CSIR GOLDEN JUBILEE

SERIES

ARTIFICIAL
INTELLIGENCE

K. D. PAVATE
 ARTIFICIAL
INTELLIGENCE

K.D. PAY ATE

Publications & Information Directorate

Dr. K.S. Krishnan Marg
New Delhi 110 012
India
Artificial Intelligence

K.D. Pavate

©Publications & Information Directorate

First Edition : May 1992
Second Edition : March 1993
Third Edition : January 1996
ISBN : 81-7236-035-5

C S I R G o l d e n Jubilee Series
Publication No. 7
Series Editor Dr. Bal Phondke
Volume Editor Parvinder S. Chawla
Cover Design Pradip Banerjee
Illustrations Pradip Banerjee, Neeru Sharma, Neeru Vijan,
Sushila Vohra, K.K. Bhatnagar and Anil Kumar
Production V. Ramachandran, K.B. Nagpal, Vinod Sharma,
Sudhir Chandra Mamgain and Radhe Shiam

Designed, Printed and Published by

Publications & Information Directorate (CSIR)
Dr. K.S. Krishnan Marg, New Delhi 110 012
India
Foreword

The Council of Scientific & Industrial Research (CSIR), estab-

lished in 1942, is committed to the advancement of scientific
knowledge, and economic and industrial development of the
country. Over the years CSIR has created a base for scientific
capability and excellence spanning a wide spectrum of areas
enabling it to carry out research and development as well as
provide national standards, testing and certification facilities.
It has also been training researchers, popularizing science
and helping in the inculcation of scientific temper in the
country.
The CSIR today is a well knit and action oriented network of
41 laboratories spread throughout the country with activities
ranging from molecular biology to mining, medicinal plants
to mechanical engineering, mathematical modelling to
metrology, chemicals to coal and so on.
While discharging its mandate, CSIR has not lost sight of the
necessity to remain at the cutting edge of science in order to
be in a position to acquire and generate expertise in frontier
areas of technology. CSIR's contributions to high-tech and
emerging areas of science and technology are recognised
among others for precocious flowering of tissue cultured
bamboo, DNA finger-printing, development of non-noble
metal zeolite catalysts, mining of polymetallic nodules from
the Indian Ocean bed, building an all-composite light re-
search aircraft, high temperature superconductivity, to men-
tion only a few.
Being acutely aware that the pace of scientific and technologi-
cal development cannot be maintained without a steady
influx of bright young scientists, CSIR has undertaken a
vigorous programme of human resource development which
• includes, inter alia, collaborative efforts with the University
Grants Commission aimed at nurturing the budding careers
of fresh science and technology graduates.
However, all these would not yield the desired results in the
absence of an atmosphere appreciative of advances in science
and technology. If the people at large remain in awe of
science and consider it as something which is far removed
from their realms, scientific culture cannot take root.
CSIR has been alive to this problem and has been active in
taking science to the people, particularly through the print
medium. It has an active programme aimed at populariza-
tion of science, its concepts, achievements and utility, by
bringing it to the doorsteps of the masses through both print
and electronic media. This is expected to serve a dual pur-
pose. First, it would create awareness and interest among the
intelligent layman and, secondly, it would help youngsters
at the point of choosing an academic career in getting a
broad-based knowledge about science in general and its
frontier areas in particular. Such familiarity would not only
kindle in them deep and abiding interest in matters scientific
but would also be instrumental in helping them to choose the
scientific or technological education that is best suited to
them according to their own interests and aptitudes. There
would be no groping in the dark for them. However, this is
one field where enough is never enough.
This was the driving consideration when it was decided to
bring out in this 50th anniversary year of CSIR a series of
p r o f u s e l y illustrated and specially written p o p u l a r
monographs on a judicious mix of scientific and technologi-
cal subjects varying from the outer space to the inner space.
Some of the important subjects covered are astronomy,
meteorology, oceanography, new materials, immunology
and biotechnology.
It is hoped that this series of monographs would be able to
whet the varied appetites of a wide cross-section of the target
readership and spur them on to gathering further knowledge
on the subjects of their choice and liking. An exciting sojourn
through the wonderland of science, we hope, awaits the
reader. We can only wish him Bon voyage and say, happy
hunting.
Preface

Computers are here to stay. They have had a big impact on

our everyday lives. With suitable software and programs
they are being used to solve a wide spectrum of problems.
For instance, they are being used in solving complex mathe-
matical problems, preparing database, forecasting economic
trends, assisting management in taking decisions and in
helping engineers to operate large factories. So, it is not
surprising when a casual onlooker says: "This computer is a
smart machine. It is very intelligent !" Now, adjectives like
'smart' and 'intelligent' are invariably associated with
human beings. A person who does all the work assigned to
him in the shortest possible time is considered smart. But,
intelligence is a quality which is much easier to recognize
than to define. One runs into serious problems while attempt-
ing to measure or assign values to it.
Over the past four decades scientists have attempted to
apply available technologies to design user-friendly com-
puter systems which exhibit a semblance to human intel-
ligence. So far they have been only partially successful.
Expert systems are available which assist their users in find-
ing solutions within certain well-defined areas of knowledge.
Today, machines can recognize spoken words and simple
sentences. There are those which analyze pictures and recog-
nize specific patterns. The new subject of Artificial Intel-
ligence (AI) is progressing rapidly. It fascinates people
because it touches on subjects which are of intimate concern
to human beings - choice, learning, cognition, etc. This book
has tried to capture the flavour of Al-linked concepts and
developments. Although written for the benefit of the
younger generation, it is hoped that this book will also create
an interest among our budding scientists and engineers.
Acknowledgments

Dr G.P.Phondke set the ball rolling by first suggesting that I

write something for the CSIR Golden Jubilee series. Even as
I was murmuring, he quickly followed it up by suggesting
the theme. It was only when he confronted me with a dead-
line which had to be strictly adhered to that I woke up to the
realization that I had made an irrevocable commitment. I
must confess that I would have missed the excitement of
writing had it not been for Dr Phondke.
The project brought me in close contact with the staff of
PID. It has been simply great working with Mrs. Parvinder
Chawla, the volume editor, and Shri S.K.Nag. Parvinder has
spared no pains in processing the manuscript and in making
valuable suggestions. The Art Section of PID, especially Shri
Pradip Banerjee, also contributed substantially by providing
illustrations to break the monotony of the text. My thanks to
all of them.
I also owe gratitude to my colleagues and friends at the
CEERI Centre in Delhi. They too have followed the progress
of this book with considerable interest. They have, in fact,
provided me with facilities to complete this writing assign-
ment. I am grateful to Dr K.V.Ramakrishnan and to Dr.
S.S.Agarwal for many useful discussions on the subject. Shri
Sadashiv Sharma helped me with the sonographs and Shri
D.J.Ray provided me with photographs. Shri S.N.Gupta as-
sisted me with both the computer and the WordPerfect pro-
gram. I must thank them for their cooperation.
Finally, I must thank my family for their whole-hearted
support, particularly, my daughter, Dr Rita, who discussed
with me on various aspects of the physiology of central
nervous system and the psychology of learning. I am also
grateful to her for reading the manuscript through the
various stages of its evolution.
K. D. Pavate
Dedicated to my wife

Suverna

for all her patience

Contents

The Genesis ... 1

Learning and Memory ... 16

Scope and Extent ... 34

A Problem Solver ... 41

Programming Intelligence ... 61

User-Friendly Systems ... 79

An Overview ... 92

Glossary ... 97
 ince the dawn of civilization

S man has been interested in

devices which could assist
him in simple arithmetical exer-
cises. Amongst the earliest known
contraptions are the abacus and
the soroban. These use beads
which are mounted on stiff wires
and can be moved up and down
on them. They are used to count
numbers. Surprisingly, they have
survived through the ages and
are being used even today in
shops in USSR, China, Singapore
and Hongkong.

The The next known advancements

were the t a b l e top c o u n t i n g
Genesis machine invented by Blaise
PASCAL (1623-1662) in 1642 and
a calculating machine b y the
eminent German mathematician-
cum-philosopher Gottfried LEIB-
N I T Z ( 1 6 4 6 - 1 7 1 6 ) in 1694.
Whereas Pascal's version could
only add and subtract, the latter
machine was capable of even
multiplying and dividing num-
bers by the process of repeated
additions or subtractions. Both
these machines operated on the
basis of toothed gear wheels
meshing into one another.

In 1801 a Frenchman named

Joseph JACQUARD (1752-1834),
i n t r o d u c e d the c o n c e p t of a
"memory" in the conventional
2 ARTIFICIAL INTELLIGENCE

Leibnitz's calculator
Abacus Analytical machine

Evolution of computers in
 THE GENESIS 3

Supercomputer

Desktop personal computer

>arallel t o h u m a n e v o l u t i o n
4 ARTIFICIAL INTELLIGENCE

weaving loom. Earlier, the workers had to memorize the

various designs or weaving patterns. The same information
was now permanently stored on a series of stiff, large-sized
cards with punched holes. These holes enabled the shuttles
with different coloured threads to operate in a sequence and
generate specific weaving patterns. By introducing a dif-
ferent set of cards, new patterns could be woven on the cloth.
Jacquard's cards were adopted almost overnight b y the entire
weaving industry in France.
A hundred years ago, Herman HOLLERITH (1869-1926)
in U.S. A extended Jacquard's principle of storing information
on punched cards to another important application. In 1891
he redesigned the cards so that holes now corresponded to
statistical data concerning individuals. He also designed and
constructed machines to punch holes and to read the data off
these cards. These machines could also perform the required
additions so as to generate useful statistical information.
Hollerith's machines enabled the entire results of the 1891
census to be available in less than two years. Had manual
methods been used, the results would have been ready only
in 1901! Hollerith's machines were there to stay for a very
long time. In fact, the company founded by him eventually
t r a n s f o r m e d i t s e l f i n t o the I n t e r n a t i o n a l B u s i n e s s
Machines (IBM), which is one of the leaders in the field of
computers even today.
Charles B ABB AGE (1791-1871), an eminent English math-
ematician commenced on a venture in the 1830s to solve
simple mathematical polynomial equations using mechani-
cal machines. His first venture was to design a "difference
engine". Later,he also designed an "analytical machine"
which could solve any complex mathematical problem. The
innovation in his design was the incorporation of a
"memory". He called this a "store" and used it to hold the
answers of various subcalculations till they were needed at a
later stage. However, Babbage was not entirely successful
with both his ventures as he lacked a precision workshop
THE GENESIS 5

with fine machining facilities. Yet, Babbage is credited with

having thought of all the basic components of a modern
computer.
By the 1930s, technology had advanced and Babbage's
ideas and concepts were revived. In 1937, Howard AIKEN
(1900-1973) at the Harvard University designed the first
electromechanical computer. It used punched cards to feed
in information and simple electric relays to store data.
Aiken's machine (Mark 1) was used to generate tables for use
by the artillery gunners of the army and also for solving
equations for those who designed the first atomic bomb.
Almost simultaneously, John MAUCHLY and Presper ECK-
ERT at the Pennsylvania University commenced work on a
similar machine but used thermionic valves rather than
relays. Since it was an all electronic calculating system it
turned out to be much faster than the Mark I. Nevertheless,
both these machines had some inherent limitations. The most
important of these being that in order to change a program
to solve another set of mathematical equations, considerable
rewiring had to be resorted to. The circuits had to be changed
and this resulted in delays. The other shortcomings were
their large size, a huge maintenance staff and an enormous
cooling mechanism.
Immediately after World War II, a large number of en-
gineers and scientists who had been working on these
machines found themselves free to exchange ideas with other
colleagues and to apply their new found expertise to solve
problems in other fields. Two very important events took
place during the late 1940s and these together enabled the
original ideas of Babbage to be engineered into some very
important products.
The first was the introduction of a new concept by John
von NEUMANN (1903-1957). He suggested that all the
detailed steps which the machine has to follow in performing
its calculations, should also be coded and entered into the
machine along with the data. Interestingly, at about the same
6 ARTIFICIAL INTELLIGENCE

time, an English mathe- DATA PROGRAM I

matician named Alan
TURING (1912-1954) also
suggested a similar idea.
This was a major
breakthrough since the con- \COMPUTER /

cept of a stored program

gave such machines consid-
erable flexibility. One no
longer had to do the rewir-
ing of machines in order to
solve a new set of mathe-
matical problems. All one
von N e u m a n n gave the concept
did now was to write out a of a stored p r o g r a m
new program or in other
words, prepare a new software and feed it into the machine
along with the data.

Another useful suggestion von Neumann proposed was

that all arithmetic be performed using the binary notation
instead of the conventional units of ten. The ones and zeros
used can be easily represented by the electronic toggles. The
first computers to be built using electronic toggles rather than
the electromagnetic relays were the EDVAC in U.S.A and the
EDSAC in England and these were the forerunners of the
modern computers.
THE GENESIS 7

The second advancement was a technological one. With

the invention of the semiconductor transistor by William
SHOCKLEY and his colleagues in 1948 it was only a matter
of time before the vacuum tubes were entirely replaced by
semiconductor devices. The foremost advantage of semicon-
ductor devices is their very low power consumption. A
revolutionary advance in semiconductor technology was
ushered in a decade later when KILBY of Texas Instruments
showed how an entire electronic circuit could be fabricated
on the surface of a silicon chip. These are known as Integrated
Circuits. \

VACUUM TUBE TRANSISTOR

INTEGRATED CIRCUIT

M1CROMIN1A TURE CIRCUIT

Shrinking c o m p u t e r components; William Shockley (inset)

8 ARTIFICIAL INTELLIGENCE

More recently, Very Large Scale Integrated circuits (VLSI)

with literally thousands of semiconductor devices are being
fabricated on single chips. The complexibility of these VLSI
circuits has grown year by year. The manufacturing tech-
nologies of such functional circuits have helped in reducing
the costs of such chips. The reliability of semiconductor chips
is also very high. Reasonably priced computers based on such
VLSI circuits are now available.

Computer and Brain

Computers are being used today to assist man in several
activities, such as navigation across oceans and the sky, and
even through space. What is more, they are used to regulate
human heart beat (pacemakers) and even control large
chemical industrial plants. No wonder computers are con-
sidered to be the machine equivalents of human brain. This
is s u p p o r t e d b y three a n a t o m i c a l a n a l o g i e s w h i c h
demonstrate a striking similarity between the functioning of
a human brain and a computer.
Firstly, just as a network of sensory neurons receive the
incoming data that arrives at the sense organs and then
transmit them as electrical impulses to the brain, so do input
devices like keyboards enter raw data into a computer.
Secondly, once the sensory impulses are delivered to the
brain, the latter processes this data and converts it into useful
information. Similarly, the Central Processing Unit (CPU) of
a computer, that is comparable to the human brain, processes
the input data according to a step by step sequence of com-
mands.
The CPU consists of three parts, namely, the Arithmetic-
cum- logic Unit (ALU), the Control Unit (CU) and the
memory. The ALU does the actual computing and performs
all logical operations. The CU analyzes the command signals
and sends instructions to various parts of the computer. Also,
it regulates the overall flow of information. Computer
THE GENESIS 9

PROCESSING

OUTPUT
INPUT

MAIN MEMORY

INPUT OUTPUT

MEMORY

Parts of a c o m p u t e r system
memory comprises data storage systems similar to human
memory which can store a wide range of information that
could be retrieved as and when required. A computer has a
R a n d o m Access M e m o r y (RAM) w h e r e data can be
read/stored and Read-Only Memory (ROM) from where
data can only be read without writing new data into it. In
addition, there are two other kinds of specialized memories
in a computer. One, the programmable ROM(PROM) that
records data only once and second, the erasable PROM
(EPROM) which has an additional facility that helps the
erasing of previous data and entry of new ones.
10 ARTIFICIAL INTELLIGENCE

Finally, in the brain, the processed information is brought

out through the motor nerves in the form of motor impulses
and is further carried to the muscles and various other body
organs. This function in a computer is performed by the
output devices such as the Visual Display Unit (VDU),
printer, etc.
The working mechanism of a computer is thus akin to that
of the brain. However, a pertinent query that springs up from
this brain-computer analogy is: Can a man-made machine
having no biological element entertain conscious thoughts
exactly as humans? Or simply, can computers simulate the
human mind?
Despite major breakthroughs in computer technology, it
has yet not been feasible to construct conscious machines -
systems that could mimic brain power. The computers have
till now only been used to do arithmetical calculations. There
is no doubt that they are efficient slaves and do whatever they
are instructed to do any number of times and without com-
plaining. Nevertheless, the abilities of a computer are not at
par with the ingenuity of the human brain. This has today
catalyzed research for the creation of machines that would
possess the intellectual abilities of human beings.

Thinking Machines
Many people have tried to compare the abilities of humans
with those of computers. Suppose you were in charge of a
large chemical manufacturing plant and it was your duty to
ensure that this factory operates at its optimum capacity with
minimum wastage of materials and human resources. You
could consult a number of experts who have operated similar
plants with success and use their combined expertise to write
a computer program in order to operate the factory. This
program would essentially imitate the thought process of a
'human expert' and manage the operation of the entire fac-
tory. Solutions to crisis situations would be anticipated and
THE GENESIS 11

Manager of a factory gets inputs from various departments

to help him take decisions
included in the program. Most people would agree that this
particular program demonstrates some (artificial) intel-
ligence in its task of operating the factory. This is a situation
where a computer simulates intelligent human behaviour as
12 ARTIFICIAL INTELLIGENCE

far as this particular assignment of operating the factory is

concerned. The same program would probably not be usable
in operating a shipyard! Some researchers have proposed
that the ultimate goal of artificial intelligence (AI) is to con-
struct a machine which could actually think like human
beings in all aspects. Such machines, known as Ultra Intel-
ligent Machines (UIM) do not exist yet, but who knows, they
may be feasible in the 21st century !
Human beings tend to think symbolically rather than
numerically. Our intelligence seems to be based on our men-
tal ability to manipulate symbols rather than numbers. On
the other hand, computers were originally designed to
process numbers. An algorithm is a step by step procedure
with well-defined starting and ending points used in solving
problems. The conventional computer architecture readily
lends itself to this step by step approach. Human reasoning
processes tend to be non-algorithmic i.e. our mental activities
consist of something much more than merely following step
by step procedures. Most of AI research has been devoted to
symbolic non-algorithmic processing techniques with the
help of computers.
In addition to symbolic processing AI researchers rely on
heuristics — assumptions based on past experience — to
solve problems. By using heuristics one does not have to
completely rethink every aspect of a problem one is faced
with. If a handy rule of thumb based on earlier experiences
exists, then it is applied to the problem (or to a particular part
of a problem).
Another aspect of AI is the use of pattern matching proce-
dure to discover relationships between activities just as
humans do. Here one tries to describe objects, events or
processes in terms of their features and relationships. People
tend almost instinctively to discover relationships between
things. They sense qualities and spot patterns that explain
how various items relate to each other. If computers are to
become more intelligent they must be able to make similar
THE GENESIS 13

associations between objects, events and processes which

people do in a natural manner.
AI continues to experience rapid changes in both concept
and scope. Three major areas of AI are expert systems, natural
language processing and robotics.

ARTIFICIAL INTELLIGENCE

NATURAL
EXPERT ROBOTICS
LANGUAGE
SYSTEM SYSTEMS

Branches of Artificial Intelligence

Expert systems: These are computer programs designed to

emulate the reasoning process of an experienced professional
in a particular area of expertise. These are also known as
knowledge based systems. These systems are designed to
assist experts and not necessarily to substitute them. Expert
systems have proved extremely useful in such diverse fields
as medical diagnosis, chemical analysis and geophysical
resource exploration. Further applications of expert systems
are likely to become available in other fields as well. Complex
instruments would soonbecome user- friendly as they would
contain programs to anticipate the difficulties a user might
face in using the machine.

Natural Language Processing : At present, the use of com-

puters is limited by communication difficulties. An effective
use of computers will be possible if people could communi-
cate with them through the use of natural languages. The
14 ARTIFICIAL INTELLIGENCE

field of natural language processing is divid ed into two parts.

Firstly, computers are trained to understand a natural lan-
guage such as ordinary English. This will enable the users to
communicate with the machine in a language with which
they are already familiar. Secondly, the machines can be
trained to produce outputs which are in English. This will
enable the user to understand what the computer has to say
very clearly. Some natural language interfaces are already
available as parts of business softwares such as spreadsheets,
database, etc.
Natural language processing consists of the following
aspects:
"•Processing English words and sentences typed onto the
computer through the keyboard. The response of the
machines, also in plain English, appear on the VDU or
through a printer.
""The machine can recognize words and sentences as
spoken by the user. Speech synthesizers can be used for the
machine to reply or to warn the user about a mistake he is
about to make.
"•Another aspect is computer vision where the machine
understands its environment with the help of tactile sensors
or TV cameras.
Robotics: A robot is an electromechanical system which can
be programmed to perform manual tasks. It can move
materials, tools, perform a variety of tasks and all these can
be programmed. There are two types of robots,viz. the fixed
place, industrial assembly robots similar to those used in
assembly of cars. The others are autonomous robots which
operate by themselves in the real world. Software is the
intelligence behind these devices. An intelligent robot in-
cludes sensory devices such as tactile sensors and TV
cameras. It can be programmed to perform different func-
tions. The system reprograms itself based on the signals
received from its sensors.
THE GENESIS 15

A robot

Closely linked with AI is the process of learning. Learning

as we all know, is a means of acquiring new knowledge.
There is already much research underway to create such
advanced programs which would enable machines to learn
just as humans do so naturally. Humans learn by the process
of memorizing facts and sequences of actions (to cycle, ride
horses, swim, etc.). Each item one learns is a piece of infor-
mation. Sequences are known as procedures. So when one
programs a computer one feeds in both factual and proce-
dural information. Another important way people learn is by
the cognitive procedure. Here a human being analyzes, or-
ganizes and correlates specific pieces of knowledge in his
mind. He subsequently learns to generalize after examining
a few specific examples. If systems can learn just as humans
do, it would be a very powerful tool as far as AUs concerned.
 human being is a remark-

A able animal as he is the

only one to be endowed
with such a large brain. One of the
major functions of this organ is to
p r o c e s s the e x t e r n a l s t i m u l i
which arrive through the senses
of vision, smell, hearing, touch
and taste. The sensory organs cor-
respond to the five different chan-
nels t h r o u g h w h i c h d a t a is
received from the outside world.

As soon as an external stimulus

which could be a photochemical,

Learning
s m e l l , s o u n d , t o u c h or a
biochemical signal arrives in the
sensory organs, the associated
and neurons undergo complex chemi-
cal changes converting the stimuli
Memory into electrical i m p u l s e s . The
electrical signals are then passed
on from one neuron to another by
release of chemical substances at
the synapse — the junctions be-
tween two neurons. On reaching
the brain, these impulses are
p r o c e s s e d to g e n e r a t e ap-
propriate responses.

While processing information,

the human brain is involved with
vital activities such as 'pattern
recognition', 'learning', 'thinking'
and ' r e m e m b e r i n g . All these
processes occur in the cortex
which is a thin layer on the sur-
face of the brain. In addition to
LEARNING AND MEMORY 17

H e b b ' s hypothesis; The release of neurotransmitter molecules

at the synapse (inset)

these basic activities, processes such as language interpreta-

tion and speech generation also take place in the brain cortex.

Learning is associated with much more than mere acquisi-

tion of factual information. It is a process which causes a
permanent change to take place in our behaviour. Physiologi-
cally, some corresponding change also takes place in the
brain with the acquisition of knowledge. Donald HEBB has
related the process of learning with some changes taking
18 ARTIFICIAL INTELLIGENCE

A cell assembly resembles a group of sky-divers holding each other in

patterns either (a) or (b); A cell assembly identifies letter ' A ' b y its three
defined strokes (inset)
place at the synapse across two neurons. The chemical trans-
mitter substance released at the synapse causes the next
neuron to fire. If the two cells activate over and over again
then the surface area of the synapse increases and the two
neurons tend to act as a pair. With increased surface area
there is an obvious strengthening of the connection at the
synapse. Hebb's theory is a plausible hypothesis of the
process of learning in the human brain.
LEARNING AND MEMORY 19

An individual cell in the visual perception region of the

brain may learn to identify say, a stroke in the letter 'A'. A
large number of cells which together are capable of identify-
ing the entire letter 'A' are collectively called a cell assembly.
It is only when this association is formed that we say we have
learnt to recognize the written letter 'A'. A cell assembly is a
sort of template formed in the brain which responds when
the appropriate visual signals are received. These groupings
can be extended to include individual words, sentences, etc.
Similar assemblies are also found to recognize acoustic
phonemes to identify the basic components of speech
sounds. Conventional learning therefore, involves
strengthening of linkages between m a n y thousands of
neurons. There is evidence to believe that an actual learning
task involves forming new links between individual cells or
between association of cells which already exist. This also
helps us to remember individual events, words, etc. by as-
sociation with separate individual pieces of information al-
ready stored in the brain.

Learning to Learn
As they grow older, all men and women acquire in their own
individualistic manner, ways of responding to unusual situa-r
tions. It is remarkable how they select their responses which
enable them to tide over a particular crisis or solve a problem.
In fact, whenever they come across a new situation they
subconsciously work out a strategy to arrive at a convenient
solution. Harry HARLOW (1905-1981) conducted experi-
ments which demonstrated that people can solve problems
in areas they are already familiar with (at least in principle if
not in detail). Harlow called these as "learning sets" as they
help human beings to learn how to leam. He suggested that
the method commonly referred to as "trial and error" is
actually an orderly development of learning and thinking
processes. When we first face an unusual situation we use the
20 ARTIFICIAL INTELLIGENCE

trial and error approach. Subsequently, when a similar situa-

tion arises at a later point of time, we find the appropriate
solution almost instantaneously (based on previous ex-
perience). This is commonly known as "insight".
Jean PIAGET (1896-1980) studied children over a long
period of time. According to him, children go through a series
of stages while developing their understanding and thinking
skills. Cognitive development is related to the development
of thinking, perception and making use of the memory.
Piaget suggested that a child's intellectual growth takes place
in a stepwise manner through the development of various
"schemas". These are internal (or mental) representation of
physical situations or actions. Usually, they are a set of rules
which the child creates, stores in his memory and sub-
sequently uses while interacting with his environment. It is
only through activity and interaction with the environment
that the child develops and modifies his schemas.

Schemas are formed by the processes known as assimila-

tion and accommodation. The manner in which a child inter-
acts with the environment depends upon his schemas at that
moment. This process of understanding the world is known
as assimilation. This process enables him to remember and to
obtain more and more information about the world. He
receives feedback from the environment which he uses to
verify the accuracy of his perceptions. When there is a mis-
match between his earlier schema and the feedback he has
received then he is in a state of 'disequilibrium'. Quite often
this is only a temporary phase and the child quickly evolves
a new schema to cope with the changed situation. The earlier
schemas may be modified to generate a new one or it is just
discarded. Continued interaction with the environment
brings new forms of stimulation to the child and this process
of changing his mental structures is known as accommoda-
tion. The assimilation of information into schemas and the
accommodation of schemas to cater to the demands of new
experiences occurs all through the child's development.
LEARNING AND MEMORY 21

Assimilation & Accommodation

22 ARTIFICIAL INTELLIGENCE

Intelligence is usually explained as being the ability to

acquire new skills and knowledge. A child develops cogni-
tive abilities in order to understand and interact more suc-
cessfully with his environment. It also enables him to form
new concepts and to grasp their significance.

Human Memory
It is important to realize that there are three basic activities
associated with human memory, viz. receiving new data
from the environment followed by its processing and conver-
sion into usable information within the brain; storing this
information in the memory; and retrieving it as and when
required.
Humans have three basic forms of memory. One is the
immediate memory. Here one retains all details of an object
or a scene or a picture for about 1 /10th of a second like a
batsman facing a ball. After that, much is forgotten. Next is
the short term memory where information is retained for a
few minutes only. This memory is also known as the working
memory and as such has a limited capacity. It is also used for
rehearsal,i.e. for mentally repeating the information over and
over again so that it enters the long term memory.
The information which is stored in the long term memory
can be broadly classified into four categories. First is the
information related to specific tasks, which includes riding a
horse or a bicycle, driving a car, typing, painting, carving, etc.
These skills often require several years of experience to per-
fect. Secondly, the information related to fear-inspired ac-
tions like being bitten by a dog, hurt by a falling wall, burnt
by a fire, etc. By proper training one can prepare oneself to
take specific actions under difficult circumstances or emer-
gency conditions. Third is 'Episodic memory', in which in-
formation regarding dated episodes or personal experiences
is stored date-wise. Finally, the 'Semantic memory', which
relates to the use of words, grammar, metaphors, and all
LEARNING AND MEMORY 33

58 60 14

T h e basic forms of h u m a n m e m o r y
24 ARTIFICIAL INTELLIGENCE

other niceties of a language.

H o w e v e r , t h e s e different
groupings are conceptual in
nature and no specific region
of the b r a i n as such is
reserved for these categories.
Information from the short
term memory is transferred to

the long term memory by the

process of repeating it a large
number of times as one learns
poetry in school.

It is common knowledge
that people tend to forget.
One e x p l a n a t i o n for this
phenomenon is that unless in-
T y p e s of l o n g term m e m o r y f o r m a t i o n is f r e q u e n t l y
retrieved and used, it tends to
LEARNING AND MEMORY 25

decay with time. Another explanation is that memory is

stable but with time other information is stuffed in and the
original contents of the memory are weakened or displaced.
Human beings have an ability to take cognizance of only
some of the incoming information while discarding the rest.
This phenomenon is known as "attention". Unless one con-
sciously pays attention, most of the information arriving
from the sensory organs is lost. For example, when one enters
a room with a large gathering of people, the first impression
is one of a buzzing confusion. However, if someone mentions
your name or happens to be discussing a subject which is of
particular interest then you suddenly concentrate and try to
follow that conversation. This is a common phenomenon and
is known as the 'Cocktail Party' effect. It is only when we pay
attention that the processed data or information is sent to the

Perception: Guessing g a m e
26 ARTIFICIAL INTELLIGENCE

short term memory and then perhaps to the long term

memory. It would be extremely taxing on the brain if we did
not have this ability to be selective.
Perception is the creation of a visual schema in our mind
based on clues arriving from our sensory organs. We trans-
late the signals from the sensory organs into individual,
meaningful and stable items in our mind. Since a schema is
an implicit mental image or a theory which we create on the
basis of worldly experiences, we are likely to have a large
number of schemas already stored in our memory. These are
created specifically to understand the world we live in.
According to Hebb, perception is an acquired charac-
teristic. The sensory experience is registered in the form of
cell assemblies. As the sensory patterns become more com-
plex and stable, the entire combination of cells are activated
together in response to a specific stimulation. These larger
sequences are known as 'phase sequences'.
However, according to the Gestalt school of psychology,
perceptual organization is an inborn characteristic. The word

A contrast to G e s t a l f s theory of perception

LEARNING AND MEMORY 27

gestalt is German for 'pattern' or 'form' and the theory em-

phasizes the ability of human beings to perceive patterns as
a whole or in their entirety. When we perceive, we give
meaning to objects by their characteristics as a whole (in toto)
and not by considering them as a jumble of parts which go
into making up a total figure. The clues from the sensory
organs along with the rules or guidelines help humans to
recognize these patterns which in turn will help in his per-
ception.

The Thought Process

As we live and grow, we interact with our fellow human
beings and also with the environment. Consequently, our
collection or store of schemas keeps on increasing. The sum
total of the schemas which we possess eventually becomes
our "knowledge". This word is used in the widest sense of its
meaning. Our schemas include a variety of visual descrip-
tions such as identification of written characters, repertoire
of written words, pictures, spoken phonemes, spoken words
and so on. These are stored in our long term memory.

The schemas mentioned above are at the elementary level.

We also possess a mental ability to link schemas together and
produce new schemas which are at a higher level. This
process is known as "thinking". For example, while driving
early in the morning we may come across a red light at the
traffic junction. Now, there is a problem. Should we risk it
and proceed without stopping. We could undergo a thought
process as follows. We visualize a situation where we do not
stop but proceed. This is followed by visualizing ourselves
being stopped by the traffic police stationed just around the
corner who either fines us on the spot or asks us to appear
before a magistrate. As yet we have seen neither the
policeman nor the magistrate (but we know that they do
exist). Not wishing to experience it, we decide to apply the
brakes to the car and halt at the red light. This process of
28 ARTIFICIAL INTELLIGENCE

Evaluating the different possibilities before arriving at a decision

LEARNING AND MEMORY 29

mentally going through a series of actions is "thinking". It

ends in our deciding on a course of actions which, hopefully,
is in the best interests of all concerned. The thinking process
need not necessarily be concerned with real objects only but
could involve abstract concepts as well!
We have yet another ability and that is to mentally classify
things and events around us. Based on our earlier experience
we learn how to group objects or events according to charac-
teristics or attributes which are common to them. For ex-
ample, there are certain types of plants which though come
in all sizes and shapes are yet grouped as trees. So, we have
a concept or an idea regarding the attributes or characteristics
which are similar and which help us to group such plants as
trees. We would certainly not classify grass or shrubs as trees.
We learn of these facts by observation, discussing with other
people and by reading books.

Concepts are symbolic for us human beings. They are very

useful and help us in comprehending complex objects and
situations. The concept of a "car" can be vividly recalled in
our minds on just hearing the sound of the engine or its horn
or the screeching of its brakes. This concept leads us, by our
thought processes to recollect about the mechanism of the car,
the need to buy petrol, the price of petrol, requirement of a
license to drive a car, new traffic laws which are currently
being enforced, etc. There can even be a hierarchial relation-
ship amongst these new concepts as to which is more impor-
tant from say, convenience of transportation or from the
speed aspects or from the running cost points of view.

Epistemology is the study of knowledge. It is concerned with

the nature, structure and origin of knowledge. Knowledge
too has been sorted and classified under many groupings. For
example:
30 ARTIFICIAL INTELLIGENCE

* Causal knowledge is the understanding of the underly-

ing cause and effect of various aspects of life.
* Procedural knowledge is knowing how to do something,
e.g. how to make tea.
* Declarative knowledge refers to knowledge being ex-
pressed as a declarative statement e.g. "true"/"false".
* Tacit knowledge is also known as the unconscious
knowledge since it cannot be expressed very specifically in

TYPES OF
KNOWLEDGE
LEARNING AND MEMORY 31

words. For example, how to ride a bicycle or a horse is

something one may know but find difficult to explain.
* Deep knowledge is acquired by exhaustive study and
understanding of a subject. Usually it is restricted to one or
two domains of knowledge as it is not possible for a single
person to study all the different areas of learning.

* Shallow knowledge is based on a superficial or an empiri-

cal understanding of a subject.

* Heuristic knowledge is a type of shallow knowledge. It

consists of various rules of thumb based on first-hand ex-
perience. They aid in solving problems but there is no guaran-
tee that they will succeed. It is in areas such as medicine and
practical engineering that heuristics play a useful role in
problem solving.

Language, an invention of man, is the symbolic vehicle of

our thought process. Besides externalizing his thoughts by
speaking and writing, man also uses language to internalize
his thoughts i.e. to think about his ideas or concepts in his
own mind or to find solutions to problems. Language, there-
fore, is very helpful in widening the scope of human
knowledge.

Yet another unique characteristic of human beings is

creativity and this includes both creative thought as well as
creative action. They both refer to generating something
original, something new which no one else had thought of
earlier. An example of the former is the concept which Max
PLANCK (1858-1947) introduced regarding energy. He sug-
gested that energy could be quantized at the atomic levels.
This novel concept which he introduced in 1895 found imme-
diate acceptance as it resolved a number of vexed problems
in physics and chemistry. Examples of creative activities are
the much appreciated wall paintings of the Ajanta caves,
sculptures of the Ellora caves, poetry by Kalidasa, etc. Re-
searchers have been intrigued as to what makes a man
32 ARTIFICIAL INTELLIGENCE

M a x Planck

creative and many theories have been put forward on this

subject but these remain speculative.

Thinking, conceptualization, problem solving and

decision taking are all intimately related to one another. They
are closely linked with man's knowledge and intelligence.
Quite often the process of thinking is followed by a decision
which may have to be taken. Some decisions have far reach-
ing consequences and once made cannot be retracted. There
LEARNING AND MEMORY 33

are also a large number of decisions which we take one day

and change the next day. This is often done on the basis of
new data or information which becomes available. When we
take a decision we can usually explain why our proposed
course of action is the best one. Nevertheless, it is not so easy
to give a detailed account of how we arrived at our con-
clusion. They are generated in the mind where emotion,
creativity and intuition are all at work. The goal of AI is to
simulate the human mind, particularly those aspects which
relates to thought process and decision making.
In many problems requiring solutions, one finds that there
are no black and white answers. Invariably, an element of
probability is involved. A large science has grown up around
this probabilistic concept in problem solving and is known
as the Fuzzy Set Theory.
 B y the late 1940's machines to
assist man with mathemati-
cal calculations had reached
an advanced stage of develop-
ment. At that time many scientists
began to ponder over the pos-
sibility of constructing machines
which could think. They got en-
couragement from the develop-
m e n t of a m a c h i n e c a l l e d
ENIGMA by Alan Turing. This
machine was designed during the
War to decipher coded German
messages. Turing subsequently
contributed to the design of a
Scope and stored program computer in U.K.
His ultimate aim was to design

Extent machines which understood the

general logical operations like
AND, NOT and OR. His idea was
that these s y m b o l s used to
describe mathematical processes
and logic could also be used to
manipulate words. Turing's ex-
tension of machines to process
non-numerical symbols has been
quoted in some quarters as the
starting point of AI. Turing died
prematurely in 1954, but the ball
had already been set rolling.

M e a n w h i l e A n t h o n y OET-
TINGER had been attempting to
use a c o m p u t e r to translate
English words into Russian and
vice versa. An English word fed
into the machine would result in
SCOPE AND EXTENT 35

a Russian equivalent to
a p p e a r on the s c r e e n .
Here, what took place was
matching of equivalent
words from the two lan-
guages. For every English
word a Russian counter-
part was selected. It was
fondly hoped that this
beginning would even-
tually lead to a machine
which could translate con-
tinuous text from one lan-
guage to another.
Unfortunately, the syntax
of the two languages had
not been taken into ac-
c o u n t . The p r o c e d u r e
which a c o m m o n man
uses in stringing words
together to form meaning-
ful sentences had yet not
been p r o p e r l y inves-
tigated at that time. This
led to serious problems in
automatic translation of
texts and often resulted in
amusing situations. For
instance, an English sen-
Alan Turing and his machine
tence, The spirit was willing
but the flesh was not came back after retranslation as The wine
was good but the meat was spoilt!

With this background, a group of eminent computer scien-

tists gathered at the Dartmouth College (U.S.A.) in the sum-
mer of 1956 to discuss about possible "thinking" machines.
Many ideas were put forth and discussed. Subsequently, they
36 ARTIFICIAL INTELLIGENCE

all went their own ways to develop new systems. Many of

them founded their own research centres to develop such
machines along with the necessary software. The term "Ar-
tificial Intelligence" came into common usage only after this
meeting.
It is important to note that there is a difference between an
AI system and AI. The former is a computer based system
that lets you solve certain types of problems. On the other
hand, AI is an academic discipline where one studies how
human beings behave intelligently and how machines could
emulate this intelligent human behaviour. An advantage of
AI is that it is a useful discipline and enables us to solve
problems in a practical manner. The procedure consists of
first studying the problem, getting to know all the facts and
the interrelated rules and then finding an appropriate solu-
tion.

Knowledge Engineering
AI c o n c e n t r a t e s on k n o w l e d g e e n g i n e e r i n g . Here
"knowledge" refers to a particular domain of expertise. This
expertise could include highly specialized or technical infor-
mation which would not normally be expected to be available
with an ordinary person. The term "engineering" refers to a
practical approach of using this knowledge to solve
problems. This is indeed a commonsense way of doing
things.
Knowledge engineering is the discipline that deals with
the way a knowledge-base is organized so that it becomes
available in the most useful manner possible. This knowledge
is collected by a knowledge engineer by searching literature
or even by asking people who are considered knowledgeable
in that particular subject. His next task is to arrange this
knowledge in some kind of an order so that it is easy to access
and retrieve quickly. This structuring of information is how-
ever, much easier said than done. One way would be to place
SCOPE AND EXTENT 37

KNOWLEDGE
ENGINEER EXPERT SYSTEM

EXPERT

USER

Development of an expert system

the information most likely needed at the top and the least
likely at the bottom, but then who is to decide? It is often
difficult to get human experts to unanimously agree as to
which aspect is more important than others. Unlike computer
systems which can scan through its entire memory in a matter
of seconds, it takes a considerable time to extract information
from a real human expert. Usually they are hesitant to tell
what they know. This is one of the reasons why developing
a knowledge-base for an expert system requires such a long
time.

Can Computers Think?

When we say a computer is thinking what we really mean is
that the computer is executing a program and is handling
information and manipulating it just as humans do. Here a
pertinent question is, do people know how they think? They
can tell what they are thinking about and what is the result
of their thought process, but certainly, they are unable to
explain the mental steps involved.
A person acquires data, information and other details from
his environment. This information keeps arriving through
38 ARTIFICIAL INTELLIGENCE

his sensory organs

on a continuous
basis. Since all data SENSE ORGANS
received are not im- SIFTING
portant, the brain
sorts them out and
rejects most of
them. But the
remaining data
which are con- SORTING

sidered vital or im-

portant are then
organized and
stored as 'know-
ledge.' Finally, the
stored knowledge
is retrieved as and INTERPRETING
when required and
is used to solve A c q u i s i t i o n a n d p r o c e s s i n g of d a t a
problems, make
decisions and assimilate additional new information which
has just arrived. When one wants to gain knowledge in a new
area of study, one makes a conscious effort in learning every-
thing about it. Once the existing body of knowledge has been
acquired one seems to know how to apply it. The unconscious
ease with which knowledge is applied is what makes intel-
ligence so difficult to quantify and hence simulate. The
reasoning mechanism in man is difficult to explain in con-
crete terms. With the goal of AI being to simulate intelligent
behaviour, computers have to be taught how to analyze
problems and to take decisions just as human beings do.

For a computer program to be considered intelligent it is

necessary that it acts as a human being would. Its thought
process however, need not be exactly same as that of a human
being. For example, an assistant in a large office spends most
of his time physically retrieving data from existing files,
SCOPE AND EXTENT 39

1 HUT ' 1

W h o ' s smarter?
40 ARTIFICIAL INTELLIGENCE

reorganizing it, preparing new reports and so on. If he does

a good job of extracting the desired information from the
huge volume of data then his superiors do not hesitate in
praising him and saying that he has used his brains (i.e. used
his thinking process). If a computer can manipulate data in a
logical manner then it too should qualify to be called intel-
ligent (or at least as intelligent as the clerks)!
Finally, the scope of AI has grown so rapidly over the past
35 years that it has already branched off into many categories.
Besides expert systems, natural language processing and
robotics, it has even been extended to machine learning that
helps a user to operate new machines and systems,
knowledge representation, data search methods and logic
programming. Each one of these have become important
subjects of study in many research centres of the world.
 r r ihe ability to solve problems
is usually taken as an index
of intelligence for human
beings. It is tempting to extend
this criterion to machines as well.
There are two kinds of problems
one comes across in real life. First-
ly, there are those which can be
solved by strictly adhering to well
established procedures or algo-
rithms. These are algebraic (or
formula) type of problems. Here
all one does is to feed data related
to some unknown variables in a
formula and the machine calcu-
A lates and produces the correct
answer. However, many

Problem problems do not quite come

under this group. They relate to a
search through a large volume of
Solver data to find the correct answer. AI
is mainly concerned with such
problems.

Soon after the Dartmouth Con-

ference, a technique called the
General Problem Solver (GPS)
was put forth by Allen NEWELL
and Herbert SIMON. Many other
powerful techniques have sub-
sequently been developed, which
come under the general category
of 'expert systems'. However,
many practical AI programs con-
tinue to use the basic concept of
the GPS technique.
42 ARTIFICIAL INTELLIGENCE

Alari Newell and Herbert Simon introduced the " I F - T H E N " rule
A JREOBLEM SOLVER 43

Newell and Simon demonstrated that many of the human

cognition or problem solving capabilities can be expressed by
production rules of the type "If-Then". For example, "IF you
are thirsty, THEN drink water". Or "IF you see a red traffic
light, THEN stop the car". These production rules apply only
to small but specific portions of knowledge which are known
as 'chunks'. Such chunks are arranged in a loose manner in
the long term memory. Moreover, Newell and Simon
believed that these chunks are linked to other allied chunks
of knowledge according to certain rules. An example of one
chunk being connected with another related chunk is: "IF you
are thirsty AND your glass is empty THEN fill up the glass
with water". Thus, they popularized the use of IF-THEN
rules to represent human knowledge.
Until the mid 1960s efforts were being made in the area of
AI to design intelligent systems which relied less on domain
knowledge and more on the methods of reasoning. Now
arriving at a solution based on a search through a huge mass
of data can be an extremely lengthy process. To overcome this
drawback, alternate search
procedures were develop-
ed and these continue to be
used in the expert systems.

One of the methods

used to simplify the search
is to draw a tree diagram
with possible solution
paths. A tree diagram con-
sists of nodes which cor-
respond to a certain 'state'
where the problem is at a
particular moment. It can
then move to other nodes
or states by various paths. ROOT NODE

The task of the problem

solver is to discover a se- A simPle tree diaSram
44 ARTIFICIAL INTELLIGENCE

Search for the missing purse

A JREOBLEM SOLVER 45

quence of state changes which initiates from a 'starting state'

(or from where the problem was originally posed) to a solu-
tion or the 'goal state'. Let us assume the problem is a search
for your missing purse. You are reasonably sure it is some-
where in the house. So you draw a diagram indicating how
you might go about the search. A possible path you take
might first lead you into your bedroom and you then search
the wardrobe, the writing table and then search the bathroom
and finally the kitchen. A pictorial description is useful since
it helps to visualize the search path even before it has been
undertaken. One could prepare many alternate search
programs in advance. These are then compared and a path
most likely to be fruitful is selected. However, a heuristic
approach for finding the missing purse would be to try and
recollect which pair of trousers you were wearing last eve-
ning and then to search its pockets.

Search Strategies
Once the tree diagram is established, one has to evolve a
strategy to proceed from the starting point to the goal or
solution. If this search net-
work is an extensive one,
then keeping track of the
path is difficult. There are
four types of strategies
which are commonly used
to find a solution to a
problem, viz. depth-first,
breadth-first, bidirection-
al and means-end.

Depth-first search: To ex-

ecute a simple search pro-
gram, one could adopt the
'depth-first' strategy. In
Depth-first search
this method each possible
46 ARTIFICIAL INTELLIGENCE

path is explored to its end before another path is tried. One

goes systematically from the beginning to the end of a path
and investigates all the nodes and their linkages to the next
nodes till the right solution has been found. This depth-first
technique is bound to find the solution because, in the worst
case, the solution will correspond to the last node. However,
to find the optimal solution one has to traverse almost all the
nodes. This is a very time-consuming method particularly
when there are long branches to be searched and they do not
contain the solution.
Breadth-first search: The
'breadth-first' strategy is
the opposite of the depth-
first method of searching.
Here all the nodes at the
same level are searched
before proceeding to the
next deeper level. The ad-
vantage of adopting the
breadth-first technique is
in situations where the
solution is close to the sur-
face. In case the solution is
deep inside, then once
again a c o n s i d e r a b l e
amount of effort will have
to be expended in arriving Breadth-first search
at the solution.

Bi-directional search: This strategy consists of two sear-

ches conducted simultaneously. One works forward from the
initial state and the other works backward from the goal state.
New levels are generated alternatively and as soon as the two
processes meet then the shortest possible route would have
been found.
Means-end analysis: In this analysis, the problem is usually
broken-down into a series of smaller sub-problems, each of
A JREOBLEM SOLVER 47

which is solved separately. This has the effect of keeping the

size of each search-space within reasonable limits. One now
moves from one sub-problem to another. The new tree will
consist of nodes which are both the goal of one sub-problem
and the root node from which the search for a solution to the
next sub-problem would commence.
Generally speaking the programmer has to choose be-
tween the depth-first and the breadth-first search methods.
The choice eventually depends on his earlier experiences in
solving similar problems. The aim of AI is to show the
programmer a superior methodology of choosing between
either of the two techniques. Heuristic methods are of great
help in solving such problems since they are based on per-
sonal experience and intuition. This search works much bet-
ter than a blind search since it is used to decide which
particular node to search next.

Knowledge-base
A knowledge-base contains all the facts, rules and procedures
related to a specific field. It is important for solving problems
in that field. In contrast to a human expert whose capacity to
work diminishes with age, a knowledge based expert system
can be made more and more efficient by collecting expertise
from not one but a number of experts. The computer program
along with the knowledge-base can be copied to create any
number of expert systems and so the expertise is never really
lost. Moreover, an expert system is available any time of the
day unlike a human expert who needs to rest between work.
Above all, a computer expert system has no preferences as to
whom it provides the expertise. The user may be a very senior
Director in an organization or a junior officer. The same
expertise is provided to both.
Before loading information into the knowledge-base, the
knowledge engineer has to define the relationships between
facts, objects and groups of objects. Let us suppose that the
48 ARTIFICIAL INTELLIGENCE

objects within a restricted knowledge-base relate to

aeroplanes. There are many different kinds of aeroplanes:
transport planes, military planes, passenger planes and so on.
The relationship between the objects which are classified as
aeroplanes is that, they satisfy the definition of being
machines with fixed wings and engines and which fly in air.
So, a passenger plane inherits this categorization and also has
an additional definition of being capable of transporting
people. This procedure is known as object-oriented program-
ming and is commonly used in developing expert systems.
The knowledge which has been obtained from an expert
has to be formalized and structured. Usually the knowledge
engineer sits down with the expert who describes a large
number of cases he has handled or problems he has solved.
From these interviews relevant knowledge is extracted.
Another way of obtaining the expertise is to observe an expert
while at work.

The knowledge engineer has to make the ideas so explicit

that the computer can accept them. An object is defined first
by assigning a name to it, then its attributes are specified and
finally values are assigned to these attributes.
Name of the object: Bicycle

Attributes: Steel frame, steel handles, pair of

pedals, two large wheels, chain
drive to rear wheel and a leather
or plastic seat.

Purpose: For personal transport.

Operation: (a) Hold the handles with both
hands; (b) Put left foot on the
pedal; (c) Push the bicycle with
right foot; (d) Climb bicycle; (s)
Pedal with both feet; (f) Steer
with handle; and (g) Reach des-
tination.
A PROBLEM SOLVER 49

A guessing g a m e defines an object for a knowledge-base

50 ARTIFICIAL INTELLIGENCE

In a conventional computer, the data and the program are

both loaded into the system. The program is in the form of
algorithms and each individual processing step which the
computer has to take is well defined. In an expert system, a
knowledge-base comprising declarative, procedural and
heuristic knowledge is created and entered into the machine.
The ways by which knowledge can be expressed are by tree
representation, semantic networks, production systems,
frames, list and predicate logic.

Tree representation: A tree is a graphical representation of

the knowledge-base. It consists of nodes which store infor-
mation and branches which connect them. The root node is
the highest node in the hierarchy of knowledge. From this
root node are paths or links which lead to other nodes. A
simple example of such a graph is a road map of a country.
The cities are the nodes and the roads are the links. Trees also
store hierarchial knowledge. An expert system which uses
this knowledge tree would ask its user relevant questions as
it moves through the tree.

PROBLEM

SUB-PROBLEM 1 SUB-PROBLEM 2 SUB-PROBLEM 3

SUB-PROBLEM 4 SUB-PROBLEM 5

SUB-PROBLEM SOLUTION

P a t h w a y s to d e c i s i o n
A PROBLEM SOLVER 51

Another application of the tree structure is in arriving at

decisions (these are called decision trees). It begins with the
statement of the problem as the root node. From this node
decision paths lead to new sub-problem nodes. Heuristic
rules are used extensively in arriving at decisions. In a way,
these rules prune the decision tree of its unproductive
branches.

Semantic networks: A semantic network is a traditional

method of representing propositional information. A
proposition is a statement that is either true or false. It is a
form of declarative statement since it represents facts. (A
proposition is sometimes called 'atomic' since it cannot be
further divided). Semantic networks are graphic repre-
sentations of knowledge based on objects and their relation-

FINISHED COMPONENT RAW

PRODUCT MATERIAL

A simple semantic network

52 ARTIFICIAL INTELLIGENCE

ships. The nodes of the networks correspond to objects and

the link between the two nodes represents the relationship
between the objects. Such networks are used to analyze the
meaning of sentences in natural language processing.
Production Systems: In an AI program where a course of
action is to be recommended it is more efficient to use a
procedural system of knowledge (i.e. facts and rules which
tell what to do with the facts). A popular procedural system
of knowledge representation is with the help of production
rules which describe condition-dependent actions and this
includes the IF-THEN rules.
Some advantages of production rules are:
* Explanations become simple. To explain a particular
reasoning all one has to do is to keep track of the rules which
have been used and display them on the screen.
* Modifications are easy. In rule-based systems the
knowledge-base can be modified by adding, deleting or
changing the rules.
* Understanding is also easy since the production rules are
extremely readable.
Frames : A frame
is a method of rep-
resenting objects
and their relation-
ships in the form of
columns and rows.
It is useful since the
method envisages
breaking down ob-
jects or situations
i n t o their con-
stituent parts. These
parts are entered in
corresponding slots
in! the -'frames. The
A PROBLEM SOLVER 53

slots can be further subdivided for more detailed structuring.

The frame is not a universally defined data structure. In each
case the frame structure and the slot have to be clearly
redefined. The frame provides greater flexibility in repre-
senting knowledge than the net. Frames were originally
developed to represent stereotyped knowledge (i.e. those
with well-defined features). Their columns and rows are
easier to understand compared to logic and production rules.

List: Knowledge can also be represented as lists like a card

catalogue in a library. The cards contain all details such as the
authors' names, title, publisher, contents, keywords, etc. One

A library catalogue list

54 ARTIFICIAL INTELLIGENCE

can go through each card, rejecting those which do not match

what one is searching for or select those which do. By proper
indexing one can make lists as efficient as the trees.

Predicate logic: The fourth way of representing knowledge

is in the form of statements or predicates. It is a formal way
of describing objects and their logical relationships. It also
includes grammar or syntax for generating valid logical state-
ments. Predicate logic has semantic rules (meanings) which
relate the symbols of the formal language to the objects and
the processing rules which can generate valid logical expres-
sions from other logical expressions (e.g., PROLOG).

Inference Engine
Just having a great deal of knowledge does not make one an
expert. One must know how to select the appropriate
knowledge and how to apply it. Similarly, a knowledge-base
alone does not make an expert system. The system must
possess another component which directs the implementa-
tion of the knowledge. This component is the inference en-
gine. It decides which heuristic search technique is to be used

USER
'xa

KNOWLEDGE
i*
USER INTERFACE

BASE

INFERENCE
M . ".-< "

ENGINE

Major components of a knowledge based system

A PROBLEM SOLVER 55

and to determine how the rules of the knowledge-base are to

be applied in solving problems.
The inference engine makes the inferences by deciding
which rules are satisfied by facts, gives priorities to these
rules and then executes the rule with the highest priority first.
The agenda is a prioritized list of rules. The explanatory
facility keeps on explaining to the user the reasons for having
taken a particular decision. (Very unlike a human expert!)

Therefore, an expert system operates in a three-stage man-

ner. Firstly, all data about the problem is entered into the
machine memory by the knowledge engineer after consult-
ations with the experts. Secondly, all rules which apply to the
particular problem are identified. Finally, the heuristic and
conflict resolving strategies are selected.

The inference engine organizes elaborate searches, helps

in sorting of information, and comparisons between a par-
ticular information against a particular model. The inference
engine then tries to use the information to find an object
which matches the input data or in other words tries to find
a solution. The inference engine uses one of the few well-
proven techniques of inferencing to arrive at a recommenda-
tion. It performs all such tasks which gives the system a
semblance to human intelligence.

There are two general strategies for inferencing which are

used in expert systems. These are the 'forward-chaining' and
the 'backward-chaining' methods. There are other strategies
as well which include means-end, backtracking, etc. A com-
bination of inferences which connect a problem to its solution
is known as a 'chain'. A chain which searches from the
problem statement to its solution is known as forward-chain-
ing. It is also known as "data-driven" since the inference
engine uses the information provided by the user to move
through facts and rules in the knowledge-base to reach a
conclusion. The inference engine in this case starts from the
set of data provided by the user and then tests all the
56 ARTIFICIAL INTELLIGENCE

FORWARD
CHAINING
GOAL

BACKWARD
CHAINING

FACTS/INITIAL KNOWLEDGE

Inferencing strategies used in expert systems

hypothesis in which this data plays a part. If at the end the

machine does not reach a conclusion, it can then ask for
additional information or would simply indicate that there is
no solution. The system operates on the basis that all rules
must be satisfied without exception. Hence the forward-
chaining method starts with the data which has been fed-in
and it tries to find a solution in conformity with the rules.

Backward-chaining is the reverse of forward-chaining. It

is a chain which proceeds from a hypothesis back to the facts
or data. It really involves working backwards from an as-
A PROBLEM SOLVER 57

sumed hypothesis to see whether the given facts can make

that possible and valid. It sees whether the data conforms to
this particular hypothesis. Backward-chaining can result in
an expert system coming up with more than one solution to
a problem.
In practice, a solution by either of these two methods often
involves working forward to a conclusion, testing that con-
clusion for reasonableness, working backward to make ad-
justments, working forward to another conclusion and so on
until an acceptable end-solution is reached. This is how
human beings solve complex problems. In the same manner,
expert systems also through combined forward and back-
ward chain methods reach the solution to a particular prob-
lem. This process is known as sideways-chaining.

Success Stories of AI
Both expert systems and conventional database programs
involve retrieval and processing of information. However,
there is a difference between the two. In medicine, for ex-
ample, a database is useful to enumerate the symptoms of
different illnesses. An expert system, on the other hand, helps
to diagnose an illness, determine its causes and suggests a
line of treatment. The database program may reorganize the
data but does not reason about it. In expert systems, it is a
human expert who has provided the information about spe-
cialized areas of knowledge such as say, water management,
diseases of the blood, methods of teaching, etc. into a com-
puter.
One of the first successful expert system was the Mycin. It
was developed in the mid 1970s by Edward FEIGENBAUM
and Edward SHORTLIFFE at the Stanford University. It is a
medical diagnosis expert system. Its purpose is to help the
doctors or physicians to diagnose and to prescribe medica-
tion to those suffering from infectious diseases of the blood.
The problem is to first decide what bacteria has caused the
58 ARTIFICIAL INTELLIGENCE

Medical diagnosis by an expert system

illness and then to prescribe an appropriate medicine which
will be effective on the bacteria and at the same time have
minimum side-effects on the patient. The physician first puts
the general data including past medical history about the
patient into the completer expert system. The Mycin then
proposes a hypothesis and then goes ahead testing it. First it
searches for clues to support that hypothesis. In case the
expert system is unable to confirm or reject the hypothesis,
then it can ask the physician additional questions and even
A PROBLEM SOLVER 59

request for certain tests to be performed. The diagnosis is

made by comparing the symptoms of the patient with those
of the various possible infections the patient might be suffer-
ing from. This process goes on till a match is found. Later,
taking into account previous medications prescribed and the
patient's known responses to certain drugs, the expert system
offers a prescription which should be just appropriate for the
patient. This program has been very much appreciated as the
user-computer interaction is in the form of a dialog. The
program gives a reason as to why it is going to ask the next
question. Even though the scope of Mycin remained
restricted to a particular type of illness it paved the way for
many similar programs in the field of medicine. Puff is an
expert system for lung diseases. Mycin and Puff work just as
well as any alert physician would.

Another expert system known as the Prospector is used in

the field of geology and mineral prospecting. It is used to

COPPER DATA
METERS

ARGILLIC

PHYLLIC

POTASSIC

PROPYUTIC

LIMIT OF SULPHIDE

Prospector helps locate mineral deposits

60 ARTIFICIAL INTELLIGENCE

predict whether certain precious minerals and natural

resources (including petroleum and gas) are found in par-
ticular regions under the earth. All information about which
particular mineral ore would be found in what specific
geological regions forms the domain of knowledge for this
program. The data consists of results of analysis of rock
samples. A number of other expert systems would be
forthcoming and interestingly, some of these would even
provide advice on matters related to tax, insurance and legal
advice!
 he first person in recorded

T history, to have defined

l o g i c was the G r e e k
philosopher ARISTOTLE. His
logic dealt with deriving the truth
(T) or falsehood (F) of statements
b a s e d on a r g u m e n t s . His
methodology is referred to as the
classical logic or syllogistic logic.
Arguments based on classical
logic are categorized as "syl-
logisms".
Syllogism is a simple example
of formal logic. The expression -
'formal' means that the logic is
Program- associated with the form of the
statement rather than the mean-
ming ing of the words. Algebra which
one learns in school is also a kind

Intelligence of formal logic. In this form of

logic, the procedure is important
than the values assigned to the
variables. In many ways syllogis-
tic logic is formalization of com-
monsense. This type of logic
continues to be the basis of legal
arguments in our courts.
An argument based on syl-
logistic logic consists of premises
(or facts) and a conclusion. The
l a t t e r is d e r i v e d from the
premises.
Consider for example the fol-
lowing elementary a r g u m e n t
relating to a child named Ram
and his love for sweets:
62 ARTIFICIAL INTELLIGENCE

Fact 1:
Ram is a boy.

Fact 2:
All boys like sweets.

Conclusion:
Ram likes sweets.

Symbolic logic
The above example can also be converted into a symbolic
form where the symbol R stands for Ram, B for Boy and S for
Sweets:

•
' *V " - F*
IT

all B
henceR^ IT

Symbolic logic was first introduced by Gottfried Leibnitz

about 300 years ago but probably it was premature and did
not catch on. Hundred and fifty years ago, George BOOLE
(1815-1864) rediscovered symbolic logic. It dealt with
abstraction of concepts into symbols and connecting them to
form logical statements. To be able to reason we have to say
things like "if 'this' is true then 'that' must be true" or "if both
this and that are true then the third thing must also be true".
To link the different facts together, one uses operators or
connectives such as 'or', 'and', 'not', etc. A symbol is usually
represented by a single letter or a string of characters but they
PROGRAMMING INTELLIGENCE 63

must commence with a letter. The Validity of logical state-

ments are best described by 'truth tables'. The connectives
enable us to determine the truth or falsity of the logical
statements.

B=

A and B —

AorB —

Gottfried Leibnitz and George Boole: Pioneers of symbolic logic

Within symbolic logic there are two distinct but interre-

lated branches. These are 'propositional' logic which deals
with the truth or falsehood of propositions made and
'predicate' logic which goes one step further by including the
relationships between individual objects or classes of objects.

Propositional Logic.
A computer operates on the basis of binary signals. The
electronic circuits inside a computer consists of components
which are capable of assuming one of the two possible stable
states. This basic concept has been used to represent longer
numbers, characters, words and even statements about the
real world in a digital format. Since propositional logic relates
to whether a particular proposition is true or false, it could
also be expressed in a digital format. Therefore, a similarity
exists between the stable states in digital electronic circuits
and propositional logic. Statements such as "It is raining
64 ARTIFICIAL INTELLIGENCE

outside" or "The traffic light is red" can only be true or false

and in a computer can be represented by a 'one' or 'zero'
states respectively.
Propositional logic can be programmed into computers
quite easily. The propositions can be represented by means
of symbols such as letters of the alphabet, just as in algebra.
If the proposition, "It is raining" can be represented by the
letter 'a' then the statement, "It is not raining" can be repre-
sented by the symbol 'not a'. In algebra, we come across
formulae such as, X=Y+Z. Now we know that Y=2 if, say, X=5
and Z=3. Similarly, in propositional logic the truth or falsity
of the proposition 'a' cannot be determined by itself. Other
related facts or propositions which we know to be true or
false, are required before we can judge, whether 'a' is true or
not.
In simple algebra, the connectives used are operators such
as +,-,x and + . Likewise, in propositional logic the connec-

'a' 'not a'

tives used between symbols are 'not', 'or', 'and', 'implies',

'equivalent', etc. The simplest connective is of course the
'not'. If 'a' is true then 'not a' is false. This can be represented
by means of a truth table which is self explanatory.
The truth tables for the other connectives 'or' and 'and' are
shown opposite. Here the symbol 'a' represents "1 like jam"
PROGRAMMING INTELLIGENCE 65

I like jam or butter. I like jam and butter.

a b a o r b a b a and b

and the symbol 'b' represents "I like butter". The truth (T)
and falsity (F) of a statement is represented by "on" and "off"
states of a bulb respectively.
The next connective is a conditional connective also
known as 'implies'. There are two parts to the statement "a

A Chicken implies egg or vice versa

66 ARTIFICIAL INTELLIGENCE

implies b". These are the 'head' which comes before the
symbol "implies" and the 'body' which comes after. The
statement says that if the head is true then the body is also
true. The statement can operate in the opposite direction also.
If the body is true then the head must be true. Suppose we
are out in the countryside for a walk (without an umbrella).
The statement, "IF it rains THEN I will be wet" can be inter-
preted as "IF it rains" implies that "I will be wet". Symbolically
this can be represented as " a implies b". The only way this
statement can be false is when 'It rains' and 'I will not be wet'.
Otherwise this statement is always true. The third statement
is, 'It does not rain' and ' I will be wet' is true because you
can get wet by falling into a pond even when it is not raining.
The fourth statement is, 'It does not rain' and i will not be
wet' and it is also true. This example which defines the "a
implies b" connective can be expressed in a truth table shown
below. The "on" and "off" states of a bulb represent the truth

a b a: - b or a implies b

(T) and falsity (F) of a statement respectively.

PROLOG and Predicate Logic

In order to generalize propositional logic, the concept of
predicate logic was developed. It happens to be a convenient
PROGRAMMING INTELLIGENCE 67

method of representing relationships within a knowledge-

base. Predicate logic is the basis of programming the AI
language, PROLOG, in which knowledge can be easily rep-
resented in the form of production rules.
In predicate logic, the items which are being discussed in
a proposition are taken out of the statement and separated
from those which explain the relationship between them. In
the simple statement, "The boy plays tennis" there are two
nouns, namely the boy and tennis. The relationship between
these two items is through the verb play. The items are known
as 'arguments' and the relationship as the 'predicate'. In
predicate logic, the statements can be expressed in a very
simple format:

Normal statement: 'The boy plays tennis.'

Predicate format: play (boy, tennis).

The relationship or the predicate is written first and then

within brackets are the two arguments separated by a
comma. According to the rules of PROLOG, the words com-
mencing with capital letters are variables and those with
lower case are constants or proper names. Also, it does not
matter whether the facts which are being declared are true or
not in real life. The predicate statement: eat (tiger,grass), is a
valid predicate statement but then it is obviously false in the
real world.
Next, let us collect some facts for our knowledge-base. As
an example these relate to who likes what. Having entered

like like

like dislike

like like
68 ARTIFICIAL INTELLIGENCE

these pieces of information into the computer memory we

can query and ask for information by typing the symbol ?-.
PROLOG will now answer logical questions, such as, ?- like
(ram,sita). In plain English this question is, "Does Ram like
Sita?". The PROLOG program will produce an answer 'yes'.
This literally means that the PROLOG program has searched
all the entries, commencing from the top of the knowledge-
base, working downwards and has found a match and so
gives the result 'yes'.
For the query; ?- like (sita,cakes), the answer is 'no' which
means that the statement cannot be proved from the facts
present in the knowledge-base.
The greatest improvement that predicate logic has over
propositional logic is in the use of variables. These are ex-
pressed by symbols which always commence with a capital
letter. A query could be made, such as; ?- like (X,icecream),
which in plain English is, "Is there any one, call him X, who
likes icecream?" The PROLOG program commences the
search from the top of the knowledge-base and comes to the
first match which is: X = ram. If a semicolon
X = is now pressed on the keyboard, the search
continues for the next person who likes
icecream.
x=
The last 'no' means that there are no more
no facts within the knowledge-base to satisfy
this query and so the search has stopped.

Once we get used to this X=

round about way of asking
questions, we can ask even X=
more complex questions
such as, ?- like (ram, X). This
x=
is asking the question, "What
x=
or whom, call it X, does Ram
like?" PROLOG will prompt- no.
ly reply:
PROGRAMMING INTELLIGENCE 69

Clearly, 'no' here means that from within the knowledge-

base there are no more items which Ram likes!
Another question one could ask is: ?- likes
(X,Y),,dislikes (Y,X), which means that,"Does
anyone called X like anyone else called Y
and this second person Y does not like X?"
The comma between the two parts on the
right hand side indicates the conjunction
'and'. PROLOG will provide the following
reply:
Interestingly, predicate statements can be extended to
combine propositions.

Proposition: "IF Ram likes icecream THEN

Ram buys icecream."

Predicate logic

Step 1: like (ram, icecream).

Step 2: like (ram, icecream) implies
buy (ram, icecream).

,
Conclusion: buy (ram,icecream).

Each part of a proposition that is complete (i.e. cannot be

broken down) is called an 'atom'. A predicate and its argu-
ments constitute an atom. For example, eat (ram,apple) is an
atom. A 'literal' is an atom with or without a 'not' in fronf of
it. Both eat (ram,apple) and not (eat (ram,apple)) are literals.
70 ARTIFICIAL INTELLIGENCE

They are known as positive and negative literals respectively.

Connectives are used with predicates as shown below.

eats ( X

eats ( Y

not (<:ats not X

eats ( or eats X or Y

eats ( eats ( ( Xand Y

eats ( :- eats ( X implies Y

Now, how does PROLOG go about finding answers? Take

the case of two simple statements such as, like (ram,icecream)
and like (sita,icecream). In the same knowledge-base a rule can
be introduced, that is, like (X,Y) :- like (X,icecream), like (Y,
icecream). For asking the question, "Do Ram and Sita like
icecream?", one keys into the computer :

?- like (ram,icecream), like (sita,icecream)

Thus predicates, conjunctions, rules and functions can be

used to manipulate predicate logic and the AI program
PROLOG. The question asked by us had one goal whereas
the question keyed into the computer has two goals. The
program searches the knowledge-base from top to bottom to
see whether the first goal, namely, like (ram,icecream) can be
found or satisfied. The moment it is found, the program
begins a search for the second goal. With these two goals
being satisfied the main goal of the original question is
answered 'yes'. If a goal fails then the entire conjecture has
failed and the answer is 'no'. This technique is known as
'backtracking' and is a powerful one in PROLOG.
PROGRAMMING INTELLIGENCE 71

Reaching Inference
The first thing to do on receiving an input is to convert it from
English into an equivalent predicate logic. Putting it in the
correct format is very important. The choice of the predicate,
the function and the constants have to be carefully made in
AI applications.
If the input sentence is a question then the answer is
somewhere in the knowledge-base but it may not be explicit.
We may have to infer the answer from the facts. This is made
clear in the following example:

Fact 1: Ram has fever.

Fact 2: Infection brings on fever.
Conclusion: Therefore Ram has an infection.

Inference and deduction are the strong points of predicate

logic. The initial facts which lead to the deduction are known
as 'axioms'. In algebra and geometry we are used to axioms
which are explicitly stated statements. However, in AI the
information received by the sensory organs constitutes the
axioms for the discussion which is current at that time. There-
fore, unlike in mathematics, in AI our set of axioms are not
stable.
Deduction is logical reasoning in which conclusions are
drawn from knowledge or facts by logic. In order to deduce
new facts from the axioms we make use of the rules for
inference. The most important rule is the 'modus ponens'
which is as follows:

Given: If A then B
Given: A is true
Deduction: B is true

The above line of argument can even be expressed as:

A, A —> B; therefore B
72 ARTIFICIAL INTELLIGENCE

We are reasonably familiar with this deduction procedure

as we use it in our every day life.
Another rule with which we are familiar is that of 'univer-
sal instantiation'. We use it quite often without being aware
of it. For example, if something is true for Delhi then it must
be true for Karol Bagh.
Yet another rule for inference is called 'abduction'. It is the
process of reasoning backwards from a true conclusion to the
premises that may have caused the conclusion. Abduction
g e n e r a t e s e x p l a n a t i o n s as the f o l l o w i n g e x a m p l e
demonstrates.

Given: If A then B
Given: B is true
Inference: A is true

This is not a legal inference as one could easily arrive at a

wrong A (in the above example). This is demonstrated in an
extreme example below:

If X has cancer of the lungs then he will be short of breath.

Y is known to be short of breath.
Therefore Y is suffering from cancer.

What has happened is that given the rule and the fact that
Y is suffering from short breath we have jumped to the
conclusion that Y is suffering from cancer. The fact that Y is
short of breath has been explained away by assuming lung
cancer. This conclusion is a drastic one and hopefully a wrong
one (at least for the sake of the doctor's reputation). Though
abduction could lead to wrong conclusions, the method is
widely followed in many fields including medicine and
economics. Abduction is the process of providing an explana-
tion. Experienced medical doctors use this method with judi-
cious caution and are able to make satisfactory diagnosis.
Another kind of inference is 'induction'. It is the process
of generalizing from a few specific cases. Suppose we meet a
PROGRAMMING INTELLIGENCE 73

boy from a particular school and he plays hockey. We meet

another boy from the same school and he too plays hockey.
Then we could infer that all boys of that school play hockey.
This process of reasoning is known as induction and it is
accepted just as we do abduction, with a certain amount of
restraint and control.

In the case of 'intuition' an inference is arrived at by

unconsciously recognizing a pattern. It is also known as the
sixth sense. It is usually correct if one is very alert and aware
of what is happening around us. If someone chokes while
eating a banana, we immediately slap him on the back,
having witnessed a similar situation long ago.

Inference by 'analogy' is arriving at a conclusion based on

a similarity to another situation. This is a common procedure
one adopts in providing explanations to phenomena in scien-
tific subjects. A similarity is drawn between a new
phenomenon and another which already has an explanation.
Based on this analogy we suggest or infer an explanation for
the new phenomenon.

LISP
LISP or List Processing was a programming language created
by John McCARTHY. It is a functional language that manipu-
lates both symbols as well as logical statements. It is a simple
language and has been extensively used, particularly in the
U.S.A. for developing expert systems. The LISP rules for
manipulation of information are easy to understand and to
learn.

LISP was developed mainly to process lists. To put items

in the form of a list is an important part of its structure.

For example, in LISP the list of all possible road transport

vehicles one can think of can be written in a line with brackets
around them. A gap is left between neighbouring elements.
74 ARTIFICIAL INTELLIGENCE

(bicycle car truck tram van scooter)

There are three ways ot assigning meanings to symbols in

LISP. First, a symbol can represent a fact, an object, a piece of
data, or it can represent an action whose result could once
again be data or knowledge. Second, a symbol can be a
variable which has a value or a meaning. The analogy here is
that symbols which represent variables are similar to the
unknowns x and y we use in algebra where the values are
determined after applying the rules of algebra. The same
variable has different values or meanings in different
problems. Finally, a symbol can represent functions. The
above list can be rewritten as: (al al a3 a4 a5 a6). Each element
of a list can be either an atom or another list. A complex list
would look like: (al (a2 a3) a4). This is a list consisting of al
which happens to be an atom, (al a3) is a list of two atoms
and a4 is another atom. The above list has three components.
On the other hand, ((al) (a2 a3) a4) is a list where al is no
longer an atom but a list. This situation of having lists within
lists is known as 'deep-nesting' and this makes it difficult to
estimate how many elements are actually contained in a list.

Both programs as well as data are represented as lists. So

what actually happens in LISP is that the programs manipu-
late lists. Also, since all programs are in the form of lists they
can manipulate the programs. This is a very useful criterion
in Al as one of its characteristics is to design programs which
learn from experience. (Learning here means modifying the
khowledge-base and the reasoning algorithms to suit new
information,}
PROGRAMMING INTELLIGENCE 75

However, since both programs and data have similar list

structures, every expression has to be evaluated. In LISP the
format of a list would be:
* The first symbol inside the left hand bracket is the name
of the function. In case it has not been defined or given a name
then the program indicates a syntax error.
* All numbers are considered as constants and retain their
values.
* All other symbols or atoms are considered variables and
are to be evaluated and their values determined.
There are a large number of functions used in LISP. Some
of the commonly used LISP functions are quote, setq, cdr, car,
cons, append, predicate, cond and defun.
Quote: This function takes an argument (which may be a
symbol, a number or a list) and returns it unaltered. In
literature it is expressed in two alternate forms as illustrated
below:

(Quote Bicycle) or (Bicycle) returns Bicycle

(Quote 1 2 3) or (1 2 3) returns 1 2 3

__
This function might appear to be very elementary but has
its uses. Its real importance is that it prevents any evaluation
of the argument from taking place. Whatever follows the
'quote' is returned as it is verbatim.
Setq: This is another very common system function. Its
effect is to bind the second symbol and the third in the list.
The function has two arguments. The first element is the
function setq. The second is a symbol variable and the third

(setq letters' (al b l cl)) returns (al b l cl)

(setq one 1) returns 1
76 ARTIFICIAL INTELLIGENCE

can be any symbolic expression. Setq returns the value of the

second argument to the variable.
The 'quote' mark in the first example ensures that the value
of the variable, 'letter', is the list (al b l cl). In the second
example the third element is a constant 1 and LISP returns
the value 1 to the variable 'one'. If one thinks carefully the
variables 'letters' and 'one' in the example above should also
have a quote in front of them to indicate no operation or
evaluation. However, convention in LISP has it that the q after
the word set (in setq). represents a quote and as such there is
no need to mention it specifically.

Similarly, in example: (setq birds '(crow sparrow ostrich)), the

variable 'birds' is a list. So if the word 'birds' is keyed into the
computer it returns the list as: (crow sparrow ostrich).

Cdr: This function is used to remove elements from a list.

The function returns a list minus its first element or atom.

(cdr' (al b l cl)) returns ( b l cl)

Here cdr has been applied to the list (al bl cl). LISP returns
the same list with the first element al removed. The value of
the list has been changed. It has been made one element
shorter and therefore, is different from the original list.
Considering the same example: (setq birds (cdr birds)), LISP
starts with the innermost part of the statement, i.e. (cdr birds).
Since there is no quote, the cdr operates on the variable 'birds'
and so it returns the list corresponding to 'birds' but with
crow eliminated. The new 'birds' list therefore, becomes
(sparrow ostrich)

Car: This function consists of an argument and a non-

empty list. It returns as its value the first element from the
list.

(car' (al bl cl)) returns al

(car' ((al) (bl) (cl))) returns (al)
PROGRAMMING INTELLIGENCE 77

The usefulness of the quote is seen here in the example. It

has prevented the argument of car from being evaluated. If
the quote were removed then LISP would try to evaluate (al
bl cl). Since this is a list it expects al to be the name of a
function. If al is not a function then an error in syntax is
indicated by the computer.
Cons. LISP functions are useful in adding new elements or
lists to existing lists. Cons enables an element to be added. It
takes two arguments, the first may be a symbol, a list or a
number, but the second must be a list. Cons inserts its first
argument as the new first element of its second argument and
returns the new list.

(cons 'a'(b c)) returns (a b c)

(cons '(a)'((b) (c))) returns ((a) (b) (c))

Cons permits one to construct new predicate function from

existing ones.
Again going back to the previous example; (setq birds (cons
'parrot birds)), cons returns as the new definition of 'birds' as:
(parrot crow sparrow ostrich).
Append: This function joins a new list on to the end of an
existing list. Here, (setq birds (append birds' (eagle myna)))
would be returned as a new list, i.e., (parrot crow sparrow
ostrich eagle myna)
Predicate: LISP has a number of predicate functions. Some
of them are atom, zerop and greaterp. These return a value
true (T) or false (F). The function atom takes one argument of
any type and returns T if the argument is a symbol or a
number and a F if the argument is a list.

(ATOM '4) returns T

(ATOM 'Al) T
(ATOM '(al b l cl)) F

The function zerop is a simple function used with a numeric

argument. It returns a T if the number is 0 and F if not.
78 ARTIFICIAL INTELLIGENCE

(ZEROP 14) returns F

(ZEROP 0) T

The function greaterp takes a list of numbers and returns T

if the numbers are in descending order, a F otherwise.

(GREATERP '(10 9 8 7)) returns T

(GREATERP '(9 3 6 5)) F

Cond: LISP enables one to write a conditional clause with

the help of the function cond.

(cond ((condition 1) (actionl action2 action3)

(condition 2) (action4 action5 action6)
(condition 3) (action7 action8 action9)))

LISP evaluates the first condition and if it is satisfied (T)

then action 1, action 2, action 3 are executed. If the first
condition turns out to be false (F) then the second condition
is evaluated and so on. The last condition is designed so as
to be always true and the corresponding actions are carried
out. The conditions and the actions can be any symbolic
expressions.
Defun: Defun is a LISP function which can be used to create
new functions.
(DEFUN function_name ((argument_list) (function body))) j

This returns as its value the name of the function being

defined.
This is only a brief introduction to LISP but it shows how
useful it can be in Al. Interestingly, in writing LISP programs
no distinction is made between the structure of data and
programs. Moreover, this language is capable of manipulat-
ing symbols and structures rather than just numerical cal-
culations.
 ver since the world's first

E general purpose computer

was put into operation, an
anxiety was expressed in many
q u a r t e r s , that one day such
machines would become far too
complex and difficult for a casual
user to operate. To begin with, he
would have to make special ef-
forts to familiarize himself with
the intricacies of the new com-
puter. This would include acquir-
ing some knowledge regarding
the details of its hardware, learn-
ing some new programming lan-
User- guages or commands used to
communicate with the computer

Friendly and above all matching the prob-

lem solving capabilities of the
machines with his actual require-
Systems ment. This trend which continued
for over three decades, made it
necessary for users to take pains
in matching their skills with those
of the computers. However, with
the advent of AI this trend is being
reversed.

With the popularization of

knowledge based systems, the
computer designers began to
think in terms of introducing new
interfaces between machines and
their users. These would enable
the two to interact on a more
familiar basis. The systems which
80 ARTIFICIAL INTELLIGENCE

emerged were labelled "user-friendly" and they are currently

proving to be very popular.

Natural Languages
Advances in knowledge based techniques have led to many
c o n c e p t u a l changes in computers. These have been
redesigned so that even 'computer illiterate' users can use
such advanced machines. This in turn has stimulated
academic research in subjects such as natural language
processing, speech recognition, synthetic speech generation,
vision systems, etc.

With knowledge-base being added, the status of a com-

puter has risen to become one of a 'partner' with its user. This

Man-machine interaction
USER-FRIENDLY SYSTEMS 81

is not surprising, since the knowledge-base is basically

derived from human experts. Both the human user and the
computer can now initiate interaction with one another by
asking questions, setting goals or generally by responding to
each other's queries. Such situations are generally known as
possessing mixed initiative. It is, therefore, even more impor-
tant that the dialogue between the two should be in a lan-
guage which the user is already familiar with. The problem
now reduces to one of making the computers familiar with
natural languages! They have to learn how to analyze senten-
ces and to extract meaning from them.

A popular way of introducing instructions and data into a

computer has been through keyboards. At first sight the
common computer languages, COBOL and BASIC appear
similar to spoken English. However, the user has to learn
them and become familiar with their rules and restrictions.
On the other hand, a natural language, say English, is learnt
at school. When we hear or read a simple sentence in English
we understand it straight away as we are familiar with its
vocabulary and rules of grammar. If it happens to be a long
or a complicated sentence with difficult words thrown in,
then we mentally analyze the sentence into its different gram-
matical clauses and then try to understand the message part
by part. 'Natural Language Understanding'is a study of the
different processes by which a computer tries to comprehend
the instructions given in ordinary English.

The core of such a 'natural language processor' is the

'Parser'. It reads each sentence and then proceeds to analyze
it. This process can be divided into three tasks viz. dividing
the signal into its acoustic-phonetic, morphological-syntactic
and semantic-pragmatic aspects. The first part comes into
operation only if the input consists of spoken words. It con-
verts the utterances or the sound waves into words. The
second establishes the syntactic form (grammatical arrange-
ment) of the sentence. The third tries to extract the meaning
from these analyzed patterns. It takes into account the com-
82 ARTIFICIAL INTELLIGENCE

C o m p u t e r programs are formal (syntactic); Human m i n d s

have mental associations (semantic)

mon usage of words in a language. This entire process is

known as 'syntactic analysis' or 'parsing' the sentence. Sub-
sequently, it can be represented in a tree type of a format
based on the parsing rules which govern the language. Ex-
tracting the meaning is known as 'semantic analysis'. The
syntactic analysis or the parse indicates whether a word is a
noun, a verb, an adjective, a determiner, etc. The meaning
becomes far easier to extract or derive after parsing. This
process can be extended to longer and complex sentences as
well.

In our normal conversation, we often use ungrammatical

structures as the following dialogue would reveal:
USER-FRIENDLY SYSTEMS 83

Doctor. When did you experience pain this week?

Patient: On Tuesday and Thursday.
Doctor. Last week ?
Patient: Wednesday.

The doctor's second question is not grammatically correct but

such expressions are normally accepted without much fuss.
No one bothers about grammar in such situations. The lis-
tener responds as though he has heard a perfectly correct
sentence. Humans accept such mistakes or errors when they
already have a gist of the discussion based on the earlier part
of the conversation. If the doctor's second question had come
after a delay of five minutes or so, the patient may not have
been able to understand the question at all. One of the most
difficult aspects of designing a natural language processor is
the complexity and flexibility of the human languages. What
happens when a machine, which has been trained to accept
statements in accordance with the correct rules and format,
has to interpret the above conversation? It may simply not
understand it. Many of the first generation natural language
processors could understand just a few types of sentences of
the natural language. The machine would understand a
spoken sentence only if it was grammatically correct. If the
rules of grammar were not constrained, then, the procedure
for analysis or parsing would become extremely lengthy.

Parsing had been investigated by linguists quite inde-

pendently of, and prior to the AI scientists. They developed
tree diagrams for parsing sentences. For example, in the
sentence: Ram ate a banana; 'Ram' constitutes a proper noun
and is the subject and 'ate a banana' constitutes the predicate.
The linguists would parse the sentence with the help of a tree
diagram.
Many AI scientists believe that the parse trees are impor-
tant intermediaries between a sentence and the mental (or
internal) representation of the concept it is supposed to cre-
ate. If in the above sentence, the determiner 'a' were to be
84 ARTIFICIAL INTELLIGENCE

RAM ATE A BANANA

Subject Predicate
Noun Verb Determiner Noun

RAM ATE THE BANANA

. Subject
Predicate
Noun Verb Determiner Noun

Parsing sentences
USER-FRIENDLY SYSTEMS 85

substituted by 'the', then the meaning or the impression

created by the sentence is different. T h e banana' makes it a
very specific banana where as 'a banana' refers to any banana
lying in the basket.
Similar parse trees could be prepared for multi-clause
sentences as well. These tree structures, in turn, have led to
the development of networks consisting of nodes or states
with arcs which tell us what is the next parse (noun, verb,
etc.) The transition networks operate by working through a
sentence from left to right. For each word there are only a
limited number of options of words which can grammatically
follow it. For example, consider the sentence, The small fat man
ate a cake. It consists of two parts viz. the noun phrase 'the
small fat man' and the verb phrase 'ate a cake'. The adjective
'small' can be followed by another adjective such as 'faf or
by a noun like 'man' but cannot be followed by a verb. Rules
for each of these phrase structures can be noted and used in
analyzing a sentence.
Each phrase structure can be written as a route from the
beginning to the end of a grammatical item. The network
with possible routes from a given starting point to a com-
pleted noun or a verb phrase is described as a transition
network.

Automatic speech recognition

Automatic speech recognition has fascinated engineers,
scientists and linguists alike. It has kept them interested for
the past half a century, first in trying to understand the
production and perception of speech sounds in human
beings and then to simulate the process using electronic
systems. They have designed machines called sonographs
which can analyze speech signals. A careful study of the
machine's output called a 'sonogram', helps to recognize the
spoken words and even extract their meaning. This con-
tinues to be an exciting challenge and vast sums of money are
86 ARTIFICIAL INTELLIGENCE

being spent on its research.

Speaking into machines
would appear to be the
most natural and efficient
means of m a n - m a c h i n e
communication.
S p e e c h is a n a t u r a l
means for human com-
munication particularly, in
situations where both the
eyes as well as the hands
are occupied or busy (eg. a
pilot of an a e r o p l a n e ) .
Automatic speech recogni-
tion systems are electronic
devices which receive the
spoken words from human
speakers and produce an
A sonograph used to output which consists of
analyze speech coded digital signals ready
to be fed into an applica-
tion unit. These signals correspond to what the Automatic
Speech Recognizer (ASR) has recognized of the speech mes-
sage. The application unit takes appropriate action on receiv-
ing these signals from the ASR. They can be used to initiate
a specific machine activity. For example, a robot may be given
oral instructions to walk ahead, to turn and so on. In the
commercially available ASRs, no comprehension of the
speech sounds as such is undertaken. The output signals
strictly correspond to the recognized words.

Human speech is characterized by three types of sounds

viz. voiced sounds where the acoustic source is the vibrating
vocal cords in the larynx, fricative sounds (like s, or sh) which
are essentially white noise. They are generated by forcing air
through narrow constrictions in the mouth. Plosive sounds
are produced by first allowing an air pressure to be build up
USER-FRIENDLY SYSTEMS 87

behind a closure and then released (as in /p/ or /b/ or /k/

etc.).
Linguistic studies of speech sounds suggests that there are
basic discrete sound segments in any language and these are

S o n o g r a m of phonemes
strung together to form words while speaking. These seg-
ments are known as 'phonemes'. Most spoken languages
possess, on an average, about 40 phonemes. Each one of these
phonemes can be characterized by a set of unique properties.

However, recognition of phonemes by computer systems

is yet not a successful proposition. The contextual influence
of continuous speech cm individual phonemes are so

S o n o g r a m s of the word 'zero' as spoken by four speakers

88 ARTIFICIAL INTELLIGENCE

considerable that it is difficult to get machines to recognize

them with certainty. There are enormous variations in the
styles of speaking by men and women. A word spoken by the
same speaker on different occasions has minor variations.
The same word spoken by different speakers also has varia-
tions depending on individual accent or dialect. In the former
case, the variations arise on account of context and position
of the phoneme within a word. In the latter case, it is on
account of factors such as the size of the vocal tract and the
individualistic style of speaking. The development of a com-
plete speaker independent speech recognition system will
require an even more comprehensive understanding of the
sources of variability in the speech signals.

Isolated word recognizers with a pause of at least 300

milliseconds between successive words are available com-

Electronic band-pass filters used in speech analysis;

A VLSI chip equivalent to the above circuit (inset)
USER-FRIENDLY SYSTEMS 89

mercially. These machines typically deal with the recognition

of a srrjall set of isolated words by a known speaker. Most
systems produce a running spectrum with the help of a bank
of band-pass filters. The spectra of the words are examined
every 20 milliseconds. The energy content in the different
frequency bands are compared with those in corresponding
bands already available and stored in the memory. The
template which has the best match with the input is selected
as the recognized word. Such machines operate very well
when the vocabulary is between 20 and 200 acoustically
d i s t i n c t w o r d s . A t e c h n i q u e k n o w n as ' d y n a m i c
programming' is extensively used in the isolated word recog-

VOCABULARY
PATTERNS 7

'COMPARE

RESPONSE
INPUT SPEECH
DECISION
FEATURES

COMPUTER

P a t t e r n r e c o g n i t i o n a p p r o a c h to isolated w o r d r e c o g n i t i o n
nition systems. It is a path finding algorithm which attempts
to seek the best alignment between two similar patterns. Time
alignment is necessary as the duration of the spoken words,
even if by the same speaker, can vary when spoken on
different occasions. If the system caters to larger vocabularies
then not only the time required to produce a match is propor-
90 ARTIFICIAL INTELLIGENCE

tionately longer but also the training period for each new
speaker would be time-consuming.
Continuous speech signals can be reduced to a sequence
of isolated words by intentionally introducing pauses be-
tween words. In this case, the individual words will be
recognized as such. This method is adequate to get the
spoken word to appear on the screen or where instructions
consisting of single word commands are to be recognized.
The speaker has to produce samples of the spoken words
before he can use the system. This is known as the 'training
phase'. Each new user has to have templates for his speech
prepared in advance before he can use the machine. How-
ever, speaker independent systems require no such in-
dividual training schemes. Such systems usually have
templates which have been averaged out over many speakers
and stored in the machine memory. The templates in such
speaker-independent word recognizers consist of- informa-
tion based on acoustic features of the words rather than on
their detailed spectra.
Continuous Speech Recognition (CSR) is much wider in
scope. Natural speech contains many inconsistencies. Coar-
ticulation (i.e. effect of one word on the next) and rhythms
make continuous speech recognition a difficult task. Such
recognizers make use of our linguistic knowledge (both syn-
tactic and semantic ) to characterize the range of acceptable
sentences. They usually consist of large vocabularies, both
in their phonetic and syllabic representations. Extensive re-
search has already been undertaken in this area of technol-
ogy, with CSR being the ultimate goal. Systems claiming 1%
error rate have already been produced. However, inde-
pendent evaluations mention error rates between 12% and
0.5% which depends on the complexity and cost of the sys-
tems.
While speech signals carry linguistic information regard-
ing the message to be conveyed, they also possess extra-
linguistic information about aspects such as the speaker's
USER-FRIENDLY SYSTEMS 91

identity, dialect, his psychological and physiological states

and also the prevailing environmental conditions such as
noise, room acoustics, etc. In order to develop high grade
speech recognition systems one has to learn how to extract
the message-bearing components from the signal while dis-
carding the rest. Understandably, considerable fundamental
research in speech science is needed before artificial speech
recognition systems can approach anywhere near human
performance.
 r t i f i c i a l I n t e l l i g e n c e is

A barely 40 years old. How-

ever, these four decades
have witnessed phenomenal ad-
vances in this field which tempts
one to say that machines would
soon perhaps be able to think and
p e r f o r m intellectual tasks as
human beings do. Ideally, the ob-
jective of Al is to design machines
or non-biological systems with
which man should be able to inter-
act exactly as he does with another
human being. An interesting ex-
ample from ancient Greece is the
Oracle of Delphi. It was a statue
which listened to the prayers of

An pilgrims and often spoke back,

providing the listener with some
wholesome advice! In actual fact,
Overview there was a hollow duct leading
from the mouth to its back where
a priestess sat. She remained out
of sight, heard, the prayers and
a n s w e r e d t h r o u g h this tube,
astonishing the visitor! Today,
thanks to Al, we have machines
w h i c h s p e a k b a c k to t h e i r
operators. You walk into some
banks in Japan or U.S.A. and
speak your account number into a
machine and it replies in a syn-
thetic speech, the exact amount of
money you have in your account.
Also, electro-mechanical robots
are now available which perform
difficult tasks such as welding.
AN OVERVIEW 93

The modern 'Oracle of Delphi'

94 ARTIFICIAL INTELLIGENCE

Such robots are currently replacing human labour in the

Japanese automobile industry.

Intense research on how the human mind functions is

essential before one asks whether its electronic equivalent can
be constructed. Though one may not be able to describe the
mind in terms of a set of rules and frames, yet there is a
possibility that the current research on parallel processing
and neural networks might lead AI scientists to simulate the
fabric and process of the mind.

Here one comes across a question as to what constitutes

knowledge. This is a very elusive concept and even more
difficult to quantify. Nevertheless, this aspect has been the
single greatest preoccupation of AI research. At present, its

KNOWLEDGE

INFORMATION

DATA

Hierarchy of k n o w l e d g e
AN O V E R V I E W 95

Human brain

Year 2500?
Human being

First commercial expert system

First expert system

Advance in AI

General problem solving

Birth of AI

-i —\ 1
1940 1960 1980 1990

Year

T r e n d s i n A I w i t h a p e e p i n t o its f u t u r e
96 ARTIFICIAL INTELLIGENCE

scope is confined to the collection of data regarding objects,

their relationships and events within restricted domains.
Expert systems, the best known manifestations of AI, have
today gained immense credibility and acceptance in many
professional fields. Several complex programs are commer-
cially available to help analyze data fed into the machine,
diagnose diseases and perform various tasks which have so
far required a high level of human expertise. Further attempts
are also being made to widen the scope of each such system
to cover wider domains.
AI is rapidly making an impact on our lives. New
machines taking advantage of AI have become irreplaceable
in commerce, industry and research. What was science fiction
a decade ago is fast turning into reality. Nonetheless, man-
made machines are yet not capable of mimicing in entirety
the sophistication and subtlety of the human mind. So far all
the knowledge-base and data which are fed into the machines
have been carefully planned in advance and various exigen-
cies taken into account. In a sharp contrast, a human being is
born into the world with an ability to learn.
Notwithstanding the inherent limitations, it is certain that
AI would bring a sea-change in information and knowledge
processing the world over in the coming decades.
Glossary
Band-pass filters: An arrangement of electronic components
which enable only signals of a specific frequency to pass
through them. At the same time, they weaken the rest of the
unwanted signals.
Binary: A number system with abase of two. It involves only
two digits : 1 and 0. In a binary code, any data is encoded
through the use of these two digits called 'bits'.
Chip: Integration of thousands of electronic devices and
components on a single, rectangular slice of silicon. It is the
key to the microelectronic revolution in computers.
Electronic toggle: A circuit having elements which are
capable of assuming either one of the two stable states, i.e.
ON or OFF states, at a given time. This device is synonymous
with a 'flip flop' circuit.
Phonemes: A set of basic sound signals used in a particular
spoken language. A change in a phoneme may alter the
meaning of the word.
Program: A set of coded instructions that direct a computer
to follow a given logical sequence to perform a particular task.
PROLOG: Short for Programming in Logic. A programming
language used in the development of expert systems. It can
be easily represented in the form of production rules. Expert
systems based on PROLOG are currently being developed in
Japan and England.
Relay: An electromagnetic switching device having multiple
electrical contacts. It is energized by passing electrical current
through its coils. Relays were used in pre-electronic com-
puters, e.g. Markl.
Software: The combination of a set of programs and
documentation that enables the computer to perform a set of
defined tasks.
98 ARTIFICIAL INTELLIGENCE

Syntax: A set of rules which enables us to correctly arrange

words when we either write or speak a sentence. It is also
known as 'grammar' of a language.
Template: A set of features which together help in correctly
identifying a particular sensory signal
Thermionic valves: A system of electrodes arranged in an
evacuated glass or a metal envelope. On heating, the negative
electrode called 'cathode' emits electrons which are attracted
by the positive electrode, 'anode'. Most valves also contain
a perforated grid interposed between the two electrodes. The
grid controls the flow of current through the valve.
 VISTAS IN BIOTECHNOLOGY
Biotechnology has taken rapid strides in
recent years and has emerged as one of the
most promising branches of science Com-
mitted to rooting out hunger by the creation
of high-yielding, disease-resistant crops and
pledged to resurrect the earth's verdant
green cover, plant biotechnology holds the
key to the world of tomorrow. Digging out
closely guarded secrets from the inner reces-
ses of the vault of life, and unleashing and
channelizing nature's power of self-renewal,
TEST
this particular brand of biotechnology has vir-
tually rewritten the story of botanical genesis
and prospects.
TUBE
This attractive, profusely illustrated book writ-
ten in a popular language highlights human
creativity and imagination in coaxing life to
FORESTS
spring from non conventional sources and A.F.MASCARENHAS
takes the reader on a trip down the woods RAJANI NADGAUDA
of tomorrow. Pages : 72
Lovely, dark and deep forests fashioned in Price : Rs. 15.00
test tubes by scientists who laboured
long because they had promises
to keep.

FORTHCOMING TITLES
Orders should be accompanied by Money
ABC of Biotechnology Order or Demand Draft made payable to
Vaccines for Tommorrow "Publications & Information Directorate
New Delhi and sent to:
Biology of Behaviour
Omnipresent HorrrpjgS Sales & Distribution Officer
[Bublications & Information Directorate, (CSIR)
^ K . S . K r i s h n a n Marg, New Delhi 110 012
 PROMISING in concept, scope and
utility, the "fifth" generation computers are at the .
threshold of ushering in an era of machines that could
simulate the human mind. In keeping with its interdis-
ciplinary nature, the instilling of intelligence into com-
puters holds avast potential in a plethora of fields.
Even in its infancy it has revolutionized database
management, information retrieval, programming lan-
guages and system design. Expert systems are fast
gaining laurels in the wide domains of medicine and in-
dustry. Equally fascinating is the insight into the future
role of increasingly intelligent robots saving precious
manpower in all walks of life.
This attractive and lavishly illustrated book, targeted
for the non-specialist, explicitly unveils the many facets
of artificial intelligence research. The ability to surpass
human thinking is apparently not far away, thanks to
computer technology that is poised for creating
machines possessing artificial intelligence.
About the Author

K.D.Pavate (b.1930), after graduating with Physics and Mathe-

matics from Fergusson College, Poona, studied Physics at the Caven-
dish Laboratory, University of Cambridge. For the next three years he
worked with the Metropolitan - Vickers Electrical Co. Ltd., Manchester.
On returning to India, he joined the Central Electronics Engineering Re-
search Institute (CEERI), Pilani.
In 1976, he moved over to Delhi as the head of the newly formed
CEERI'Centre. Here he established R & D groups to work on various
user-oriented aspects of electronics. These included development of
audiovisual systems, electronic communication and animation systems
for museums, electronic systems to assist the National Literacy Mis-
sion, development of small capacity EPABX systems, secure speech
communication, speech analysis and speech synthesis technologies,
among others.
Pavate's interest in recent years has shifted to science communica-
tion and in writing articles on science and technology. Artificial Intel-
ligence is his first popular science book.

ISBN : 8 1 - 7 2 3 6 - 0 3 5 - 5