arbitrary
blocks:
ON(A,B) : block A is on block B.
ONTABLE(A) : block A is on the table.
CLEAR(A) : block A has nothing on
it. HOLDING(A): the arm holds block A.
ARMEMPTY : the arm holds nothing.
• Initial state: The initial configuration of blocks.

• Actions:
UNSTACK(A,B) : pick up clear block A from block B
and hold it in the arm
STACK(A,B) : place block A held in the arm onto clear
block B
PICKUP(A) : lift clear block A with the empty arm
PUTDOWN(A) : place block A held in the arm onto a
free space on the table.
• Goal state: A certain configuration of the blocks.

• Path cost: Number of actions taken to reach the goal state.

Solution
The blocks world problem has always a trivial solution: All blocks not already correctly
positioned for the goal state be set off onto the table (one at a time with the mechanical arm),
and then reassembled in the proper order on top of any blocks already correctly positioned.
To illustrate the idea consider the blocks world problem specified in Figure 2.14. A
solution consists of the following 12 moves in that order:
2.3. EXAMPLE PROBLEMS 21

1. UNSTACK(2, 3)
2. PUTDOWN(2)
3. UNSTACK(4, 5)
4. PUTDOWN(4)
5. UNSTACK(5, 6)
6. STACK(5, 4)
7. PICKUP(2)
8. STACK(2, 5)
9. PICKUP(3)
10. STACK(3, 1)
11. PICKUP(6)
12. STACK(6, 3)

2.3.6 Monkey and banana problem

The monkey and banana problem is a famous toy problem in artificial intelligence,
particu- larly in logic programming and planning.

Description
A monkey is in a room. A bunch of bananas is hanging from the ceiling. The monkey
cannot reach the bananas directly. There is box in the room. The ceiling is just the right
height so that a monkey standing on the box can get hold of the bananas. The monkey
knows how to move around and carry things around. How can the monkey get the
bananas? What is the best sequence of actions for the monkey? (See Figure 2.15.)

Figure 2.15: Monkey and banana

Remark
The purpose of the problem is to raise the question: Are monkeys intelligent? Both
humans and monkeys have the ability to use mental maps to remember things like where
to go to find shelter, or how to avoid danger. They can also remember where to go to
gather food and water, as well as how to communicate with each other. Monkeys have the
ability not only to remember how to hunt and gather but to learn new things, as is the
case with the monkey and the bananas: despite the fact that the monkey may never
have been in an identical
2.3. EXAMPLE PROBLEMS 22

situation, with the same artifacts at hand, a monkey is capable of concluding that it needs
to make a ladder, position it below the bananas, and climb up to reach for them.

Problem formulation
Let the initial location of the monkey be A, that of the bananas be B and that of the box
be C. Initially, the monley and box have height “low”; but when the monkey climbs on
the box, it will have height “high”.

• States: Each state is represented by the location and height of the monkey. bananas
and the box.

• Initial state:
Location(monkey, A), Location(bananas, B), Location(box, C),
Position(monkey, low), Position(bananas, high), Position(box, low)

• Actions: Let x and y be locations. The following is a simplified description of the

available actions. Clearly, to apply these actions certain preconditions are to be
met.
GO(x, y): Go from location x to y.
PUSH(object, x, y): Push object from location x to y.
CLIMBUP(object, x): Climb up on the object at location x.
CLIMBDOWN(object, x): Climb down from the object at location x.
GRASP(object, x, height): Grasp object located at x in position height)

• Transition model: Given a state and action, this returns the resulting state.

• Goal test: The goal state can be specified as follows:

Location(monkey, A), Location(bananas, B), Location(box, C),
Position(monkey, low), Position(bananas, low), Position(box, low)

• Path cost: Number of actions.

Solution
Step 1. GO(A,C)

Step 2. PUSH(box, C, B)

Step 3. CLIMBUP(box, B)

Step 4. GRASP(bananas, B, high)

Step 5. CLIMBDOWN(box, B)

Step 6. GO(B, A)
2.3. EXAMPLE PROBLEMS 23

Figure 2.16: A road map of Romania

2.3.7 A map problem

Description
Given a road map of Romania (see Figure 2.16), find the shortest route from Arad to
Bucharest. (In the figure, the numbers written by the sides of the line segments are the
distances between the cities the line segment connects.)

Problem formulation
Using the idea of states the problem can be formulated as follows:
• States: Each city in the map represents a state.
• Initial state: The initial state is the city of Arad.
• Actions: Travel to a connected nearby city.
• Transition model: Given a state and action, this returns the resulting state.
• Goal test: The city of Bucharest is reached.
• Path cost: The total distance traveled in reaching the goal state.

2.3.8 Water jug problem

Problem statement
We have two jugs of capacity 4 liters and 3 liters, and a tap with an endless supply of
water. The objective is to obtain 2 liters of water exactly in the 4-liter jug with the
minimum steps possible.

Problem formulation
• States: Let x denote the number of liters of water in the 4-liter jug and y the
number of liters of water in the 3-liter jug. Now x = 0, 1, 2, 3, or 4 and y = 0, 1,
2, or 3. The ordered pair (x, y) represents a state.
2.3. EXAMPLE PROBLEMS 24

Figure 2.17: The water jug problem

• Initial state: The ordered pair (0, 0).

• Actions: Each action is represented in the form “(x, y) → (u, v)” where (x, y)
represents the state before the application of the action and (u, v) represents the
state after the application of the action. The possible actions are given in Table 2.1.
The table also specifies the conditions under which the various actions can be
applied.

Sl No. State before State after Description of op-

action action eration
1 (x, y) → (4, y) Fill 4-liter jug
if x < 4
2 (x, y) → (x, 3) Fill 3-liter jug
if y < 3
3 (x, y) → (0, y) Empty 4-liter jug
if x > 0 on the ground
4 (x, y) → (x, 0) Empty 3-liter jug
if x > 0 on the ground
5 (x, y) → (4, y − (4 − Pour water from 3-
if x + y ≥ 4 x)) liter jug into 4-liter
and y > 0 jug until 4-liter jug
is full
6 (x, y) → (x − (3 − Pour water from 4-
if x + y ≥ 3 y), 3) liter jug into 3-liter
and x > 0 jug until 3-liter jug
is full
7 (x, y) → (x + y, 0) Pour all water
if x + y ≤ 4 from 3-liter jug
and y > 0 into 4- liter jug
8 (x, y) → (0, x + y) Pour all water
if x + y ≤ 3 from 4-liter jug
and x > 0 into 3- liter jug

Table 2.1: Actions in the water jug problem

2.4. PROBLEM CHARACTERISTICS 25

• Transition model: The production system given in Table 2.1 also specifies the
tran- sition model.

• Goal test: The state (2, n) where n is some integer.

• Path cost: Number of operations performed.

Solution
One solution to the water jug problem is given below:
Rule 2
Rule Rule Rule Rule Rule 7
7 2 5 3
(0, 0) −−−→ (0, 3) −−−→ (3, 0) −−−→ (3, 3) −−−→ (4, 2) −−−→ (0, 2) −−−→ (2, 0).

2.4 Problem characteristics

2.4.1 Characteristics
The following are the important characteristics of a problem.

1. Problem is decomposable to smaller or easier problems.

2. Solution steps can be ignored or undone.

3. The problem universe is predictable.

4. There are obvious good solutions without comparisons to all other possible solutions.

5. Desired solution is a state of the universe or a path to a state.

6. Requires lots of knowledge; or, uses knowledge to constrain solutions.

7. Problem requires periodic interaction between humans and computer.

2.5 Examples illustrating problem characteristics

2.5.1 Problem is decomposable to smaller or easier problems
Example 1
Consider the problem: Evaluate the integral
∫ 2
(x + sin2 x) dx.

We show that this problem is decomposable to smaller subproblems (see Figure 2.18).
Since ∫ 2 ∫
(x + sin2 x) dx = ∫x2 dx + sin2 xdx,
the problem can be decomposed into two simpler subproblems.
∫
• Problem (i): Evaluate x2 dx.
∫
• Problem (ii): Evaluate sin2 xdx.
 2.5. EXAMPLES ILLUSTRATING PROBLEM CHARACTERISTICS 26
∫
(x2 + sin2 x)
dx
∫
sin2 xdx
∫
x2 dx
∫1
2 (1 − cos 2x) dx
1 3
3x

∫1 ∫ 1
2 dx —2 cos 2xdx

1x
2
—41 sin 2x

Figure 2.18: Decomposing a problem into smaller subproblems

The first subproblem can be easily solved to get

∫
x2 dx = 31 x3.

Since
∫ ∫
sin2 xdx = 12 (1 − cos 2x)
∫ ∫
dx = 21 dx − cos
2xdx,
Problem (ii) can be divided into two subproblems:
∫
• Problem (iii): Evaluate 12 dx.
∫
• Problem (iv): Evaluate − 12 cos 2xdx.
These subproblems can be easily solved as
∫ 1
2
dx =2 1
∫ 1
1
2 cosx2xdx = − 2 sin 2x

Combining the solutions of the subproblems we get a solution of the given problem.

Example 2
Consider a simple blocks world problem specified by Figure 2.19.

Figure 2.19: A simple blocks world problem

The following are the solution steps :
UNSTACK(C,A)
2.5. EXAMPLES ILLUSTRATING PROBLEM CHARACTERISTICS 27

PUTDOWN(C)
PICKUP(B)
STACK(B,A)
STACK(C,B)
Since, the steps are interdependent the problem cannot be decomposed into two
independent problems.

2.5.2 Solution steps can be ignored or undone.

Example 1
Suppose our goal is to prove a theorem in mathematics. On the way, we prove some
prelim- inary results, say Result 1, Result 2, Result 3, and then finally we prove the
theorem. The steps in the proof look like this:
Result 1
Result 2
Result 3
Theorem
Suppose, we later realize that Result 2 is not actually needed in proving the theorem. Then,
we may ignore Result 2 and present the steps of the proof as follows:
Result 1
Result 3
Theorem
In this example, the solution steps can be ignored.

Example 2
Consider the 8-puzzle. The solution involves a sequence of moves. In the process of
finding a solution, after some moves, we realize that a certain previous move has be
reversed. The previous move can be undone by backtracking the moves. In this example,
the solution steps can be ignored.

Example 3
Consider the problem of playing chess. If a chess playing programme makes a wrong
move, the rules of the game do not permit undoing the move. In this example, the
solution steps cannot be undone or ignored.

2.5.3 The problem universe is predictable.

Example 1
In the 8-puzzle, every time we make a move we know exactly what will happen. This
means that it is possible to plan an entire sequence of moves. Thus in this problem, the
universe is predictable.
2.5. EXAMPLES ILLUSTRATING PROBLEM CHARACTERISTICS 28

Example 2
In a game like bridge, this is not possible because a player does not know where the
various cards are and what the other players will do on their turns. In this problem, the
universe is unpredictable.

2.5.4 There are obvious good solutions without comparison to all other pos-
sible solutions.
Example 1
Consider a mathematical problem. In general, there may be many methods for solving the
problem. Any method is a good method without comparison to other methods provided it
solves the problem. In general, any “any-path problem” is an example of a problem
having obvious good solutions without comparison to other possible solutions.

Example 2
A “best-path problem” is a problem having no obvious good solutions. The travelling
sales- man problem is an example for a best-path problem. The travelling salesman
problem can be formulated as follows: “Given a list of cities and the distances between
each pair of cities, what is the shortest possible route that visits each city exactly once
and returns to the origin city?”

2.5.5 Desired solution is a state of the universe or a path to a state.

Example 1
In the missionaries and cannibals problem, if we organise the various states in the form of
a tree, it can be seen that the solution to the problem is a path connecting the various
states (see Figure 2.13).

Example 2
In the cryptarithmetic puzzle, the solution to the problem is a state of the problem,
namely, the state representing the assignment of digits to the letters in the puzzle.

2.5.6 Requires lots of knowledge; or, uses knowledge to constrain solutions.

Example 1
Consider the problem of playing chess. The amount of knowledge required to play chess
is very little: just the rules of the game! Additional knowledge about strategies may be
used to make intelligent moves!!

Example 2
Consider the problem of scanning daily newspapers to decide which are supporting and
which are opposing a political party in an upcoming election. It is obvious that a great
deal of knowledge is required to solve this problem.
2.6. PRODUCTION SYSTEMS 29

2.5.7 Problem requires periodic interaction between human and computer.

Example
Even in a so called “fully automated system” situations constantly arise that call for
human intervention. When the machines get thrown off track, or become faulty, experts
have to be summoned to step in and troubleshoot the problems.

2.6 Production systems

The term “production system” refers to many things. It may refer to a computer program
which is used to provide a solution for a problem using a set of rules. It may also refer to
a programming language for writing such programs. Further it can also be thought of as a
model of computation that can be applied to implement search algorithms or as a model
of human problem solving.
However, here, we define a production system as a structure having certain well
defined components as described below.

2.6.1 Definition
A production system (also called production rule system) consists of the following:

1. Set of rules
This is the most important component of a production system. Briefly, a production
rule is a rule of the form “C → A” which may be interpreted as “given condition C,
take action A”. Thus, each rule in the set of rules consists of a left side (a pattern)
that determines the applicability of the rule and a right side that describes the
operation to be performed if the rule is applied.

2. Knowledge databases
The databases contain whatever information is appropriate for the particular task.
Some parts of the database may be permanent.
The database includes a working memory whose elements may pertain only to the
solution of the current problem. Working memory contains a description of the
cur- rent state of the world in the problem-solving process. The description is
matched against the conditions of the production rules. When a condition matches,
its action is performed. Actions are designed to alter the contents of working
memory.

3. Control strategy
The control strategy includes a specification of the order in which the rules are to
be applied and a conflict resolution strategy.

(a) Order of application of rules

There are two methods of execution for a rule system:
• Forward chaining
In forward chaining systems, the execution of rules starts from facts in
databases and moves towards some goal or conclusion (see Figure 2.20).
2.6. PRODUCTION SYSTEMS 30

Figure 2.20: Recognise-act (or, select-execute) cycle: Forward chaining model

• Backward chaining
In backward chaining, the execution of rules starts from the goals and
con- clusions and works backwards to see if there is any data from which
the goals and conclusions can be derived (see Figure 2.21).
(b) Conflict resolution strategy
A system with a large number of rules and facts may result in many rules
being true for the same fact; these rules are said to be in conflict. The conflict
res- olution strategy is a way of resolving the conflicts that manages the
execution order of these conflicting rules. Some of the commonly used
conflict resolution strategies are listed below:
• Choose arbitrarily.
• Choose the lowest numbered rule.
• Choose the rule containing the largest number of conditions.
• Choose the least recently used rule.
• Choose a rule where the condition is a new fact.
• Choose the rule with the highest priority (weight).

4. Interpreter
The interpreter repeats the following operations: all rules whose conditions are sat-
isfied are found (rule matching), one of them is selected (conflict resolution), and
its action is called (rule firing).
The interpreter (also referred to as rule applier) can be viewed as a “select-execute
loop”, in which one rule applicable to the current state of the data base is chosen,
and then executed. Its action results in a modified data base, and the select phase
begins again. This cycle is also referred to as a “recognize-act cycle”.
2.6. PRODUCTION SYSTEMS 31

Figure 2.21: Backward chaining model

2.6.2 Production system example: 3 × 3 knight’s tour

A knight’s tour is a sequence of moves of a knight on a chessboard such that the knight
visits every square exactly once. In the 3 × 3 knight’s tour problem, instead of the usual
8 × 8 board in chess, we consider a 3 × 3 board as in Figure 2.22.

Problem statement
Using the production system model, determine whether a path exists from square 1 to
square 2 in the 3 × 3 knight’s tour problem (see Figure 2.22).

1 2 3
4 5 6
7 8 9

Figure 2.22: A 3 × 3 chess board

2.6. PRODUCTION SYSTEMS 32

Solution
Step 1. Problem formulation as a production system
(i) Knowledge database
Since there is no information other than that contained in the statement of the
problem, the database contains only the working memory. The working memory
contains the following two facts:

{knight on square 1, knight on square 2}

We denote this initial working memory by the set {1, 2}.

(ii) Production rules

The set of production rules is given in Table 2.2.

Rule No. Condition Action

1 knight on square 1 → knight on square 8
2 knight on square 1 → knight on square 6
3 knight on square 2 → knight on square 9
4 knight on square 2 → knight on square 7
5 knight on square 3 → knight on square 4
6 knight on square 3 → knight on square 8
7 knight on square 4 → knight on square 9
8 knight on square 4 → knight on square 3
9 knight on square 6 → knight on square 1
10 knight on square 6 → knight on square 7
11 knight on square 7 → knight on square 2
12 knight on square 7 → knight on square 6
13 knight on square 8 → knight on square 3
14 knight on square 8 → knight on square 1
15 knight on square 9 → knight on square 2
16 knight on square 9 → knight on square 4

Table 2.2: The set of production rules for the 3 × 3 knight’s tour problem
.

(iii) Conflict resolution strategy

Select and fire the first rule in the conflict set that does not lead to a repeated state.

(iv) Interpreter
Interpreter is the select-execute loop.
2.6. PRODUCTION SYSTEMS 33

Step 2. Application of the rules

Table 2.3 shows the sequence of the applications of the rules starting from the state repre-
sented by “knight on square 1” in the working memory.

Step no. Working Current Conflict Fire Remarks

memory square set rule
0 {1, 2} 1 – – Initial state
1 {1, 2} 1 1, 2 1 –
2 {8, 2} 8 13, 14 13 –
3 {3, 2} 3 5, 6 5 –
4 {4, 2} 4 7, 8 7 –
5 {9, 2} 9 15, 16 15 –
6 {2, 2} 2 – – Halt

Table 2.3: Sequence of the applications of the production rules in the 3 × 3 knight’s tour
problem

Step 3. Conclusion
There exists a path from the initial state “knight on square 1” to the goal state “knight on
square 2”.

2.6.3 Production system example: The water jug problem

We have already seen the formulation of the water jug problem in the standard format of
a problem in artificial intelligence (see Section 2.3.8). We now formulate the same
problem in the format of a production system and then provide a solution to the problem
using the solution method of production systems.

Problem statement
We have two jugs of capacity 4 liters and 3 liters, and a tap with an endless supply of
water. The objective is to obtain 2 liters of water exactly in the 4-liter jug with the
minimum steps possible.

Solution
Step 1. Problem formulation as a production system
As in section 2.3.8, we represent a state of the problem by an ordered pair (x, y) of
integers where x denotes the number of liters of water in the 4-liter jug and y the number
of liters of water in the 3-liter jug.

1. Knowledge database
Since there is no information other than that contained in the statement of the prob-
lem, the database contains only the working memory. The working memory
contains
2.6. PRODUCTION SYSTEMS 34

the following two facts:

{(0, 0), (2, 0)}
where (0, 0) is the initial state and (2, 0) is the goal state.

2. Production rules
The set of production rules is given in Table 2.4. This is identical with Table 2.1
except for the differences in the column titles.

Sl No. Condition Action Description

1 (x, y) → (4, y) Fill 4-liter jug
if x < 4
2 (x, y) → (x, 3) Fill 3-liter jug
if y < 3
3 (x, y) → (0, y) Empty 4-liter jug
if x > 0 on the ground
4 (x, y) → (x, 0) Empty 3-liter jug
if x > 0 on the ground
5 (x, y) → (4, y − (4 − Pour water from 3-
if x + y ≥ 4 x)) liter jug into 4-liter
and y > 0 jug until 4-liter jug
is full
6 (x, y) → (x − (3 − Pour water from 4-
if x + y ≥ 3 y), 3) liter jug into 3-liter
and x > 0 jug until 3-liter jug
is full
7 (x, y) → (x + y, 0) Pour all water from
if x + y ≤ 4 3-liter jug into 4-
and y > 0 liter jug
8 (x, y) → (0, x + y) Pour all water from
if x + y ≤ 3 4-liter jug into 3-
and x > 0 liter jug

Table 2.4: Production rules of the water jug problem

3. Conflict resolution strategy

Select and fire the first rule in the conflict set that does not lead to a repeated state.

4. Interpreter
Interpreter is the select-execute loop.

Step 2. Application of the rules

The various steps in the solution procedure obtained by using the forward chaining
method is shown in Table 2.5.
2.6. PRODUCTION SYSTEMS 35

Step Working mem- Current Conflict set Fire Remarks

No. ory state rule
0 {(0, 0), (2, 0)} (0, 0) – – Initial state
1 {(0, 0), (2, 0)} (0, 0) 1, 2 1 –
2 {(4, 0), (2, 0)} (4, 0) 2, 3, 4, 6 2 –
3 {(4, 3), (2, 0)} (4, 3) 3, 4, 5, 6 3 –
4 {(0, 3), (2, 0)} (0, 3) 1,7 7 –
5 {(3, 0), (2, 0)} (3, 0) 1, 2, 3, 4, 6, 8 2 –
6 {(3, 3), (2, 0)} (3, 3) 1, 3, 4, 5, 6 5 –
7 {(4, 2), (2, 0)} (4, 2) 2, 3, 4, 5, 6 3 –
8 {(0, 2), (2, 0)} (0, 2) 1, 2, 7 7 –
9 {(2, 0), (2, 0)} (2, 0) – – Goal state

Table 2.5: Sequence of the applications of the production rules to reach the goal state in
the water jug problem

Let us examine the working of the conflict resolution strategy in this example.
Consider Step No. 6 in Table 2.5. The current state is (3, 3). If we examine the rules, it
can be seen that this state satisfies the conditions of the Rules 1, 3, 4, 5, and 6 and hence
the conflict set is {Rule 1, Rule 3, Rule 4, Rule 5, Rule 6}. The conflict resolution
strategy we have adopted is “select and fire the first rule that does not lead to a repeated
state”. If we select Rule 1 and fire it, the resulting state is (4, 3) and we have obtained this
state in Step No. 3. Similarly if we select Rule 3 and fire it the resulting state is (0, 3)
which we have already obtained in Step No. 4. Proceeding like this, we see that that the
first rule in the conflict set that does not lead to a repeated state is Rule 5 and we select it
and fire it to obtain the non-repeated state (4, 2).

3. Conclusion
The operations specified in Table 2.5 specify a method of getting 2 liters of water in the
4-liter jug. However, this need not necessarily be the optimum solution in the sense that
there may be a solution with less number of operations.

2.6.4 Production system

characteristics (Classes of
production systems)
Definitions
1. Monotonic and nonmonotonic production systems
A monotonic production system is a production system in which the application 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 production system which is not
monotonic is called a nonmonotonic production system.

2. Partially commutative production systems

2.6. PRODUCTION SYSTEMS 36

A partially commutative production systems is a production system with the

property that if the application of a particular sequence of rules transforms state x to
state y then any allowable permutation of those rules also transforms state x into
state y.

3. Commutative production systems

A production system that is both monotonic and partially commutative is called a
commutative production system.

Remarks
1. It can be shown that, depending on how the operators are chosen, the 8-puzzle and
the blocks world problem are partially commutative systems.

2. Production systems that are not partially commutative are useful in which
irreversible changes occur. For example, the process to produce a desired chemical
compound may involve may irreversible steps.

3. Commutative production systems are useful for solving ignorable problems like the
problem of proving a theorem in mathematics.

4. Nonmonotonic partially commutative production systems are useful for problems

in which the order of operations is not important. For example, robot navigation is
such a problem.

Table 2.6 summarises the characteristics of some of the well known production systems.

Monotonic Nonmonotonic
Partially commutative Theorem proving Robot navigation
Not partially commutative Chemical synthesis Bridge

Table 2.6: The four categories of production systems

2.6.5 Advantages and disadvantages of production systems

Advantages
Some of the advantages of production system in artificial intelligence are the following.
1. Provides excellent tools for structuring AI programs.

2. The system is highly modular because individual rules can be added, removed or
modified independently.

3. There is separation of knowledge and control.

4. The system uses pattern directed control which is more flexible than algorithmic
con- trol.

5. Provides opportunities for heuristic control of the search.

6. Quite helpful in a real-time environment and applications.

 2.7. SAMPLE QUESTIONS 37

Disadvantages
The following are some of the disadvantages.
1. There is a lot of inefficiency in production systems. For example, there may be
situations where multiple rules get activated during execution and each of these
rules may trigger exhaustive searches.
2. It is very difficult to analyze the flow of control within a production system.
3. There is an absence of learning due to a rule-based production system that does not
store the result of the problem for future use.
4. The rules in the production system should not have any type of conflict operations.
When a new rule is added to the database it should ensure that it does not have any
conflict with any existing rule.

2.7 Sample questions

(a) Short answer questions
1. Describe the 8-puzzle problem.
2. State the missionaries and cannibals problem in AI.
3. What is a production system in AI?
4. Explain the blocks world problem.
5. What is a cryptarithmetic puzzle? Illustrate with an example.
6. Briefly explain the concepts of forward chaining and backward chaining in the con-
text of a production system.
7. What is the role of a “conflict resolution strategy” in a production system?
8. What are the different classes of production systems?
9. What actions are available in the blocks world problem?
10. Illustrate with an example the concept of decomposability of a problem into
smaller or easier problems.
11. Give an example of a problem in which some of the solution steps can be ignored
or undone.
12. Define a monotonic production system.
13. Consider the following familiar set of production rules:

1 IF green THEN walk

2 IF red THEN wait
3 IF green AND blinking THEN hurry
4 IF red OR green THEN traffic light works
2.7. SAMPLE QUESTIONS 38

Which of the above rules will be put into a conflict set by the system if the working
memory contains two facts: green, blinking? Apply some conflict resolution
strategy to fire a rule.

(a) Long answer questions

1. Explain the components of a problem in AI.
2. State the missionaries and cannibals problem and describe its various components
as a problem in AI.
3. State the monkey and banana problem and describe its various components as a
prob- lem in AI.
4. Briefly explain the characteristics of a problem in AI.
5. Explain the water jugs problem and give the system of production rules for the
prob- lem.
6. Describe a production system in AI. What are the merits and demerits of
production systems?
7. Define a production rule system and illustrate it with an example.
8. Write a note on the water jug problem using production rules.
9. Solve the following cryptarithmetic problems:
(a) A A
B B
C C
+

A B C

(b) E A T
+ T H A T

A P P L E

(c) F O R T Y
T E N
+ T E N

S I X T Y

(d) C R O S S
+ R O A D S

D A N G E R

10. Formulate the water jug problem as a production rule system.

 Chapter 3

Blind search strategies

Search algorithms are one of the most important areas of artificial intelligence. Search is an
important component of problem solving in artificial intelligence and, more generally, in
computer science, engineering and operations research. Combinatorial optimization, deci-
sion analysis, game playing, learning, planning, pattern recognition, robotics and theorem
proving are some of the areas in which search algorithms play a key role.
In this chapter we consider a certain class of search algorithms called blind search
algorithms.

3.1 Preliminaries
In describing the various strategies discussed in this and the next chapter, we make extensive
use the mathematical notions of graphs and trees. In this short section we quickly review
the necessary concepts.

3.1.1 Graphs
A graph G is an ordered pair (V, E) where the elements of V are called vertices or nodes
and the elements of E are called edges or sides. Each element of E can be thought of an
arc joining two vertices in V . Sometimes the arcs may have specific directions in which
case the graph is called a directed graph. There may be multiple edges joining two given
edges. If there are more than one edge joining two vertices, the graph is called a
multigraph. There may also be loops in graphs, a loop being an edge joining a vertex
with itself. In the sequel, unless otherwise explicitly stated, by “graph” we shall always
mean an undirected graph without loops and without multiple edges.
Graphs can be pictorially represented as in Figure 3.1. The set of vertices of the graph
in the figure is V = {v1, v2, v3, v4, v5} and the set of edges is E = {e1, e2, e3, e4, e5, e6,
e7}.

3.1.2 Trees
A graph is said to be connected if there is a path along the edges joining any two vertices in
the graph. A path which returns to the starting vertex is called a cycle. A tree is a
connected graph without cycles.

39
3.2. SEARCH IN GENERAL 40

e1
v1 v2
e2
e7 e6 e4 v3
e3
v5 v4
e5

Figure 3.1: A graph without loops and without multiple edges

v2 v5 e d h k
v1 v2 c g j
v6 v3
v3 a b f
v5 v4 v1 v4 i

(a) Connected graph (b) Disconnected graph (c) Tree

Figure 3.2: Different types of graphs

Examples
Rooted trees
In search strategies, we generally consider rooted trees where one vertex is designated as
the root and all edges are directed away from the root. In pictorial representations of
rooted trees, the root vertex is positioned near the top of a page and the remaining
vertices are positions below the root vertex at different levels (see, for example, Figure
3.3).

Figure 3.3: A rooted tree

3.2 Search in general

3.2.1 Solution as a sequence of actions
In Section 2.2.1 we discussed the following problem: “The problem consists of a 2 × 2
board with three numbered tiles and a blank space. A tile adjacent to the blank space can
3.2. SEARCH IN GENERAL 41

slide into the space. The objective is to reach the goal shown in Figure 2.4 by sliding the
tiles one at a time from start as in Figure 2.4.”
We have also seen that the solution to the problem is the following sequence of actions:

Up → Left → Down → Right

This, that the solution is a sequence of actions, is true in general. Assuming that the envi-
ronment has certain properties1, the solution to any problem (of the type considered in
AI) is a fixed sequence of actions.

3.2.2 Search algorithm

The process of looking for a sequence of actions that reaches the goal is called search. A
search algorithm takes a problem as input and returns a solution in the form of an action
sequence. Once a solution is found, the actions it recommends can be carried out. This is
called the execution phase.

3.2.3 Search tree

Having formulated some problems, we need to solve them. Since a solution is an action
sequence, search algorithms work by considering various possible action sequences. The
possible action sequences starting at the initial state form a search tree with the initial
state at the root; the branches are actions and the nodes correspond to states in the state
space of the problem. The root node of the tree corresponds to the initial state.

Example
Consider a person intending to travel from Arad to Bucharest in Romania (see Figure
3.4). Partial search trees for finding a route from Arad to Bucharest are shown in Figure
3.5.

Figure 3.4: A road map of Romania

1
The environment must be fully observable, discrete, known and deterministic.
3.3. BLIND SEARCH 42

Figure 3.5: Partial search trees for finding a route from Arad to Bucharest

3.2.4 Search strategy

The first step is to test whether this is a goal state. Then we need to consider taking
various actions. We do this by expanding the current state; that is, applying each legal
action to the current state, thereby generating a new set of states. Now we must choose
which of these possibilities to consider further. A search strategy specifies how to
choose which state to expand next.
Figure 3.5 shows some of the early stages in the construction of the search tree for
finding a path from Arid to Bucharest. In the figure, nodes that have been expanded are
shaded, nodes that have been generated but not yet expanded are outlined in bold and
nodes that have not yet been generated are shown in faint dashed lines.

3.3 Blind search

A blind search (also called an uninformed search) is a search that has no information
about its domain. The only thing that a blind search can do is distinguish a non-goal state
from a goal state. The blind search algorithms have no domain knowledge of the problem
state. The only information available to blind search algorithms is the state, the successor
function, the goal test and the path cost. Strategies that know whether one non-goal state is
“more promising” than another are called informed search or heuristic search strategies.
Blind search is useful in situations where there may not be any information available
to us. We might just be looking for an answer and won’t know we have found it until we
see it. In real world, we will have millions of nodes and we will have no idea where to
look for our target. It is also useful to study these blind searches as they form the basis for
some of the intelligent searches.
3.4. BREADTH-FIRST SEARCH 43

3.3.1 Example
Figure 3.4 shows a map of Romania.

• Let us assume that we are currently at Arad and we want to get to Bucharest.

• To construct a search tree we start by designating the initial state as the root node.
In the present problem, the initial state is the city of Arad and so we designate Arad
as the root node (Figure 3.5 (a)).

• There are three possible actions that we can take while we are at Arad, namely, go
to Sibiu, go to Timisoara and go to Zerind. These actions define three branches of
the search tree emanating from the root node and ending at the nodes named as
Sibiu, Timisoara and Zerind. These three branches form level 1 of the search tree
(Figure
3.5 (b)). A blind search will have no preference as to which node it should explore
first.

• Suppose we arbitrarily move to Sibiu. Then the possible actions are: go to Arad,
go to Rimnicu Vicea, go to Fagaras and go to Oradea. These actions produce four
branches from the node Sibiu (Figure 3.5 (c)).

• The process is continued.

Remarks
The search tree shown in Figure 3.5(c) includes the path from Arad to Sibiu and back to
Arad again! We say that Arad is a repeated state in the search tree, generated in this case
by a loopy path. Considering such loopy paths means that the complete search tree for
Romania is infinite because there is no limit to how often one can traverse a loop. On the
other hand, the map has only 20 states. However, there are mechanisms which ensure that
such loopy paths are generated while creating a search tree.

3.4 Breadth-first search

Breadth-first search is a simple strategy in which the root node is expanded first, then all
the successors of the root node are expanded next, then their successors, and so on. In
general, all the nodes are expanded at a given depth in the search tree before any nodes at
the next level are expanded.

Remark
Figure 3.6 shows the order in which the nodes in a tree are explored while applying the
breadth-first search algorithm.

3.4.1 Algorithm
The algorithm returns the goal state or failure.

1. Create a variable called NODE-LIST and set it to the initial state.

2. Until a goal state is found or NODE-LIST is empty do:

 3.4. BREADTH-FIRST SEARCH 44

Figure 3.6: The order in which the nodes are explored while applying the BFS algorithm:
0→1→2→3→4→5→6

(a) Remove the first element from NODE-LIST and call it E. If NODE-LIST was
empty, quit and return failure.
(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.

3.4.2 Examples
Example 1
Let the search tree be as in Figure 3.7 the initial and goal states being 1 and 8 respectively.
Apply the breadth-first search algorithm to find the goal state.

Figure 3.7: The search tree for a simple problem

Solution
The details of the various steps in finding a solution to the problem are depicted in Table
3.1. It can be seen that the nodes are visited in the following order:

1 → 2 → 3 → 4 → 5 → 6 → 7 → 8.
3.4. BREADTH-FIRST SEARCH 45

Node gener-
Step no. NODE-LIST Node E Comment
ated from E
1 1 - - Initial state
2 - 1 - -
3 - 1 2 -
4 2 1 - -
5 2 1 3 -
6 2, 3 1 - -
7 2, 3 1 4 -
8 2, 3, 4 1 - -
9 3, 4, 2 - -
10 3, 4 2 5 -
11 3, 4, 5 2 - -
12 3, 4, 5 2 6 -
13 3, 4, 5, 6 2 - -
13 4, 5, 6 3 - No node generated from E
14 5, 6 4 - -
15 5, 6 4 7 -
16 5, 6, 7 4 - -
17 5, 6, 7 4 8 Goal state. Return 8 and quit.

Table 3.1: Table showing the details of the various steps in the solution to the problem
specified in Figure 3.7

Example 2
Consider the water jug problem discussed in Section 2.6.3 and the associated production
rules given Table 2.1. We now construct the search tree using the breadth-first search
algo- rithm.

Stage 1. Construct a tree with the initial state (0, 0) as its root.
(0, 0)

Figure 3.8: Stage 1 in the construction of the breadth-first search tree of the water
jug problem
Stage 2. Construct all the children of the root by applying each of the applicable rules to
the initial state. Then we get a tree as in Figure 3.9.
3.5. DEPTH-FIRST SEARCH 46

(0, 0)

(4, 0) (0, 3)

Figure 3.9: Stage 2 in the construction of the breadth-first search tree of the water
jug problem
Stage 3. Now for each leaf node in Figure 3.9, generate all its successors by applying all
the rules that are appropriate. The tree at this point is as shown in Figure 3.10.
(0, 0)

(4, 0) (0, 3)

(4, 3) (0, 0) (1, 3) (4, 3) (0, 0) (3, 0)

Figure 3.10: Stage 3 in the construction of the breadth-first the search tree of the
water jug problem
Stage 4. The process is continued until we get a node representing the goal state, namely,
(2, 0).

3.5 Depth-first search

3.5.1 Algorithm
1. If the initial state is a goal state, quit and return success.

2. Otherwise, do the following until success or failure is signaled:

(a) Generate a successor E of the initial state. If there are no more successors,
signal failure.
(b) Call depth-first search with E as the initial state.
(c) If success is returned, signal success. Otherwise continue in this loop.

Remark
Figure 3.11 shows the order in which the nodes in a tree are explored while applying the
depth-first search algorithm.

3.5.2 Examples
Example 1
Let the search tree be as in Figure 3.12 the initial and goal states being 1 and 7
respectively. Apply the depth-first search algorithm to find the goal state.
3.5. DEPTH-FIRST SEARCH 47

Figure 3.11: The order in which the nodes are explored while applying the DFS algorithm:
0→1→3→4→2→5→6

2 5

3 4 6

7 8

Figure 3.12: A graph

Solution
We denote by DFS(x) an invocation of the depth-first search algorithm with x as the
initial vertex. The various steps in the implementation of the algorithm are shown in
Figure 3.13. Note that the various nodes are visited in the following order:

1→2→3→4→5→6→7

Call DFS(1)
1. Node 1 is not a goal state.
2. Do until successor or failure is signaled.
(a) Choose 2 as successor of node 1: E = 2.
(b) Call DFS(2).
1. Node 2 is not a goal state.
2. Do until success or failure is signaled.
(a) Choose 3 as successor of 2: E = 3
(b) Call DFS(3).
1. Node 3 is not a goal state.
2. Do until successor or failure is signaled.
(a) Node 3 has no successor. Signal failure.
DFS(3) ends.
(a) Choose 4 as successor of 2: E = 4
(b) Call DFS(4).
1. Node 4 is not a goal state.
 3.5. DEPTH-FIRST SEARCH 48

2. Do until successor or failure is signaled.

(a) Node 4 has no successor. Signal failure.
DFS(4) ends.
DFS(2) ends.
(a) Choose 5 as successor of node 1: E = 5
(b) Call DFS(5).
1. Node 5 is not a goal state.
2. Do until success or failure is signaled.
(a) Choose 6 as successor of 5: E = 6
(b) Call DFS(6).
1. Node 6 is not a goal state.
2. Do until successor or failure is signaled.
(a) Choose 7 as a successor of 6: E = 7
(b) Call DFS(7).
1. Node 7 is a goal state. Quit and return success.
DFS(7) ends.
DFS(6) ends.
DFS(5) ends.
DFS(1) ends.

Figure 3.13: Finding a path from node 1 to node 7 in Figure 3.12 using the DFS

Example 2
Consider the water jug problem discussed in Section 2.6.3 and the associated production
rules given Table 2.1. We now construct the search tree using the depth-first search algo-
rithm.

Stage 1. Construct a tree with the initial state (0, 0) as its root. The initial state is not a
good state.
(0, 0)

Figure 3.14: Stage 1 in the construction of the depth-first search tree of the water
jug problem

Stage 2. We generate a successor E for (0, 0). We choose the successor obtained by the
application of Rule 1 in the production rules. Then we get a tree as in Figure
(0, 0)

(4, 0)
3.15.
Figure 3.15: Stage 2 in the construction of the depth-first search tree of the water
jug problem
Stage 3. Now for the leaf node in Figure 3.15, generate a successor by applying an appli-
cable rule. The tree at this point is as shown in Figure 3.16.
3.6. BREADTH-FIRST SEARCH VS. DEPTH-FIRST SEARCH 49

(0, 0)

(4, 0)

(4, 3)

Figure 3.16: Stage 3 in the construction of the depth-first search tree of the water
jug problem
Stage 4. The process is continued until we get a node representing the goal state, namely,
(2, 0).

3.6 Breadth-first search vs. depth-first search

The Table 3.2 summarises the main advantages and disadvantages of breadth-first search
and depth-first search.

Table 3.2: Breadth-first search vs. depth-first search

Sl No Breadth-first Depth-first
1 Requires more memory because Requires less memory because
all of the tree that has so far only the nodes in the current
been generated must be stored. path are stored.
2 All parts of the search tree must May find a solution without ex-
be examined at level n before amining much of the search
any node on level n + 1 can be space. It stops when one solu-
examined. tion is found.
3 Will not get trapped exploring a May follow an unfruitful path
blind alley. for a long time, perhaps for
ever.
4 If there is a solution, May find a long path to a so-
Guaranteed to find a solution if lution in one part of the tree
there exists one. If there are when a shorter path exists in
multiple so- lutions then a some other unexplored path of
minimal solution (that is, a the tree.
solution that takes a minimum
number of steps) will be found.
5 Breadth-first search uses queue Depth-first search uses stack
data structure. data structure.
6 More suitable for searching ver- More suitable when there are
tices which are closer to the so- lutions away from source.
given source.
3.7. DEPTH-FIRST ITERATIVE DEEPENING (DFID) SEARCH 50

3.7 Depth-first iterative deepening (DFID) search

Iterative deepening combines the benefits of depth-first and breadth-first searches. Like
breadth-first search, the iterative deepening search is guaranteed to find a solution if one
exists. In general, iterative deepening is the preferred uninformed search method when
the search space is large and the depth of the solution is not known.

3.7.1 Algorithm
1. Set SEARCH-DEPTH = 1.
2. Conduct a depth-first search to a depth of SEARCH-DEPTH. If a solution is found,
then return it.
3. Otherwise, increment SEARCH-DEPTH by 1 and go to step 2.

Iteration 1 Iteration 2 Iteration 3 Iteration

4 Figure 3.17: Iterative deepening

Figure 3.18: The order in which the nodes are explored in iterative deepening
3.8. BLIND SEARCH IN GRAPHS 51

3.7.2 Advantages and disadvantages

1. The DFID will always find the shortest solution path to the goal state if it exists.

2. The maximum amount of memory used is proportional to the number of nodes in

the solution path.

3. All iterations but the final one are wasted.

4. The DFID is slower than the depth-first search only by a constant factor.

5. DFID is the optimal algorithm in terms of space and time for uninformed search.

3.8 Blind search in graphs

The algorithms we have discussed in the previous sections of this chapter can only be ap-
plied for searching the nodes of a tree. Since a tree is a graph with some special proper-
ties, these algorithms cannot be directly applied to search the nodes of a graph. However,
the breadth-first and depth-first search algorithms can be modified to create algorithms
for searching the nodes in a graph.
Suppose we are searching a graph G. The set of all leaf nodes available for expansion
at any given point is called the frontier. (Many authors call it the open list.)

3.8.1 Graph-search algorithm

1. Initialize the frontier using the initial state of problem.

2. Initialize the explored set to be empty.

3. Do the following until success or failure is returned:

(a) If the frontier is empty then return failure.

(b) Choose a leaf node and remove it from the frontier.
(c) If the node contains a goal state then return the corresponding solution.
(d) Add the node to the explored set.
(e) Expand the chosen node, adding the resulting nodes to the frontier only if not
in the frontier or explored set.

3.8.2 Breadth-first and depth-first search in graphs

In the graph-search algorithm given in Section 3.8.1, we may use different data structures
to store the nodes in the frontier. If we use the queue data structure (first-in first-out
(FIFO) queue) the resulting algorithm is the breadth-first search algorithm where as if we
use the stack data structure (last-in first-out (LIFO) queue) the resulting algorithm is the
depth-first search algorithm.
3.8. BLIND SEARCH IN GRAPHS 52

(a) Queue data structure (b) Stack data

structure Figure 3.19: Queue and stack data structures

3.8.3 Example

Figure 3.20: A graph

Consider the graph shown in Figure 3.20. Let us explore the graph starting from the
node A. If we use the queue data structure to store the frontier set, then one of the orders
in which the nodes are explored is shown in Table 3.3. A different order in which the
nodes are explored is shown in Table 3.4.

Iteration No. Frontier set Explored set

(Queue (FIFO))
Initial A ∅
Iteration 1 E B {A}
Iteration 2 D C E {A, B}
Iteration 3 D C {A, B, E}
Iteration 4 D {A, B, E, C}
Iteration 5 {A, B, E, C, D}

Table 3.3: Frontier set using queue data structure, and explored set
3.9. SAMPLE QUESTIONS 53

Iteration No. Frontier set Explored set

(Queue (FIFO))
Initial A ∅
Iteration 1 B E {A}
Iteration 2 D B {A, E}
Iteration 3 C D {A, E, B}
Iteration 4 C {A, E, B, E, D}
Iteration 5 {A, E, B, D, C}

Table 3.4: Frontier set using queue data structure, and explored set
If we use the stack data structure (LIFO) to store the frontier set, then one of the
orders in which the nodes are explored is shown in Table 3.5.

Iteration No. Frontier set Explored set

(Stack (LIFO))

Initial A ∅

E
Iteration 1 {A}
B

D
Iteration 2 { A, E }
B

C
Iteration 3 { A, E, D }
B

Iteration 4 B { A, E, D, C }
Iteration 5 { A, E, D, C, B}

Table 3.5: Frontier set using stack data structure, and explored set

3.9 Sample questions

(a) Short answer questions
1. Define search tree and illustrate with an example.

2. What is meant by blind search? Give examples of blind-search methods.

3.9. SAMPLE QUESTIONS 54

3. Compare and contrast unformed and uninformed search methods.

4. List some of the uninformed search methods.

5. State the advantages of the breadth-first search method.

6. Will the breadth-first search method always find the optimal solution? Why? Illus-
trate with an example.

7. Compare and contras BFS and DFS methods.

8. Give an outline of the DFID search algorithm.

9. Let the search tree be as in Figure 3.21. If the BFS algorithm is used, what is the
order in which the various nodes are explored starting from node 0.

Figure 3.21: A search tree

10. Let the search tree be as in Figure 3.21. If the DFS algorithm is used, what is the
order in which the various nodes are explored starting from node 0.

11. Consider the following graph:

Figure 3.22: A search tree

Starting at root node 1, give the order in which the nodes will be visited by the
breadth-first and depth-first algorithms.

12. If the graph in Figure 3.23 is explored using the breadth first algorithm starting
from node Q, what is the order in which the nodes will be explored? what will be
the order if it is explored using the depth first algorithm?

M N O
R Q P

Figure 3.23: A graph

3.9. SAMPLE QUESTIONS 55

13. If the graph in Figure 3.24 is explored using the breadth first algorithm starting
from node 0, what is the order in which the nodes will be explored? what will be
the order if it is explored using the depth first algorithm?

Figure 3.24: A graph

(b) Long answer questions

1. Distinguish breadth-first and depth-first search methods.

2. Briefly describe the breadth-first and the depth-first search for trees.

3. Write the breadth-first search algorithm and illustrate it with an example.

4. Write the depth-first search algorithm and illustrate it with an example.

5. Explain a graph-search algorithm.

6. Explain various uninformed search strategies.

 Chapter 4

Heuristic search strategies

In computer science and artificial intelligence, a heuristic1 is a technique designed for

solving a problem more quickly when classic (that is, well known) methods are too slow,
or for finding an approximate solution when classic methods fail to find any exact
solution. This is achieved by trading optimality, completeness, accuracy, or precision for
speed. In a way, it can be considered a shortcut.
The objective of a heuristic is to produce a solution in a reasonable time that is good
enough for solving the problem at hand. This solution may not be the best of all the actual
solutions to this problem, or it may simply approximate the exact solution. But it is still
valuable because finding it does not require a prohibitively long time.
In this chapter we consider informed (heuristic) search strategies, that is, strategies
that use problem-specific knowledge beyond the definition of the problem itself,
strategies that can find solutions more efficiently than an uninformed strategy.

4.1 Heuristic function

A heuristic function, also called simply a heuristic, is a function that ranks alternatives in
search algorithms at each branching step based on available information to decide which
branch to follow.

58 A

47 B 34 C 41 D

Figure 4.1: Values of a heuristic function

For example, let us assume that we are at node A in the search tree of some search
problem (see Figure 4.1). Let us also assume that there are three different branches, say
1
The word heuristics comes from Greek word heuriskein which means “to find, to discover”.

56
4.1. HEURISTIC FUNCTION 57

AB, AC and AD, along which we can continue the search for the goal node. We have to
decide which one of these nodes is to be selected. From the circumstances of the problem
let it be possible to obtain an estimate of the cheapest path from the nodes to the goal
node. Let these estimates be as shown alongside each of the nodes in the figure. For
example, the estimate of the cheapest cost from node C to the goal node is 34.
One way of choosing the next node is to chose that node for which the estimated cost
of the cheapest path to goal is least. As per this criterion, we choose to proceed along the
branch AC. (Of, course, there are other ways in which the estimates can be used to
choose the next branch.) These estimated costs define a heuristic function for the
problem. The methods of constructing such functions from the circumstances of a
problem are illustrated in Section 4.1.4.

4.1.1 Definition
The heuristic function for a search problem is a function h(n) defined as follows:

h(n) = estimated cost of the cheapest path from the

state at node n to a goal state.

Remark
Unless otherwise specified, it will be assumed that a heuristic function h(n) has the
follow- ing properties:

1. h(n) ≥ 0 for all nodes n.

2. If n is the goal node, then h(n) = 0.

4.1.2 Admissible heuristic functions

A heuristic function is said to be admissible if it never overestimates the cost of reaching
the goal, that is, the cost it estimates to reach the goal is not higher than the lowest
possible cost from the current point in the path.
Let h(n) be a heuristic function for a search problem. h(n) is the estimated cost of
reaching the goal from the state n. Let h∗(n) be the optimal cost to reach a goal from n.
We say that h(n) is admissible if

h(n) ≤ h∗(n) for all n

4.1.3 Remarks
1. There are no limitations on the function h(n). Any function of our choice is accept-
able. As an extreme case, the function h(n) defined by

h(n) = c for all n

where c is some constant can also be taken as a heuristic function. The heuristic
function defined by
h(n) = 0 for all n
is called the trivial heuristic function.
4.1. HEURISTIC FUNCTION 58

2. The heuristic function for a search problem depends on the problem in the sense
that the definition of the function would make use of information that are specific
to the circumstances of the problem. Because of this, there is no one method for
constructing heuristic functions that are applicable to all problems.

3. However, there is a standard way to construct a heuristic function: It is to find a

solution to a simpler problem, which is one with fewer constraints. A problem with
fewer constraints is often easier to solve (and sometimes trivial to solve). An
optimal solution to the simpler problem cannot have a higher cost than an optimal
solution to the full problem because any solution to the full problem is a solution to
the simpler problem.

4.1.4 Examples
In the following subsections we present several examples of heuristic functions. In each
of these examples, the heuristic function is defined using knowledge about the
circumstances of the particular problem. Also, these functions are solutions of problems
obtained by re- laxing some of the constraints in the problem. Moreover, it can be seen
that all the presented heuristic functions are admissible.

1. h(n) in path search in maps

Consider the problem of finding the shortest path from Arid to Bucharest in Romania (see
Figure 3.4).
In the problem, the traveler is required to travel by roads shown in the map. Let us
consider a problem obtained by relaxing this condition. Let the traveler be allowed to
travel by any method. In this relaxed problem, obviously, the shortest path from from a
city n to Bucharest is the air distance between n and Bucharest. We may use this
solution as the definition of a heuristic function for the original problem.
If n is any of the cities in Romania, we may define the heuristic function h(n) as fol-
lows:

h(n) = The straight line distance (air distance) from n to Bucharest.

Table 4.1 shows the values of h(n). Note that h(n) ≥ 0 for all n and h(Bucharest) = 0.
Now h∗(n) is the the length of the shortest road-path from n to Bucharest. Since, the
air distance never exceeds the road distance we have

h(n) ≤ h∗(n) for all n.

Thus, h(n) as defined above is an admissible heuristic function for the problem

2. h(n) in missionaries and cannibals problem

Recall the description of the missionaries and cannibals problem given in Section 2.3.3.
To define a heuristic function for the missionaries and cannibals problem, let us
consider a simpler problem wherein we ignore the condition that at any location (on either
bank of the river or in the boat) the number of cannibals should not exceed the number of
missionaries. Let k be the number of persons at the initial side of the river. If k = 2 we
need only 1 river crossing to take both persons to the other side of the river. If k = 3 we
need 3 river
 4.1. HEURISTIC FUNCTION 59

City (n) h(n) City (n) h(n)

Arad 366 Mehadia 241
Bucharest 0 Neamt 234
Craiova 160 Oradea 380
Dobreta 242 Pitesti 100
Eforie 161 Rimnicu Vilcea 193
Fagaras 176 Sibiu 253
Giurgiu 77 Timisoara 329
Hirsova 151 Urziceni 80
Iasi 226 Vaslui 199
Lugoj 244 Zerind 374

Table 4.1: Heuristic function for Romania travel problem

crossings, if k = 4 also we need only 3 river crossings to take all people to the other side
of the river. If k = 5 or k = 6 we need 5 river crossings to take all people to the other
side of the river. See below for details of the crossings in the case when k = 6.

Crossing 1 Take 2 to destination

side Crossing 2 Return to initial side
Crossing 3 Take 2 to destination
side Crossing 4 Return to initial side
Crossing 5 Take 2 to destination
side
In general, in the simplified scenario, it takes k or k − 1 river crossings depending on
whether k is odd or even to take k persons to the other side of the river.
We may choose the following function as a heuristic function for the missionaries and
cannibals problem. If (x, y, z) is a state of the problem (see Section 2.3.3), we define

0 if x + y =

h(x, y, z) =  x0 + y if x + y is odd


x + y − 1 if x + y > 0 and x + y is even

Note that x + y is the number of people on the initial side of the river.
Note that h(n) ≥ 0 for all n and h(Goal state) = 0. Also, it is obvious that this is an
admissible heuristic function for the problem.
Another admissible heuristic function for the problem is given below:
1
h(n) = 2 (x + y).

3. h(n) in the 8-puzzle problem

There are several restrictions, though not explicitly stated, on the movements of the tiles in
the 8-puzzle problem like the following (see Section 2.3.2):
4.1. HEURISTIC FUNCTION 60

1. A tile can be moved only to a nearby vacant slot.

2. Tiles can only be moved in the horizontal or vertical direction.

3. One tile cannot slide over on another tile.

We can get different admissible heuristic functions for the 8-puzzle problem by solv-
ing simpler problems by relaxing or removing some of these restrictions. We define two
different heuristic functions h1(n) and h2(n) for the 8-puzzle problem. h1(n) is obtained
by solving the problem without all the three restrictions given above. h2(n) is obtained by
solving the problem obtained by relaxing first restriction to the restriction that “a tile may
be moved to any nearby slot and deleting the third restriction”.

• Hamming distance2 heuristic

h1(n) = Number of tiles out of position.

Example

5 8 1 2 3
4 2 1 4 5 6
7 3 6 7 8

(a) Given state (n) (b) Goal state

Figure 4.2: An initial state and a goal state of 8-puzzle

Consider the state shown in Figure 4.2(a) and compare the positions of the tiles
with their positions in the goal state shown in Figure 4.2(b). It can be seen that
only two tiles, namely 4 and 7 (ignoring the empty cell), are in their positions.
Hence, for the state we have
h1(n) = 6.

• Manhattan distance heuristic

We first consider the Manhattan distance between two cells in a grid of cells. Figure
4.3 illustrates the concept using 8 × 8 grid.
2
The terminology comes from information theory. In information theory, the Hamming distance between
two strings of equal length is the number of positions at which the corresponding symbols are different. In
other words, it measures the minimum number of substitutions required to change one string into the other.
For example, the Hamming distance between the strings “1011011010” and “1010011110” is 2.
4.1. HEURISTIC FUNCTION 61

(C) 0 1 2 (D) 3
(B) 7
6
5 (E) 0
(A) 0 1 2 3 4 1
2
(F) 3

Figure 4.3: In the 8 × 8 grid, the Manhattan distance between the cells (A) and (B)
is 7, between (C) and (D) is 3 and between (E) and (F) is 3.

The Manhattan distance heuristic function is defined as follows:

h2(n) = Sum of the Manhattan distances of every numbered tile to its goal
position.

Example

For the initial state n shown in Figure 4.2 we have

h2(n) = Manhattan distance of 1 from goal position+
Manhattan distance of 2 from goal position+
···+
Manhattan distance of 2 from goal position
=3+1+3+0+2+1+0+3
= 13.

4. h(n) in blocks world problem

We define a heuristic function for the blocks world problem (see Section 2.3.5) as follows:

h(n) = Add one point for every block that is resting on the thing it is
supposed to be resting on. Subtract one point for every block
that is sitting on the wrong thing.
This is an example of a heuristic function not satisfying the conditions h(n) ≥ 0 and
h(Goal node) = 0.

Example
To illustrate the idea, consider a blocks world problem with the initial and goal states as
in Figure 4.4. In the initial state, only block numbered “1” is resting on the thing it is
supposed to be resting on. All other blocks are sitting on the wrong things. Hence we
have:

h(Initial state) = +1 − 1 − 1 − 1 − 1 − 1 = −4.

4.2. BEST FIRST SEARCH 62

Initial state Goal state

Figure 4.4: A blocks world

problem

Since in the goal state every block is sitting on the right thing, we have

h(Goal state) = 1 + 1 + 1 + 1 + 1 + 1 = 6 /= 0.

5. h(n) in water jug problem

For the water jug problem given in Section 2.6.3, we may choose the following as a
heuristic function. For a state specified by the ordered pair (x, y), we define

h(x, y) = |x − 2| + |y − 2|.

This function satisfies the condition h(n) ≥ 0 for all n. It does not satisfy the condition
h(Goal state) = 0, because as a goal state is specified by a pair of the form (2, y), we have

h(Goal state) = h(2, y) = |y − 2|.

4.2 Best first search

If we consider searching as a form of traversal in a graph, an uninformed search
algorithm blindly traverses to the next node in a given manner without considering the
cost associated with that step. An informed search, on the other hand, uses an evaluation
function f (n) to decide which among the various available nodes is the most promising
(or “best”) before traversing to that node. An algorithm that implements this approach is
known as a best first search algorithm. It should be emphasised that, in general, the
evaluation function is not the same as the heuristic function.
There are different best first search algorithms depending on what evaluation function
is used to determine the most promising node. One commonly used evaluation function is
the heuristic function. The corresponding algorithm is sometimes called the the greedy
best first search algorithm. Another variant known as the A٨ best search algorithm uses
the sum of the heuristic function and the function that specifies the cost of traversing the
path from the start node to a particular node as the evaluation function.

4.3 Greedy best first search

4.3.1 The algorithm
The greedy best first search algorithm uses the following notations and conventions:
4.3. GREEDY BEST FIRST SEARCH 63

OPEN : A list which keeps track of the current

“immediate” nodes available for traversal.
CLOSED : A list that keeps track of the nodes already traversed.
h(n) : The heuristic function used as the evaluation func-
tion f (n), that is, f (n) = h(n).

Algorithm
1. Create two empty lists: OPEN and CLOSED.

2. Start from the initial node (say N) and put it in the ordered OPEN list.

3. Repeat the next steps until goal node is reached.

(a) If OPEN list is empty, then exit the loop returning FALSE.
(b) Select the first/top node (say N) in the OPEN list and move it to the CLOSED
list. Also record the information of the parent node of N.
(c) If N is a goal node, then move the node to the CLOSED list and exit the loop
returning TRUE. The solution can be found by backtracking the path.
(d) If N is not the goal node, generate the immediate successors of N, that is, the
immediate next nodes linked to N and add all those to the OPEN list.
(e) Reorder the nodes in the OPEN list in ascending order according to the values
of the evaluation function f (n) at the nodes.

4.3.2 Examples
Example 1
Use the greedy best first search algorithm to find a path from A to C in the graph shown in
Figure 4.5.

A B

D C

Figure 4.5: An example graph

Use the following heuristic function:

Node n S A B C D
h(n) 6 2 3 0 2
4.3. GREEDY BEST FIRST SEARCH 64

Solution
Table 4.2 shows the various stages in the execution of the algorithm with details of the
current node, children of the current node, the contents of the OPEN list and the contents
of the CLOSED list.

Table 4.2: Various stages in the execution of the greedy best first search algorithm for the
graph in Figure 4.5

Current Children OPEN CLOSED Remarks

(N) of N (node, parent)
S – – – Initial state
– – S – Initial state added
to OPEN.
S – – (S, –) S made current
node and added to
CLOSED.
S A, B – (S, –) Children of S
S – A, B (S, –) Add children to
OPEN and order
according to val-
ues of heuristic
function.
A – B (S, –), (A, S) A made current
node and added to
CLOSED.
A B, D B (S, –) (A, S) Children of A
A – D, B (S, –) (A, S) Add children to
OPEN and order
according to val-
ues of heuristic
function.
D – B (S, –), (A, S), D made current
(D, A) node and added to
CLOSED.
D A, C B (S, –), (A, S), Children of D
(D, A)
D – C, B (S, –), (A, S), Add children to
(D, A) OPEN and order
according to val-
ues of heuristic
function.
4.3. GREEDY BEST FIRST SEARCH 65

C – B (S, –), (A, S), C made current

(D, A), (C, D) node and added to
CLOSED.
C – – – C is goal state.

From the list of nodes in the CLOSED list at the stage when the goal state has been
reached we see that the path from S to C is

“C ← D ← A ← S”,

or equivalently,
“S → A → D → C”.

Example 2
Using the greedy best first search algorithm, find an optimal path from A to F in the
search graph shown in Figure 4.6. In the figure, the numbers written alongside the nodes
are the values of the heuristic function and the numbers written alongside the edges are
the costs associated with the edges.

9 2 0
B 5 2
D F
6 11
8
4 C 2 3
3 8

A E G
7 3 5
10 1

Figure 4.6: An example graph

Solution
Table 4.3 shows the various stages in the execution of the algorithm with details of the
current node, children of the current node, the contents of the OPEN list and the contents
of the CLOSED list.

Table 4.3: Various stages in the execution of the greedy best first search algorithm for the
search tree in Figure 4.6

Current Children OPEN CLOSED Remarks

(N) of N (node, parent)
A – – – Initial state
4.3. GREEDY BEST FIRST SEARCH 66

– – A – Initial state added

to OPEN.
A – – (A, –) A made current
and added to
CLOSED.
A B, C, E – (A, –) Children of A
A – E, C, B (A, –) Add children to
OPEN and order
according to val-
ues of heuristic
function.
E – C, B (A, –), (E, A) E made current and
added to CLOSED
E C, D, G C, B (A, –), (E, A) Children of E
E – G, D, C, B (A, –), (E, A) Add children to
OPEN and order
according to val-
ues of heuristic
function.
G – D, C, B (A, –), (E, A), G made current
(G, E) and added to
CLOSED.
G F D, C, B (A, –), (E, A), E is child of G, but
(G, E) is in CLOSED and
so not considered.
G – F, D, C, B (A, –), (E, A), Add children to
(G, E) OPEN and order
according to val-
ues of heuristic
function.
F – D, C, B (A, –), (E, A), F made current and
(G, E), (F, G) added to CLOSED.
F – – – Goal state reached.

From the list of nodes in the CLOSED list at the stage when the goal state has been
reached we see that the path from A to F is “F ← G ← E ← A”, or equivalently, “A → E →
G → F ”.

Example 3
Using the greedy best first search algorithm find the optimal path from Arid to Bucharest
in Figure 4.7 using the heuristic function defined by Table 4.1.
4.3. GREEDY BEST FIRST SEARCH 67

Figure 4.7: A road map of Romania

Solution

Current Children OPEN CLOSED (node,

parent)
Arad-366 - -
- - Arad-366 -
Arad-366 - - (Arad-366,-)
Arad-366 Timisoara- - (Arad-366,-)
329, Sibiu-
253, Zerind-
374
Arad-366 - Timisoara-329, (Arad-366,-)
Sibiu-253, Zerind-
374
Arad-366 - Sibiu-253, (Arad-366,-)
Timisoara-329,
Zerind-374
Sibiu-253 - Timisoara-329, (Arad-366,-),
Zerind-374 (Sibiu-253, Arad-
366)
Sibiu-253 Oradea-380, Timisoara-329, (Arad-366,-),
Fagaras-176, Zerind-374 (Sibiu-253, Arad-
Rimnicu 366)
Vicea-193
Sibiu-253 - Oradea-380, (Arad-366,-),
Fagaras-176, (Sibiu-253, Arad-
Rimnicu Vicea- 366)
193, Timisoara-
329, Zerind-374
4.3. GREEDY BEST FIRST SEARCH 68

Sibiu-253 - Fagaras-176, (Arad-366,-),

Rimnicu Vicea- (Sibiu-253, Arad-
193, Timisoara- 366)
329, Zerind-374,
Oradea-380
Fagaras-176 - Rimnicu Vicea- (Arad-366,-),
193, Timisoara- (Sibiu-253, Arad-
329, Zerind-374, 366), (Fagaras-176,
Oradea-380 Sibiu-253)
Fagaras-176 Bucharest-0 Rimnicu Vicea- (Arad-366,-),
193, Timisoara- (Sibiu-253, Arad-
329, Zerind-374, 366), (Fagaras-176,
Oradea-380 Sibiu-253)
Fagaras-176 - Bucharest-0, (Arad-366,-),
Rimnicu Vicea- (Sibiu-253, Arad-
193, Timisoara- 366), (Fagaras-176,
329, Zerind-374, Sibiu-253)
Oradea-380
Fagaras-176 - Bucharest-0, (Arad-366,-),
Rimnicu Vicea- (Sibiu-253, Arad-
193, Timisoara- 366), (Fagaras-176,
329, Zerind-374, Sibiu-253)
Oradea-380
Bucharest-0 - Rimnicu Vicea- (Arad-366,-),
193, Timisoara- (Sibiu-253, Arad-
329, Zerind-374, 366), (Fagaras-
Oradea-380 176, Sibiu-253),
(Bucharest-0,
Fagaras-176)
Goal node

The greedy best first search algorithm yields the following path from Arad to Bucharest:
Arad → Sibiu → Fagaras → Bucharest

Remarks
It may be noted that the above solution is not the optimal path from Arad to Bucharest. The
cost of the above path is
140 + 99 + 211 = 450
where as there is path with lower cost, namely,
Arad → Sibiu → Rmnicu Vilcea → Pitest → Bucharest
whose total cost is only
140 + 80 + 97 + 101 = 418.
4.4. A∗ BEST FIRST SEARCH 69

4.4 A∗ best first search

Let h(n) be a heuristic function associated with the search problem and g(n) be the cost of
traversing the path from the start node to the node n. In the A∗ best first search3, we use
the following evaluation function:

f (n) = g(n) + h(n).

4.4.1 Examples
The following examples illustrate the basic idea of the A∗ algorithm.

Example 1
Use the A∗ best first search algorithm to find a path from A to C in the graph shown in
Figure 4.8.

S
1 4
2
A B

11 5 2

D C
3

Figure 4.8: An example graph

Use the following heuristic function:

Node n S A B C D
h(n) 6 2 3 0 2

Solution
Step 1. We start with S.
Nodes A and B are immediate successors of S. We have

f (A) = g(A) + h(A) = 1 + 2 = 3

f (B) = g(B) + h(B) = 4 + 3 = 7
min{f (A), f (B)} = min{3, 7} = 3 = f (A)

So we move to A and the path is S → A.

3
The notation A∗ is borrowed from the statistical literature. Statisticians use a hat to indicate an estimate
for a quantity, and often use a star to indicate an estimate that’s optimal with respect to some criterion.
4.4. A∗ BEST FIRST SEARCH 70

Step 2. Nodes B and D can be reached form A. We have

f (B) = g(B) + h(B) = (1 + 2) + 3 = 6

f (D) = g(D) + h(D) = (1 + 11) + 2 =
14 min{f (B), f (D)} = min{6, 14} = 6 = f (B)

So we move to B and the path so far is S → A → B.

Step 3. C can be reached from D and C is the goal state.

Hence the required path is S → A → B → C.

Example 2
Consider the graph shown in Figure 4.9. The numbers written on edges represent the dis-
tance between the nodes. The numbers written on nodes represent the heuristic value.
Find the most cost-effective path to reach from start state A to final state J using A*
Algorithm.

Figure 4.9: A search graph

Solution
Step 1. We start with node A.
Node B and Node F can be reached from node A.
A* Algorithm calculates f (B) and f (F ).

f (B) = 6 + 8 = 14
f (F ) = 3 + 6 = 9

Since f (F ) < f (B), so it decides to go to

node F. Path: A → F

Step 2. Node G and Node H can be reached from node F.

4.4. A∗ BEST FIRST SEARCH 71

A* Algorithm calculates f (G) and f (H).

f (G) = (3 + 1) + 5 = 9
f (H) = (3 + 7) + 3 = 13

Since f (G) < f (H), so it decides to go to node G.

Path: A → F → G

Step 3. Node I can be reached from node G.

A* Algorithm calculates f (I).

f (I) = (3 + 1 + 3) + 1 = 8

It decides to go to node I.
Path: A → F → G → I
Step 4. Node E, Node H and Node J can be reached from node I.
A* Algorithm calculates f (E), f (H) and f (J).

f (E) = (3 + 1 + 3 + 5) + 3 = 15
f (H) = (3 + 1 + 3 + 2) + 3 = 12
f (J) = (3 + 1 + 3 + 3) + 0 = 10

Since f (J) is least, so it decides to go to node J.

Path: A → F → G → I → J

Step 5. This is the required shortest path from node A to node J.

4.4.2 A∗ best first search algorithm

Algorithm
We use the following notations:

OPEN : Nodes that have been generated, but not yet

examined
CLOSED : Nodes that have already been examined
g(n) : The cost of getting from start node to node
n h(n) : The heuristic function
f (n) : g(n) + h(n)
start : The initial node
BESTNODE : Node in OPEN with lowest f value
SUCCESSOR : A successor of BESTNODE
OLD : A node in OPEN/CLOSED
cost(X,Y) : Cost of getting from node X to successor node
Y.
4.5. GREEDY BEST FIRST VS A∗ BEST FIRST 72

1. (a) Set start as the only member of OPEN.

(b) Set g(start) = 0.
(c) Set f (start) = h(start).

2. Until a the goal node is found, repeat the following procedure:

(a) If there are no nodes in OPEN report failure.

(b) Otherwise do the following:
i. Pick the node in OPEN with the lowest f value. Call it BESTNODE.
ii. Remove BESTNODE from OPEN.
iii. If BESTNODE is a goal, exit and report a solution.
iv. If BESTNODE is not a goal, generate the successors of BESTNODE. Let
SUCCESSOR denote a successor of BESTNODE. For each
SUCCESSOR do the following:
A. Set SUCCESSOR to point back to BESTNODE.
B. Set g(SUCCESSOR) equal to

g(BESTNODE) + cost(BESTNODE, SUCCESSOR).

C. Examine whether SUCCESSOR is in the list OPEN.

• Let SUCCESSOR be in OPEN.
Call that node OLD. Delete SUCCESSOR from the list of
BESTN- ODE’s successors and add OLD. If g(OLD) via its
current parent is greater than or equal to g(SUCCESSOR) via
BESTNODE then re- set OLD’s parent link to point to
BESTNODE, record new cheaper path in g(OLD) and update f
(OLD).
• Let SUCCESSOR be not in OPEN and let SUCCESSOR be in
CLOSED. If so, call the node in CLOSED as OLD and add OLD to the
list of BESTNODE’s successors. Check to see if the new path or the
old path is better (as in ) and reset the parent link and g and f . If we
have found a better path to old, the improvements should be
propagated4
to OLD’s successors.
• Let SUCCESSOR be not in OPEN and not in CLOSED. Then put
it in OPEN and add it the list of BESTNODE’s successors.
Compute

f (SUCCESSOR) = g(SUCCESSOR) + h(SUCCESSOR).

4.5 Greedy best first vs A∗ best first

4.6 Simple hill climbing

Hill climbing is an iterative algorithm that starts with an arbitrary solution to a problem,
then attempts to find a better solution by making an incremental change to the solution.
If the change produces a better solution, another incremental change is made to the new
solution, and so on until no further improvements can be found. The solution obtained by
4
How exactly the improvements can be propagated is a a little tricky. We omit the details.
4.6. SIMPLE HILL CLIMBING 73

Sl. No. Greedy best first A∗ best first

1 The greedy best first is not If the heuristic function is ad-
com- plete, that is, the missible, the A∗ best first is
algorithm may not find the goal complete.
always. If the heuristic function is ad-
2 In general, the greedy best first missible, the A∗ best first is
may not find the optimal path always yields the optimal path
from the initial to the goal state. from the initial state to goal
state.
In general, A* best first uses
3 In general, greedy best first more memory than greedy best
uses less memory than A∗ best first.
first.

Table 4.5: Greedy best first vs A∗ best first

this method may not be the global optimum; it will only be a local optimum. In this
sense, it is a local search method. The difference between a local and a global maxima is
illustrated in Figure 4.10. In the figure, the x-axis denotes the nodes in the state space and
the y-axis denotes the values of the objective function corresponding to particular states.

Figure 4.10: Local and global maxima and minima

When we perform hill-climbing, we are short-sighted. We cannot really see far. We

make local best decisions with the hope of finding a global best solution. See Figure 4.11

Figure 4.11: Hill climbing

4.6. SIMPLE HILL CLIMBING 74

4.6.1 Simple hill climbing algorithm

In the following the word “operator” refers to some applicable action. It usually includes
information about the preconditions for applying the action and also about the effect of
the action.

1. Evaluate the initial state. If it is also a goal state, then return it and quit. Otherwise
continue with the initial state as the current state.

2. Loop until a solution is found or until there are no new operators left to be applied
in the current state:

(a) Select an operator that has not yet been applied to the current state and apply
it to produce a new state.
(b) Evaluate the new state.
i. If it is a goal state then return it and quit.
ii. If it is not a goal state but if it is better than the current state then make it
the current state.
iii. If it is not better than the current state then continue in the loop.

4.6.2 Examples
Example 1
Apply the hill climbing algorithm to solve the blocks world problem shown in Figure 4.12:

Figure 4.12: A blocks world problem

Solution
To use the hill climbing algorithm we need an evaluation function or a heuristic function.
We consider the following evaluation function:

h(n) = Add one point for every block that is resting on the thing it is
supposed to be resting on. Subtract one point for every block
that is sitting on the wrong thing.
4.6. SIMPLE HILL CLIMBING 75

We call “initial state” “State 0” and “goal state” ”Goal”. Then considering the blocks A, B,
C, D, E, F, G, H in that order we have
h(State 0) = −1 − 1 + 1 + 1 + 1 + 1 + 1 + 1
=4
h(Goal) = +1 + 1 + 1 + 1 + 1 + 1 + 1 + 1
=8
There is only one move from State 0, namely, to move block A to the table. This
produces the State 1 with a score of 6:
h(State 1) = 1 − 1 + 1 + 1 + 1 + 1 + 1 + 1 = 6.
There are three possible moves form State 1 as shown in Figure 4.13. We denote these three

Figure 4.13: A blocks world problem

states State 2(a), State 2(b) and State 2 (c). We also have
h(State 2(a)) = −1 − 1 + 1 + 1 + 1 + 1 + 1 + 1 = 4
h(State 2(b)) = +1 − 1 + 1 + 1 + 1 + 1 + 1 − 1 = 4
h(State 2(c)) = +1 − 1 + 1 + 1 + 1 + 1 + 1 − 1 = 4
Hill climbing will halt because all these states have lower scores than the current state.
The process has reached a local maximum that is not a global maximum. We have
reached such a situation because of the particular choice of the heuristic function. A
different choice of heuristic function may not produce such a situation.

Example 2
Given the 8-puzzle shown in Figure 4.14, use the hill-climbing algorithm with the
Manhat- tan distance heuristic to find a path to the goal state.

Solution
By definition, the Manhattan distance heuristic is the sum of the Manhattan distances of
tiles from their goal positions. In Figure 4.14, only the tiles 5, 6 and 8 are misplaced and
their distances from the goal positions are respectively 2, 2 and 1. Therefore,
h(Initial state) = 2 + 2 + 1 = 5.
4.7. STEEPEST-ASCENT HILL CLIMBING 76

1 2 3 1 2 3
4 8 4 5 6
7 6 5 7 8

Initial state Goal

state Figure 4.14: An 8-

puzzle

Figure 4.15 shows the various states reached by applying the various operations specified
in the hill-climbing algorithm. The numbers written alongside the grids are the values of
the heuristic function.
Note that the problem is a minimisation problem. The value of the heuristic function
at the goal state is 0 which is the minimum value of the function. Note further that once
we find an operation which produces a better value than the value in current state we do
proceed to that state without considering whether there is any other move which produces
a state having a still better value.

Figure 4.15: The search path obtained by applying the hill-climbing algorithm to an 8-
puzzle

4.7 Steepest-ascent hill climbing

In steepest-ascent hill climbing, we consider all the moves from the current state and
selects the best as the next state. In the basic hill climbing, the first state that is better
than the
4.8. ADVANTAGES AND DISADVANTAGES OF HILL CLIMBING 77

current state is selected.

4.7.1 Algorithm
1. Evaluate the initial state. If it is also a goal state, then return it and quit. Otherwise
continue with the initial state as the current state.
2. Loop until a solution is found or until a complete iteration produces no change to
current state:

(a) Let SUCC be a state such that any possible successor of the current state
will be better than SUCC.
(b) For each operator that applies to the current state do:
i. Apply the operator and generate a new state.
ii. Evaluate the new state. If it is a goal state, then return it and quit. If
not, compare it to SUCC. If it is better then set SUCC to this state. If it
is not better, leave SUCC alone.
(c) If SUCC is better than current state, then set current state to SUCC.

4.8 Advantages and disadvantages of hill climbing

Advantages
The following are some of the advantages of the hill climbing algorithms.
• Hill climbing is very useful in routing-related problems like travelling salesmen
prob- lem, job scheduling, chip designing, and portfolio management.
• It is good in solving optimization problems while using only limited computation
power.
• It is sometimes more efficient than other search algorithms.
• Even though it may not give the optimal solution, it gives decent solutions to
compu- tationally challenging problems.

Disadvantages
Both the basic hill climbing and the steepest-ascent hill climbing may fail to produce a
solution. Either the algorithm terminates without finding a goal state or getting into a
state from which no better state can be generated. This will happen if the programme has
reached either a local maximum, a ridge or a plateau.
• Local optima (maxima or minima): A local maximum is a peak that is higher
than each of its neighboring states, but lower than the global maximum (see Figure
4.10). A solution to this problem is to backtrack to an earlier solution and try going
in a different direction.
• Ridges: A ridge is a sequence of local maxima. Ridges are very difficult to
navigate for a hill-climbing algorithm (see Figure 4.17). This problem may be
overcome by moving in several directions at once.
 4.9. SIMULATED ANNEALING 78

• Plateaus: A plateau is an area of the state space where the evaluation function is
flat. It can be a flat local maximum, from which no uphill path exists, or a shoulder,
from which progress is possible (see Figure 4.16). One solution to handle this
situation is to make a big jump in some direction to get to a new part of the search
space.

Figure 4.16: Local and global maxima

Figure 4.17: A ridge: The ridge is the curved line at the top joining the local maxima

4.9 Simulated annealing

Simulated annealing is a variation of hill climbing. In simulated annealing, some down-hill
moves may be made at the beginning of the process. This to explore the whole space at
the beginning so that we do not land in a local maximum, or plateau or ridge instead of
the global maximum. The algorithm is somewhat analogous to a process known as
“annealing” in metallurgy.
the general principle of simulated annealing is illustrated in Figure 4.18
4.9. SIMULATED ANNEALING 79

Figure 4.18: General principle of simulated annealing

4.9.1 Annealing
The name of the algorithm comes from annealing in metallurgy. Annealing is a heat
treat- ment that alters the physical and sometimes chemical properties of a material to
increase its ductility and reduce its hardness, making it more workable. It involves
heating a material above a certain temperature, maintaining a suitable temperature for an
appropriate amount of time and then cooling.

4.9.2 Algorithm
Notations and conventions
The following notations and conventions are used in the algorithm.
• A variable BEST-SO-FAR which takes a problem state as a value.

• A certain function of time denoted by T (t) or simply T . This function is called the
annealing schedule.

• AS per standard usage, in the context of the simulated annealing algorithm, the
heuristic function will be called an objective function.
• It will be assumed that the problem is a minimisation problem

Algorithm
1. Evaluate the initial state. If it is also a goal state, then return it and quit. Otherwise,
continue with the initial state as the current state.
2. Initialise BEST-SO-FAR to the current state.

3. Initialise T according to the annealing schedule.

4. Loop until a solution is found or until there are no new operators left to be applied
in the current state.
4.10. SAMPLE QUESTIONS 80

(a) Select an operator that has not yet been applied to the current state and apply
it to produce a new state.
(b) Evaluate the new state. Compute

∆E = (value of current state) − (value of new state).

• If the new state is a goal state, then return it and quit.

• If it is not a goal state but is better than the current state, then make it the
current state. Also set BEST-SO-FAR to this new state.
• If it is not better than the current state then make it the current state with
probability e−∆E/T .
5. Return BEST-SO-FAR as the answer.

4.10 Sample questions

(a) Short answer questions
1. Define a heuristic function and an admissible heuristic function and illustrate it
with an example.

2. Give a heuristic function for the problem of path search in maps. Is it admissible?
Why?

3. Give an admissible and nontrivial heuristic for the missionaries and cannibals prob-
lem.

4. Give a nontrivial heuristic for the blocks world problem. Is it admissible?

5. Define the Hamming distance and Manhattan distance heuristic functions for the 8-
puzzle.

6. Given the 8-puzzle specified in Figure 4.19, what are the values of the Hamming
dis- tance and Manhattan distance heuristic functions at the initial state?

8 7 1 2 3
1 3 2 4 5 6
6 5 4 7 8

(a) Initial state (b) Goal state

Figure 4.19: An initial state and a goal state of 8-puzzle

7. Define a nontrivial heuristic function for the blocks world problem and compute its
value at the initial state for the problem shown in Figure 4.20.
4.10. SAMPLE QUESTIONS 81

Figure 4.20: A block world problem

(b) Long answer questions

1. Compare and contrast the greedy best first and the A∗- best first algorithms.
2. Write the simple hill climbing algorithm and illustrate it with an example.

3. Write the steepest ascent hill climbing algorithm.

4. Explain the simulated annealing algorithm.

5. Explain the following types of hill climbing:

(a) Simple hill climbing.

(b) Steepest-ascent hill climbing

6. What are the advantages and disadvantages of the hill climbing algorithm?

7. Consider a blocks world with 4 blocks A, B, C and D with the start and goal states
given in Figure 4.21. Solve the problem using the hill climbing algorithm and a
suitable heuristic function. Show the intermediate decisions and states.

A D
D C
C B
B A

Start Goal

Figure 4.21: A block world problem

8. Using the greedy best first search algorithm, find an optimal path from S to G in
the search graph shown in Figure 4.22. Show in a tabular form the various stages in
the execution of the algorithm with details of the current node, children of the
current node, the contents of the OPEN list and the contents of the CLOSED list.
4.10. SAMPLE QUESTIONS 82

Figure 4.22: A search graph

Use the heuristic
function:

Node (n) A B C D E F H I S G
h(n) 12 4 7 3 8 2 4 9 13 0

9. Solve the previous problem using the A∗ best first search algorithm.
10. Consider the graph shown in Figure 4.23. In the graph, “S” is the start state and
“G1, G2, G3” are three goal states. When traversing the graph, one can move only
in the direction indicated by the arrows. The numbers on the edges are the step-cost
of traversing that edge. the numbers in the nodes are estimated cost to the nearest
goal state. Apply each of the algorithms below to find a solution path from the start
state to a goal state and to calculate the total cost to the goal.

(a) Depth-first search

(b) Greedy best-first search

Figure 4.23: A search graph

11. Use the graph in Figure 4.24 to find the path using A∗ (S is the start state and G!,
G2 are the goal states). Find the path from to start to goal using the greedy best
first
4.10. SAMPLE QUESTIONS 83

search also.

Figure 4.24: A search graph

Chapter 5

Game playing

A multiagent environment is an environment in which each agent needs to consider the

actions of other agents and how they affect its own welfare. In this chapter we cover
com- petitive environments, in which the agents’ goals are in conflict, giving rise to
adversarial search problems which are often known as games.

5.1 Games considered in AI

In artificial intelligence, the most commonly considered games are those that have the
fol- lowing properties:

• Deterministic: A game is deterministic if an action of a player leads to completely

predictable outcomes. Chess is an example of a determinist game: the rules allow
for no variation of outcome and there are no physical factors involved. Football,
on the other hand, is an indeterministic game: players cannot reliably kick the ball
exactly the same way each time, and the small variations of that action lead to
chaotic outcomes.

• Turn-taking: A turn-taking game is one where the players play one at a time in
alternating turns.

• Two-player

• Zero-sum: A zero-sum game1 is defined as one where the total payoff to all
players is the same for every instance of the game.

• Perfect information: Perfect information implies knowledge of each player’s

utility functions, payoffs and strategies. Chess is an example of a game with
perfect infor- mation as each player can see all the pieces on the board at all times.
Other examples of games with perfect information include tic-tac-toe, checkers,
and Go.

Example 1: Chess
The best known example of a game considered in AI is chess. The reader is assumed to
be familiar with how the game of chess is played. The initial position is shown in Figure
5.1.
1
The terminology of zero-sum game comes from mathematical game theory a branch of applied mathematics
extensively used in economics.

84
5.2. GAME TREE 85

Figure 5.1: Chess board in the initial position

In chess, the outcome is a win, loss, or draw, with values +1, 0, or 1/2. It is a zero-
sum game because every game has the total payoff of either 0 + 1, 1 + 0 or 1/2 +
1/2.

Example 2: Tic-tac-toe

Figure 5.2: Tic-tac-toe

Tic-tac-toe or noughts and crosses is a paper-and-pencil game for two players, X and
O, who take turns marking the spaces in a 3 × 3 grid. The player who succeeds in placing
three of his/her marks in a diagonal, horizontal, or vertical row is the winner (see Figure
5.2).

5.2 Game tree

We illustrate the idea with an example. Figure 5.3 shows part of the game tree for the
tic-tac-toe game.

Example
Consider the game of tic-tac-toe and let the players be named “MAX” and “MIN”. MAX
plays by placing an X and MIN plays by placing an O in a cell in the grid.
In the figure the initial state is represented by a blank grid at the top. From the initial
state, MAX has nine possible moves. These are indicated in the figure by the nine
children of the initial state. Play alternates between MAX’s placing an X and MIN’s
placing an
O. This is continued until we reach leaf nodes corresponding to terminal states such that
one player has three in a row or all the squares are filled. The number on each leaf node
indicates the utility value of the terminal state from the point of view of MAX; high
values are assumed to be good for MAX and bad for MIN (which is how the players get
their names).
5.3. SOLVING A GAME 86

Figure 5.3: A partial game tree for the game of tic-tac-toe

5.3 Solving a game

By solving a two-player zero sum game we mean determining what actions are to be
taken by the players when they are at particular states in playing the game. We
assume that the players are “rational” by which we mean that each player would play to
secure the optimal advantage. Since it is a zero-sum game, one player’s gain is
opponent’s loss. As a convention, the player who is trying to maximize the gain is called
the maximizing player and is MAX and the player who is trying to minimize the loss is
called the minimizing player and is MIN.
The general method to solve such game problems is based on the “minimax strategy”
whereby the player MAX chooses a move that would maximize his minimum gain and
player MIN chooses a move which would minimize his maximum loss. The proceudre
based on this strategy known as the “Minimax algorithm” is described in detail in Section
5.6 below.

Example 1
Solve a zero-sum two-player game represented by the game tree shown in Figure 5.4.

Solution
The steps in arriving at a solution are described below.

1. The first player is MAX and he is at state A.

2. There are three possible moves for MAX now: a1, a2, a3.

3. Suppose A chooses a1. The resulting state is B.

 5.3. SOLVING A GAME 87

Figure 5.4: A two-ply game

(a) Then the next move is of MIN. There are are three possible moves for MIN
from B, namely, b1, b2, b3. The respective losses are 3, 12, 8.
(b) Being rational, MIN tries to minimize his loss and co chooses b1 with a
possible loss of 3. This results in a gain of 3 for MAX.

4. Suppose A chooses a2. The resulting state is C.

(a) Then the next move is of MIN. There are are three possible moves for MIN
from B, namely, c1, c2, c3. The respective losses are 2, 4, 6.
(b) Being rational, MIN tries to minimize his loss and co chooses c1 with a
possible loss of 2. This results in a gain of 2 for MAX.

5. Suppose A chooses a3. The resulting state is D.

(a) Then the next move is of MIN. There are are three possible moves for MIN
from B, namely, d1, d2, d3. The respective losses are 14, 5, 2.
(b) Being rational, MIN tries to minimize his loss and co chooses d3 with a
possible loss of 2. This results in a gain of 2 for MAX.

6. Now, if MAX chooses a1 the minimum gain is 3, if MAX chooses a2 the minimum
gain is 2 and a3 is chosen the minimum gain is 2.

7. Hence to maximize the minimum gain, MAX chooses

a1. The solution can be stated as follows:

“Initially, MAX chooses a1 and then MIN chooses b1. This guarantees a mini- mum
gain of 3 for MAX and a maximum loss of 3 for MIN.”

Example 2
Solve the game represented by the game tree in Figure 5.5. Assume that the maximising
player plays first.
5.4. FORMULATION OF A GAME AS A SEARCH PROBLEM 88

Figure 5.5: A two-ply game

Solution
We have:

Minimum loss for MIN from move from B = min{9, −6, 0} = −6

Minimum loss for MIN from move from C = min{8, −2} = −2
Minimum loss for MIN from move from D = min{−4, −3} = −4
Maximum gain for MAX from move from A = max{−6, −2, −4} = −2.

The solution can be stated as follows:

“Initially, MAX chooses the move AC and then MIN chooses CI. This guaran-
tees a minimum gain of -2 for MAX and a maximum loss of -2 for MIN.”

5.4 Formulation of a game as a search problem

We consider games with two players whom we call MAX and MIN. MAX moves first,
and then they take turns moving until the game is over. At the end of the game, points are
awarded to the winning player and penalties are given to the loser.
A game can be formally defined as a kind of search problem with the following ele-
ments:

• S0: The initial state, which specifies how the game is set up at the start.

• PLAYER(s): Defines which player has the move in a state s.

• ACTIONS(s): Returns the set of legal moves in the state s.

• RESULT(s, a): The transition model, which defines the result of a move a in state
s

• TERMINAL-TEST(s): A boolean function of the state s. It’s value is true when

the game is over and false otherwise. States where the game has ended are called
terminal states.

• UTILITY(s, p): A utility function (also called an objective function or payoff

func- tion), defines the final numeric value for a game that ends in terminal state s
for a player p.
 5.5. THE MINIMAX VALUE 89

A zero-sum game is defined as one where the total payoff to all players is the same for every
instance of the game.
The initial state, ACTIONS function, and RESULT function define the game tree for
the game-a tree where the nodes are game states and the edges are moves.

Example

Figure 5.6: A two-ply game

To digest the above notations, let us examine what the above elements are in the game
represented by the tree in Figure 5.6.

S0 A
PLAYER(B) MIN
ACTIONS(B) {b1, b2, b3}
RESULT(C, c2) I
TERMINAL-TEST(B) False
TERMINAL-TEST(K) True
UTILITY(E, MIN ) 3

5.5 The minimax value

The minimax value at a state s, denoted by MINIMAX(s) is defined as follows:

MINIMAX(s)
UTILITY(s) if TERMINAL-TEST(s)

 max MINIMAX(RESULT(s, a)) if PLAYER(s) =
=  a ACTIONS(s)
∈ MAX
a∈ min (s) MINIMAX(RESULT(s, a)) if PLAYER(s) = MIN
ACTIONS

Example
Consider a two-player game with the game tree as shown in Figure 5.6. The ∆-nodes are
“MAX nodes” in which it is MAX’s turn to play and the ∇-nodes are the “MIN nodes”.
The terminal nodes sho the utility values for MAX. The minimax values of the other
nodes
5.6. THE MINIMAX ALGORITHM 90

are calculated as follows:

MINIMAX(B) = min MINIMAX(RESULT(B, a))
a∈{b1,b2,b3}
= min{MINIMAX(E), MINIMAX(F), MINIMAX(G)}
= min{3, 12, 8}
= 3.

MINIMAX(C) = min
a∈{c1,c2,c3}
MINIMAX(RESULT(C, a))
= min{MINIMAX(H), MINIMAX(I), MINIMAX(J)}
= min{2, 4, 6}
= 2.

MINIMAX(D) = min
a∈{d1,d2,d3}
MINIMAX(RESULT(D, a))
= min{MINIMAX(K), MINIMAX(L), MINIMAX(M)}
= min{14, 5, 2}
= 2.

MINIMAX(A) = max
a∈{a1,a2,a3}
MINIMAX(RESULT(A, a))
= min{MINIMAX(B), MINIMAX(C), MINIMAX(D)}
= min{3, 2, 2}
= 3.

5.6 The minimax algorithm

The minimax algorithm computes the minimax decision from the current state.

5.6.1 The algorithm

function MINIMAX-DECISION (state) returns an action

return arg maxa∈ACTIONS(s) MIN-VALUE (RESULT (state, a))
function MAX-VALUE (state) returns a utility value
if TERMINAL-TEST(state) then return UTILITY(state)
v ← −∞
for each a in ACTIONS(state) do
v ← MAX (v, MIN-VALUE (RESULT (s, a)))
return v
function MIN-VALUE(state) returns a utility value
if TERMINAL-TEST(state) then return UTILITY(state)
v←∞
for each a in ACTIONS(state) do
v ← MIN(v, MAX-VALUE(RESULT (s, a)))
return v
5.6. THE MINIMAX ALGORITHM 91

Remarks
1. The algorithm is presented using the pseudocode employed by Russel and Norvig
in their book “Articial Intelligence: A Modern Approach”.
2. The algorithm determines the action to be taken when the game is at state. To
determine the action to be taken at the initial state the function value MINIMAX-
DECISION(initial state) has to be computed.
3. In the algorithm the notation “←” denotes the assignment operator: “x ← y” means
assign y to x.
4. The notation arg maxa∈Sf (a) computes the element a of set S that has the maximum
value of f (a). For example, let

S = {1, 2, 3, 4, 5}
f (1) = 28, f (2) = 39, f (3) = 38, f (4) = 32, f (5) =
30
then

arg maxa∈S f (a) = 2.

5.6.2 Example
Problem
Consider a perfect two player game with the game tree shown in Figure 5.7. Assuming
that the root is a maximizing node, apply the minimax algorithm to determine the optimal
initial action for the MAX player.

Figure 5.7: A game tree

Solution
We are required to compute MINIMAX-DECISION(A). According to the minimax algo-
rithm, this is given by
MINIMAX-DECISION(A) = arg max MIN-VALUE(RESULT(A,a)).
a∈ACTIONS(A)

Now from the game tree we can see that

ACTIONS(A) = {AB, AC}.

5.6. THE MINIMAX ALGORITHM 92

and also that

RESULT(A, AB)=B, RESULT(A, AC)=C.
Thus

MINIMAX-DECISION(A) = arg max {MIN-VALUE(B), MIN-VALUE(C)}.

a∈ACTIONS(A)

1. Next we compute MIN-VALUE(B). Since B is not a terminal state and since ACTIONS(B)=
{BD, BE}, MIN-VALUE(B) is computed as follows:
v←∞
v ← MIN(v, MAX-VALUE(RESULT(B, BD))
v ← MIN(v, MAX-VALUE(RESULT(B, BE))
MIN-VALUE(B) ← v.

(a) v ← ∞
(b) We have to compute
MAX-VALUE(RESUL(B,BD))=MAX-VALUE(D).
Since ACTIONS(D)= { DH, DI} and since D is not a terminal state MAX-
VALUE(D) is computed as follows:
w ← −∞.
w ← MAX(w, MIN-VALUE(RESULT(D,
DH)) w ← MAX(w, MIN-
VALUE(RESULT(D, DI)) MAX-VALUE(D)
← w.
(i) w ← −∞.
(ii) We compute
MIN-VALUE(RESUL(D, DH))=MAX-VALUE(H).
Since H is a terminal state, we have
MAX-VALUE(H) = UTILITY(H) = -
1.
Thus we have
w ← MAX(−∞, -1)
w ← −1
(iii) We compute
MIN-VALUE(RESUL(D, DI))=MAX-VALUE(I).
Since I is a terminal state, we have
MAX-VALUE(I) = UTILITY(I) =
4.
Thus we have
w ← MAX( -1, 4)
w←4
(iv) Finally we get
MAX-VALE(D)← 4.
v ← MIN(v, MAX-VALUE(RESULT(B, BD))).
that is, v ← MIN(∞, 4), that is v ← 4.
(c) We compute
MAX-VALUE(RESUL(B,BE))=MAX-VALUE(E).
5.7. MINIMAX WITH ALPHA-BETA PRUNING 93

Since E is not a terminal state, and since ACTIONS(E)= {EJ, EK}, MAX-
VALUE(E) is computed as follows:
w ← −∞
w ← MAX(w, MIN-VALUE(RESULT(E, EJ))
w ← MAX(w, MIN-VALUE(RESULT(E,
EK)) MAX-VALUE(E) ← w.
(i) w ← −∞
(ii) Let us compute the
assignment We compute
MIN-VALUE(RESUL(E, EJ))=MAX-VALUE(J).
Since J is a terminal state, we have
MAX-VALUE(J) = UTILITY(J) =
2.
Thus we have
w ← MAX(−∞, 2)
w←2
(iii) We compute
MIN-VALUE(RESUL(E, EK))=MAX-VALUE(K).
Since K is a terminal state, we have
MAX-VALUE(K) = UTILITY(K) =
6.
Thus we have
w ← MAX(2, 6)
w←6
v ← MIN(v, MAX-VALUE(RESULT(B, BE)))
that is, v ← MIN(4, 6), that is v ← 4.
(d) We now get
MIN-VALUE(B) ← 4.

2. In a similar way we can compute MIN-VALUE(C) ← −3.

Finally
MINIMAX-DECISION(A) = arg max {MIN-VALUE(B), MIN-VALUE(C)}
a∈ACTIONS(A)
= arg max {4, −3}
a∈ACTIONS(A)
= 4.
The action in the set ACTIONS(A) which produces this minimax value is AB and so the
player must start with the action AB.

5.7 Minimax with alpha-beta pruning

AlphaâA˘ S¸ beta pruning is a search algorithm that seeks to decrease the number of
nodes that are evaluated by the minimax algorithm in its search tree. It is an adversarial
search algorithm used commonly for machine playing of two-player games (Tic-tac-toe,
Chess, Go, etc.). It stops evaluating a move when at least one possibility has been found
that proves the move to be worse than a previously examined move. Such moves need
not be evaluated further.
5.7. MINIMAX WITH ALPHA-BETA PRUNING 94

5.7.1 Basic idea

The algorithm maintains two values, α and β, which respectively represent the minimum
score that the maximizing player is assured of and the maximum score that the
minimizing player is assured of. Initially we set α = −∞ and β = +∞. Whenever the
maximum score that the minimizing player (that is, the β-player) is assured of becomes
less than the minimum score that the maximizing player (that is, the α-player) is
assured of (that is, β < α), the maximizing player need not consider further descendants
of this node, as they will never be reached in the actual play.

5.7.2 Example
Example 1
Figure 5.8 represents a game tree showing which nodes can be pruned. Suppose the values

Figure 5.8: Alpha-beta pruning: Example

of the leaf nodes are given or are computed given the definition of the game. The
numbers at the bottom show some of these values. The other values are irrelevant, as we
show here.

1. The value of node h is 7, because it is the minimum of 7 and 9.

2. Just by considering the leftmost child of i with a value of 6, we know that the value
of i is less than or equal to 6.

3. Therefore, at node d, the maximizing agent will go left. We do not have to evaluate
the other child of i.

4. Similarly, the value of j is 11, so the value of e is at least 11, and so the minimizing
agent at node b will choose to go left.

5. The value of l is less than or equal to 5, and the value of m is less than or equal
to 4; thus, the value of f is less than or equal to 5, so the value of c will be less
than or equal to 5.

6. So, at a, the maximizing agent will choose to go left.

Notice that this argument did not depend on the values of the unnumbered leaves.
Moreover, it did not depend on the size of the subtrees that were not explored.
5.7. MINIMAX WITH ALPHA-BETA PRUNING 95

Example 2
Consider the game tree shown in Figure 5.6. We show below why the leave c2 and c3 can
be pruned without affecting the final decision.
Figure 5.9 shows the various stages in the calculation of the optimal decision for the
game tree in Figure 5.6. At each point, we also show the range of possible values for each
node.

Figure 5.9: Stages in the calculation of the optimal decision for the game tree in Figure 5.6

(a) The first leaf below B has the value 3. Hence, B, which is a MIN node, has a
value of at most 3.

(b) The second leaf below B has a value of 12; MIN would avoid this move, so the
value of B is still at most 3.

(c) The third leaf below B has a value of 8; we have seen all B’s successor states,
so the value of B is exactly 3. Now, we can infer that the value of the root is at
least 3, because MAX has a choice worth 3 at the root.

(d) The first leaf below C has the value 2. Hence, C, which is a MIN node, has a
value of at most 2. But we know that B is worth 3, so MAX would never
choose
C. Therefore, there is no point in looking at the other successor states of C. This
is an example of alpha-beta pruning.

(e) The first leaf below D has the value 14, so D is worth at most 14. This is still
higher than MAX’s best alternative (that is, 3), so we need to keep exploring
D’s successor states. Notice also that we now have bounds on all of the
successors of the root, so the root’s value is also at most 14.
 5.7. MINIMAX WITH ALPHA-BETA PRUNING 96

(f) The second successor of D is worth 5, so again we need to keep exploring. The
third successor is worth 2, so now D is worth exactly 2. MAX’s decision at the
root is to move to B, giving a value of 3.

Example 3
Figure 5.10 illustrates alpha-beta pruning. The grayed-out subtrees are not explored (when
moves are evaluated from left to right). The max and min levels represent the turn of the
player and the adversary, respectively.

Figure 5.10: Alpha-beta pruning: Example

5.7.3 Algorithm
The minimax with alpha-beta pruning algorithm is given below.

Algorithm
01. function alphabeta(node, α, β, Player)
02. if node is a terminal node then
03. return the value of node
04. if Player == MAX then
05. value := −∞
06. for each child of node do
07. value := max(value, alphabeta(child, α, β, MIN))
08. α := max(α, value)
09. if α ≥ β then
10. break (* β cutoff *)
11. return value
12. else
13. value := +∞
14. for each child of node do
15. value := min(value, alphabeta(child, α, β, MAX))
16. α := min(β, value)
17. if α ≥ β then
18. break (* α cutoff *)
19. return value

The algorithm is initially invoked by issuing the command:

5.8. SAMPLE QUESTIONS 97

alphabeta(InitialState, −∞, +∞, MAX)

5.7.4 Additional example

Figure 5.11: Alpha-beta pruning: Example

Solution

Figure 5.12: Alpha-beta pruning: Stages (1) and (2)

Figure 5.13: Alpha-beta pruning: Stages (3) and (4)

5.8 Sample questions

(a) Short answer questions
1. Explain the concept of a game tree with an example.
5.8. SAMPLE QUESTIONS 98

Figure 5.14: Alpha-beta pruning: Stages (5) and (6)

Figure 5.15: Alpha-beta pruning: Stages (7) and (8)

Figure 5.16: Alpha-beta pruning: Stages (9) and (10)

Figure 5.17: Alpha-beta pruning: Stages (11) and (12)

2. Define a zero-sum game.

3. Define the minimax value at a state s in a game.

4. Compute MINIMAX(S) in the game tree shown in Figure 5.18.

5. Why the alpha-beta pruning method is better than the minimax search method in
5.8. SAMPLE QUESTIONS 99

MAX S

MIN A B

Terminal C D E F G
5 3 8 4 2
Figure 5.18: A game tree

solving a game?

(b) Long answer questions

1. Explain the minimax algorithm for computing the minimax decision.
2. Explain with an example the alpha-beta pruning algorithm.
3. Write the alpha-beta pruning algorithm.
4. Consider a two-player game in which the minimax search procedure is used to
com- pute the best move for the first player. Assume the following game tree in
which the scores are from the first player’s point of view. Suppose the first player
is the max- imising player and needs to make the first move. What move should be
chosen at this point?

Figure 5.19: A game tree

5. Determine which of the branches in the game tree shown in Figure 5.20 will be
pruned if we apply alpha-beta pruning to solve the game.

MAX A

MIN B C

MAX D E F G
Terminal H I J K L M N O
3 5 6 9 1 2 0 -1

Figure 5.20: A game tree

Chapter 6

Knowledge representation

6.1 Knowledge
Knowledge is awareness or familiarity gained by experiences of facts, data, and
situations. Knowledge is a familiarity, awareness, or understanding of someone or
something, such as facts (propositional knowledge), skills (procedural knowledge), or
objects (acquaintance knowledge). By most accounts, knowledge can be acquired in
many different ways and from many sources, including but not limited to perception,
reason, memory, testimony, scientific inquiry, education, and practice. The philosophical
study of knowledge is called epistemology.

6.1.1 Knowledge in AI
The following are the kinds of knowledge which need to be represented in AI systems:

• Object:
All the facts about objects in our world domain. E.g., Guitars contains strings,
trum- pets are brass instruments.

• Events:
Events are the actions which occur in our world.

• Performance:
It describe behavior which involves knowledge about how to do things.

• Meta-knowledge:
It is knowledge about what we know.

• Facts:
Facts are the truths about the real world and what we represent.

• Knowledge base:
It is the totality of information, structured or unstructured, stored in a system.

100
6.2. REPRESENTATION 101

6.1.2 Different types of knowledge

1. Declarative knowledge
Declarative knowledge, also known as descriptive knowledge, is to know about
some- thing such as concepts, facts, and objects. It is expressed in declarative
sentences.

2. Procedural knowledge
Procedural knowledge, also known as imperative knowledge, is a type of
knowledge which is responsible for knowing how to do something. It can be
directly applied to any task and it includes rules, strategies, procedures, agendas,
etc.

3. Meta-knowledge
Knowledge about the other types of knowledge is called meta-knowledge.

4. Heuristic knowledge
Heuristic knowledge is knowledge of experts in a field or subject. Heuristic knowl-
edge is a “rule of thumb” knowledge based on previous experiences, awareness of
approaches, and which are good to work but not guaranteed.

5. Structural knowledge
It describes relationships between various concepts such as kind of, part of, and
grouping of something. It describes the relationship that exists between concepts
or objects.

6.2 Representation
By “representation” we mean a relationship between two domains where the first is
meant to “stand for” or take place of the second. Usually the first domain is more
concrete, or accessible in some way than the second. For example, a drawing of a
milkshake and a hamburger on a sign might stand for a less visible fast food restaurant.

6.3 Knowledge representation

Knowledge representation is a field of artificial intelligence that focuses on designing
computer representations that capture information and knowledge about the world that
can be used to solve complex problems. Knowledge representation is not just storing data
into some database, but it also enables an intelligent machine to learn from that
knowledge and experiences so that it can behave intelligently like a human. Virtually all
knowledge representation languages have a reasoning or inference engine as part of the
system.
A knowledge representation structure (or, knowledge representation system) is a
particular set of definitions, rules and procedures for setting up a representation that
captures information and knowledge about the world.

6.4 Desirable properties of knowledge representation systems

The list below gives the important desirable properties of knowledge representation sys-
tems. Unfortunately there is no single knowledge representation systems that that have
6.5. SOME KNOWLEDGE REPRESENTATION SYSTEMS 102

all the desired properties. Because of this, there are different knowledge representations
systems having varying capabilities.

1. Representational adequacy
This the ability to represent all kinds of knowledge that are needed in a domain.

2. Inferential adequacy
This is the ability to derive new knowledge from old.

3. Inferential efficiency
This is the ability to incorporate into the knowledge base additional information.

4. Acquisitional efficiency
This is the ability to acquire new knowledge easily.

6.5 Some knowledge representation systems

There are several knowledge representation systems such as the following.

• Semantic networks

• Frames

• Conceptual dependencies

• System based on logic

6.6 Semantic networks

A network is an interconnected group or system. The word “semantic” means “related to
meaning, especially meaning in language”. Thus a “semantic network” is an
interconnected system or group related to meaning. Such a system can be represented by
a directed labelled graph. Semantic networks are a logic-based formalism for knowledge
representation.

6.6.1 Definition
A semantic network is a graph constructed from a set of vertices (or nodes) and a set of
directed and labelled edges. The vertices or nodes represent concepts or objects, and the
edges represent relations between the nodes.

6.6.2 Examples
Example 1
Consider the knowledge contained in the following sentence:

S: “Sparrow is a bird.”
6.6. SEMANTIC NETWORKS 103

There are two concepts in the sentence, namely, “Sparrow” and “Bird”. The relation be-
tween these concepts is indicated by “is a”. We represent the two concepts by two nodes
in a graph and the relation between them by a directed edge with the label “is-a”. (We
have represented the “is a” relation by “is-a” as it is the convention in the theory of
semantic networks.) The knowledge contained in S can now be represented by the graph
shown in Figure 6.1. This is the semantic network representation of the knowledge
contained in S.

Sparrow is-a
Bird

Figure 6.1: Semantic network representation of “Sparrow is a bird”

Example 2
Consider the knowledge contained in the following sentences:
S: “Tweety is a bird.”, “Birds are animals.”
The sentences has three objects “Tweety”, “Birds” and ”Animals”. Tweety is the name of
a particular bird. So the sentence “Tweety is a bird” means that the bird Tweety is an
instance of the class of things indicated by the word “Birds”. The sentence “Birds are
animals” means that all members of the class of things indicated by ”Birds” are also
members of the class of things indicated by “Animals”.
To represent the knowledge contained in S by a semantic network, we need three ver-
tices to denote the three objects. These vertices are to be connected by directed edges
with appropriate labels. The knowledge contained in S is represented by the graph
exhibited in Figure 6.2.

subset-of
Tweety instance-of Birds Animals

Figure 6.2: Semantic network representation of the sentences “Tweety is a bird.” and “Birds
are animals.”
We may represent the nodes as rectangles as in Figure 6.3.
instance-of subset-of
Tweety Birds Animals

Figure 6.3: Semantic network representation of the sentences “Tweety is a bird.” and “Birds
are animals.”

Example 3
Consider the sentence:
S: “A bird has feathers.”
This sentence means that the object “Feather” is part of the the object “Bird”. In a
semantic network representation of the sentence, the edge joining the nodes is labelled
“HAS-A” (or “HASA”, or “PART-OF”) as in Figure 6.4. Such a relation is referred to as
a part-whole relation. The link or edge is from whole to part.
6.6. SEMANTIC NETWORKS 104

HAS-A
Bird Feather

Figure 6.4: Semantic network representation of “A bird has feathers”

Example 4
Consider the sentence:

S: “Bill is taller than John.”

is-taller-than
Bill John

Figure 6.5: Inappropriate semantic network representation of “Bill is taller than John”

This representation of the sentence does not represent the knowledge expressed by the
sen- tence because the relation “is-taller-than” cannot be taken as a basic relation. What
the sentence actually means is that the height of Bill is greater than than the height of
John. Thus the proper semantic representation of the sentence “Bill is taller than John” is
the one given Figure 6.6. In this figure we have used the basic relation “is-greater-than”
and also the relation “height”.

is-greater-than
h1 h2

height height

Bill John

Figure 6.6: Semantic network representation of “Bill is taller than John”

Example 5
Consider the sentence:

S: “Sara has brown eyes.”

This sentence has three concepts “Sara”, “Brown” and “Eyes”. “Brown” is a property of
“Eyes” and, in a semantic representation of the sentence, the property can be denoted by
the label “Property” or more specifically by the label “Colour”. The semantic network is
shown in Figure 6.7.

HAS Eyes Colour

Sara Brown

Figure 6.7: Semantic network representation of the sentence “Sara has brown eyes.”
6.6. SEMANTIC NETWORKS 105

Example 6
Consider the knowledge represented by the following sentences:

• Tweety and Sweety are birds.

• Tweety has a red beak.

• Sweety is Tweety’s child.

• A crow is a bird.

• Birds can fly.

The semantic network representation of the sentences is given in Figure 6.8. Note the use
of the labels “Property”, “Child-of” and “Color” in the network.

Figure 6.8: A semantic network

Example 7
Consider the following sentences:

• Motor bike is a two wheeler.

• Scooter is a two wheeler.

• Two wheeler is a moving vehicle.

• Moving vehicle has a brake.

• Moving vehicle has a engine.

• Moving vehicle has electrical system.

• Moving vehicle has fuel system.

The semantic network representation of the sentences is given in Figure 6.9.

6.6. SEMANTIC NETWORKS 106

IS-A IS-A
Scooter Two-wheeler Motor-bike

IS-A
HAS-A HAS-A
Breaks Moving vehicle Engine

HAS-A HAS-A

Electrical system Fuel system

Figure 6.9: A semantic network

Example 8
Figure 6.10 shows another simple semantic network. Note the use of the label “is” instead
of “IS-A”. This is perfectly permitted as there are no standardisation of the notations for
labels. Also note the labels “lives in” and “works in”.

Thrissur

is lives in

City is
John Person

is works in

Cochin

Figure 6.10: A simple semantic network

Example 9
Figure 6.11 shows a semantic network with several objects and categories.

Figure 6.11: A semantic network

6.6. SEMANTIC NETWORKS 107

Example 10
Figure 6.12 shows a semantic network with two objects (John and Mary) and four
categories (Mammals, Persons, FemalePersons and MalePersons).

Figure 6.12: A semantic network

Example 11
Figure 6.13 shows a semantic network with “IS-A” and “AKO” (A-KIND-OF) links. The
link AKO is used to relate one class (or category) to another. AKO relates generic nodes
to generic nodes while the IS-A relates an instance or individual to a generic class.

Figure 6.13: A semantic network

Example 12
Figure 6.14 illustrates a semantic network where the principal concept is “financial prod-
uct,” which can be a loan or an investment. An example of a long-term loan is a
mortgage, and an example of a low-risk investment is a government bond. Note the
variety of labels used to represent the various relations. Note also the presence of two-
way links connecting nodes.
6.6. SEMANTIC NETWORKS 108

Figure 6.14: A semantic network

6.6.3 Remarks
It appears that there are no universally agreed conventions on the use of geometrical
shapes to represent nodes. Ovals, circles and rectangles are all used to represent nodes.
Also, there is no complete set of standard terminology to label the edges in semantic
graphs. However, the following are some of the most common semantic relations used in
semantic networks.

• IS-A (ISA)

• IS-PART-OF

• IS-SUBSET-OF

• A-KIND-OF (AKO)

6.6.4 Different kinds of semantic networks

The following are six of the most common kinds of semantic networks:

1. Definitional networks
These emphasize the subtype relation between a concept type and a newly defined
subtype.
6.6. SEMANTIC NETWORKS 109

Figure 6.15: A definitional semantic network

2. Assertional networks
Assertional semantic networks, also known as propositional semantic networks
are designed to represent assertions or propositions. Figure 6.16 shows a semantic
net- work representation of the proposition “All men are mortal”. In the figure,
node M1 corresponds to the proposition that “All men are mortal.” Node V1 is the
variable cor- responding to “all men.” Node M2 is the proposition that corresponds
to the arbitrary man being a member of the class of men.

Figure 6.16: An assertional semantic network

3. Implicational networks
These use implication as the primary relation for connecting nodes.

Figure 6.17: An implicational semantic network

6.6. SEMANTIC NETWORKS 110

4. Executable networks
These include some mechanism which can perform inferences, pass messages, or
search for patterns and associations.

Figure 6.18: An executable semantic network

5. Learning networks
These build or extend their representations by acquiring knowledge from examples.

Figure 6.19: A learning semantic network

6. Hybrid networks
These combine two or more of the previous techniques.

6.6.5 Intersection search

One of the earliest ways that semantic networks were used was to find relationships
between objects by spreading activation from each of two nodes and seeing where the
activations met. This process is called intersection search.
For example, given the semantic network shown in Figure 6.20 consider the question:
“What is the relation between Chicago Cubs and Brooklyn Dodgers?”
By spreading activations from the nodes Chicago Cubs and Brooklyn Dodgers, we
see that they meet at the node Baseball Players. Thus the question can be answered as
“They are both teams of baseball players”
6.7. ADVANTAGES AND DISADVANTAGES OF SEMANTIC NETWORKS 111

Figure 6.20: A learning semantic network

6.6.6 Representing nonbinary predicates

To represent a proposition involving a nonbinary relation in semantic nets, we create a
node to representing the particular proposition and then relate the elements of the
proposition to the newly created node.

Example 1
Consider the proposition: “John gave the book to Mary.” This is a proposition involving
three elements.
This proposition can be considered as an instance of the category of events called
”Give”. Let us call this instance as EV7. There are three components to this event,
namely, an agent (the giver), a beneficiary and an object. The object “the book” is a
particular in- stance of the the category “Book”. Let us denote this particular instance of
Book by BK23. The proposition can b now be represented by the semantic net shown in
Figure 6.21.

Figure 6.21: A semantic network representing the sentence: “John gave the book to Mary.”

Example 2
Consider the proposition: “The score in Cubs vs. Dodgers game was 5 - 3.” This is also a
proposition involving three elements. This may be represented as a semantic network as
in Figure 6.22. In the figure, “G23” is the particular instance of the category “Game”
played by Cubs and Dodgers.

6.7 Advantages and disadvantages of semantic networks

6.7.1 Advantages
1. They give an adaptable method of representing knowledge because many different
types of object can be included in the network.
 6.8. FRAMES 112

Figure 6.22: A semantic network representing the sentence: “The score in Cubs vs. Dodgers
game was 5 - 3.”

2. The network is graphical and therefore relatively easy to understand.

3. Can be used as a common communication tool between the knowledge engineer

and the human expert.

4. Efficient in space requirement.

5. Easily clusters related knowledge.

6.7.2 Disadvantages
1. There is no standard definition of link names.

2. Semantic networks are not intelligent, dependent on creator.

3. Links are not all alike in function or form.

4. Links on objects represent only binary options.

5. Undistinguished nodes that represent classes and that represent individual objects.

6. Processing is inefficient for large networks.

7. Do not represent performances effectively.

8. It is difficult to express some properties using semantic networks, like negation,

dis- junction, etc.

6.8 Frames
6.8.1 Definition
A frame is a collection of attributes and possible values that describe some entity in the
world. A frame system is a collection of frames that are connected to each other by the
fact that the value of an attribute in one frame may be another frame.

Remarks
1. The basic characteristic of a frame is that it represents related knowledge about a
narrow subject.

2. A frame system is a good choice for describing a mechanical device, for example a
car.
 6.8. FRAMES 113

3. Just as with semantic nets, there are no standards for defining frame-based systems.
4. A frame is analogous to a record structure, corresponding to the fields and values
of a record are the slots and slot fillers of a frame.
5. A frame is basically a group of slots and fillers that defines a stereotypical object.
6. A frame is also known as slot-filter knowledge representation in artificial intelli-
gence.

6.8.2 Frame structure

The frame contains information on how to use the frame, what to expect next, and what to
do when these expectations are not met. Some information in the frame is generally
unchanged while other information, stored in “terminals”, usually change. Terminals can
be considered as variables. Top level frames carry information, that is always true about
the problem in hand, however, terminals do not have to be true. Their value might change
with the new information encountered. Different frames may share the same terminals.
Each piece of information about a particular frame is held in a slot. The information
can contain:
• Facts or data
– Values (called facets)
• Procedures (also called procedural attachments)
– IF-NEEDED: deferred evaluation
– IF-ADDED: updates linked information
• Default values
– For data
– For procedures
• Other frames or subframes

6.8.3 Examples
Example 1
Table 6.1 is a frame for a book.

Slot Value
Title Artificial Intelligence
Genre Computer Science
Author Peter Norvig
Edition Third Edition
Year 1996
Page 1152

Table 6.1: A frame for a book

6.8. FRAMES 114

Example 2
Table 6.2 is the frame representation of the statement: “Peter is an engineer as a profession,
and his age is 25. He lives in city London, and the country is England.”

Slot Value
Name Peter
Profession Doctor
Age 25
Marital status Single
Weight 78

Table 6.2: A frame

Example 3
Table 6.3 displays another frame. In this frame, he car is the object, the slot name is the
attribute, and the filler is the value.

Slots Fillers
Manufacturer General Motors
Model Chevrolet Caprice
Year 1979
Transmission Automatic
Tyres 4
Colour Blue
engine Petrol

Table 6.3: A frame

Example 4
Table 6.4 defines a frame with name ALEX. The various slots in the frame, their names,
their values and their types are all specified in the table.
6.8. FRAMES 115

Slot Value Type

ALEX - (This Frame)
NAME Alex (key value)
ISA BOY (parent frame)
SEX Male (inheritance value)
AGE IF-NEEDED: (procedural
Subtract(current,BIRTHDATE); attachment)
HOME 100 Main St. (instance value)
BIRTHDATE 8/4/2000 (instance value)
FAVORITE_FOOD Spaghetti (instance value)
CLIMBS Trees (instance value)
BODY_TYPE Wiry (instance value)
NUM_LEGS 1 (exception)

Table 6.4: Example frame

The parent frame BOY of the frame named ALEX is shown in Table 6.5.

Slot Value Type

BOY - (This Frame)
ISA Person (parent frame)
SEX Male (instance value)
AGE Under 12 yrs. (procedural attachment
- sets
HOME A Place constraint)
NUM_LEGS Default = 2 (frame)
(default, inherited
from Person frame)

Table 6.5: Example frame

Example 5
Figure 6.23 shows a simplified frame system. There are two types of attributes that can
be associated with a class or set: those that are about the set itself and those inherited by
each element of the set. In the figure, the latter type of attributes are prefixed by an
asterisk (*).

6.8.4 Frame languages

A frame language is a technology used for knowledge representation in artificial intelli-
gence. The earliest Frame based languages were custom developed for specific research
projects and were not packaged as tools to be re-used by other researchers. One of the
first general purpose frame languages was KRL (Knowledge Representation Language).
One of
6.9. CONCEPTUAL DEPENDENCY 116

Figure 6.23: A frame system

the most influential early Frame languages was KL-ONE. KL-ONE spawned everal
subse- quent Frame languages. One of the most widely used successors to KL-ONE was
the Loom language developed by Robert MacGregor at the Information Sciences
Institute.

6.9 Conceptual dependency

Conceptual dependency (CD) is a theory of knowledge representation developed by
Roger Schank and his teammates at Yale University in the 1970’s. It is a theory to
represent natural language sentences in such way that:

• It is independent of the language in which the sentences are stated.

• It facilitates drawing inferences from sentences.

6.9.1 Simple examples of representations of sentences in CD theory

Example 1
Consider the English sentence:
6.9. CONCEPTUAL DEPENDENCY 117

“John ran”.

Let us analyse this sentence.

This sentence has tow words “John” and “ran”. The first word refers to a particular
person or a particular object. The second word is a verb and hence denotes an action. The
action is running, that is, moving from one place to another. In the terminology of CD
theory, we say that it is an example of a physical transfer of an object from one place to
another. We also say that it is an action of the PTRANS type. (PTRANS stands for
physical transfer.) In the sentence, “ran” depends on “John” and “John” depends on
“ran”. There is a mutual dependency.
In the notations and terminology of CD theory, we represent the sentence “John ran.”
in the following form:
p
John ⇐⇒ PTRANS
Here, the two-sided double-line arrow indicates mutual dependency and the letter “p”
above the arrow indicates that the action is in the past tense.
Further, in CD theory, we think of “John” as a member of the category of “real world
objects” which is denoted by “PP” which is an abbreviation for “picture provider”. Also,
PTRANS is a member of the category of real world actions which is denoted by “ACT”.
The representation of the sentence “John ran.” given above can be thought of as obtained
by an application of the following rule of dependency in CD theory:

PP ⇐⇒ ACT

Example 2
Consider the English
sentence:
“John gave Mary a book.”

Let us analyse this sentence.

There are four concepts in the sentence: “John”, ”gave”, ”Mary”, ”book”. In the sen-
tence the actor is “John”, the action is “gave” and the object of the action is ”book”. The
actor and the action are mutually dependent, but the action and object are not. Action de-
pends on the object, but the object does not depend on the action. So, the sentence ”John
gave a book” is represented as follows:

P O
John ⇐⇒ gave ←− book

Now, consider the action “gave”. It has two meanings: simply a physical transfer, or an
ownership transfer. In the former case it is an example of PTRANS. In the latter case it is
an abstract transfer which in CD theory is indicated by ATRANS. Assuming that it is the
latter, we may represent “John gave a book.” as

P O
John ⇐⇒ ATRANS ←− book

Next, let us consider the dependency of “Mary” on the other concepts in the sentence.
The dependency of “Mary” on “John” is one of a “recipient-donor” relationship and the
“book” is dependent on this relation. So the dependency can be represented as follows:
6.9. CONCEPTUAL DEPENDENCY 118

to
R Mary
book
John
from

In the diagram, R indicates a recipient-donor relationship. Putting all the components to-
gether, we obtain the CD representation of the sentence “John gave Mary a book.” as
shown below:
to
p O R Mary
John book
ATRANS John
from

Example 3
Consider the English sentence:

“John ate the ice cream with a spoon.”

The verb “ate” which is the past tense of “eat” is conceptualised as “ingesting” or “taking
in” and is indicated by the primitive action INGEST. In the sentence, the spoon is the
instrument of the action. In CD theory, an instrument of action is indicated by the letter
“I”. The sentence can be represented in the following form.
John
p O I
John INGEST ice cream
do
O
spoon

Remark
In any domain in which we build a conceptual representation system, we will have to
choose an appropriate level for primitive actions.

6.9.2 The building blocks of conceptual dependency

In the examples of conceptual representations discussed above, we have seen several
prim- itive actions like PTRANS, ATRANS and INGEST, the primitive category PP, the
symbol for the past tense, symbols indicating relations like o for object relation and R for
recipient- donor relation. In this section we have collected all these building blocks of the
conceptual dependency theory.

1. Primitives conceptual categories

The primitive conceptual categories available in the CD theory are given Table 6.6.
6.9. CONCEPTUAL DEPENDENCY 119

Primitive Meaning
ACT Real world actions. There are only eleven of these ACTs
(see Table 6.7).
PP Real world objects (picture producers). Only physical
objects are PPs.
AA Modifiers of actions (action aiders) (Attributes of actions)
PA Attributes of an object (picture aiders). PAs take the form:
STATE(VALUE). That is, a PA is an attribute characteristic
(like color or size) plus a value for that characteristic (red or
10 cm).
T Times
LOC Location

Table 6.6: Primitive conceptual categories in CD theory

2. The primitive acts of CD

The CD theory recognises 11 primitive acts. Table 6.7 gives these primitive acts, their
meanings and examples.

Table 6.7: Primitives of conceptual dependency theory

Primitive Meaning Example

ATRANS Transfer of an abstract relationship give
(Abstract TRANSfer)
PTRANS Transfer of the physical location of an object go
(Physical TRANSfer)
PROPEL Application of physical force to an object push
(PROPELling an object by applying physical force)
MOVE Movement of a body part by it’s owner kick
(MOVEment)
GRASP Grasping of an object by an actor throw
(GRASPing)
INGEST Ingesting of an object by an animal eat
(INGESTing)
EXPEL Expulsion of something from the body of an animal cry
(EXPElling)
MTRANS Transfer of mental information tell
(Mental TRANSfer)
MBUILD Building new information out of old decide
(Mentally BUILDing)
6.9. CONCEPTUAL DEPENDENCY 120

SPEAK Production of sounds say

(SPEAKing by producing sounds)
ATTEND Focusing of sense organ toward a stimulus listen
(ATTENDing to a stimulus)

3. Conceptual tenses
There are definite symbols in CD theory to indicate the tenses of verbs like, for example,
“p” for past tense. The available symbols are given in Table 6.8.

Symbol Tense Symbol Tense

p Past ? Interrogative
f Future / Negative
t Transition nil Present
ts Start transition delta Timeless
tf Finished transition c Conditional
k Continuing

Table 6.8: Conceptual tenses in conceptual dependency theory

4. Indicators of dependencies

(arrows) Arrow type

Meaning
−→, ←− Direction of dependency
Two-way links between the actor (PP) and the action (ACT)
Dependency between PP and PA

Table 6.9: Meanings of arrows in conceptual dependency theory

5. Symbols for conceptual cases

Symbol Meaning
o Objective case
R Recipient case
I Instrumental case (e.g., eat with a
spoon) D Directive case (e.g., going home)

Table 6.10: Symbols above arrows

6. States
States of objects are described by scales which have numerical values. For example, the
state of health of a person is represented by a numerical value which goes from −10 to
6.9. CONCEPTUAL DEPENDENCY 121

+10, “−10” indicating “dead” and “+10” indicating “perfect health”. The intermediate
values may represent various states of health as indicated below:

State of health Numerical value

dead −10
gravely ill −9
sick −9 to −1
all right 0
tip top +7
perfect health +10

Various other states like the states of “fear”, “anger”, “mental state” (sad, happy, etc.),
hunger, etc. have also been defined the CD theory with appropriate numerical values.
There are also states that have absolute measures like, “length”, “mass”, “weight”,
“speed”, “size”, etc. Further, there are also states which indicate relationships between
objects like “possession” indicated by “POSS” in CD theory.

6.9.3 Rules
The conceptual categories combine in certain specified ways. The rules for combination
of conceptual categories are called the conceptual syntax rules. There are used to
represent natural language statements. The various rules are summarised in Table 6.11.

Table 6.11: Rules in CD theory

Sl. Rule Usage Meaning

No.
P
1 PP ACT John PTRANS John ran.
PP PA John height (> average
2 John is tall.
3 PP PA John doctor John is a doctor.
xPP x
boy
 
4 A nice boy
P A ni ce
PP dog
POSS-BY
John’s dog (POSS-
5 PP John BY denotes “pos-
session by”.)
o p o
6 ACT ←− PP John PROPEL cart John pushed the
⇐⇒ ←−
cart.
6.9. CONCEPTUAL DEPENDENCY 122

p o John
John ATRANS
o PP Mary
ACT o
7 PP book John took the book
from Mary.
John
I
John INGEST
o DO
I ice cream o
ACT spoon John ate ice cream
8
with a spoon.
P D field
John PTRANS bag
PP o
D
ACT fertilizer John fertilized the
9 PP field (“D” denotes
the directive case.)
PP size > x
PP plants
10 PA size = x The plants grew.

11 Bill shot Bob.

Ram DO
health = −10
Ravan P Ram killed Ravan.
12 health > −10
(Here “DO” an un-
specified activity.)
yesterday
T

13 John PTRANS John ran yesterday.

(“T” denotes time.)
o D home
John PTRANS John While going home,
John John saw a frog.
CP
o R (Here “CP” de-
14 John MTRANS frog eyes
notes “Conceptual
Processor” which
is a symbol for an
unspecified object.)
woods
PP
John o R CP
MTRANS frog
15 ears John heard a frog in
the woods.
6.10. SAMPLE QUESTIONS 123

6.9.4 Advantages and disadvantages of the CD theory

Advantages
1. CD theory gives a representation of sentences which is independent of the words
used in the statement.

2. CD can act as Interlingua and can facilitate translation from one language to
another based on the idea of the statement and not just translation of the words used
in the statement.

3. We can fill missing pieces in the fixed structures by fetching them from the context.

Disadvantages
1. It is only a theory of representation of events.

2. CD theory requires that all knowledge be decomposed into low-level primitives.

This may be impossible in certain situations.

3. The CD representation is sometimes very inefficient. To represent a sentence it

may take several pages of CD diagrams.

4. Complex representations require a lot of storage.

5. There are no representations of relationships which can help us make inferences

from the links.

6. The set of 11 primitive acts is incomplete. All knowledge should be expressed into
these acts which is sometimes inefficient and even impossible.

7. Many concepts are not recognized in CD.

8. In the generalized representation, the finer meaning is lost. For example there is a
difference between giving and gifting something which is not captured by CD.

9. It is difficult to construct the original sentence from its CD representation.

6.10 Sample questions

(a) Short answer questions
1. Define a semantic network and give an example.

2. What are the different types of knowledge?

3. What is meant by a knowledge representation system? Name some knowledge rep-

resentation systems.

4. State some of the desirable properties of knowledge representation systems.

5. Explain the intersection search method.

6.10. SAMPLE QUESTIONS 124

6. Define a frame and give an example.

7. What makes the conceptual dependency theory important?

8. Give a simple example for representation of sentences in CD theory.

9. Give three primitive acts in CD theory with meanings and examples.

10. How are states of objects represented in CD theory?

11. Give two examples of rules in CD theory.

12. Construct semantic net representations of each of the following:

(a) Raja is a bank manager.

(b) Raja works in SBI located in Defence Colony.
(c) Raja is 26 years old.
(d) Raja has blue eyes.
(e) Raja is senior in service to John.

13. Construct semantic net representations of each of the following:

(a) Dave is Welsh, Dave is a lecturer.

(b) Paul leant his new Compact Disc to his best friend.

14. Give a CD representation of the sentence: “I gave a book to the man.”

15. Give a CD representation of the sentence: “Mohan took the book from Laila.”

16. Express the statement “He gave the pen” through conceptual dependency.

17. What are the kinds of knowledge represented in AI systems?

18. List some frame languages in AI.

(b) Long answer questions

1. Describe the building blocks of the CD theory.

2. Draw a semantic network representing the following knowledge:

“Every vehicle is a physical object. Every car is a vehicle. Every car has four wheels.
Electrical system is a part of car. Battery is a part of electrical system. Pollution
system is a part of every vehicle. Vehicle is used in transportation. Swift is a car.”

3. Describe the different types of semantic networks with examples.

4. What are the advantages and disadvantages of the CD theory.

5. Represent the relationships between quadrangle, parallelogram, rhombus, rectangle

and square in the form of a semantic network.
6.10. SAMPLE QUESTIONS 125

6. Represent the following knowledge in a semantic network:

Dogs are Mammals Birds have Wings Mammals are Animals
Bats have Wings Birds are Animals Bats are Mammals
Fish are Animals Dogs chase Cats Worms are Animals
Cats eat Fish Cats are Mammals Birds eat Worms
Cats have Fur Fish eat Worms Dogs have Fur

7. Represent the following sentences by a semantic network:

Birds are animals. Birds have feathers, fly and lay

eggs. Albatros is a bird. Donald is a bird.
Tracy is an albatros.

8. Represent the following sentences by a semantic

network: Palco is a calico. Herb is a tuna.
Charlie is a tuna. All tunas are fishes.
All calicos are cats. All cats like to eat all kinds of fishes.

9. Represent the following sentences by a semantic

network: Circus elephants are ele-

phants.
Elephants have heads. Elephants have
trunks. Heads have mouths. Elephants are animals.
Animals have hearts. Circus elephants are
performers. Performers have costumes. Costumes are
cloths.
Horatio is a circus elephant.

10. Represent the following sentences by a semantic

network: Every truck is a vehicle.
Every trailer truck is a truck that has
as part a trailer,
an unloaded weight, which is a weight measure,
a maximum gross weight, which is a weight measure, a
cargo capacity, which is a volume measure,
and a number of wheels, which is the integer 18.

11. What is conceptual dependency? How do you represent conceptual dependency?

12. What is CD? List primitive acts with meaning.

6.10. SAMPLE QUESTIONS 126

13. What do you understand by conceptual dependency? Give the conceptual depen-
dency representations of the following sentences:

(a) John shot Mary.

(b) John fertilized the field.
(c) Mary drove her car to the office.

14. Represent the following sentences using conceptual dependency diagrams.

(a) John pushed the cart.

(b) The plants grew.
(c) John ran yesterday.
(d) John is a teacher.
(e) While going home, I saw a frog.

15. Explain the primitive conceptual categories in CD theory.

 Chapter 7

Knowledge representation using logic

In the last chapter we discussed knowledge representation using semantic networks and
conceptual dependency diagrams. In the present chapter we consider a third method of
knowledge representation, namely, representation using logic, especially first order predi-
cate logic.
In the first section we give a quick review of the basic ideas of propositional logic as
a preparation for studying first order predicate logic. The chapter concludes with a
summary of the various knowledge representation methods.

7.1 Propositional logic

7.1.1 Propositions
A proposition is a statement which is either true or false but not both. If a proposition is
true we say that its truth value is “True” and if the proposition is false, we say its truth
value is “False”. The truth value “True” is also denoted by T or 1 and the truth value
“False” is sometimes denoted by F or 0.

Examples
1. The statement “New Delhi is the capital of India.” is a proposition and its truth
value is True,

2. The statement “India became a republic in 1947.” is a proposition and its truth
value is False.

3. The sentence “It may rain tomorrow.” is not a proposition because it has no
definite truth value.

Propositions are denoted by capital letters like A, P , X, etc.

7.1.2 Operations on propositions: Logical operators

Complex statements can be constructed by applying operations on propositions. Such state-
ments are called propositional formulas or propositional expressions. A table giving the
truth values of a propositional expression is called the truth table of the expressions. The
basic operations on propositions are described below.

127
7.1. PROPOSITIONAL LOGIC 128

1. Negation
The negation of a proposition P is denoted by ¬P . By definition, the truth value of ¬P is
False if the truth value of P is True and the truth value of ¬P is True if the truth value of
P is false. The truth table of ¬P is exhibited in Table 7.1.

P ¬P
T F
F T

Table 7.1: Truth table of negation

2. Disjunction
The disjunction of two propositions P and Q is denoted by P ∨ Q. The truth value of
P ∨ Q is True if P is True, or Q is True or both P and Q are True. The truth value of P
∨ Q is False in all other cases. The truth values are given Table 7.2.

P Q P ∨Q
T T T
T F T
F T T
F F F

Table 7.2: Truth table of disjunction

3. Conjunction
The conjunction of two propositions P and Q is denoted by P ∧ Q. The truth value
of P ∧ Q is True only if both P and Q are True. The truth value of P ∨ Q is False in all
other cases. The truth values are given Table 7.3.

P Q P ∧Q
T T T
T F F
F T F
F F F

Table 7.3: Truth table of conjunction

7.1. PROPOSITIONAL LOGIC 129

4. Implication
The implication of proposition Q from P is denoted by P ⇒ Q. The truth value of P ⇒
Q is False only if P is True and Q is False. The truth value of P ⇒ Q is True in all
other cases. The truth values are given Table 7.4.

P Q P ⇒Q
T T T
T F F
F T T
F F T

Table 7.4: Truth table of implication

5. Two-way implication (or, biconditional)

The two-way implication of P from Q is denoted by P ⇔ Q. The truth value of P ⇔
Q is True only if both P and Q have the same truth values. The truth values are given
Table 7.5.

P Q P ⇔Q
T T T
T F F
F T F
F F T

Table 7.5: Truth table of implication

7.1.3 Tautologies and contradictions

A logical expression is called a tautology if its truth value is always True whatever be the
truth values of its component propositions. A logical expression is called a contradiction
if its truth value is always False whatever be the truth values of its component
propositions.

Examples
For example it can be seen that the expression (¬P ) ∨ (P ∨ Q) is a tautology. We
prove this by constructing a truth table of the expression.
7.1. PROPOSITIONAL LOGIC 130

P Q ¬P P ∨ Q (¬P ) ∨ (P ∨ Q)
T T F T T
T F F T T
F T T T T
F F T F T

Table 7.6: Truth table of (¬P ) ∨ (P ∨ Q)

Also it can be seen that ¬P ∧ (P ∧ Q) is a contradiction from Table 7.7.

P Q ¬P P ∧ Q ¬P ∧ (P ∧ Q)
T T F T F
T F F F F
F T T F F
F F T F F

Table 7.7: Truth table of (¬P ) ∧ (P ∧ Q)

7.1.4 Identities
Two logical expressions X and Y are said to be logically equivalent if the expression
X ⇔ Y is a tautology. If X and Y are logically equivalent we write X ≡ Y and say that
X ≡ Y is a logical identity.

Example
Prove the identity:
(P ⇒ Q) ≡ ((¬P ) ∨ Q)
We have to show that the expression

(P ⇒ Q) ⇔ ((¬P ) ∨ Q)

is a tautology. This we do by constructing the truth table.

P Q P ⇒Q ¬P (¬P ∨ Q) (P ⇒ Q) ⇔ ((¬P ) ∨ Q)
T T T F T T
T F F F F T
F T T T T T
F F T T T T

Table 7.8: Truth table of (P ⇒ Q) ⇔ ((¬P ) ∨ Q)

7.1. PROPOSITIONAL LOGIC 131

Some important identities

Let P , Q, R, S be arbitrary propositions and let T denote a proposition whose truth
value is True and F one whose truth value is False. Then we have the following
identities.

1. P ∨ (¬P ) ≡ T

2. P ∧ (¬P ) ≡ F

3. P ∨ P ≡ P

4. P ∧ P ≡ P

5. P ∨ Q ≡ Q ∨ P

6. P ∧ Q ≡ Q ∧ P

7. P ∨ (Q ∨ R) ≡ (P ∨ Q) ∨ R

8. P ∧ (Q ∧ R) ≡ (P ∧ Q) ∧ R

9. (P ⇒ Q) ≡ ((¬P ) ∨ Q)

10. ¬(P ∨ Q) ≡ (¬P ) ∧ (¬Q) (De Morgan’s law)

11. ¬(P ∧ Q) ≡ (¬P ) ∨ (¬Q) (De Morgan’s law)

7.1.5 Inferences
Arguments
An argument in propositional logic is a sequence of propositions. All but the final
propo- sition are called premises and the final proposition is called the conclusion. An
argument is valid if the truth of all its premises implies that the conclusion is true.
An example of an argument is given in Figure 7.1. It means that if the statements P
∨Q, P ⇒ R and Q ⇒ R are true then R is also true.

P ∨Q
Premises
P⇒R
Q⇒R
Conclusion ∴ R

Figure 7.1: An argument in propositional logic

Proof of validity of an argument

A proof of validity of a given argument is a sequence of statements, each of which is
either a premise of that argument or follows from preceding statements of the sequence
by an elementary valid argument, such that the last statement in the sequence is the
conclusion of the argument whose validity is being proved.
7.1. PROPOSITIONAL LOGIC 132

7.1.6 Rules of inference

The elementary valid arguments that are used in constructing proofs of validity of argu-
ments are called the rules of inference. Rules of inference are templates for building valid
arguments.
Let us create a rule of inference. Constructing truth tables, one can prove that the
following is a tautology:
(P ∧ (P ⇒ Q)) ⇒ Q
This means that if P and P ⇒ Q are true then Q is also true which in turn implies that
the following argument is valid:

P
P ⇒Q
∴ Q
This is a well known rule of inference is known as “modus-ponens”. Table 7.9 lists the
various rules of inference and their names.

Example of proof of validity

Consider the argument given in Figure 7.1. The proof of the validity of this argument is
shown in Table 7.10.

Step no. Formula Derivation

1 P ∨Q Premise
2 P⇒R Premise
3 Q⇒R Premise
4 ¬P ∨ R Step 2, logical identity
5 Q∨R Steps 1,4 (Resolution)
6 ¬Q ∨ R Step 3, logical identity
7 R∨R Steps 5, 6 (Resolution)
8 R Step 7, logical identity

Table 7.10: Proof of validity of the argument in Figure 7.1

7.1.7 Conjuctive normal form

The atomic sentences in propositional logic consist of a single proposition symbol. A
literal is either an atomic sentence (a positive literal) or a negated atomic sentence (a
negative literal). A clause is a disjunction of literals. A sentence expressed as a
conjunction of clauses is said to be in conjunctive normal form (CNF) or clause form.
7.1. PROPOSITIONAL LOGIC 133

Sl. No. Inference rule Name

P
1 P ⇒Q Modus ponens
∴ Q

¬Q
2 P ⇒Q Modus tollens
∴ ¬P

P ⇒Q
3 Q⇒ Hypothetical syllogism
∴ RP⇒
R
P ∨Q
4 ¬P Disjunctive syllogism
∴ Q

P
5 Addition
∴ P ∨Q

P ∧Q
6 Simplification
∴ P

P
7 Q Conjunction
∴ P ∧Q

P ∨Q
8 ¬P ∨ R Resolution
∴ Q∨R

Table 7.9: Rules of inference

Examples

Sentences Examples
Atomic sentence P, Q, R, . .
.
Literals P, ¬P, Q, ¬Q, R, ¬R, . . .
Clauses P ∧ Q ∧ ¬R, ¬Q ∧ R, . . .
CNF (P ∧ Q ∧ ¬R) ∨ (¬P ∧ Q)
7.1. PROPOSITIONAL LOGIC 134

Conversion to CNF
Every sentence of propositional logic is logically equivalent to a conjunction of clauses. We
now describe a procedure for converting to CNF.

1. Eliminate “⇔” by replacing α ⇔ β with (α ⇒) ∧ (β ⇒ α).

2. Eliminate “⇒” by replacing α ⇒ β with ¬α ∨ β.

3. We “move ¬ inwards” by repeated application of the following equivalences:

¬(¬α) ≡ α
¬(α ∧ β) ≡ (¬α ∨ ¬β)
¬(α ∨ β) ≡ (¬α ∧ ¬β)

4. Apply the distributivity laws distributing ∨ over ∧ wherever possible, that is, we
apply the equivalence:

α ∨ (β ∧ γ) ≡ (α ∨ β) ∧ (α ∨ γ).

Example
Transform the following sentence into CNF:

P ⇔ (Q ∨ R)

Solution
Given sentence:
P ⇔ (Q ∨ R)
Eliminating “⇔” we get the equivalent sentence:

(P ⇒ (Q ∨ R)) ∧ ((Q ∨ R) ⇒ P )

Eliminating “⇒” we have

(¬P ∨ (Q ∨ R)) ∧ (¬(Q ∨ R) ∨ P )

Moving ¬ inwards in ¬(Q ∨ R) we have

(¬P ∨ (Q ∨ R)) ∧ ((¬Q ∧ ¬R) ∨ P )

Using distributivity of ∨ over ∧ in ((¬Q ∧ ¬R) ∨ P ) we have

(¬P ∨ (Q ∨ R)) ∧ ((¬Q ∨ P ) ∧ (¬R ∨ P ))

By associativity we have

(¬P ∨ Q ∨ R) ∧ (¬Q ∨ P ) ∧ (¬R ∨ P )

This is the CNF of the given sentence.

7.1. PROPOSITIONAL LOGIC 135

7.1.8 Resolution-refutation in propositional logic

A proof technique is a procedure for showing that a statement logically follows from a
set of premises. Resolution is a proof technique that works on conjunctive normal form
expression. The proof technique known as resolution depends on the the rule of inference
known as “resolution” (see Table 7.9). The basic idea of resolution can be stated thus:

1. Select two clauses that contain conflicting terms (that, one term in one of the
clauses is the negation of a term in the other clause).

2. Combine those two clauses.

3. Cancel out the conflicting terms.

What makes the resolution proof technique so important is that this one simple technique
is capable of performing the same kinds of reasoning tasks as modus ponens, modus
tollens, syllogisms and many other logical operations.
A resolution refutation proof is proof by contradiction using resolution. Like every
proof by contradiction, we begin by assuming the opposite of what we wish to prove, and
then show that this “fact” would lead to a contradiction.
The algorithm of the resolution-refutation proof technique is given below.

Algorithm: Propositional resolution

Let be required to produce a proof of a proposition P from a set of premises F .

Step 1. Convert all propositions in F to clause form.

Step 2. Negate P and convert the result to clause form. Add it to the set of clauses
obtained in Step 1.

Step 3. Repeat until either a contradiction is found or no progress can be made.

(a) Select two clauses. Call these the parent clauses.

(b) Resolve them together. The resulting clause is called the resolvent.
(The resolvent is the disjunction of all literals of both the parent clauses
with the following exception: If there are any pairs of literals L and ¬L
such that one of the parent clauses contains L and the other contains ¬L,
select one such pair and eliminate both L and ¬L from the resolvent.)
(c) If the resolvent is the empty clause then a contradiction has been found.
(d) If it is not, then add it to the set of clauses available to the procedure.

If a contradiction has been found, we conclude that P is a valid conclusion from the
premises F . Otherwise, that is if the algorithm terminates in such a status that no progress
can be made, we conclude that P is not a valid conclusion from F .

Example
Using resolution-refutation determine whether the argument shown in Figure 7.1 is valid.
7.2. FIRST ORDER PREDICATE LOGIC (FOPL) 136

Solution
The premises are
P ∨ Q, P ⇒ R, Q ⇒ R,
Converting these to the clause form we get

P ∨ Q, ¬P ∨ R, ¬Q ∨ R.

We now add the negation of the conclusion to the premises:

P ∨ Q, ¬P ∨ R, ¬Q ∨ R, ¬R.

The various stages in the resolution are shown in Table 7.11.

Step no. Formula Derivation

1 P ∨Q Premise
2 ¬P ∨ R Premise
3 ¬Q ∨ R Premise
4 ¬R Negated conclusion
5 Q∨R Steps 1, 2 (Resolution)
6 ¬Q Steps 3, 4 (Resolution)
7 R Steps 5, 6 (Resolution)
8 Empty clause Steps 4, 7 (Resolution)

Table 7.11: Proof by resolution-refutation of the argument in Figure 7.1

7.2 First order predicate logic (FOPL)

In propositional logic we are concerned with the truth values of stand alone statements
like “New Delhi is capital of India” and “105 is not a prime number” and of complex
statements obtained by applying various logical connectives to such sentences. We are
also concerned with the different inference rules which will help us derive new sentences
from given premises.
In first order predicate logic, also called simply predicate logic, we are concerned
with statements which will be applicable to the elements in some set of objects and about
the truth values of such statements. For example, consider the set of children shown in
the Figure 7.2. With reference to this set of children, we may may like to know the truth
values of the following statements:
“All children wear boots.”
“All children are boys.”
“There is a girl among the children.”
The predicate logic is a theory for studying such problems.
7.2. FIRST ORDER PREDICATE LOGIC (FOPL) 137

Figure 7.2: A domain whose objects are children

To study the first of the three sentences given above, we consider the sentence “x
wears boots”. Here we assume that x is the name of an unspecified arbitrary child in
the photo- graph. As such it has no truth value because we have not specified what x is. If
we replace x by the name of a child in the photograph, say “John”, we get the
sentence “John wears boots.” which has a truth value namely True. The statement
“All children wear boots” is true if all sentences obtained by replacing x in “x wears
boots” by the names of each and every child in the photograph is true. This is indeed
the case and so we conclude that the statement “All children wear boots” is true.
To discuss the the third of the three statements, we consider the sentence “x is a girl”
For the statement “There is a girl among the children” to be true it is enough that there is
at least one name say “Sara” among the set of names of children which makes the
sentence “Sara is a girl” true. This is indeed the case and so the truth value of the
sentence “Thers is a girl among the children” is True.
Sentences like “x wears boots” and “x is a girl” are called predicates. The predicate
“x wears boots” may be symbolically represented as “wearsboots(x)”, or
“WearsBoots(x)” or simply “W (x)”. Similarly, the sentence “x is a girl” may be
represented as “girl(x)” or “Girl(x)” or simply “G(x)”.

Why first order?

The “first order predicate logic” is so called only to distinguish it from the “second order
predicate logic”. Second-order logic was introduced by Frege in 1879 who also coined
the term “second order”. In first order predicate logic, we study truth values of predicates
which are essentially statements of properties of elements of sets. In the second order
logic, we study properties of sets of such properties of sets. To see an example of such a
property, let P (x) be some arbitrary predicate. Then the following is a statement in second
order logic:
”For all predicates P , and for each object x,
either P (x) is true or ¬ (Px) is true.”

In these notes we are concerned with only first order logic.1

1
For a detailed introduction to second order logic, one may refer to “Second-
order and Higher-order Logc”, Stanford Encyclopedia of Philosophy, available ar
https://plato.stanford.edu/entries/logic-higher-order/.
7.2. FIRST ORDER PREDICATE LOGIC (FOPL) 138

Figure 7.3: A model containing five objects, two binary relations, three unary relations
(indicated by labels on the objects), and one unary function, left-leg.

7.2.1 Domain of discourse

In discussing FOPL, we assume a certain domain of discourse, or simply a domain (also
called a “world”or a “universe”). The domain contains all the objects under
consideration. The predicate logic requires that the domain be non-empty, that is, that it
contains at least one object.
The domain is completely arbitrary. For example, Figure 7.3 shows a domain with
five objects, namely, King Richard, King John, the left legs of Richard and John and a
crown.

7.2.2 Predicates
A predicate is a statement (specifying some relation) that contains variables (predicate
variables), and they may be true or false depending on the values of these variables. The
domain is the set of values that variables can take.

Definition
A predicate is a sentence that contains a finite number of variables and becomes a
statement when specific values are substituted for the variables.

Examples
The following are predicates in appropriate domains:

1. “x is a boy.”

2. “x wears boots.”

3. “x is a king.”

7.2.3 Basic elements

• The basic elements of first-order logic are the symbols that stand for objects, rela-
tions, and functions.
 7.3. QUANTIFIERS 139

• The symbols are of three kinds: constant symbols, which stand for objects;
predicate symbols, which stand for relations; and function symbols, which stand
for functions.

• Each predicate and function symbol comes with an arity that fixes the number of
arguments.

A predicate with arity 1 is called a unary predicate. The predicate “x is a king” is unary
predicate. A predicate with arity 2 is called a binary predicate. The predicate “x is a
brother of y” is a binary predicate. In general, a predicate with arity n is called a n-ary
predicate.

Notation
We adopt the convention that the symbols begin with uppercase letters. For example, for the
world specified in Figure 7.3, we might use the constant symbols Richard and John; the
predicate symbols Brother, OnHead, Person, King, and Crown; and the function symbol
LeftLeg. The choice of names is entirely up to the user.

Terms and atomic sentences

A term is a logical expression that refers to an object. Constant symbols are therefore terms.
An atomic sentence (or atom for short) is formed from a predicate symbol optionally
followed by a parenthesized list of terms, such as Brother(Richard, John).

7.3 Quantifiers
7.3.1 Universal quantification
Consider the statement:

“All kings are persons."

To write this in the notations of FOPL, we introduce two predicates as follows:

“King(x)” denoting “x is a king.”

“Person(x)” denoting “x is a person.”

In the notations of first-order logic we write the sentence “All kings are persons.” as
follows:
∀ x King(x) → Person(x)
The symbol ∀ is usually pronounced “For all . . .” and is called the universal quantifier.
Thus, the above sentence says:

“For all x, if x is a king, then x is a person.âA˘ ˙I

The symbol x is called a variable. By convention, variables are lowercase letters. A

variable is a term all by itself, and as such can also serve as the argument of a function -
for example, LeftLeg(x). A term with no variables is called a ground term.
Intuitively, the sentence ∀ x P , where P is any logical expression, says that P is
true for every object x.
7.3. QUANTIFIERS 140

7.3.2 Existential quantification

Universal quantification makes statements about every object. Similarly, we can make a
statement about some object in the universe without naming it, by using an existential
quan- tifier.
For example consider the statement:
King John has a crown on his head.
We need two predicates:

Crown(x) : “x is a crown”
OnHead(x, John) : “x is on the head of
John.” To say that “King John has a crown on his head”, we write
∃ x Crown(x) ∧ OnHead(x, John)
∃x is pronounced “There exists an x such that . . .” or “For some x . . .”. The symbol “∃”
is called the existential quantifier. Notice that, by our definition, the sentence would
also be true in a model in which King John was wearing two crowns. This is entirely
consistent with the original sentence “King John has a crown on his head.”
Intuitively, the sentence ∃ x P says that P is true for at least one object x.

7.3.3 Nested quantifiers

We will often want to express more complex sentences using multiple quantifiers. We
illustrate the idea by examples:
• “Brothers are siblings” can be written as
∀x ∀y Brother(x, y) ⇒ Sibling(x, y)

• “Everybody loves somebody” means that for every person, there is someone that
person loves:
∀x∃y Loves(x, y)

• The statement “There is someone who is loved by everyone” we write as

∃y∀x x Loves(x, y).

The order of quantification is very important.

7.3.4 Connections between ∀ and ∃

The two quantifiers ∀ and ∃ are intimately connected with each other, through negation. We
have the following logical equivalences:
1. ∀x ¬P ≡ ¬∃x P
2. ¬∀x P ≡ ∃x ¬P
3. ∀x P ≡ ¬∃x ¬P
4. ∃x P ≡ ¬∀x ¬P
7.3. QUANTIFIERS 141

7.3.5 Examples of quantified statements

We now consider several examples of quantified statements. In these examples, the mean-
ings of the predicates are obvious from the contexts.

1. “If something is a cube, it is not a tetrahedron.”

∀ x Cube(x) ⇒ ¬ Tetrahedron(x)

2. “Everybody loves a lover.”

∀ x ∀ y [Person(x) ∧ Lover(y) ⇒ L(x, y)]
where L(x, y) denotes x loves y.

3. “If two persons speak the same language, they understand each other.”
∀ x, y, l SpeaksLanguage(x, l) ∧ SpeaksLanguage(y, l)
⇒ Understands(x, y) ∧ Understands(y, x)

4. “There is a barber who shaves all men in town who do not shave themselves.”
∃x Barber(x) ∧ (∀ y Man(y) ∧ ¬ Shaves(y, y) ⇒ Shaves(x, y))

5. “Every person who buys a policy is smart.”

∀ x Person(x) ∧ (∃ y, z Policy(y) ∧ Buys(x, y, z)) ⇒ Smart(x)
(Buys(x, y, z) denotes x buys y from z.)

6. “Only one student took French as elective.”

∃ x Student(x) ∧ Takes(x) ∧ (∀ y (y /= x) ⇒ ¬ Takes(y)
(Takes(x) means x takes French as elective.)

7. “The best score in French is always higher than the best score in German in all
semesters.”
∀ s ∃ x ∀ y Score(x, French, s) > Score(y, German, s)
(Score(x, g, s) denotes the score obtained by student x in subject g in semester s.)

8. “A person born in the UK, each of whose parents is a UK citizen or a UK resident,

is a UK citizen by birth.”

∀x Person(x) ∧ Born(x)∧

∀ y Parent(y, x) ⇒ ((∃ r Citizen(y, r) ∨ Resident(y)) ⇒ Citizen(x, Birth)

(Parent(x, y) denotes x is a parent of y and Citizen(x, r) denotes x is a citizen
of UK by reason r.)

9. “Everyone who loves all animals is loved by someone”

∀ x [∀ y Animal(y) ⇒ Loves(x, y)] ⇒ [∃ y Loves(y, x)]

10. “Every farmer who owns a donkey beats it.”

∀
x Farmer(x) ⇒ ∀ y ((Donkey(y) ∧ Owns(x, y)) ⇒ Beats(x, y))
7.4. WELL-FORMED FORMULAS IN FOPL 142

7.4 Well-formed formulas in FOPL

7.4.1 Well-formed formulas
Well-formed formulas (wffs) in FOPL are constructed using the following rules:

1. True and False are wffs.

2. Each propositional constant (i.e. specific proposition) is a wff

3. Each atomic formula (i.e. a specific predicate with variables) is a wff.

4. If A and B are wffs, then so are ¬A, (A ∨ B), (A ∧ B), (A ⇒ B), and A ⇔ B.

5. If x is a variable (representing objects of the universe of discourse), and A is a

wff, then so are ∀x A and ∃x A.

7.4.2 Clausal forms (also called conjuctive normal forms (CNFs))

1. A literal is either an atomic formula or the negation of an atomic formula.

2. A clause is the disjunction of literals.

3. A formula in clausal form (or, CNF) consists of a conjunction of clauses.

7.4.3 Algorithm to convert a wff to the clausal form (or, CNF)

1. Eliminate the two-way implication (biconditional) ⇔
Replace a ⇔ b by (a ⇒ b) ∧ (b ⇒ a).

2. Eliminate implications
Replace a ⇒ b by ¬a ∨ b.

3. Move ¬ inwards
Make replacements as follows:
¬(a ∨ b) by ¬a ∧ ¬b
¬(a ∧ b) by ¬a ∨ ¬b
¬ ∀ x P by ∃ x ¬ P
¬ ∃ x P by ∀ x ¬ P

4. Standardize variables
Change variables so that each quantifier binds a unique variable.
For example, for sentences like (∃ x P (x))∨(∃ x Q(x)) which use the same
variable name twice, change the name of one of the variables.
Skolemize2
5.
Skolemization is the process of removing existential quantifiers by elimination.
2
The terminology is in honour of Thoralf Albert Skolem (1887 - 1963), a Norwegian mathematician who
worked in mathematical logic and set theory.
7.4. WELL-FORMED FORMULAS IN FOPL 143

(a) Existentially quantified variables which are not inside the scope of a universal
quantifier are replaced by creating new constants. For example, ∃ x P (x) is
changed to P (c), where c is a new constant.
(b) Replace every existentially quantified variable y with a term f (x1, , xn)
whose function symbol f is new. The variables of this term are the variables
that are universally quantified and whose quantifiers precede that of y.
• For example, in the formula ∀ x ∃ y ∀ z P (x, y, z), Skolemization replaces
y with f (x), where f is a new function symbol, and removes the
quantifi- cation over y. The resulting formula is ∀x ∀z P (x, f (x), z).
The Skolem term f (x) contains x but not z, because the quantifier to be
removed ∃ y is in the scope of ∀ x, but not in that of ∀z
• As a second example, when we Skolemize

we get
∀ x ∀ y ∃ z ∀ u ∃ v P (x, y, z, u, v)

∀ x ∀ y ∀ u P (x, y, F (x, y), u, G(x, y, u))

where F (x, y) and G(x, y, u) are Skolem functions.
The general rule is that the arguments of the Skolem function are all the universally
quantified variables in whose scope the existential quantifier appears.

6. Remove universal quantifiers

At this point, all remaining variables will be universally quantified. We now drop
the universal quantifiers.

7. Distribute ∨ over ∧
Convert the resulting expression into a conjunction of disjunctions by using
distribu- tivity.
Replace a ∨ (b ∧ b) by (a ∨ b) ∧ (a ∨ b).
8. CNF of the given wff
The sentence is now in CNF.

7.4.4 Examples
Example 1
Convert the following wff to the clausal form:

∀ x [∀ y A(y) ⇒ L(x, y)] ⇒ [∃ y L(y, x)]

Solution
Step 1. Eliminate implications
∀ x ¬[∀ y ¬A(y) ∨ L(x, y)] ∨ [∃ y L(y, x)]
that is,
∀ x [¬∀ y ¬A(y) ∨ L(x, y)] ∨ [∃ y L(y, x)]
7.4. WELL-FORMED FORMULAS IN FOPL 144

Step 2. Move ¬ inwards

The sentence goes through the following transformations:
∀ x [∃ y ¬(¬A(y) ∨ L(x, y))] ∨ [∃ y L(y, x)]
∀ x [∃ y ¬¬A(y) ∧ ¬L(x, y)] ∨ [∃ y L(y, x)]
∀ x [∃ y A(y) ∧ ¬L(x, y)] ∨ [∃ y L(y, x)]
Step 3. Standardize variables
∀ x [∃ y A(y) ∧ ¬L(x, y)] ∨ [∃ z L(z, x)]
Step 4. Skolemize
∀ x [A(F (x)y) ∧ ¬L(x, F (x))] ∨ [L(G(x), x)]
(F and G are Skolem functions.)
Step 5. Drop universal quantifiers
[A(F (x)y) ∧ ¬L(x, F (x))] ∨ [L(G(x), x)]
Step 6. Distribute ∨ over ∧
[A(F (x) ∨ L(G(x), x)] ∧ [L(x, F (x)) ∨ L(G(x), x)]
Step 7. This the CNF of the given sentence.

Example 2
Convert the following wff to the clausal form:
∀ h i
x [E(x)] ⇔ [∀y E(T (x, y))]

Solution
Step 1. Remove biconditional
∀ x ([E(x)] ⇒ [∀y E(T (x, y))]) ∧ ([∀y E(T (x, y))] ⇒ [E(x)])
Step 2. Remove implications
∀ x (¬[E(x)] ∨ [∀y E(T (x, y))]) ∧ (¬[∀y E(T (x, y))] ∨ [E(x)])
Step 3. Move ¬ inwards
∀ x (¬[E(x)] ∨ [∀y E(T (x, y))]) ∧ ([∃y ¬ E(T (x, y))] ∨ [E(x)])
Step 4. Standardize variables
∀ x (¬[E(x)] ∨ [∀y E(T (x, y))]) ∧ ([∃z ¬ E(T (x, z))] ∨ [E(x)])
Step 5. Skolemize
∀ x (¬[E(x)] ∨ [∀y E(T (x, y))]) ∧ ([¬ E(T (x, F (x)))] ∨ [E(x)])
where F (x) is a Skolem function.
Step 6. Drop universal quantifiers
(¬[E(x)] ∨ [E(T (x, y))]) ∧ ([¬ E(T (x, F (x)))] ∨ [E(x)])
Step 7. This is the CNF of the given sentence.
7.5. INFERENCE RULES IN FOPL 145

Example 3
Transform to the CNF the following wff:

∀ x [Romans(x) ∧ know(x, Marcus)] ⇒

[hate(x, Caesar) ∨ (∀ y ∃ z hate(y, z) ⇒ thinkcrazy(x, y))]

This is a symbolic representation of the sentence: “All Romans who know Marcus
either hate Caesar or think that anyone who hates anyone is crazy.” (see RKN, p.109) To
simplify the writing of the various steps we shall write R for “Romans”, K for “know”, H
for “hate”, T for “thinkcrazy”, M for “Marcus” and C for “Caesar”.
In the simplified notations, the given sentence can be written as

∀ x [R(x) ∧ K(x, M )] ⇒ [H(x, C) ∨ (∀ y (∃ z H(y, z)) ⇒ T (x, y))]

Solution
Step 1. Remove implications
∀ x ¬[R(x) ∧ K(x, M )] ∨ [H(x, C) ∨ (∀ y ¬ (∃ z H(y, z)) ∨ T (x, y))]

Step 2. Move ¬ inwards

∀ x [¬R(x) ∨ ¬K(x, M )] ∨ [H(x, C) ∨ (∀ y (∀ z ¬H(y, z)) ∨ T (x, y))]

Step 3. Standardise variables

The variables are already standardised.

Step 4. Skolemize
No change.

Step 5. Remove universal quantifiers

[¬R(x) ∨ ¬K(x, M )] ∨ [H(x, C) ∨ ((¬H(y, z)) ∨ T (x, y))]

Step 6. This is the CNF of the given sentence.

7.5 Inference rules in FOPL

7.5.1 Rules borrowed from propositional logic
All the inference rules for propositional logic (see Tables 7.9) like modus ponens, modus
tollens, resolution, etc. can be applied in FOPL as well.

7.5.2 Universal instantiation (UI)

The rule of Universal Instantiation (UI for short) says that we can infer any sentence ob-
tained by substituting a ground term (a term without variables) for the variable in a
predicate formula involving universal quantifiers.
In short, UI says that if ∀ x P (x) is true then P (c) is true, where c is a constant in
the domain.
7.5. INFERENCE RULES IN FOPL 146

Example
Consider the following predicate formula:

∀ x King(x) ∧ Greedy(x) ⇒ Evil(x)

If this is true then we can infer that the following are all true.

King(John) ∧ Greedy(John) ⇒ Evil(John)

King(Richard) ∧ Greedy(Richard) ⇒ Evil(Richard)
King(Father(John)) ∧ Greedy(Father(John)) ⇒ Evil(Father(John))

7.5.3 Existential instantiation (EI)

The rule for Existential Instantiation (EI for short) says that in a predicate formula involving
existential quantifiers, the variable may be replaced by a single new constant symbol.
In short, UI says that if ∃ x P (x) is true then P (k) is true, where k is a new constant.

Example
For example, from the sentence

∃ x Crown(x) ∧ OnHead(x, John)

we can infer the sentence

Crown(C) ∧ OnHead(C, John)

as long as C does not appear elsewhere in the knowledge base.

7.5.4 Generalised modus ponens (GMP)

Given atomic sentences P1, P2, . . . , PN , and implication sentence

(Q1 ∧ Q2 ∧ · · · ∧ QN ) ⇒ R,

where Q1, Q2, . . . , QN and R are atomic sentences, and

subst(θ, Pi) = subst(θ, Qi) for i = 1, ..., N,

we can derive the new sentence:

Subst (θ, R)
where Subst (θ, R) denotes the result of applying a set of substitutions defined by θ to the
sentence R.
7.5. INFERENCE RULES IN FOPL 147

Remarks
1. Recall the modus ponens inference rule of propositional logic:

P
P ⇒Q
∴ Q

2. A substitution list is usually written in the following form:

θ = {v1/t1, v2/t2, . . . , vn/tn}

This means that we are required to replace all occurrences of variable symbol vi by
term ti. Substitutions are made in left-to-right order in the list.
For example, let the substitutions be

θ = {x/IceCream, y/Ziggy}

and the statement be

R = eats (x, y)
then the substitution Subst (θ, R) yields

eats (Ziggy, IceCream).

7.5.5 The resolution inference rule

To state the resolution inference rule in FOPL. we require the concept of unification.

Unification
Unification is the process of finding substitutions that make different logical expressions
look identical.
For example, consider the logical expressions P (x, y) and P (a, f (z)). If we
substitute x by a and y by f (z) in the first expression, the first expression will be identical
to the second expression. We say that the unifier of the expressions is the set θ = {a/x, f
(z)/y}.
As another example, consider the sentences:

Knows(John, x), Knows(x, Elizabeth)

Since both sentences contain the same variable, we apply standardisation and change the x
in Knows(x, Elizabeth) to y and get the following sentences:

Knows(John, x), Knows(y, Elizabeth)

Now, it is easy to see that the two sentences are unified by the substitutions

θ = {x/Elizabeth, y/John}.

For unification of two sentences, the predicate symbol must be same and number of
arguments must also be the same.
7.6. INFERENCE METHODS IN FOPL 148

Outline of a procedure for unification

1. Predicate symbols must match.
2. For each pair of predicate arguments:
(a) Different constants cannot match.
(b) A variable may be replaced by a constant.
(c) A variable may be replaced by another variable.
(d) A variable may be replaced by a function as long as the function does not
con- tain an instance of the variable.

The resolution inference rule

Given sentence P1 ∨ . . . ∨ Pn and sentence Q1 ∨ . . . ∨ Qm where each Pi and Qj is a
literal, if Pj and ¬Qk unify with substitution list θ then we can derive the resolvent
sentence:
Subst (θ, P1 ∨ . . . ∨ Pj−1 ∨ Pj+1 ∨ . . . ∨ Pn ∨ Q1 ∨ . . . Qk−1 ∨ Qk+1 ∨ . . . ∨ Qm)

Example
Consider the clauses:

P (x, f (a)) ∨ P (x, f (y)) ∨ Q(y), and ¬P (z, f (a)) ∨ ¬Q(z).

Let us denote
P1 = P (x, f (a)), Q1 = ¬ P (z, f (a)).
We have
¬, Q1 = ¬(¬ P (z, f (a))) = P (z.f (a)).
Clearly, P1 and ¬ Q1 unify under the substitution

θ = {x/z}.

Hence using the resolution inference rule we get the resolvent clause

P (z, f (y)) ∨ Q(y) ∨ ¬Q(z).

7.6 Inference methods in FOPL

7.6.1 Propositionalisation: Reduction to propositional inference
The following are the various steps of the method.
1. Using EI rule, we replace an existentially quantified statement by one instantiation.
2. Using UI rule, we replace a universally quantified statement by all possible
instanti- ations.
3. We apply the methods of propositional logic to make valid inferences.
This procedure cannot always be applied because it may produce infinite sets of proposi-
tions.
7.6. INFERENCE METHODS IN FOPL 149

7.6.2 Resolution-refutation method in FOPL

This is very similar to the resolution-refutation method of proof in propositional logic.

Algorithm: Propositional resolution

Let be required to produce a proof of a statement P from a set of statements F .

Step 1. Convert all statements in F to clause form.

Step 2. Negate P and convert the result to clause form. Add it to the set of clauses
obtained in Step 1.

Step 3. Repeat until either a contradiction is found, no progress can be made, or a pre-
determined effort has been expended.

(a) Select two clauses. Call these the parent clauses.

(b) Resolve them together. The resulting clause is called the resolvent.
(The resolvent is the disjunction of all literals of both the parent clauses
with the following exception: If there are any pairs of literals T1 and / T2
such that one of the parent clauses contains T1 and the other contains T2 and
if T1 and T2 are unifiable, then neither T1 T2 should appear in the resolvent.
We call T1 and T2 complementary literals. Use the substitution produced by
unifica- tion to create the resolvent. If there is more than one pair of
complementary literals, only one pair should be omitted from the resolvent.)
(c) If the resolvent is the empty clause then a contradiction has been found.
(d) If it is not, then add it to the set of clauses available to the procedure.

If a contradiction has been found, we conclude that P is a valid conclusion from the
premises F . Otherwise, that is if the algorithm terminates in such a status that no progress
can be made, we conclude that P is not a valid conclusion from F .

Remark
If the clauses to be resolved are chosen in certain systematic ways, the resolution
procedure will find a contradiction if one exists. However it may take a long time.

7.6.3 Examples
The various steps in applying the above algorithm to a given simple problem can be sum-
marised as follows:

1. Convert the facts into statements in FOPL.

2. Convert FOPL statements into CNF.

3. Negate the statement which needs to be proved and to the premises.

4. Draw the resolution graph showing the various unifications.

 7.6. INFERENCE METHODS IN FOPL 150

Example 1
Prove the validity of the following argument:

1. Cats like fish.

2. Cats eat everything they like.

3. Josephine is a cat.

4. Therefore, Josephine eats fish.

Solution
First, we express the given sentences and the negated goal G in first-order logic in the CNF.

1. ¬ cat(x) ∨ likes(fish)

2. ¬cat(y) ∨ ¬likes(y, z) ∨ eats(y, z)

3. cat(jo)

4. eats(jo, fish) (conclusion)

The resolution proof that Josephine eats fish is given in Figure 7.4.

¬ eats(jo, fish) ¬ cat(y) ∨ ¬ likes(y,z) ∨ eats(y,z)

¬ cat(jo) ∨ ¬ likes(jo, fish) cat(jo)

¬ cat(x) ∨ likes(x, fish) ¬ likes(jo, fish)

¬ cat(jo) cat(jo)

contradiction

Figure 7.4: An example of a resolution-refutation proof

Example 2
Prove that the conclusion
hate(Marcus, Caesar)
can be derived from the following premises:

1. man(Marcus)

2. Pompeian(Marcus)

3. ¬ Pompeian(x1) ∨ Roman(x1)

4. ruler(Caesar)

5. ¬Roman(x2) ∨ loyalto(x2, Caesar) ∨ hate(x2, Caesar)

 7.6. INFERENCE METHODS IN FOPL 151

6. loyalto(x3, f (x3))

7. ¬man(x4) ∨ ¬ruler(y1) ∨ ¬tryassassinate(x4, y1) ∨ loyalto(x4, y1)

8. tryassassinate(Marcus, Caesar)

Solution
The various stages in the derivation are shown in Figure 7.5.

¬ hate (Marcus, Caesar) 5

3 ¬ Roman (Marcus) ∨ loyalto (Marcus, Caesar)

¬ Pompeian (Marcus) ∨ loyalto (Marcus, Caesar) 2

7 loyalto (Marcus, Caesar)

1 ¬ men (Marcus) ∨ ¬ ruler (Caesar) ∨ ¬ tryassassinate (Marcus,

Caesar)

¬ ruler (Caesar) ∨ ¬ tryassassinate (Marcus, Caesar) 4

¬ tryassassinate (Marcus, Caesar) 8

contradiction

Figure 7.5: An example of a resolution-refutation proof

Example 3
Use resolution-refutation method to check the validity of the following argument:
Everyone who loves all animals is loved by someone.
Anyone who kills an animal is loved by no one. Jack
loves all animals.
Either Jack or Curiosity killed the cat, who is named Tuna.
Therefore, Curiosity killed the cat.

Solution
First, we express the given sentences, some background knowledge, and the negated goal
G in first-order logic:

A. ∀ x [∀ y Animal(y) ⇒ Loves(x, y)] ⇒ [∃ y Loves(y, x)]

B. ∀ x [∃ z Animal(z) ∧ Kills(x, z)] ⇒ [∀ y ¬ Loves(y, x)]

 7.6. INFERENCE METHODS IN FOPL 152

C. ∀ x Animal(x) ⇒ Loves(Jack, x)

D. Kills(Jack, Tuna) ∨ Kills(Curiosity, Tuna)

E. Cat(Tuna)

F. ∀ x Cat(x) ⇒ Animal(x)

G. ¬Kills(Curiosity, Tuna) (negation of conclusion)

Now we apply the conversion procedure to convert each sentence to CNF:

1. A1: Animal(F (x)) ∨ Loves(G(x), x)

2. A2: ¬Loves(x, F (x)) ∨ Loves(G(x), x)

3. B: ¬Loves(y, x) ∨ ¬Animal(z) ∨ ¬Kills(x, z)

4. C: ¬Animal(x) ∨ Loves(Jack, x)

5. D: Kills(Jack, Tuna) ∨ Kills(Curiosity, Tuna)

6. E: Cat(Tuna)

7. F: ¬Cat(x) ∨ Animal(x) Ânˇ G.

8. G: ¬Kills(Curiosity, Tuna)

The resolution proof that Curiosity killed the cat is given in Figure 7.6.

Cat(Tuna) ¬Cat(x)∨Animal(x) Kills(Jack,Tuna)∨Kills(Curiosity,Tuna) ¬Kills(Curiosity,Tuna)

Animal(Tuna) ¬Loves(y,x)∨Animal(z)∨¬Kills(x,z) Kills(Jack,Tuna)

¬Loves(y, x)∨ Kills(x, Tuna)

¬Loves(y,Jack) ¬Loves(x, F (x))∨Loves(G(x), x)) ¬Animal (x)∨Loves(Jack, x)

¬Animal(F (Jack))∨Loves(G(Jack),Jack) Animal(F (x))∨Loves(G(x), x)

Loves(G(Jack), Jack)

Contradiction

Figure 7.6: An example of a resolution-refutation proof

7.6. INFERENCE METHODS IN FOPL 153

Example 4
Given the premises:

“The law says that it is a crime for an American to sell weapons to hostile
nations. The country Nono, an enemy of America, has some missiles, and all
of its missiles were sold to it by Colonel West, who is American.”

prove that West is a criminal.

Solution
First, we represent the facts as first-order clauses.

1. “. . . it is a crime for an American to sell weapons to hostile nations”:

American(x) ∧ Weapon(y) ∧ Sells(x, y, z) ∧ Hostile(z) ⇒ Criminal(x)

2. “Nono . . . has some missiles.”

∃ x Missile(x) ∧ Owns(Nono, x)

3. “All of its missiles were sold to it by Colonel West”:

Missile(x) ∧ Owns(Nono, x) ⇒ Sells(West, x, Nono)

4. We also know that missiles are weapons:

Missile(x) ⇒ Weapon(x)

5. We must know that an enemy of America counts as “hostile”:

Enemy(x, America) ⇒ Hostile(x)

6. “West, who is American . . .”:

American(West)

7. “The country Nono, an enemy of America . . .”:

Enemy(Nono, America)

8. “West is a criminal.”
Criminal(West)

We next write these sentences in the CNF as follows:

1. ¬American(x) ∨ ¬Weapon(y) ∨ ¬Sells(x, y, z)∨ / Hostile(z) ∨ Criminal(x)

2. ¬Missile(x) ∨ ¬Owns(Nono, x) ∨ Sells(West, x, Nono)

3. ¬Enemy(x, America) ∨ Hostile(x)

4. ¬Missile(x) ∨ Weapon(x)

5. Missile(M 1)

6. Owns(Nono, M 1)
 7.7. SUMMARY OF KNOWLEDGE REPRESENTATION METHODS 154

7. American(West)

8. Enemy(Nono, America)

9. Criminal(West)

The various satges of the resolution proof are shown in Figure 7.7

¬Am(x)∨We(y)∨Se(x, y, z)∨¬Ho(z)∨Cr(x) ¬Cr(West)

Am(West) ¬Am(West)∨¬We(y)∨¬Se(West,y,z)∨¬Ho(
z)
¬Mi(x)∨We(x) ¬We(y)∨¬Se(West,y,z)∨¬Ho(z
)
Mi(M1) ¬Mi(y)∨¬Se(West,y,z)∨¬Ho(z)

¬Mi(x)∨¬Ow(Nono,x)∨Se(West,x,Nono) ¬Se(West,M1,z)∨¬Ho(z)

Mi(M1) ¬Mi(M1)∨¬Ow(Nono,M1)∨¬Ho(Nono)

Ow(Nono,M1) ¬Ow(Nono,M1)∨¬Ho(Nono)

¬En(x, America)∨Ho(x) ¬Ho(Nono)

En(Nono, America) ¬En(Nono, America)

Contradiction

Figure 7.7: An example of a resolution-refutation proof

7.7 Summary of knowledge representation methods

In this and the previous chapter we have seen several knowledge representation methods that
are extensively used in artificial intelligence. In this section we make a brief summary of
these methods.

1. Knowledge and knowledge representation

Knowledge is a familiarity, awareness, or understanding of someone or something, such as
facts, skills, or objects. Knowledge can be acquired in many different ways and from
many sources, including perception, reason, memory, testimony, scientific inquiry,
education, and practice. There are different types of knowledge like declarative
knowledge, procedural knowledge, heuristic knowledge, structural knowledge, etc.
Knowledge representation is a field of artificial intelligence that deals with designing
computer representations that capture information and knowledge about the world that
can be used to solve problems. Knowledge representation is not just storing data into
some
7.7. SUMMARY OF KNOWLEDGE REPRESENTATION METHODS 155

database, but it also enables an intelligent machine to learn from that knowledge and
expe- riences so that it can behave intelligently like a human.
A knowledge representation structure (or, knowledge representation system) is a par-
ticular set of definitions, rules and procedures for setting up a representation that captures
information and knowledge about the world.
There are several knowledge representation systems such as the following:

(a) Semantic networks

(b) Frames

(c) Conceptual dependencies

(d) System based on logic

2. Semantic networks
Semantic networks are graphs in which vertices or nodes represent concepts or objects,
and the edges represent relations between the nodes. It is a knowledge representation
format that can be used to store the “meanings” of words so that human like use of these
meanings is possible.
Figure 7.8 shows a semantic network with four objects (John, Mary, 1 and 2) and four
categories (Mammals, Persons, FemalePersons and MalePersons).

Figure 7.8: A semantic network

3. Frames
The frame contains information on how to use the frame, what to expect next, and what
to do when these expectations are not met. Some information in the frame may remain
unchanged while others may change. The changing information are stored in “terminals”.
Terminals can be considered as variables. Different frames may share the same terminals.
Table 7.12 defines a frame with name ALEX. The various slots in the frame, their
names, their values and their types are all specified in the table.

4. Conceptual dependency
Conceptual dependency (CD) is a theory of knowledge representation to represent
natural language sentences in such way that the representation is independent of the
language in which the sentences are stated and it facilitates drawing inferences from
sentences.
7.7. SUMMARY OF KNOWLEDGE REPRESENTATION METHODS 156

Slot Value Type

ALEX - (This Frame)
NAME Alex (key value)
ISA BOY (parent frame)
SEX Male (inheritance value)
AGE IF-NEEDED: (procedural
Subtract(current,BIRTHDATE); attachment)
BIRTHDATE 8/4/2000 (instance value)
CLIMBS Trees (instance value)

Table 7.12: Example frame

The principal conceptual categories in CD theory are “real world actions” (denoted
by ACT), “real world objects” (PP), “modifiers of actions” (AA), “attributes of objects”
(PA), “times” (T) and “location” (LOC). The CD theory recognises 11 primitive ACTs
like “MOVE” (movement of a body part by its owner), “PROPEl” (application of
physical force to an object), etc.
The English sentence:
“John ate the ice cream with a spoon.”
is represented in CD theory as in Figure 7.9.

John
p O I
John INGEST ice cream
do
O
spoon

Figure 7.9: Representation of the sentence ”John ate the ice cream with a spoon.” in CD
theory

5. Knowledge representation using logic

Logic is the systematic study of valid rules of inference, that is, the relations that lead to
the acceptance of one proposition (the conclusion) on the basis of a set of other proposi-
tions (premises). Knowledge representation schemes are making use of logic to construct
computable models for given domains.
Now let us examine how logic can be used to form representations of the world and
how a process of inference can be used to derive new representations about the world and
how these can be used by an intelligent agent to deduce what to do.
For this we require a formal language to represent knowledge in a computer tractable
form and a reasoning process to manipulate this knowledge to deduce non-obvious facts.
7.8. SAMPLE QUESTIONS 157

Logic makes statements about the world which are true (or false) if the state of affairs
it represents is the case (or not the case). Compared to natural languages and
programming languages logic combines the advantages of natural languages and formal
languages. Logic is concise, unambiguous, context insensitive, expressive and effective
for inferences.
A logic is defined by the syntax that describes the possible configurations that
constitute sentences, semantics that determines what facts in the world the sentences refer
to and a proof theory which is a set of rules for generating new sentences that are
necessarily true given that the old sentences are true.
There are two kinds of logic: propositional logic and first-order logic. For example,
knowledge contained in the following sentences can be represented by propositional
logic:

If the program is efficient, it executes quickly. Either the program is

efficient, or it has a bug. However, the program does not execute quickly.
Therefore it has a bug.

We write
E: the program is efficient
Q: the program executes quickly
B: the program has a bug
The given argument can be represented as follows:

(E ⇒ Q) ∧ (¬E ⇒ B) ∧ ¬Q ⇒ B

The knowledge contained in the following statements can be represented using predi-
cate logic.

1. Cats like fish.

2. Cats eat everything they like.

3. Josephine is a cat.

4. Therefore, Josephine eats fish.

This argument can be represented using the notations of predicate logic as follows:

1. ∀ x cat(x) ⇒ likes(fish)

2. ∀ y ∀ z¬cat(y) ∨ ¬likes(y, z) ∨ eats(y, z)

3. cat(jo)

4. eats(jo, fish) (conclusion)

7.8 Sample questions

(a) Short answer questions
1. Define the domain of discourse in predicate logic and illustrate with an example.

2. Explain the existential and universal quantifiers.

7.8. SAMPLE QUESTIONS 158

3. Explain the sentences: “∀ x : x is a boy” and “∃ y : y has legs”.

4. Define a well-formed formula in FOPL.

5. When do we say that a logical statement is in clausal form?

6. Explain universal and existential instantiations in FOPL.

7. Explain the generalised modes ponens.

8. Explain the concept of unification in predicate logic.

9. Give a procedure for unification.

10. What is the resolution inference rule in FOPL?

11. Given the following statements:

“If it is a pleasant day you will do strawberry picking.” “If
you are doing strawberry picking you are happy.”
Use refutation resolution to conclude that “If it is pleasant day you are happy.”

(b) Long answer questions

1. Convert the following sentences to CNF: ∀ x P (x) ⇒ ∀ y ∃ x Q(Y, x).

2. From the sentence “Heads I win, tails you lose,” prove using resolution refutation
that “I win.”

3. Translate the following sentences into predicates in first order logic.

(a) Every gardener likes the sun.

(b) You can fool some of the people all of the time.
(c) You can fool all of the people some of the time.
(d) All purple mushrooms are poisonous.
(e) No purple mushroom is poisonous.
(f) There are exactly two purple mushrooms.
(g) There is one and only one Pope.

4. Translate the following into FOPL:

(a) No student loves Bill.

(b) Bill has at least one sister.
(c) Bill has at most one sister.
(d) Bill has exactly one sister.
(e) Bill has at least two sisters.
(f) Only one student failed History.
(g) No student failed Chemistry but at least one student failed History.

5. Prove the following argument by resolution-refutation:

7.8. SAMPLE QUESTIONS 159

• John likes all kind of food.

• Apple and vegetable are food.
• Anything anyone eats and not killed is food.
• Anil eats peanuts and is still alive.
• Harry eats everything that Anil eats.
• Therefore, John likes peanuts.

6. Give an algorithm to convert a well-formed formula to clausal form.

7. Describe the resolution-refutation algorithm in predicate logic.

8. Tony, Shi-Kuo and Ellen belong to the Hoofers Club. Every member of the
Hoofers Club is either a skier or a mountain climber or both. No mountain climber
likes rain, and all skiers like snow. Ellen dislikes whatever Tony likes and likes
whatever Tony dislikes. Tony likes rain and snow. Is there a member of the
Hoofers Club who is a mountain climber but not a skier?

9. Anyone passing his history exams and winning the lottery is happy. But anyone
who studies or is lucky can pass all his exams. John did not study but John is
lucky. Anyone who is lucky wins the lottery. Is John happy?

10. Consider the following axioms:

• All hounds howl at night.

• Anyone who has any cats will not have any mice.
• Light sleepers do not have anything which howls at night.
• John has either a cat or a hound.

Use resolution refutation to show that if John is a light sleeper, then John does not
have any mice.
Chapter 8

Planning

Planning is the task of finding a procedural course of action for a system to reach its
goals while optimizing overall performance measures. It is the first and foremost activity
to achieve desired results. Planning is a fundamental property of intelligent behavior.
This makes planning an important sub-area of artificial intelligence.
Planning techniques have been applied in a variety of tasks including robotics, pro-
cess planning, web-based information gathering, autonomous agents and spacecraft
mission control.

8.1 Overview
8.1.1 Decomposition of problems
When trying to solve a complicated problem, it may be necessary to work on small pieces
of the problem separately and then combine the partial solutions at the end into a
complete problem solution.
The ability to decompose a problem into small pieces to find a solution to the
problem is important because of the following reasons.

1. The decomposition may help us to avoid having to recompute the entire problem
state when we move from one state to next. This in turn help us to consider only
that part of the state that may have changed.

• For example, if we move from one room to another, it does not affect the
loca- tions of the doors and the windows in the two rooms.
• Consider the problem of guiding a robot in an ordinary house. The description
of a single state is very large since it must describe the locations of each and
every object in the house as well as that of the robot. A single action by the
robot will only change a small part of the total state. Instead of writing rules
that describe transformations one entire state into another, we would like to
write rules that describe only the affected parts of the state description.

2. The decomposition may divide a single difficult problem into several easier
subprob- lems.

160
8.2. COMPONENTS OF A PLANNING SYSTEM 161

8.1.2 Nearly decomposable problems

It may not be always possible to divide a problem into completely separate subproblems.
However, problems may be nearly decomposable by which we mean that they can be di-
vided into subproblems that have only a small amount of interaction.
For example, consider the problem of moving all furniture out of a room. This can be
divided into smaller problems each involving moving one piece of furniture. But if there
is a bookcase behind a couch, we must move the couch first before moving the book case.
Thus the subproblem of moving the couch and moving the bookcase interact.

8.1.3 Planning
Planning is the process of decomposing a problem into appropriate subparts and of
record- ing and handling interactions among the subparts as they are detected during the
problem solving process.
The word planning also refers to the process of computing several steps of a
problem- solving procedure before executing them.

8.1.4 Impact of predictability on planning

The success of the planning approach depends critically on the predictability of the prob-
lem’s domain or universe. If the universe is unpredictable we cannot be sure of the
outcome of a solution step. In such cases we can consider only a set of possible outcomes
possibly in some order according to the likelihood of the outcome occurring. Possibly the
plan have to include paths for all possible outcomes of each step. But there may a great
many outcomes for each step. In such situations it would be a great waste of effort to
formulate plans for all possibilities.
If the universe is unpredictable, we may take one step at a time and not really try to
plan ahead. Or, we we may try to produce a plan that is likely to succeed. If the plan
fails, we may start the planning process over again!

8.2 Components of a planning system

The following the components of a planning system:

1. Choosing rules to apply

2. Applying rules

3. Detecting a solution

4. Detecting dead ends

5. Repairing an almost correct solution

8.2.1 Choosing rules to apply

The commonly used technique can be described thus:

1. Isolate a set of differences between the desired goal state and the current state.
 8.2. COMPONENTS OF A PLANNING SYSTEM 162

2. Identify those rules that reduce the differences.

3. If several rules are hound, use heuristic methods to chose one.

8.2.2 Applying rules

In simple systems application of rules is easy. Each rule simply specified the problem state
that would result from the application of the rule.
In complex systems rules specify only a small part of the problem state. In such cases
there are several methods for the application of rules.

1. Describe each of the changes to the state description the application of the rule
makes. Add a new rule that everything else is remaining unchanged. To do this, we
may require the explicit statements of a large number of premises or axioms.

2. Describe each operator by a list of new predicates (called ADD) that the operator
causes to become true and a list of old predicates (called DELETE) that the
operator causes to become false. There is also a third list (called
PRECONDITION) which specifies the predicates that must be true for the operator
to be applied. Any predicate not included in either ADD or DELETE is is assumed
to be unaffected by the rule.

8.2.3 Detecting a solution

A planning system can be considered to have succeeded in finding a solution to a problem, if
it has found a sequence of operators that transforms the initial problem state into the goal
state. How does the system know whether it has succeeded in finding a solution? In
simple problems the system can easily determine whether it has succeeded in finding a
solution by checking whether the problem state and goal state match.
But in complex problems, in a description of a problem state, the entire state may not
be explicitly described; the state may be specified by a description of a smaller set of
relevant properties. So in such cases there should be a method to determine whether two
problem states match. This method will depend on how the various states are represented
in the planning system.
For example, suppose that a planning system is represented using predicate logic. Let
there be a predicate P (x) as part of the goal. To see whether P (x) has been satisfied at
any stage, it is enough to prove that we can derive P (x) from the assertions describing
that particular state. If we can construct such a proof, the problem-solving process can be
ended.

8.2.4 Detecting dead-ends

The planning system must be able to detect dead-end, that is, it must be able to determine
whether a path it is exploring can never lead to a solution or appears unlikely to lead to a
solution.
The mechanisms for detecting a solution are used to detect dead-ends also.

8.2.5 Repairing an almost correct solution

Consider the problem of solving a nearly decomposable problem.
We assume that the problem is completely decomposable and solve each subproblem
separately and then combine the solutions of the subproblems to obtain a solution of the
8.3. GOAL STACK PLANNING 163

original problem. Since the problem is only nearly decomposable, the combined solution
may not be the correct solution; it will only be an almost correct solution. The following
are some of the ways in which an almost correct solution may be repaired.
1. Throw out the almost correct solution and look for another one and hope that it is
better!
2. Consider the situation arising from the execution of the sequence of operations
speci- fied in the proposed almost correct solution. The difference between this and
the goal will much less than the difference between the initial state and the goal.
The problem solving system may be applied again to eliminate the difference.
3. Find exactly what went wrong and try a direct patch.
4. Apply the least commitment strategy. This means, leave the almost correct
solutions incompletely specified until the last possible moment and when as much
information as possible is available complete the specifications in such a way that
no conflicts arise.

8.3 Goal stack planning

8.3.1 Stacks
As the name indicates, in goal stack planning, we make use the stack data structure. A
stack is an Abstract Data Type (ADT), commonly used in most programming languages.
It is named stack as it behaves like a real-world stack, for example â A˘ S¸ a stack of
books or a pile of plates, etc.

Figure 8.1: A stack of books and a pile of plates

A real-world stack allows operations at one end only. For example, we can place or
remove a book or a plate from the top of the stack only. Likewise, the Stack ADT allows
all data operations at one end only. At any given time, we can only access the top
element of a stack.

8.3.2 Goal stack

In goal stack planning, we make use of stack called a goal stack usually denoted by GS.
The goal stack GS contains both subgoals and actions that have been proposed to satisfy
those subgoals. It relies on a database, denoted by DB, that describes the current
situation, and a set of actions described by PRECONDITION, ADD and DELETE lists.
8.3. GOAL STACK PLANNING 164

8.3.3 Basic idea

At each succeeding step of the problem solving process, a subgoal of the stack is pursued.
An action that could cause it to be true is sought. If an appropriate action is found, its
precondition is established as the top subgoal of the stack in order for that action to be
applicable. When a sequence of actions that satisfies a subgoal is found, that sequence is
applied to the current situation, yielding a new current situation. Then, all the subgoals
that are satisfied by the new current situation are popped off the stack. Next, another
subgoal of the stack is explored and an attempt is made to satisfy it, starting from the
situation that was produced as a result of satisfying the first subgoal.
This process continues until the goal stack is empty. Then, as one last check, the
original goal is compared to the final situation derived from the application of the chosen
actions. If any conditions on the goal are not satisfied in that situation (which they might
not be if they were achieved at one point and then undone later), then those unsolved
subgoals are reinserted onto the stack and the process resumed.

8.3.4 The STRIPS planner

The reasoning strategy used by STRIPS is goal stack planning. STRIPS (Stanford
Research Institute Problem Solver) is an automated planner developed by Richard Fikes
and Nils Nilsson in 1971. The same name was later used to refer to the formal language
of the inputs to this planner. This language is the base for most of the languages for
expressing automated planning problem instances in use today; such languages are
commonly known as action languages.

8.3.5 Algorithm
1. Push the compound predicate describing the goal state in to the Stack.
2. Push the individual predicates of the goal state in to the Stack.
3. Loop till the Stack is empty
(a) Pop an element E from the Stack.
(b) If E is a predicate
i. If E is True
A. Do nothing.
ii. Else
A. Push the relevant action into the Stack.
B. Push the individual predicates of the precondition of the action into
the Stack.
(c) Else if E is an action “a”
i. Apply the action “a” to the current state.
ii. Add the action “a” to the plan.

8.3.6 Illustration: Blocks world

To illustrate the goal stack planning approach, we consider the blocks world domain (see
Section 2.3.5 also).
8.3. GOAL STACK PLANNING 165

1. The objects of the blocks world

• A flat surface on which blocks can be placed.

• A number of square blocks, all of the same size.

• A robot arm that can manipulate the blocks.

• Blocks can be stacked one upon another.

2. Predicates and relations to describe the state of the blocks

world PredicateMeaning

ON(x, y) Block x is on block y

ONTABLE(x) Block x is on table.
CLEAR(x) There is nothing on top of block x
HOLDING(x) The arm is holding block x
ARMEMPTY The arm is holding nothing.

3. Actions
The following are the available actions in the blocks world.

Action Meaning

UNSTACK(x, y) Pick up block x from its current position on block y and

hold it in the arm
STACK(x, y) Place block x held in the arm on block y.
PICKUP(x) Pick up block x from table and hold it.
PUTDOWN(x) Put block x held in the arm down on the table.

The PRECONDITION list (the list of predicates that should be true to apply the action),
ADD list (the list of predicates that become true after the action) and DELETE list (the
list of predicates that become false after the action) associated with the various actions
are given below.
8.3. GOAL STACK PLANNING 166

Action PRECONDITION ADD list DELETE list

list

UNSTACK(x, y) ARMEMPTY, HOLDING(x), ARMEMPTY

CLEAR(x), ON(x, y) CLEAR(y)
STACK(x, y) HOLDING(x), ARMEMPTY,
CLEAR(y) ON(x, y),
CLEAR(x)
PICKUP(x) ARMEMPTY,
HOLDING(x) ARMEMPTY
CLEAR(x).
PUTDOWN(x) HOLDING(x). ARMEMPTY,
ONTABLE(x),
CLEAR(x)

4. Problem
Given the initial and the final states of a blocks world with four blocks as shown in Figure
8.2, use goal stack planning algorithm to obtain a plan for achieving the goal state.

Figure 8.2: A simple blocks world problem: Initial state on the left and the goal state on the
right

5. Solution
Step 1. The initial state can be described by the statement:

ON (B, A)∧ONTABLE(A)∧ONTABLE(C)∧ONTABLE(D)∧ARMEMPTY

The goal state can be specified by the statement:

ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Step 2. We create a stack and call it GS. We initialise GS to be empty.

(empty)

Step 3. We push the goal state to the stack:

8.3. GOAL STACK PLANNING 167

ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Step 4. The goal state can be divided into four components, namely,

ON (C, A), ON (B, D), ONTABLE(A), and ONTABLE(D)

Of these the last two are true in the initial state. We push the first two
components to the stack GS.

ON (C, A)
ON (B, D)
ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Note: The choice of the order in which the components are stacked is indeed im-
portant. For example, the two new components may be stacked in the following
order also. We have chosen the order as above in a purely arbitrary manner.
How- ever, to make an informed choice, we may need additional knowledge
about the world in which we are operating.

ON (B, D)
ON (C, A)
ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Step 5. Consider the top item in GS which is ON (C, A). Check whether it is true in cur-
rent state. It is not. Find an action which makes it true. the action is STACK(C, A).
We replace ON (C, A) by STACK(C, A) to get the following GS.

STACK(C, A)
ON (B, D)
ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Step 6. To apply the action STACK(C, A) the preconditions CLEAR(A)∧HOLDING(C)

must be true. This precondition and also its components are also pushed to the
stack GS.
8.3. GOAL STACK PLANNING 168

CLEAR(A)
HOLDING(C)
CLEAR(A) ∧ HOLDING(C)
ON (C, A)
ON (B, D)
ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Note: As in Step 4, here also the order in which the components CLEAR(A)
and HOLDING(C) are stacked has been chosen arbitrarily. The command
HOLDING(C) may be stacked above CLEAR(A).

Step 7. We now check whether CLEAR(A) is true. It is not. the only operator that make
it true is UNSTACK(B, A). So we replace CLEAR(A) by UNSTACK(B, A).

UNSTACK(B, A)
HOLDING(C)
CLEAR(A) ∧ HOLDING(C)
STACK(C, A)
ON (B, D)
ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Step 8. To apply UNSTACK(B, A), the precondition ON (B, A)∧CLEAR(B)∧ARMEMPTY

must be true. We add this precondition and its components to the stack GS.
 8.3. GOAL STACK PLANNING 169

ON (B, A)
CLEAR(B)
ARMEMPTY
ON (B, A) ∧ CLEAR(B) ∧ ARMEMPTY
UNSTACK(B, A)
HOLDING(C)
CLEAR(A) ∧ HOLDING(C)
STACK(C, A)
ON (B, D)
ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Step 9. The top element ON (B, A) is true and so we pop it off from the stack. The
next elemt CLEAR(B) is also true and so we pop it off also. At the current
state, the arm is not holding anything and so ARMEMPTY is also true and
we pop it off also. Sine each of the components of the next statement is true, the
compounded statement is also true. Thu it is also popped off from the stack
GS. At this stage the stack is as follows.

UNSTACK(B, A)
HOLDING(C)
CLEAR(A) ∧ HOLDING(C)
STACK(C, A)
ON (B, D)
ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Step 10. The top element of the stack is UNSTACK(B, A). The preconditions for apply-
ing this action are satisfied.

(a) We now apply this operator to get a new world state.

(b) We record that UNSTACK(B, A) is the first operator of the solution
se- quence.
(c) The state of the world is specified by the following statements:

ONTABLE(A), ONTABLE(C), ONTABLE(D),

HOLDING(B), CLEAR(A)
 8.4. HIERARCHICAL PLANNING 170

(d) The goal stack GS is now

HOLDING(C)
CLEAR(A) ∧ HOLDING(C)
STACK(C, A)
ON (B, D)
ON (C, A) ∧ ON (B, D) ∧ ONTABLE(A) ∧ ONTABLE(D)

Step 11. The process is now repeated.

8.4 Hierarchical planning

8.4.1 The basic approach
In order to solve real world problems, the problem solver may have to generate long
plans. To do this efficiently, the problem solver may ignore some of the details of the
problem until a solution to the main issues have been found. This can be done by creating
a hierarchy of of the tasks to be performed and then attempting to find a solution starting
from the tasks at the highest level of the hierarchy. Creating a plan in this way is known
as hierarchical planning.

Example
• For example, let it be required to travel at short notice to Mumbai with minimum
expenses for an interview. The task of the highest priority is to check the
availability of an airline at affordable cost. After finding a plan to solve this
problem, the person may find plans for solving the next level problems like
preparing baggage and hiring a taxi. Attempts to solve these problems lead to
problems at the third level: detailed plans for preparing the baggage and hiring a
taxi. These will in turn lead to problems at the next lower level.

8.4.2 The ABSTRIPS planner system

ABSTRIPS, a planning system developed in 1974 and built on top of the STRIPS
planning system, is a planning system has a method for the implementation of
hierarchical planning. In ABSTRIPS, the hierarchies are constructed by assigning
criticalities which are numbers indicating relative difficulty to the preconditions of the
operator. First a plan is found that satisfies only the preconditions of the the operators of
the highest criticality values. The plan is then refined by considering the preconditions at
the next level of criticality and in- serting steps into the plan to achieve these
preconditions. The process is repeated to the lowest level of criticality. For this system
to work, there must be a scheme to assign the appropriate critcality values to the various
terms in the precondition. Because this process explores entire plans at one level of
detail before looking any of the lower details, it has
8.5. SAMPLE QUESTIONS 171

been called length-first search.

8.5 Sample questions

(a) Short answer questions
1. Define planning in AI.

2. List the components of planning.

3. give examples of planning strategies.

(b) Long answer questions

1. Explain the components of planning.

2. Explain goal stack planning.

3. Describe the goal stack planning algorithm.

4. Explain hierarchical planning.

 Chapter 9

Learning

9.1 What is learning?

Learning denotes changes in the system that are adaptive in the sense that they enable the
system to do the same task or tasks drawn from the same population more efficiently and
more effectively the next time.1

9.2 Forms of learning based on feedback

Consider a class of problems in which from a collection of input-output pairs a machine
learns a function that predicts the output for new inputs.

1. Unsupervised learning
In unsupervised learning the agent learns patterns in the input even though no explicit feed-
back is supplied. The most common unsupervised learning task is clustering: detecting
potentially useful clusters of input examples. For example, a taxi agent might gradually
develop a concept of “good traffic days” and ”bad traffic days” without ever being given
labeled examples of each by a teacher.

2. Reinforcement learning
In reinforcement learning the agent learns from a series of reinforcements - rewards or
punishments. For example, the lack of a tip at the end of the journey gives the taxi agent
an indication that it did something wrong. The two points for a win at the end of a chess
game tells the agent it did something right.

3. Supervised learning
In supervised learning the agent observes some example input-output pairs and learns a
function that maps from input to output.
1
H A Simon, ”Why Should Machines Learn?”, Chapter 2 in Machine Learning, An Artificial Intelligence
Approach, Tioga Publishing Company, 1983.

172
9.3. FORMS OF LEARNING BASED ON DIFFERENT WAYS OF ACQUIRING
KNOWLEDGE173
4. Semi-supervised learning
In semi-supervised learning we are given a few labeled examples and must make what
we can of a large collection of unlabeled examples. Imagine that you are trying to build a
system to guess a person’s age from a photo. We gather some labeled examples by
snapping pictures of people and asking their age. That is supervised learning. But in
reality some of the people lied about their age and to uncover them is an unsupervised
learning problem.

9.3 Forms of learning based on different ways of acquiring knowl-

edge
9.3.1 Rote learning
Rote learning is simply storing data in memory. This helps programmes to perform better
in the future. We store computed values so that we do not have to recompute them later.
In complex learning system this requires capabilities for organised storage of information
and generalisation.

Example learning system

A programme to play checkers developed by Arthur Samuel2 in 1963 used the rote
learning technique to improve its performance.

9.3.2 Learning by taking advice

The idea of advice taking in AI based learning was proposed as early as 1958 by John
McCarthy3. However very few attempts were made in creating such systems until the late
1970s.
There are two basic approaches to advice taking:

• Take high level, abstract advice and convert it into rules that can guide
performance elements of the system. Automate all aspects of advice taking.

• Develop sophisticated tools such as knowledge base editors and debugging. These
are used to aid an expert to translate his expertise into detailed rules. Here the
expert is an integral part of the learning system.

Example learning system

FOO (First Operational Operationaliser) is a programme developed in 1983 which
accepts advice for playing hearts, a card game. It tries to convert high level advice
(principles, problems, methods) into effective executable (LISP) procedures. Hearts is a
game of partial information with no known algorithm for winning.
Arthur Lee Samuel (1901 âA˘ S¸ 1990) was an American pioneer in the field of computer gaming and
2

artifi- cial intelligence. He popularized the term “machine learning” in 1959. The Samuel Checkers-playing
Program was among the world’s first successful self-learning programs, and as such a very early
demonstration of the fundamental concept of artificial intelligence (AI).
John McCarthy (1927 âA˘ S¸ 2011) was an American computer scientist and cognitive scientist. McCarthy
3

was one of the founders of the discipline of artificial intelligence. He co-authored the document that coined
the term “artificial intelligence” (AI), and was very influential in the early development of AI.
9.3. FORMS OF LEARNING BASED ON DIFFERENT WAYS OF ACQUIRING
KNOWLEDGE174
General advice can be given such as “Avoid taking points”. A human must convert
into a FOO representation of the form

(avoid (take-points me) (trick))

FOO operationalises the advice by translating it into expressions it can use in the game.

(achieve (not (during

(scenario
(each p1 (players) (play-card
p1)) (take-trick (trick-winner)))
(take-points me))))

9.3.3 Learning in problem solving

There are three basic methods in which a system can learn from its own experiences.

1. Learning by parameter adjustment

Many programs rely on an evaluation procedure to summarise the state of search etc.
Game playing programs provide many examples of this. However, many programs have
a static evaluation function.
So the basic idea of idea of parameter adjustment is to:

• Start with some estimate of the correct weight settings.

• Modify the weight in the program on the basis of accumulated experiences.

• Features that appear to be good predictors will have their weights increased and
bad ones will be decreased.

Example learning system

Samuel’s Checkers programs employed used a static function of the form

c1t1 + c2t2 + · · · + c16t16

where the t terms are values of the 16 features and the c terms are weights. As learning
progresses, the c values change.

2. Learning with macro-operators

The basic idea here is similar to rote learning: "Avoid expensive recomputation".
Sequences of actions that can be treated as a whole are called macro-operators. For
example making dinner can be described by the macro-actions lay the table, cook dinner,
serve dinner. We could treat laying the table as on action even though it involves a
sequence of actions.
9.3. FORMS OF LEARNING BASED ON DIFFERENT WAYS OF ACQUIRING
KNOWLEDGE175
Example learning system
The STRIPS problem-solving system employed macro-operators in it’s learning phase.
The STRIPS system has a learning component: After each problem solving episode, the
system stores the computed plan as a macro-operator.
Consider a blocks world example in which ON(C,B) and ON(A,TABLE) are true.
STRIPS can achieve ON(A,B) in four steps:

UNSTACK(C, B), PUTDOWN (C), PICKUP (A), STACK(A, B)

STRIPS now builds a macro-operator MACROP with preconditions ON(C,B),

ON(A,TABLE), postconditions ON(A,B), ON(C,TABLE) and the four steps as its body.
MACROP can now be used in future operation. But it is not very general. The above can be
easily generalised with variables used in place of the blocks.

3. Learning by chunking
Chunking involves ideas similar to macro-operators and originates from psychological ideas
on memory and problem solving. Its computational basis is in production systems.

Example learning system

The idea of learning by chunking has been used in the SOAR system which uses
production rules to represent its knowledge. SOAR is an architecture for general
intelligence developed by John E Laird and others in 1987.

9.3.4 Inductive learning

This involves the process of learning by example – where a system tries to induce a
general rule from a set of observed instances.
Inductive learning involves classification, that is, assigning to a particular input the
name of a class to which it belongs. Classification is important to many problem solving
tasks. A learning system has to be capable of evolving its own class descriptions.But it
is sometimes difficult to construct class definitions by hand. This is particularly true in
domains that are not well understood or that change rapidly. The task of constructing
class definitions is called induction or concept learning.

Example learning systems

The following are examples of concept learning programmes.

• Winston’s learning programme: This programme operated in the blocks world do-
main. Its goal was to construct representations of concepts in the blocks domain.

• The version space approach introduced by Mitchell in 1977 attempts to produce a

description that is consistent with all positive examples but no negative examples
in the training set.

• A third approach to concept learning is the induction of decision trees as

exemplified by the ID3 programme of Quinlan (1986). ID3 is a program that can
build trees automatically from given positive and negative instances. Basically
each leaf of a
 9.3. FORMS OF LEARNING BASED ON DIFFERENT WAYS OF ACQUIRING
KNOWLEDGE176

Figure 9.1: A decision tree

decision tree asserts a positive or negative concept. To classify a particular input we

start at the top and follow assertions down until we reach an answer (see Figure
9.1).

9.3.5 Explanation based learning

Humans appear to learn quite a lot from one example. The basic idea of explanation based
learning is to use results from one example problem solving effort the next time around.
An explanation based learning accepts four kinds of input:

• A training example: what the learning program “sees” in the world.

• A goal concept: A high level description of what the program is supposed to learn.

• An operational criterion: A description of which concepts are usable.

• A domain theory: A set of rules that describe relationships between objects and
actions in a domain.

From this, EBL computes a generalisation of the training example that is sufficient to de-
scribe the goal concept and also satisfies the operationality criterion.

9.3.6 Discovery
Discovery is a restricted form of learning in which one entity acquires knowledge without the
help of a teacher. Discovery can be of two types:

• Theory driven discovery

• Data driven discovery

Theory driven discovery: Example learning system

The Automated Mathematician (AM) is one of the earliest successful discovery systems. It
was created by Douglas Lenat in 1970’s.
AM has two inputs: One, a description of some concepts of set theory, for example, set
union, intersection, the empty set; two, information on how to perform mathematics.
Given the above information AM discovered:

• Integers: it is possible to count the elements of a set.

 9.3. FORMS OF LEARNING BASED ON DIFFERENT WAYS OF ACQUIRING
KNOWLEDGE177
• Addition: The union of two disjoint sets and their counting function.

• Multiplication: Having discovered addition and multiplication as laborious set-

theoretic operations more effective descriptions were supplied by hand.

• Prime Numbers: Factorisation of numbers and numbers with only one factor were
discovered.

• Golbach’s Conjecture: Even numbers can be written as the sum of 2 primes. E.g. 28
= 17 + 11.

Data driven discovery: Example learning system

Many discoveries are made from observing data obtained from the world and making sense
of it, for example, discovery of planets, quantum mechanics, discovery of sub-atomic par-
ticles.
BACON4 is an attempt to provide such an AI system. BACON was able able to de-
rive the ideal gas law. BACON has also been applied to Kepler’s third law, Ohm’s law,
conservation of momentum and Joule’s law.

9.3.7 Analogy
Analogy involves a mapping between what might appear to be two dissimilar concepts.
For example, consider the following statement: “Finding a good man is like finding
a needle in a haystack.” This involves a mapping between two worlds: One, a world of
men, good men and searching for good men; two, a world of small objects, a haystack
and searching for a needle.
There are two methods of analogical problem solving in AI: transformational analogy
and derivational analogy.

Transformational analogy
Carbonell (1983) describes a T-space method to transform old solutions into new ones.

• Whole solutions are viewed as states in a problem space – the T-space.

• T-operators prescribe methods of transforming existing solution states into new ones.

• Reasoning by analogy becomes a search in T-space – means-end analysis.

Derivational analogy
Derivational analogy is a method of solving problems based upon the transfer of past experi-
ence to new problem situations. The experience transfer process consists of recreating
lines of reasoning, including decision sequences and accompanying justifications, that
proved effective in solving particular problems requiring similar initial analysis. The
derivational analogy approach is advocated as a means of implementing reasoning from
individual cases in expert systems.
4
Named after the philosopher of science Sir Francis Bacon (1561 - 1626).
9.4. NEURAL NET LEARNING 178

9.4 Neural net learning

9.4.1 Neurons and perceptrons
An Artificial Neuron Network (ANN), popularly known as Neural Network is a computa-
tional model based on the structure and functions of biological neural networks. It is like
an artificial human nervous system for receiving, processing, and transmitting
information. Figure 9.2 shows the structure of a neuron in the human nervous system.

Figure 9.2: The structure of a neuron

In an ANN, each neuron receives one or more inputs and produces one or more
outputs. In its simplest form known as a perceptron a neuron takes a weighted sum of its
inputs and produces an output 1 if the sum is greater than a certain pre-defined threshold
value and 0 otherwise. The relation between the inputs and the output of a perceptron can
be represented as in Figure 9.3. In the figure the output is defined as follows:

Figure 9.3: The model of a perceptron

(
1 if w1x1 + w2x2 +·····+ wnxn > w
y = 0 otherwise

where w is some threshold value.

9.4. NEURAL NET LEARNING 179

9.4.2 Neural networks

A neural networks consists of a number of neurons structured into different layers as in
Figure 9.4 where each circle represents a neuron.
Basically, there are three different layers in a neural network:

1. Input layer: All the inputs are fed in the model through this layer.

2. Hidden layers: There can be more than one hidden layers which are used for
process- ing the inputs received from the input layers.

3. Output layer: The data after processing is made available at the output layer.

Figure 9.4: The different layers of a neural network

9.4.3 Learning in neural networks

In neural networks, learning is the process of modifying the values of the weights and the
threshold. The learning takes place through an iterative process of “going and return” by
the layers of neurons. The “going” is a forward propagation of information and the
“return” is a backward propagation of information.

9.4.4 Applications
Artificial neural networks have found applications in many disciplines. Application areas
include:

1. vehicle control, trajectory prediction

2. natural resource management

3. pattern recognition (radar systems, face identification, signal classification, 3D re-

construction, object recognition and more)

4. sequence recognition (gesture, speech, handwritten and printed text recognition)

5. medical diagnosis

6. finance (e.g. automated trading systems)

 9.5. GENETIC ALGORITHMS 180

7. data mining

8. visualization

9. machine translation

10. e-mail spam filtering

9.5 Genetic algorithms

A genetic algorithm is an algorithm inspired by the process of natural selection in the evo-
lution of organisms. Genetic algorithms are commonly used to generate high-quality
solu- tions to optimization and search problems.

Figure 9.5: A genetic algorithm flowchart

9.5.1 Outline
In a genetic algorithm, a population of candidate solutions (called individuals, creatures, or
phenotypes) to an optimization problem is evolved toward better solutions. Each
candidate solution has a set of properties (called its chromosomes or genotype) which can
be mutated and altered; traditionally, solutions are represented in binary as strings of 0s
and 1s.
The evolution usually starts from a population of randomly generated individuals, and
is an iterative process. The population in each iteration called a generation. In each gener-
ation, the fitness of every individual in the population is evaluated; the fitness is usually
the value of the objective function in the optimization problem being solved. The more
fit in- dividuals are randomly selected from the current population, and each individual’s
genome (the set of chromosomes) is modified (recombined and possibly randomly
mutated) to form a new generation. The new generation of candidate solutions is then
used in the next it- eration of the algorithm. Commonly, the algorithm terminates when
either a maximum
9.6. SAMPLE QUESTIONS 181

number of generations has been produced, or a satisfactory fitness level has been reached
for the population.
The chromosome representation, selection, crossover (or, recombination), mutation,
and fitness function computation are the key elements of the genetic algorithm.
A genetic algorithm flowchart is shown in Figure 9.5.

9.6 Sample questions

(a) Short answer questions
1. What is learning?

2. Describe supervised, unsupervised and reinforcement learning.

3. List the different types of learning based on feedback.

4. Explain inductive learning.

5. Explain explanation based learning.

6. Explaining discovery learning.

7. Describe the use of analogy in learning.

(b) Long answer questions

1. Explain the different types of learning based on different ways of acquiring knowl-
edge.

2. Explain different types of learning in problem solving.

3. Describe the basic ideas of neural net learning.

4. With the help of a flowchart, explain the basic concepts of the genetic algorithm.
 Chapter 10

Expert systems

10.1 Introduction
An expert system is a computer program that uses artificial intelligence methods to solve
problems within a specialized domain that ordinarily requires human expertise. The first
expert system was developed in 1965 by Edward Feigenbaum and Joshua Lederberg of
Stanford University. Their system known as Dendral was designed to analyze chemical
compounds. Expert systems now have commercial applications in fields as diverse as
med- ical diagnosis, petroleum engineering, and financial investing.
In order to accomplish feats of apparent intelligence, an expert system relies on two
components: a knowledge base and an inference engine. A knowledge base is an
organized collection of facts about the system’s domain. An inference engine interprets
and evaluates the facts in the knowledge base in order to provide an answer. Typical
tasks for expert systems involve classification, diagnosis, monitoring, design, scheduling,
and planning for specialized endeavours.
Facts for a knowledge base must be acquired from human experts through interviews
and observations. This knowledge is then usually represented in the form of “if-then”
rules. The knowledge base of a major expert system includes thousands of rules. A
probability factor is often attached to the conclusion of each rule and to the ultimate
recommendation, because the conclusion is not a certainty.

10.2 Architecture of an expert system

Figure 10.1 shows the typical architecture of an expert system. The various components are
explained below.

10.2.1 User interface

The user interacts with the system through a user interface which may use menus, natural
language or any other style of interaction.
The user interface is closely linked to the explanation component described below.
This is the dialogue component of the system. One side of the dialogue involves the user
ques- tioning the system and on the other side the system must be able to question the
user to establish the existence of evidence. The dialogue component has two functions;
determines which question to ask next and keep record of the previous questions.

182
10.2. ARCHITECTURE OF AN EXPERT SYSTEM 183

Figure 10.1: Typical architecture of an expert system

The dialogue could be one of three styles:

• System controlled, where the system drives the dialogue through questioning the user.

• User controlled, where the user drives the consultation by providing information to
the system.

• Mixed control, where both user and system can direct the consultation.

10.2.2 Knowledge base

This element of an expert system system is so critical to the way the most expert system
are constructed that they are also popularly known as knowledge based systems.
A knowledge base contains both declarative knowledge (facts about objects, events
and situations) and procedural knowledge (information about course of action). Although
many knowledge representation techniques have been used in expert system the most
common form is the rule-based production system approach.
In rule-based systems the procedural knowledge, in the form of heuristic “if- then”
rules, is completely integrated with the declarative knowledge. However, not all rules
pertain to the system’s domain. There are rules called meta-rules which pertain to other
rules (or even to themselves). Here is an example of a meta-rule from MYCIN:
“If there are rules which do not mention the current goal in their premise or if there
are rules which mention the current goal in their premise, then it is definite that the
former should be done before the latter.”

10.2.3 Inference engine

By simply having access to a great deal of knowledge does not make us an expert, we
must know how or when to apply the appropriate knowledge. Similarly just having a
knowledge base does not make an expert system intelligent. The system must have
another component which directs the implementation of knowledge base. This
component of the system is the inference engine.
The inference engine uses the information provided to it by the knowledge base and
the user to infer new facts. It also decides which heuristic search techniques are to be
used in determining how the rules in the knowledge are to be applied to the problem. In
fact an inference engine “runs” an expert system, determining which rules are to be
invoked,
10.2. ARCHITECTURE OF AN EXPERT SYSTEM 184

accessing the appropriate rules, executing the rules and determining when an acceptable
solution has been found.
The reasoning process can be simple or complicated, depending on the structure of
knowledge base. If the knowledge base consists of simple rules and facts, forward
chaining suffices. However, for a knowledge which consists of rules and unstructured
logic (facts, data and variables), both sophisticated forward and backward chaining with
well thought- out search strategies may be required. Other methods such as problem
reduction, pattern matching, unification etc. are also used to implement the reasoning
process.
Mostly, the knowledge is not inter twined with the control structure. This has a value-
added advantage, the same inference engine which works well in one expert system may
work just as well with a different knowledge base to make another expert system, thus
reducing expert system developing time.
For example, inference engine of MYCIN is available separately as EMYCIN
(essential MYCIN). EMYCIN has been used with different knowledge base to create many
new expert system, eliminating the need to develop a new inference engine. This will be
taken up after having studied expert system shells.

10.2.4 Case specific data

The case specific data includes both data provided by the user and partial conclusions
(along with certainty measures) based on this data. In a rule-based system the case
specific data will be the elements in working memory.

10.2.5 Explanation system

The explanation system allows the program to explain its reasoning to the user. It is not
ac- ceptable for an expert system to take decisions without being able to provide an
explanation for the decisions it has taken. Users using these expert systems need to be
convinced of the validity of the conclusion drawn before applying it to their domain.
They also need to be convinced that the solution is appropriate and applicable in their
circumstances.
Knowledge engineers building the expert system also need to examine the reasoning
behind decisions in order to access and evaluate the mechanisms being used. If
explanation component is not provided it would not be possible to judge whether the
expert system is working as desired or intended.

10.2.6 Knowledge base editor

Most expert systems provide a mechanism for editing the knowledge base. In the
simplest case, this is just a standard text editor for modifying rules and data by hand. But
many tools include other facilities in their support environment. For example, EMYCIN
uses automatic book-keeping.
Another common facility in knowledge base editors is syntax checking, where the
edi- tor uses knowledge about the grammatical structure of the expert system language to
help the user input rules with the correct spelling and format. When the user enters an
ungram- matical rule or command, the editor catches it and explains what is wrong.
An extremely useful but generally unavailable facility for knowledge base editors is
consistency checking where the system checks the semantics or meanings of the rules and
data being entered to see if they conflict with existing knowledge in the system. When a
10.3. ROLES OF INDIVIDUALS WHO INTERACT WITH EXPERT SYSTEMS 185

conflict occurs, the editor helps the user resolve the conflict by explaining what caused it
and describing ways to remove it.
Another potentially useful but generally unavailable facility for knowledge base edi-
tors is knowledge extraction where the editor helps the user enter new knowledge into the
system. It combines syntax and consistency checking with sophisticated prompting and
explanation to let even naive users add or modify rules.

10.2.7 Expert shell

One important feature of expert systems is the way they separate domain specific
knowledge from more general purpose reasoning and representation techniques. The
general purpose bit (in the dotted box in figure 10.1) is referred to as an expert system
shell. The shell provides the inference engine, the user interface, an explanation system
and sometimes a knowledge base editor. There are numerous commercial expert system
shells, each one appropriate for a slightly different range of problems.

10.3 Roles of individuals who interact with expert systems

• Domain expert The domain experts are the individuals who are currently experts
in solving the problem which the system is intended to solve.

• Knowledge engineer
The knowledge engineer is the person who encodes the expert’s knowledge in a
declarative form that can be used by the expert system.

• User
The user is the person who will be consulting with the system to get advice which
would have been provided by the expert.

• System engineer
The system engineer builds the user interface, designs the declarative format of the
knowledge base, and implements the inference engine.

10.4 Languages and tools

Expert system tools are programming systems which simplify the job of constructing an
expert system. They range from very high-level programming language to low-level
support facilities. We divide expert system tools into four major categories as shown in
Figure 10.2. We will describe these tools and explain briefly how they are used.

10.4.1 Programming languages

Most important programming languages used for expert system applications are generally
either problem-oriented languages such as FORTRAN and PASCAL, or symbol-manipulation
10.4. LANGUAGES AND TOOLS 186

Figure 10.2: Types of expert systems

languages such as LISP and PROLOG. Symbol-manipulation languages are designed for
ar- tificial intelligence applications; for example, LISP has mechanism for manipulating
sym- bols in the form of list structures. A list is simply a collection of items enclosed by
paren- thesis, where each item can be either a symbol or another list. List structures are
useful building blocks for representing complex concepts.
The most popular and widely used programming language for expert system applica-
tions are LISP and PROLOG. Symbol-manipulation languages like these are more suit-
able for work in artificial intelligence, although a few expert systems have been written in
problem-oriented languages like FORTRAN and PASCAL.
Expert systems are also written in CLIPS (C Language Integrated Production System
developed in mid 1980s). The major advantages of these languages as compared to
conven- tional programming languages, is the simplicity of the addition, elimination or
substitution of new rules and memory management capabilities.

10.4.2 Knowledge engineering languages

A knowledge engineering language is a sophisticated tool for developing expert systems.
It consists of an expert system building language integrated into an extensive support
environ- ment. A knowledge engineering language is a type of programming language
designed to construct and debug expert system. Knowledge engineering languages can be
categorized as either skeletal systems or general-purpose systems.
10.5. TYPICAL EXPERT SYSTEMS 187

A skeletal knowledge engineering language is a stripped down expert system, that is,
an expert system with its domain-specific knowledge removed leaving only the inference
en- gine and support facilities. The general-purpose knowledge engineering language can
han- dle many different problem areas such as knowledge extraction, giving inference or
making user interface though its use is rather tedious. These languages have a range of
generality and flexibility. Table 10.1 shows some well known knowledge engineering
languages.

Table 10.1: Knowledge engineering languages for building expert systems

10.4.3 System-building aids

The system-building aids consist of programs which help acquire and represent the
domain expert’s knowledge and programs which help design the expert system under
construc- tion. These programs address very difficult tasks; many are research tools just
beginning to evolve into practical and useful aids, although a few are offered as full-
blown commercial systems.
Compared with programming and knowledge engineering languages, relatively few
system-building aids have been developed. Those which exist fall into two major cate-
gories; design aids and knowledge acquisition aids. The AGE system exemplifies design
aids, while TEIRSIAS, MOLE and SALT exemplify knowledge acquisition, TIMM
system construction and SEEK knowledge refinement aids.

10.4.4 Tool support environment facilities

These are software packages which come with each tool to make it more user friendly
and more efficient. The different components of a tool support environment are shown in
Figure 10.3.

10.5 Typical expert systems

1. Dendral
10.5. TYPICAL EXPERT SYSTEMS 188

Figure 10.3: Components of a support environment for expert system tools

Dendral was a project in artificial intelligence of the 1960s. Its primary aim was to
study hypothesis formation and discovery in science. For that, a specific task in
sci- ence was chosen: help organic chemists in identifying unknown organic
molecules, by analyzing their mass spectra and using knowledge of chemistry. It
began at Stan- ford University in 1965 and spans approximately half the history of
AI research. Den- dral is considered the first expert system because it automated
the decision-making process and problem-solving behavior of organic chemists.

2. MYCIN
MYCIN was an early backward chaining expert system that used artificial intelli-
gence to identify bacteria causing severe infections, such as bacteremia and
meningi- tis, and to recommend antibiotics, with the dosage adjusted for patient’s
body weight. The name was derived from the antibiotics themselves, as many
antibiotics have the suffix "-mycin". The Mycin system was also used for the
diagnosis of blood clotting diseases. MYCIN was developed over five or six years
in the early 1970s at Stanford University. It was written in Lisp.
MYCIN operated using a fairly simple inference engine and a knowledge base of
approximately 600 rules. It would query the physician running the program via a
long series of simple yes/no or textual questions. At the end, it provided a list of
possible culprit bacteria ranked from high to low based on the probability of each
diagnosis, its confidence in each diagnosis’ probability, the reasoning behind each
diagnosis, and its recommended course of drug treatment.

3. DART (Diagnostic Assistance Reference Tool)

DART is a joint project of the Stanford University and IBM that explores the
applica- tion of artificial intelligence techniques to the diagnosis of computer faults.
It assists a
10.6. BENEFITS AND LIMITATIONS 189

technician in finding the faults in a computer system. The primary goal of the
DART Project was to develop programs that capture the special design knowledge
and di- agnostic abilities of the experts and to make them available to field
engineers. The practical goal is the construction of an automated diagnostician
capable of pinpoint- ing the functional units responsible for observed malfunctions
in arbitrary system configurations.

4. XCON
XCON (eXpert CONfigure) is a system with the ability to select specific software
to generate a computer system as per user preference, written in 1978.

5. PXDES
This system could easily determine the type and the degree of lung cancer in
patients based on limited data.

6. CaDet
This is a clinical support system that identifies cancer in early stages.

7. PIP
PIP (Process Invention Procedure) is an hierarchical expert system for the synthesis
of chemical process flowsheets. It uses a combination of qualitative knowledge,
that is, heuristics and quantitative knowledge, that is, design and cost calculations,
arranged in a hierarchic manner. A hybrid, expert system control architecture was
developed for PIP that allows these two types of knowledge-bases to interact with
each other.

8. INTERNIST
INTERNIST-I was designed between 1972 and 1973 to provide computer assisted
diagnosis in general internal medicine by attempting to model the reasoning of
clini- cians.

10.6 Benefits and limitations

Some problems like scheduling manufacturing can not be adequately dealt with using
math- ematical algorithms. Such problems can be dealt with in an intelligent way using
expert systems.
Many expert systems require large, lengthy and expensive development efforts.
Expert systems lack the breadth of knowledge and the understanding of fundamental
principles of a human expert. In fast moving fields such as medicine or computer
science, keeping the knowledge base up to date is a critical problem. Expert systems can
only represent limited forms of IF-THEN type of knowledge. There are no adequate
representations for deep causal models or temporal trends. Expert systems cannot yet
replicate knowledge which is intuitive, based on analogy or on common sense.
Expert system do best in automating lower-end clerical functions. They can provide
electronic check lists for lower-level employees in service bureaucracies such as banking,
insurance, sales and welfare agencies. The applicability of expert system to managerial
problems generally involve drawing facts and interpretations from divergent sources,
eval- uating facts and comparing one interpretation of the facts with another and do not
involve
10.7. SAMPLE QUESTIONS 190

analysis or simple classification. They typically perform very limited tasks which can be
performed by professional, in a few minutes and hours.

10.7 Sample questions

(a) Short answer questions
1. What is an expert system?

2. Give a definition of “Expert System”.

3. List the languages and tools used in developing expert systems.

4. Give the names of some of the programming languages used to develop expert sys-
tems.

5. Give the names of some of the knowledge engineering languages used in

developing expert systems.

(b) Long answer questions

1. Outline how Expert Systems can be distinguished from more conventional
computer systems.

2. With the help of a neat diagram, describe the architecture of an expert system.

3. Briefly describe some of the well known expert systems.

4. What are the benefits and limitations of expert systems.

 Chapter 11

Fuzzy logic

11.1 What is fuzzy logic?

The phrase “fuzzy logic” has two meanings.

• As multivalued logic
In one sense the term “fuzzy logic” means the extension of ordinary logic, in which
the set of truth values is the two-element set {0, 1}, to the case where the set of
truth values is any finite or infinite subset of the closed interval [0, 1] (the set of all
numbers between 0 and 1 including both). In this sense, fuzzy logic is sometimes
referred to “many-valued logic” or “multivalued logic”. There are many ways in
which this extension can be carried out and consequently there are many different
multivalued logics.

• As fuzzy sets
In the second sense, the term “fuzzy logic” refers to the use of fuzzy sets in the repre-
sentation and manipulation of vague information for the purpose of making
decisions or taking actions. It is the theory of fuzzy sets. A fuzzy set is a
generalisation of the concept of a set.

In Section 11.2 of this chapter we discuss fuzzy logic in the first sense and in the re-
maining sections we discuss fuzzy logic in the second sense.

11.2 Fuzzy logic as multivalued logic

As a typical multivalued logic, we consider a three-valued logic.

11.2.1 Three-valued logic

Thinking of 0 as representing “false” and 1 as representing “true”, we add a third truth value
1
u representing “undecided” or “unknown”. It is common to use 2 instead of u. Thus, in
the three-valued logic the set of truth values is the three-element set {0, u, 1}.

191
11.2. FUZZY LOGIC AS MULTIVALUED LOGIC 192

11.2.2 Logical connectives in three valued logic

As in classical two-valued logic, in multivalued logic also propositions are combined us-
ing the logical connectives “AND”, “OR” and “NOT” (denoted respectively by ∧, ∨
and
¬) to form compound propositions. The truth values of such compound propositions are
determined by reference to truth tables. In the case of three-valued logic and in general in
any multivalued logic different researchers have defined the truth tables is different ways.
In these logics, the implication “⇒r” and the biconditional “⇔” are also defined by truth
tables.

Lukasiewicz’s definition of connectives

Here is the specification for a famous three-valued logic, that of Lukasiewicz.

∨ 0 u 1 ∧ 0 u 1 ¬
0 0 u 1 0 0 0 0 0 1
u u u 1 u 0 u u u u
1 1 1 1 1 0 u 1 1 u

⇒ 0 u 1 ⇔ 0 u 1
0 1 1 1 0 1 u 0
u u 1 1 u u 1 u
1 0 u 1 1 0 u 1

Table 11.1: Logical connectives in Lukasiewicz logic

Kleene logic
The Kleene logic has the same tables for AND, OR, and NOT as the Lukasiewicz logic
given above, but differs in its definition of implication.

⇔ 0 u 1
0 1 1 1
u u u 1
1 0 u 1

Table 11.2: Definition of implication in Kleene logic

The only difference between the two definitions is that in Lukasiewicz logic “undecided
implies undecided” is “true” where as in Kleene logic is “unknown”.
11.3. FUZZY SETS 193

11.2.3 Compound propositions in three-valued logic

Starting with a set of primitive or atomic propositions V we can build up compound
propo- sitions using this set and logical connectives. The set F of such compound
propositions is the same as in classical two-valued logic. The truth values of such
propositions can be computed using the truth tables of the logical connectives.

Example
Using the definitions of logical connectives in Lukasiewicz logic, construct the truth table
of the compound proposition (P ∧ Q) ⇒ P .

Solution

P Q P ∧ Q (P ∧ Q) ⇒ P
0 0 0 1
0 u 0 1
0 1 0 1
u 0 0 1
u u u 1
u 1 u 1
1 0 0 1
1 u u 1
1 1 1 1

11.2.4 Other multivalued logics

There are several multivalued logics like the following:

1. Bochvar’s three-valued logic

2. Belnap’s four-valued logic

3. Godel’s family Gk of multivalued logics where k is a positive integer greater than 2

4. A four-valued logic established by IEEE with the standard IEEE-1364

5. A nine-valued logic established by IEEE with the standard IEEE 1164

11.3 Fuzzy sets

11.3.1 Introduction
Fuzziness or vagueness can be found in many areas of daily life. This is more frequently
seen in which human judgment, evaluation, and decisions are important. The reason for
this is that our daily communication uses “natural languages” and a good part of our
thinking is
11.4. CRISP SETS 194

done in it. In these natural languages the meaning of words is very often vague. Examples
are words such as “birds” (how about penguins, bats, etc.?), “red roses”, but also terms
such as “tall men”, “beautiful women”, etc. The term “tall men” is fuzzy because the
meaning of “tall” is fuzzy and dependent on the context (height of observer, culture, etc.).

11.3.2 Example
Imagine the set of all young men residing in a small town. Since the number of people
residing in the town is not very large it is indeed possible to prepare a list of all persons in
the town. But can we prepare a list of all young men in the town? It is not possible
because the attribute of being “young” is not well defined, is vague and is imprecise. We
say that he attribute is a “fuzzy” attribute. However, all of us will agree that a person
aged 80 years cannot be a member of the “set of young men”! Similarly a three-year old
boy cannot also be a member of the set. Also, all of us would agree that a 25-year old
man residing in the city should be a member of the set.
To describe the set of young men in the town, we may arbitrarily assign a number to
each person which may indicate the “youngness” of the person. We assign the smallest
number to a person having the least “youngness” property and the highest value to those
who have the highest “youngness” property. The assignment of numbers is purely arbi-
trary. However it should reflect a human being’s understanding the notion of
“youngness”: a higher value must indicate a higher level of “youngness”. A set to whose
elements are assigned numbers like this is called a fuzzy set. We shall consider a formal
definition later. In practice, however, we usually assign values in the range 0 to 1,
including both.

Figure 11.1: The graph of the measure “youngness”, “middle-agedness” and “oldness”
vs. “age”

11.4 Crisp sets

In contrast to fuzzy sets, we shall refer to sets studied in classical set theory as “crisp
sets”. In the case of crisp sets, given an object, we can say unambiguously whether the
object is in the crisp set or not. In the case of crisp sets also, we can assign numbers to
objects: Assign 1 if the object is in the set and assign 0 if it is not in the set. In this sense,
crisp sets, that is sets considered in elementary set theory can also be regarded as special
fuzzy sets.
11.5. DEFINITION AND EXAMPLES 195

Figure 11.2: Fuzzy set vs. crisp set

11.5 Definition and examples

11.5.1 Definition
A fuzzy set is a pair (U, m) where U is a set (assumed to be non-empty) and m : U → [0,
1] a function. The set U is called universe of discourse. The function m is called the
member- ship function. For each x ∈ U , the value m(x) is called the grade of
membership of x in (U, m).
The fuzzy set (U, m) is often denoted by a single letter A. In such a case, the member-
ship function is denoted by µA.

Types of membership
Let x ∈ U . Then x is said to be

• not included in the fuzzy set (U, m) if m(x) = 0 (no member),

• fully included if m(x) = 1 (full member),

• partially included if 0 < m(x) < 1 (fuzzy member).

Support, kernel, α-cut

Let A = (U, m) be a fuzzy set. The support of A is the crisp set

Supp (A) = {x ∈ U : m(x) > 0}.

The kernel of A is the crisp set

Kern (A) = {x ∈: m(x) = 1}.

The α-cut (also called the α-level cut of A), denoted by A≥α or Aα, is defined as the
crisp set
A≥α = {x ∈ U : µA(x) ≤ α}.

Notation
Let A = (U, m) be a fuzzy set and let U be a finite set. Let the support of A be Supp (U ) =
{x1, x2, . . . , xn}. Then the fuzzy set A is sometimes written as a set of ordered pairs as
follows:
A = {(x1, m(x1)), (x2, m(x2)), · · · , (xn, m(xn))}
11.5. DEFINITION AND EXAMPLES 196

The following notation

A = {m(x1)/x1, m(x2)/x2, . . . , m(xn)/xn}

as well as the notation

A = m(x1)/x1 + m(x2)/x2 + . . . + m(xn)/xn

are also used to denote the fuzzy set A. Still others write

m(x1) m(x2) m(xn)

A= 1 , 2 , . . . , n .
x x x

Note that in these representations, the symbols “/” and “+” are just symbols and they do
not represent the corresponding arithmetical operations.

Crisp set as a fuzzy set

Let A be a crisp subset of the universe of discourse U . A can be treated as a fuzzy set
A = (U, m) where the membership function m is defined as
(
1 if x ∈ A
m(x) = 0 if x /∈ A

The membership function of a crisp set is called the characteristic function of the set.

11.5.2 Examples
Example 1
Consider the set
U = {5, 10, 15, 20, 30, 35, 40, 45, 60, 70}
and the function m : U → [0, 1] defined by the following table:

x 5 10 15 20 30 35 40 45 60 70
m(x) 1 1 1 1 0.6 0.5 0.4 0.2 0 0

A = (U, m) is a fuzzy set. The support of A is

Supp (A) = {5, 10, 15, 20, 30, 35, 40, 45}

and the kernel of A is

Kern (A) = {5, 10, 15, 20}.
The fuzzy set may be represented as

A = {1/5, 1/10, 1/15, 1/20, 0.6/30, 0.5/35, 0.4/40, 0.2/45}

or as

A = 1/5 + 1/10 + 1/15 + 1/20 + 0.6/30 + 0.5/35 + 0.4/40 + 0.2/45.

11.5. DEFINITION AND EXAMPLES 197

Example 2
A realtor wants to classify the houses he offers to his clients. One indicator of comfort of
these houses is the number of bedrooms in it. Let

U = {1, 2, 3, 4, ..., 10}

be the set of available types of houses described by x = number of bedrooms in a house.

Then the fuzzy set “comfortable type of house for a 4-person family” may be described
as

A = {0.2/1, 0.5/2, 0.8/3, 1.0/3, 0.7/5, 0.3/6)}.

Example 3
The fuzzy set defined as “integers close to 10” may be specified by

A = 0.1/7 + 0.5/8 + 0.8/9 + 1/10 + 0.8/11 + 0.5/12 + 0.1/13

Example 4
Let the universe U = R be the set of all real numbers. Then the fuzzy set A described
as “the set of real numbers close to 10” can be specified by the membership function
1
µA(x) = .
1 + (x − 10)2

11.5.3 Some special membership functions

Let the universe of discourse U = R be the set of all real numbers. The following
mem- bership functions define certain special fuzzy sets A = (U, m) which are useful
in some applications.

1. Triangular membership function


0 x≤a
 x−a
 b − a<x≤
µA(x) = a x −
b
b<x≤c
c−b

b
 0 c<x

Figure 11.3: Graph of the triangular membership function

11.5. DEFINITION AND EXAMPLES 198

2. Trapezoidal membership function

0 x≤a
x−a
 b − a<x≤b
µA(x) =
a1 b< x ≤ c
x−c
 d−b c<x≤d

0 d<x

Figure 11.4: Graph of the trapezoidal membership function

3. Gaussian membership function

1 x − m!
k

s
µA(x) = exp −
2

Figure 11.5: Graph of the Gaussian membership function

4. Sigmoid membership function

1
µA(x) =
1 + e−a(x−m)
11.6. SET-THEORETIC OPERATIONS FOR FUZZY SETS 199

Figure 11.6: Graph of the sigmoid membership function

11.6 Set-theoretic operations for fuzzy sets

11.6.1 Definitions
Let A and B be two fuzzy sets in a universe U .

1. Equality
A and B are said to be equal, denoted by A = B, if µA(x) = µB(x) for all x in U .

2. Union
The union of A and B, denoted by A ∪ B, is the fuzzy set C whose membership
function is defined by

µC(x) = max{µA(x), µB(x)}, x ∈ U.

3. Intersection
The intersection of A and B, denoted by A∩B, is the fuzzy set D whose membership
function is defined by

µD(x) = min{µA(x), µB(x)}, x ∈ U.

4. Complement
The complement of A, denoted by A, is the fuzzy set E whose membership function
is defined by
µE(x) = 1 − µA(x), x ∈ U.

Remarks
1. The definitions of the membership functions of the union, intersection and comple-
ment of fuzzy sets may be given in the following forms also:

µA∪B(x) = max{µA(x), µB(x)}

µA∩B(x) = min{µA(x), µB(x)}
µA(x) = 1 − µA(x)
11.6. SET-THEORETIC OPERATIONS FOR FUZZY SETS 200

2. The support of a fuzzy set is a crisp set. We have

Supp (A ∪ B) = Supp (A) ∪ Supp (B)

Supp (A ∩ B) = Supp (A) ∩ Supp (B)
Supp (A) = Supp (A)

3. If A is a fuzzy set in a universe U then in general we have

A ∪ A /= U, A ∩ A /= ∅.

However, equalities hold if A is a crisp set considered as a fuzzy set.

11.6.2 Examples
Example 1
Given the fuzzy sets

A = {0.3/2, 0.4/3, 0.1/4, 0.8/5, 1.0/6}

B = {0.7/4, 0.5/5, 1.0/6, 0.02/7, 0.75/8}

find A ∪ B and A ∩ B.

Solution
We have:

Supp (A) = {2, 3, 4, 5, 6}

Supp (B) = {4, 5, 6, 7, 8}

Let us compute A ∪ B.

Supp (A ∪ B) = Supp (A) ∪ Supp (B)

= {2, 3, 4, 5, 6, 7, 8}
We have
µA∪B(x) = max{µA(x), µB(x)}.
11.6. SET-THEORETIC OPERATIONS FOR FUZZY SETS 201

Hence

µA∪B(2) = max{µA(2), µB(2)}

= max{0.3, 0} (2 /∈ Supp (B))
= 0.3
µA∪B(3) = max{µA(3), µB(3)}
= max{0.4, 0} (3 /∈ Supp (B))
= 0.4
µA∪B(4) = max{µA(4), µB(4)}
= max{0.1, 0.7}
= 0.7
···= ···
µA∪B(8) = max{µA(8), µB(8)}
= max{0, 0.75} (8 /∈ Supp (A))
= 0.75

Therefore
A ∪ B = {0.3/2, 0.4/3, 0.7/4, 0.8/5, 1.0/6, 0.02/7, 0.75/8}.
Now, let us calculate A ∩ B.
Supp (A ∩ B) = Supp (A) ∩ Supp (B)
= {4, 5, 6}
We
µA∩B(x) = min{µA(x), µB(x)}.
have
µA∩B(4) = min{µA(4), µB(4)}
Hence = min{0.1, 0.7}
= 0.1
µA∩B(5) = min{µA(5), µB(5)}
= max{0.8, 0.5}
= 0.5
µA∩B(6) = min{µA(6), µB(6)}
= max{1, 0, 1, 0}
= 1.0

A ∩ B = {0.1/4, 0.5/5, 1.0/6}

Therefore

Example 2
Let the universe be U = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10} and let
A = {0.2/1, 0.4/2, 0.6/3, 0.8/4, 1.0/5}.
Compute A¯ .
11.7. SOME MORE DEFINITIONS 202

Solution
By definition
µA(x) = 1 − µA(x), for all x ∈ U.
Therefore, we
have:

A = {0.8/1, 0.6/2, 0.4/3, 0.2/4, 1.0/6, 1.0/7, 1.0/8, 1.0/9, 1.0/10}.

11.7 Some more definitions

1. Fuzzy empty set
A fuzzy set A is said to be empty if µA(x) = 0 for all x ∈ U , that is, if Supp (A) = ∅.

2. Fuzzy subset
A fuzzy set A is said to be a subset of a fuzzy set B, denoted by A ⊆ B, if µA(x) ≤
µB(x) for all x ∈ U .

3. Disjoint fuzzy sets

Two fuzzy sets A and B are said to be disjoint if µA(x) = 0 or µB(x) = 0 for all
x ∈ U , that is, if Supp (A ∩ B) = ∅.

Examples
1. Let
A = {0.12/2, 0.23/4}, B = {0.15/2, 0.37/4, 0.15/6},
Then A ⊆ B.

2. Let U = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10} and let

A = {0.1/1, 0.2/2, 0.3/3}, B = {0.2/6, 0.4/7, 0.6/8}.

Then

Supp (A) = {1, 2, 3}

Supp (B) = {6, 7, 8}
Supp (A ∩ B) = Supp (A) ∩ Supp (B)
=∅

Hence A and B are disjoint fuzzy sets.

11.8 Properties of set operations on fuzzy sets

Let A, B, C be fuzzy sets in some universe X. Then we have:

1. Commutativity
A∪B=B∪B
A ∩ B = B ∩
A.
 11.9. FUZZY VARIABLES 203

2. Associativity
A ∪ (B ∪ C) = (A ∪ B) ∪ C)
A ∩ (B ∩ C) = (A ∩ B) ∩ C
3. Distributivity
A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C),
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C)
4. Identity
A ∪ ∅ = A, A ∪ X = A
A ∩ ∅ = ∅, A ∩ X = A
5. Involution
(A) = A.
6. De Morgan’s laws

A∪B=A∩B
A ∩ B = A ∪ B.

Example
Given the universe U = {a, b, c, d, e, f, g, h} and the fuzzy sets
A = {0.1/a, 0.2/b, 0.3/c, 1.0/d, 0.1/e, 0.5/h}
B = {0.75/b, 0.2/c, 0.35/d, 0.4/e, 0.5/f }
C = {1.0/e, 0.9/f, 0.8/g, 0.7/h}
verify the distributivity properties.

Solution

B ∪ C = {0.75/b, 0.2/c, 0.35/d, 1.0/e, 0.9/f, 0.8/g, 0.7/h}

A ∩ (B ∪ C) = {0.2/b, 0.2/c, 0.35/d, 0.1/e, 0.5/h}
A ∩ B = {0.2/b, 0.2/c, 0.35/d, 0.1/e}
A ∩ C = {0.1/e, 0.5/h}
(A ∩ B) ∪ (A ∩ C) = {0.2/b, 0.2/c, 0.35/d, 0.1/e, 0.5/h}
Hence we have
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C).
In a similar way, the other distributivity property can be verified.

11.9 Fuzzy variables

An ordinary variable, or a crisp variable, is a variable whose possible values are elements of
a given set. For example, a real variable is a variable whose possible values are elements
of the set of real numbers.
11.10. EXAMPLE FOR APPLICATION OF FUZZY SETS 204

11.9.1 Definition
A fuzzy variable x is a variable whose possible values are fuzzy sets in some universe of
discourse.

11.9.2 Examples
Example 1
We define a fuzzy variable x as follows. Let the universe of discourse be

U = {1, 2, 3, 4, 5}.

The possible values of x are fuzzy sets in U . A possible value of x is

A = {0.1/1, 0.7/2, 0.8/3, 0.2/4, 0/5},

or
B = {1.0/1, 0.8/2, 0.6/3, 0.4/4, 0.2/5}.

Example 2
Consider a variable which denotes the age of a person in years. We may denote this
variable by age. As a crisp variable, age can assume nonnegative integers as a values.
We may take age as a fuzzy variable. To do this, we have to define a universe of
discourse U which we may take as the set of all nonnegative integers. As a fuzzy variable
the possible values can that can be assigned to age are fuzzy sets in U . These fuzzy sets
may denoted by linguistic terms indicative of the age of a person, such as infant, young,
middle-aged, old, etc. Thus the fuzzy variable age can be thought of as a variable whose
possible values are such linguistic terms. However, it is to be emphasised that the value is
not the linguistic term but the fuzzy set represented by the term.

11.10 Example for application of fuzzy sets

We explain the working of a conventional fuzzy room cooler shown schematically in
Figure 11.7.

The device
The cooler is implemented using a fan encased in a box with wool or hay that is contin-
uously moistened with a trickle of water. A motorised pump controls the rate of flow of
water. Two sensors measures the fan motor speed and the temperature.

Objective
The basic aim is to achieve a smooth control and to save water. We have to control the
rate of flow of water based on fan motor speed (measured in rpm’s) and temperature
(measured in ◦C). To make the problem precise, let it be required to find the rate of flow
of water if the temperature is 42◦C and fan motor speed is 31 rpm.
11.10. EXAMPLE FOR APPLICATION OF FUZZY SETS 205

Figure 11.7: A sectional view of a fuzzy room cooler

Variables
The input variables of the system are “temperature” and “fan-motor-speed”. We take
them as fuzzy variables. We assign the fuzzy sets denoted by “Cold, Cool, Moderate,
Warm, Hot” as the values of the variable temperature and the fuzzy sets denoted by
”Slack, Low, medium, Brisk, Fast” as the values of fan-motor-speed. The output variable
is “water-flow- rate” and it may take the fuzzy sets “Strong-Negative (SN), Negative (N),
Low-Negative (LN), Medium (M), Low-Positive (LP), Positive (P), High-Positive” as its
values.

Membership functions
Based on observations and experimentation we have to construct the membership
functions to the various fuzzy sets introduced above. We assume that the membership
functions are as shown in Figure 11.8.

Rules for action

The values for the output variable depends on the values of the input variable. In logic
controllers, the values of the output variable are computed based on certain rules. These
rules are formulated by the designer based on real life experiences. For example, for the
room cooler the rules could be the following and probably many more such rules.

R.1 If temperature is Hot and fan-motor-speed is Slack then water-flow-rate is HP.

R.2 If temperature is Hot and fan-motor-speed is Low then water-flow-rate is HP.

R.3 If temperature is Hot and fan-motor-speed is Medium then water-flow-rate is P.

R.4 If temperature is Hot and fan-motor-speed is Brisk then water-flow-rate is HP.

R.5 If temperature is Warm and fan-motor-speed is Medium then water-flow-rate is

LP.

R.6 If temperature is Warm and fan-motor-speed is Brisk then water-flow-rate is P.

 11.10. EXAMPLE FOR APPLICATION OF FUZZY SETS 206

Figure 11.8: Graphs of membership functions

R.7 If temperature is Cool and fan-motor-speed is Low then water-flow-rate is N.

R.8 If temperature is Moderate and fan-motor-speed is Low then water-flow-rate is

M.

Fuzzification
We have to express the observed values of 42◦C and 31 rpm as values of the fuzzy variables
temperature and fan-motor-speed.
From the graphs of the membership functions of the values of the fuzzy variable tem-
perature, we see that a value of 42◦C corresponds to the fuzzy set Warm with a
membership grade of 0.142 and also to the fuzzy set Hot with a membership grade of 0.2.
Similarly from From the graphs of the membership functions of the values of the
fuzzy variable fan-motor-speed, we see that a value of 31 rpm corresponds to the fuzzy
set Medium with a membership grade of 0.25 and also to the fuzzy set Brisk with a mem-
bership grade of 0.286.
This process of expressing crisp values as values of fuzzy variables is known as
fuzzifi- cation.
11.10. EXAMPLE FOR APPLICATION OF FUZZY SETS 207

Application of rules
Since we have two possible fuzzy values for temperature and two possible fuzzy values
for fan-motor-speed we have the combinations of values as shown Table 11.3. The table
also shows the applicable rules, the value of the output variable obtained by applying the
rule and further the membership grade of the output value. Note that the membership
grade of the output variable water-flow-rate is taken as the minimum of the membership
grades of the two input variables.

temperature fan-motor-speed Applicable rule water-flow-rate

Hot/0.2 Medium/0.25 R3 P/0.2
Hot/0.2 Brisk/0.286 R4 HP/0.2
Warm/0.142 Medium/0.25 R5 LP/0.142
Warm/0.142 Brisk/0.286 R6 LP/0.142

Table 11.3: Output values and their membership grades

Figure 11.9: Defuzzification

Figure 11.10: Defuzzification

11.11. SAMPLE QUESTIONS 208

Defuzzification
From the above set of possible fuzzy values for the output variable, we have to derive a
crisp value for the output variable. This process is known as defuzzification. One method
for doing this known as the “centre of gravity method”.
To obtain the crisp value, we consider value P with membership grade 0.2 and
consider the graph of the membership function of the fuzzy variable P . In this graph, we
consider the shaded region shown in Figure 11.9. In a similar way, by considering the the
other possible output values and their member ship grades, we construct shaded regions
in the graphs of the respective membership functions. In this way, we get four shaded
regions. We next construct in one graph a composite region consisting of all the four
shaded regions as in Figure 11.10.
The x-coordinate of the centre of gravity, or the centroid, of this composite region is
the defuzzified crisp value of the output variable water-flow-rate.

11.11 Sample questions

(a) Short answer questions
1. What is fuzzy logic?

2. Give an example of a three-valued logic.

3. Define fuzzy set and give an example.

4. What is a membership function of a fuzzy set?

5. Define the support and kernel of a fuzzy set and illustrate with examples.

6. Explain why a crisp set can also be considered as a fuzzy set.

7. Define the union and intersection of fuzzy sets. Illustrate with examples.

8. Define the complement of a fuzzy set. Illustrate with an example.

9. Define disjoint fuzzy sets and give an example.

10. Using an example of a fuzzy set A verify that (A) = A.

11. Define a fuzzy variable and give an example.

12. What is the purpose of defuzzycation? Name at least one method used for defuzzy-
cation.

(b) Long answer questions

1. Given the following fuzzy sets

A = {0.1/1, 0.3/2, 0.45/3}, B = {0.15/1, 0.34/2}

find A ∪ B and A ∩ B.
11.11. SAMPLE QUESTIONS 209

2. Let
A = 0.5/1 + 0.9/2 + 1/5, B = 0.7/2 + 0.9/3 + 0.1/4.
Compute A ∪ B and A ∩ B.
3. Suppose U = {1, 2, 3, 4, 5} and A = 0.5/1 + 0.3/3 + 1/5. Compute:

(a) A
(b) A ∩ A
(c) A ∪ A
4. Let U = {1, 2, 3, 4, 5, 6, 7, 8} and let
A = {0.4/1, 0.7/3, 0.5/5, 0.8/7}, B = {0.5/1, 0.2/2, 1.0/6, 0.1/8}.
Verify De Morgan’s laws.
5. Consider two fuzzy subsets of the set X = {a, b, c, d, e} defined by
A = {1/a, 0.3/b, 0.2/c0.8/d}, B = {0.6/a, 0.9/b, 0.1/c, 0.3/d, 0.2/e}.
Compute the following: Supp (A), Supp(B), A ∪ B, A ∩ B, A and B.
6. Let
A = 0.2/1 + 0.5/2 + 0.7/3 + 1.0/4 + 0.8/5 + 0.4/6 + 0.2/7.
Compute the α-level set of A for α = 0.5.
7. Given the fuzzy sets

0.12 02.5 03.8 41.0 50.7 60.3 0.32 04.4 05.6 06 .8 81.0 91.0101.0
C= , , , , , ,D = , , , , , ,

compute C ∩ D and C ∪ D.
8. Let
A = 0.1/2 + 0.5/3 + 0.4/4 + 0.5/5, B = 0.1/2 + 0.6/3 + 1/4 + 0.6/5 + 0.1/6.
Is it true that B /⊂ A?
9. Consider these (very subjective) membership functions for the height of a person:

Figure 11.11: Graphical representations of membership functions

11.11. SAMPLE QUESTIONS 210

Compute the graphical representation of the membership function of the following

fuzzy sets:

(a) Tall ∩ Small

(b) (Tall ∪ Medium) − Tall

10. Construct a fuzzy set (universal set and membership function) for the concept
“early in the morning”. (Hint: Choose the universal set as {1, 2, . . . , 23, 24}
(times of a day) and define a membership function to indicate “early in the
morning”.)
Index

A∗ best first search, CLIPS, 186

69 8-puzzle, 12
CNF, 132
ABSTRIPS, 170 commitment strategy, 163
acquisitional efficiency, 102 conceptual category, 118
action, 10 primitive -, 118
activation, 110 conceptual dependency, 116
actuator, 6 rules of -, 121
addition, 133 states, 120
admissible heuristic functions, conceptual tenses, 120
57 agent, 6 conclusion, 131
rational - , 7 conflict resolution, 29
agent function, conjunction, 128
7 conjunctive normal form,
agent program, 7 132 connected graph, 39
almost correct solution, constraint satisfaction, 15
161 alpha-beta pruning, 93 contradiction, 129, 135
analogy, 177 control strategy, 29
derivational - , 177 crisp set, 194
transformational - , crisp variable, 203
177 crossover, 181
annealing, 78 cryptarithmetic puzzle, 15
argument, 131 cybersecurity, 4
valid - , 131 cycle, 39
arity, 139
atomic sentence, 139 DART, 3, 188
autonomous planning, 3 Dartmouth College, 1
autonomous scheduling, 3 dead-end, 162
DEEP BLUE, 3
basic element, 138 defuzzification, 208
best first search, Dendral, 182, 187
62 biconditional, denial of service, 4
129 depth-first search, 46
blocks world, 19 deterministic game, 84
branch, 41 directed graph, 39
breadth-first search, 43 discovery, 176
disjunction, 128
CaDet, 189 domain of discourse, 138
CD, 116 driverless car, 3
center of gravity method, 208
chromosome, 180 edge, 39
clausal form, 142
clause, 132
clause form, 132 211
INDEX 212

Edward Feigenbaum, 182 Graph-search algorithm,

effector, 6 51 greedy best first search,
evaluation function, 62 62
existential instantiation, 146
expanding a state, 42 Hamming distance, 60
expert system, 182 heuristic, 56
architecture of -, 182 - function, 56
expert shell, 185 hidden layer, 179
inference engine, 183 hierarchical planning, 170
knowledge base, 183 hill climbing, 72
user interface, 182
implication, 129
False, 127 inference, 131
first order predicate logic, 136 rules of - , 132
FOO, 173 inference rule
FOPL, 136 - in FOPL, 145
frame, 112 inferential adequacy, 102
frame language, 115 inferential efficiency, 102
frontier, 51 initial state, 10
full member, 195 input layer, 179
fuzzification, 206 INTERNIST, 189
fuzzy logic, 191 interpreter, 30
fuzzy member, 195 intersection search, 110
fuzzy set, 195 iterative deepening, 50
operations on - , 199
fuzzy variable, 203 John McCarthy, 1
Joshua Lederberg, 182
game, 84
- with perfect information, kernel, 195
84 deterministic - , 84 KL-ONE, 116
turn-taking - , Kleene logic, 192
84 two-player - , knowledge, 100
84 zero-sum - , - representation, 101
84 - base, 100
game with perfect information, declarative - ,
84 Garry Kasparov, 3 101 heuristic - ,
General Problem Solver, 101
2 generating new states, meta-knowledge, 100
42 genetic algorithm, 180 procedural - , 101
genome, 180 structural - , 101
genotype, 180 Knowledge-based systems, 2
Geometry Theorem Prover, 2 KRL, 115
global maximum, 73
goal stack, 163 learning, 172
goal stack planning, 163 - by chunking, 175
goal test, 11 - by induction, 175
graph, 10, 39 - by parameter adjustment, 174
connected - , 39 - by taking advice, 173
directed - , 39 - with macro-operators,
174 explanation based - ,
176 reinforcement -, 172
INDEX 213

rote - , 173 natural language processing, 4

semi-supervised - , 173 negation, 128
supervised - , 172 neural network, 178, 179
unsupervised - , 172 neuron, 178
level cut, 195 NLP, 4
literal, 132 node, 39, 41
local optimum, 73, 77 nonbinary relation, 111
local search, 73
logic open list, 51
Belnap’s - , 193 operator, 74
Bochvar’s - , 193 optimal solution, 11
fuzzy - , 191 output layer, 179
Godel’s - , 193
Kleene - , 192 path, 10
Lukasiewicz - , path cost, 11
192 multivalued -, perceptron, 178
191 three-valued - phenotype, 180
, 191 PIP, 189
logical equivalence, 130 planning, 160
logical identity, 130 components of - , 161
logistic planning, 3 goal stack - , 163
loop, 39 hierarchical -, 170
Lukasiewicz, 192 plateau, 78
predicate, 137, 138
machine translation, 4 predicate logic, 136
macro-operators, 174 premise, 131
MAGEN, 3 problem
Manhattan distance, 60 8-puzzle - , 12
map problem, 23 blocks world - , 19
membership function, characteristics of - , 25
195 Gaussian - , 198 cryptarithmetic - , 15
sigmoid - , 198 decomposable - , 25, 160
trapezoidal - , 198 definition of -, 7
triangular - , 197 ignorable - , 25
membership grade, 195 map -, 23
meta-knowledge, 100 missionaries and cannibals - ,
MEXAR2, 3 13 monkey and banana - , 21
minimax, 89 nearly decomposable - , 161
missionaries and cannibals, predictable - , 25
13 modes-ponens, 132 sliding tile -, 7
modus ponens vacuum clear world, 11
generalised - , 146 water jug - , 23
modus tollens, 133 problem-solving agent, 7
monkey and banana problem, 21 production rule, 29
multigraph, 39 production system, 29
mutation, 181 commutative - , 36
MYCIN, 184, 188 monotonic - , 35
nonmonotonic - , 35
NASA, 3
INDEX 214

partially commutative - , intersection -, 110

35 PROLOG, 186 simulated annealing, 78
proof of validity, 131 steepest-ascent hill climbing - , 76
proposition, 127 uninformed -, 42
propositional expression, 127 search algorithm, 39,
propositional formula, 127 41 second order logic,
propositional semantic network, 137 select-execute
109 propositionalisation, 148 loop, 30
PXDES, 189 self-driving car, 3
quantifier, 139 semantic network, 102
existential - , 140 assertional -, 109
universal - , 139 definitional -, 108
executable -, 110
rational agent, 1, 7 hybrid -, 110
recombination, 181 implicational -, 109
Remote Agent program, 3 learning -, 110
representation, 101 propositional - , 109
representational adequacy, 102 shoulder, 78
resolution, 133 simple hill climbing, 72
resolution-refutation, 135 Simplification, 133
- in FOPL, 149 simulated annealing, 78
resolvent, 135 Skolem function, 143
ridge, 77 Skolem term, 143
robo-car, 3 Skolemization, 142
Robotic vehicle, 3 sliding tile problem,
robotics, 4 7 slot-filler, 113
Roger Schank, 116 solution, 11
room cooler, 204 spam fighting, 3
Roomba, 4 speech recognition,
root, 40 3 speech to text, 3
root node, 41 state space, 10
rule applier, 30 steepest-ascent hill climbing,
76 STRIPS, 164
SAINT, 2 STUDENT, 2
search, 41 successor, 10
A∗ best first - , 69 support, 195
- strategy, 42 syllogism
- tree, 41 disjunctive - , 133
adversarial -, hypothetical - , 133
84 best first - , symbols in FOPL, 139
62 blind -, 42
breadth-first -, 43 tautology, 129
depth-first -, 46 term, 139
depth-first iterative deepening -, 50 terminal states, 88
greedy best first - , 62 transition model, 10, 88
heuristic -, 42 tree, 39
hill climbing - , 72 trivial heuristic, 57
informed -, 42 True, 127
truth table, 127
INDEX 215

truth value, 127

turn-taking game, 84
two-player game, 84
two-way implication, 129

unary predicate, 139

unification, 147
universal instantiation, 145
universe, 138
utility function, 88

vacuum cleaner world, 11

vertex, 39

Walter Pitts, 1
Warren McCulloch, 1
water jug problem, 23
well-formed formula,
142
wff, 142
working memory, 29
world, 138

XCON, 189

zero-sum game, 84
This page is intentionally left blank.
Lecture Notes in Artificial Intelligence
Dr V N Krishnachandran

Department of Computer Applications

(MCA) Vidya academy of Science and
Technology
Thrissur – 680 501