WHAT IS

ARTIFICIAL INTELLIGENCE?

What Is Artificial Intelligence?

ABOUT THIS BOOK

Today, as artificial intelligence (AI) technology begins to emerge from
the laboratory and to venture into the commercial marketplace, it is
being greeted with a strange mixture of welcome and anxiety, an
uneasy blend of anticipation and trepidation. Unfortunately, it is not
being met by a great deal of understanding, a situation that this book
may help to remedy.
Reading this book will not, of course, make you an " AI
expert," any more than you could become an expert in physics by
reading a single physics book. The aim of this book is to familiarize you
with various aspects of artificial intelligence, including:
The fundamental concepts of AI;
Where AI has been, its current status, and where it is going;
The different technologies that comprise AI; and
The potential impact of AI on your life.
The book also is designed to whet your appetite for more
knowledge about AI. Reading this book will provide you with a solid
foundation if you wish to delve deeper into specific features of the
exciting world of artificial intelligence.

ABOUT THIS CHAPTER

Al rewarchors do not It would be easier to impart a clear understanding of AI if a concise
agree on a single defini- and generally-accepted definition were available; unfortunately,
tion for artificial intelli-
there is wide disagreement within the field itself as to exactly what
gence; therefore, several
definitions are presented
constitutes artificial intelligence. It is not unusual to find scientists
in this chapter. who consider themselves to be working in the field of AI who are not
considered to be doing so by some of their colleagues. Conversely,
there are scientists working in areas that are traditionally"
'

considered to be part of AI who refuse to apply that label to their

work.
Nonetheless, it is helpful to examine some of the definitions of
artificial intelligence that have been suggested. By considering what
various people in the field have .said about AI, it is possible to find

similarities that can help you grasp the essence of what artificial
intelligence is all about.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE''

This chapter explores the iniphcations of several useful

definitions of artificial intelligence and discusses some of he current
t

technologies which are generally considered to comprise Al.

A DEFINITION OF ARTIFICIAL INTELLIGENCE

The following definition of Al may seem vague at first glance:
"Artificial intelligence is the study of how to
make computers do things at which, at the
moment, people are better.'"
Elaine Rich, Artificial Intelligence

However, a further exploration of Rich's definition shows it to be

quite thought-provoking.

What Computers Can Do Better Than People

Some of the functions that Implicit in Rich's definitionthe idea that there are indeed things that
is
computers do better than computers do better than people, which probably does not surprise
people are performing
you.
numerical calculations,
storing information, and
Here are discussions of some of the things that machines do
repeating operations. better than people.

Numerical Computation
One small, special-purpose computer with which you're undoubtedly
familiar the hand-held calculator {Figure 1-1). If you were asked to
is

multiply 6218 by 93.37, would you rather perform the computation in

your head or use a calculator? Which technique would be faster? More
accurate?
Figure 1-1.
Tl-66 Programmable
Calculator

'
Elaine Rich, /lrt(/V(;o//Hte//(:a«''«f (New York: Mctiraw Hill, 1SW:J), p. 1.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE?

Obviously, even a tiny calculator can outperform a human

when it comes to mathematical computations. (Computers were, in
fact,invented specifically for this purpose.) Larger computers are so
fastand accurate thai they can perform calculations that literally
would be impossible for one person to complete in a lifetime using any
other means.

Information Storajic
likely that many of the bills you receive each month are processed
It is

by a computer. A typical business computer might contain, for

example, the names and addresses of [Link] of customers,
accompanied by complete records of their financial transactions
(Figure 1-2).

you didn't have a computer, do you suppose that it would be

If

possible for you to remember aW of that data? Of course not. Yet a

computer can "remember" volimiinous amounts of information and
'

recall any of it at your command.

Repetitive Operations
We frequently instruct computers to perform the same tasks, day in
and day and over again. Fortunately for us, computers don't
out, over
get bored. you use a computer to print 1000 copies of a customer
If

report, for example, the quality of the last report is as good as that of
the first. If you tried to copy those reports manually, the quality
probably would start to diminish quickly after the first few copies.
Figure 1-2.
Sample Records Stored
^
f
In a Computer I

NAME ADDRESS AMOUNT OWED

SANDY ADAMS 6682 FIELD AVE , ARNOLD, MD 21012

JOE ALLEN 1225 CONNECTICUT AVE . AUSTIN. TX 78703
ELIZABETH BROWN 47 LIBERTY LANE, WASHINGTON DC 20008
RICHARD COX 10362 MAPLE AVE., WICHITA, KS 67212
GEORGE DIXON 8123 PECAN LANE, DALLAS, TX 75230
DOROTHY EVANS 555 TRAILS END ST AUSTIN TX 78703
GENE FRANKLIN 3 HESTGROVE DR , ROWLAND HTS , CA 91748
JENNIFER GREEN 7119 WAVERLY AVE , SAN ANTONIO. TX 78213
 WHAT IS
ARTIFICIAL INTELLIGENCE?

('()in])iitors Are ".Just Machini's"

The kinds of things that computers do better than people, such as
those just discussed, are activities that we call mechatiicnl—
"mindless" activities that are obviously capable of being performed
by machinery. Ever since the industrial revolution, humans have
accepted the fact that machines are often superior in i)eiforming
many purely mechanical physical activities.
Similarly, in the last 40 years, we've come to accept the fact
that computers— computing machines— can outperform humans in
purely mechanical tnetUal activities. That fact has not damaged our
collective self-image, however, because we remain secure in the belief
that there are also many things that people can do better than
computers.
What People Can Do Better Than Computers
People have traditionally outperformed computers in activities that
involve intelligence.
We do much more than just process information; we
understand We "make sense" out of what we see and hear; we
it.

come up with new ideas seemingly out of thin air; we use common
sense to make our way through a world that sometimes .seems highly
illogical.

One definition of AI is If people are more intelligent than computers and if, as in
based on the goal of Rich's definition, AI tries to improve the performance of computers in
making computers exhibit activities that people do better, then the goal of AI is to make
intelligent behavior.
computers more intelligent. This concept forms the basis of a second
defirtitionof AI:

"ArtificiaJ intelligence is the part of computer

science concerned with designing intelligent
computer systems, that is, systems that exhibit
the characteristics we associate with intelligence
in human behavior."^

Avron Barr and Edward A. Feigenbaum, The Handbook of

Artificial Intelligence

Intelligent Behavior
The goal of AI, according to Barr and Feigenbaum, is to develop

intelligent important to notice that they define an

computers. It is

intelligent computer as one that emulates intelligent behavior in

humans.
But exactly what do we consider to be intelligent behavior in
humans? In other words, what is intelligence?

^ Avron Barr and Edward A. Feigenbaum, The Handtxxjk of Artificial Intelligence, 3

vols. (Los Altos, CA: William Kaufman, 1981-82), 1:.3.

UNDERSTANDING AFITIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE''

That question is not nearly as easy to answer as it might seem.

Referring to his teaching experiences as director of the AI Laboratory
at the Massachusetts Institute of Technology' (MIT), Patrick Winston
notes that "defining intelligence usually takes a semester-long
struggle, and even after that I am not sure we ever get a definition
really nailed down."
However, to explain what an "intelligent" computer should
be able to do, we need a better understanding of what we mean by
intelligence.

WHAT IS INTELLIGENCE?
Knowing some of the Why do we reser\e the term intelligence for humans? Why don't we
characteristics of intelli consider computers to be intelligent? What is itthat people can do
gence can help clarify better than computers that makes people intelligent but not
what is required to make
computers?
computers intelligent.
While an exact definition of intelligence has proven to be
extremely elusive, the following characteristics are suggested by
Douglas Hofstadter in a list of "essential abilities for intelligence"
from his Pulitzer Prize- winning book Godel, Escher, Bach: A n E^et-nal
Golden Braid.'
u "To respond to situations very flexibly"— You do not necessarily
'

respond the same way each time you are confronted with an
identical problem. If you did, you would be exhibiting mechanical,
rather than intelligent, behavior.
'To make sense out of ambiguous or contradictory messages"— You
'

are able to understand many statements that appear to be

ambiguous or contradictor^' largely because your knowledge and
experience allow you to place them in context.
To recognize the relative importance of different elements of a
'

situation"— Although you are bombarded with an over^vhelming

amount of information each day, you "make sense" of your world
by assigning different levels of importance to different events.
"To between situations despite differences which
find similarities
may separate them"— By recognizing similarities, you can base your
future actions on what you have learned in the past. Two situations
do not have to be identical for you to apply the lessons of your
experience.
'To draw distinctions between situations despite similarities which
may link them"— Although two situations may appear to be similar
on the surface, you are able to notice differences which may lead
you to adjust your reactions accordingly.

'Patrick H. Winston, "Perspective," in The Al Business, eds. Patrick H. Winston and

Karen A. [Link] (Cambridge, MA: MIT Press, 1984), p. 1
*
[)ouglas R. Hofstadter, Gndel. Escher. Bach. An Eternal Golden Braid(New York:
Vintage, 1980), p. 26.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE''

Thest' abilities share at least one attribute: they all come very
easily to people. In fact, we often group these abilities under the
heading of cuntmon sense. The implication is that there is nothing
special about possessing these kinds of mental abilities; they are, in
fact, common.
Cognitive Science
When human intelligence Oddly enough, the very abilities that come most naturally to people,
is being simulated, tasks such as those just mentioned, have proven to be among the most
that are difficult for
difficult to simulate on a computer. This surprised some of the early AI
humans to do may prove
researchers, who assumed that skills that were easy for people also
easy to program on a
computer. could be programmed easily into a computer. In many cases, the
reverse turned out to be true.
The more difficult a task is for you to do, the more deliberate
and conscious thought you have to devote to it. If you have to
concentrate on the precise steps that are necessary to produce a
certain result, it may not be difficult to program those steps into a
computer.
Activities which are based Mathematicians can describe in great detail the technique for
on rules and skills are easy multiplying two numbers together, for example, and accountants can
to program on a computer;
describe accounting procedures with precision. Multiplication and
while activities based on
accounting are just two examples of activities that are difficult for
problem-solving and
decision-making are dif f i
many people; but because they can be described in intricate detail,
cult to program. they are performed easily by computers.
On the other hand, if an activity comes so naturally to you that
you don't have to think about it at all, you may have great difficulty in
describing exactly how you did it. After all, it may seem so trivial that
you may have never thought about the mental processes involved.
What did you have for dinner last night? Now, can you list the
mental steps you went through to remember what you ate? How do
you coordinate the intricately interconnected series of muscular
contractions necessary to pick up a coffee cup? Can you describe the
process of reading and understanding this book? Can you explain how
black marks on pieces of paper are transformed to knowledge in your
brain?
The field that investigates these details of the mechanics of
human intelligence is known as cognitive science. The research
conducted by cognitive scientists that helps to explain the workings of
human intelligence is a great help to those in AI who are trying to
simulate that intelligence on a computer.
Modeling Human Performance
How do you program a computer so that it exhibits intelligent
behavior?

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE''

Researchers in AI have used many different techniques to

make computers more intelligent; many of them are discussed
throughout this book. One commonly used technique is to determine
the process used by humans to produce a particular type of intelligent
behavior, and then to simulate that process on a computer.
Al researchers help test As cognitive scientists determine the processes that produce
cognitive science theories human intelligence in a given situation, these processes may be
by trying to simulate
programmed into a computer in an attempt to simulate that behavior.
models of human behavior
on a computer. This Al technique is called rmxleling or simulation. In effect, you are
creating a model of intelligent human behavior to trj' to sim ulate that
behavior on a computer to determine if the computer will exhibit the
same intelligent behavior as a human.
As shown in Figure 1-3, the link between cognitive science
and computer modeling is bidirectional. Cognitive scientists develop
theories of human intelligence that are programmed into computer
models by Al researchers; the computer models are then used to test
the validity of those theories. The feedback from the computer models
allows the cognitive scientists to refine their theories, which can be
used to implement better models, and so on.

How Important Is the Process?

How important is it to simulate the processes of human intelligence? Is
it the goal of AI to simulate intelligent behavior with a computer by

any mearts? Or is it truly artificial intelligence only if you simulate

intelligence by using the same techniques as a human?

Figure 1-3.
The "Feedback Loop" In
Al Computer Modeling

_:::>,

Here
 WHAT IS
ARTIFICIAL INTELLIGENCE''

Al rcM'archiTN (li>iinit'i' m As with other aspects of the definition of art ificial

whether or not inlclli intelligence, there is disagreement in the Al community ahoiit this
gence must be simulated
point. Some scientists believe that the goal of artificial intelligence is
using the same proeethur
simply to simulate intelligent behavior on a computer, using any
a huntan would use.
techniques that prove to be effective. (Some of these people prefer the
term "machine intelligence" to "artificial intelligence. ') Others claim
that it is not Al if you simulate intelligence using procedures other

than those that might be used by humans.

For example, suppose you wanted to program a computer to
play a good game of chess. You could write the program to imitate the
thought processes of a human chess expert, and most people would
agree that your program demonstrated artificial intelligence.
Alternatively, you might write the progjram to consider the relative
merits of 10, 000 different moves before making a move, although no
human would ever evaluate that many possibilities.
Even if both programs played championship [Link], there still
would be some disagreement as to whether the second program
should be categorized as Al. Some scientists would say that the second
program was exhibiting intelligent human behavior, so it is Al; others
would say that the program is not Al because it uses techniques that
are not representative of intelligent human thought processes.
People and Computers
Developing a computer Some researchers propose an ultimate goal for Al that is even more
that can think may depend ambitious than having the computer simulate human processes or
not only on the definition
exhibit intelligent human behavior. These people claim that Al
of intelligence but also on
eventually will produce computers that "think"— machines that
the definition of the brai n
actually are intelligent.
Will computers ever be as intelligent as people? That's an
extremely difficult question to answer for several reasons. As we have
seen, an examination of our concept of human intelligence reveals
that, while we may possess an instinctive understanding of the nature
of intelligence, it is a difficult concept to define with great precision. If
we cannot even agree on what constitutes human intelligence, how
will we know when we have successfully achieved artificial
intelligence?
Whether or not you believe that a computer will ever possess
the intelligence of a human brain may hinge not only on how you
define intelligence, it may depend also on how you define brain.

Meat Machines
Our notions of ourselves as living, conscious beings are intimately
linked to our notions of brains and thought. "Cogito, ergo sum," said
Descartes—
'

' I think therefore

, I am " —basing the proof of his very
existence on his awareness of his own mental processes.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE?

But what is the nature of the brain? Can thought, feelings, and
emotions be represented by a set of rules that can be reproduced in a
machine? Is the brain nothing more than an incredibly sophisticated
computer?
EJased on ideas dat ing back The debate over the essence of human rationality is literally
to Plato and Aristotle, thousands of years old. Plato and Aristotle were among the first to
some people think that the divide human capabilities into two distinct areas: the physical body
human brain works like a
and the rational mind. If the mind is considered to be separate from
machine. Thus, the brain '
'

could be called a "meat our physical , ' animal ' nature , it is not a large step to believe that the
machine." mind operates as a useful machine.
view is the belief that the workings of the
Tlie mechanistic
mind can be desciibed terms of the electro-chemical functioning of
in
the brain. Containing about 100 billion cells with complex
interrelations that are still only dimly understood, the brain would
have to be considered an extraordinarily intricate machine and
certainly an unusual one in that it is composed of living material—
"meat machine."
Although some scientists Is the development of intelligent machines just a matter of
say that some day a com- achieving a more complete understanding of the workings of this meat
puter will be intelligent,
machine and programming a computing machine accordingly? Will a
many people feel that a
computer will never think
computer so programmed not just simulate intelligence— not just
likethevdo. appear to be intelligent— but actually be intelligent in the same sense
that people are intelligent? Will a computer someday be able to say
"Computo, ergo sum"? (See Figure 1-4 for comments about thinking
machines from Doug Lenat, head of AI at MCC, the Microelectronics
and Computer Technology Corporation).
No matter what the scientists may say, many of us
instinctively feel that the workings of our minds can never be
programmed into a machine and that no computer ever will have a
mind of its own. While conceding that someday it may be possible to
program a computer so that it appears to be intelligent, it is difficult to
accept the proposition that a machine actually can be made to think in
the sense that people think, to understand information rather than
just to process data.
Philosopher Hubert Dreyfus, a persistent critic of AI, refuses
to subscribe to the mechanistic view of the human mitid " If one .

thinks of the importance of the sensory-motor skills in the

development of our ability to recognize and cope with objects, or of
the role of needs and desires in structuring all social situations, or
finally of the whole cultural background of human self-
interjjretation," he maintains, "the idea that we can simply ignore this
know-how while formalizing our understanding as a complex system
of fact and rules is highly implausible.""^

^ Hubert Dreyfus, What Computers Can Do: The Limits of Artificial Intelligence
't

(New York: Harper & Row, 1972).

UNDERSTANDING ARTIFICIAL INTELLIGENCE
 WHAT IS
ARTIFICIAL INTELLIGENCE''

Figure 1-4.
An Interview with Doug
Lenal on "People vs. Q: Is the question "Can a machine think?" a nneaningfui question?
Computers" LENAT: It's a fair question, suppose. think the answer is: sure, it's
I I possible.
If you claim that It's not a fair question, you'll have to explain to me why you
believe that other people think. You'll get to the fact that you can say things to
them, and they act as if they were thinking; the responses they give are (at least
on occasion) thoughtful responses; and the behavior they exhibit over time
shows that they are occasionally thinking, and learning from what they've seen,
and reacting intelligently, and so forth.
To the extent that could build machines that more or less acted as
I

appropriately in various situations, would be hard to say that they think any
it

more or less than you would say that other people think.
Q: So you think that machines will some day be able to think, just as people
think?
LENAT: That's right. Remember, down deep we're just machines of a sort, too
We're really just meat machines instead of metal machines. Although that
sounds extremely callous and antihumanistic, don't think needs to be. I it

It's very much like automobiles. took a course on automotive repair a

I

couple of years ago, and came away with a profound sense of awe and respect
I

for automobiles. never dreamed they were nearly as complicated as found

I I

them to be.
think the same thing is true with people. It doesn't matter whether they are
I

"just machines." The more you find out about them for instance, by trying to —
simulate their behavior —
the more respect you actually gain for people. You end
up believing that people are actually much more sophisticated than you believed
when you started. In that sense, is a humbling experience rather than a
it

"shallowing" experience, in terms of your respect for human beings.

OTHER DEFINITIONS OF Al

Other definitions of Al aru The two definitions of AI presented previously concentrated on the
concerned with symtwlic comparison between the abilities of humans and the abilities of
processing, heuristics, and
computers. The following definitions of artificial intelligence focus on
pattern matching.
the difference between programming techniques used in Al and more
conventional methods of programming.

Symbolic Processing
A third definition men- According to Bruce Buchanan and Edward Shortliffe, symbolic
tions that AI solves prob- processing is an essential characteristic of artificial intelligence:
lems by symt)olic and
nonalgorithmic methods, "Artificial intelligence is that branch of computer
instead of numeric and science dealing with symbolic, nonalgorithmic
algorithmic methods.
methods of problem solving."*
Bruce G. Buchanan and Edward H. Shortliffe, Rule-Based
Expert Systems

* Bruce G. Buchanan and Edward H . Shortliffe, Rule-Based Expert Systems (Reading,

MA: Addison-Wesley, 1984), p. 3.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE'

This definition focuses on two characteristics of computer programs:

Numeric vs. Symbolic— As has been noted, computers originally
were designed specifically to process numbers. People, however,
tend to think ^i/mlxjlicatly rather than numerically: our intelligence
seems to be based, in part, on our mental ability to manipulate
si/mbols, rather than just numbers. (Symbolic processing is explored
further in Chapter 7.)
Algorithmic vs. Nonalgorithmic— An algorithm is a step-by-step
procedure with well-defined starting and ending points, which is
guaranteed to reach a solution to a specific problem. Computer
architecture readily lends itself to this step-by-step approach;
computer programs traditionally have been based on algorithms.
Many human reasoning processes, however, tend to be
nonalgorithmic; in other words, our mental activities consist of
more than just following logical, step-by-step procedures.
Much AI research continues to be devoted to symbolic and
nonalgorithmic processing techrtiques in an attempt to emulate more
closely human reasoning processes with a computer.

Heuristics
AI researchers rely on In an encyclopedia article, Bruce Buchanan includes heuristics as a
heuristics (rules of thumb) key element of artificial intelligence:
to solve problems.
"Artificial intelligence is the branch of computer
science that deals with ways of representing
knowledge using symbols rather than numbers
and with rules-of-thumb, or heuristic, methods for
processing information."^

Bruce G. Buchanan, Encyclopedia Britannica

Like the previous definition, this definition notes that symbolic

processing is an important AI concept. It also introduces a concept that
we have not yet discussed: heuristic processing.
A heurisi ic Ls a "rule of thumb' ' that helps you to determine
how to proceed. While you may not be familiar with the term, you
frequently use heuristics, consciously or otherwise, to help you decide
what you are going to do, as illustrated in Fiffure 1-5.
By using a heuristic, you do not have to rethink completely
every problem with which you are faced. If you have a handy rule of
thumb that applies to your situation, it may suggest to you exactly how
to proceed.

''
Encyclopedia Britannica, 1985 Yearbook of Science and the Future, s. v.

Intelligence: Toward Machines that Think," by Bruce G. Buchanan.

UNDERSTANDING AFTTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE?

Figure 1-5.
Samples of
Subconscious and
Conscious Heuristics

SUBCONSCIOUS HEURISTIC CONSCIOUS HEURISTIC

Heuristic programming incorporates this same rule-of -thumb

approach into the process of using AI to solve problems with
computers.

Pattern Matching
A fifth definition of AI Another definition of artificial intelligence focuses on pattern-
focuses on the use of matching techniques:
pattern-matching tech-
niques in an attempt to "In simplified terms, artificial intelligence works
discover relationships be- with pattern-matching methods which attempt to
tween activities just as describe objects, events, or processes in terms of
humans do.
their qualitative features and logical and
computational relationships."^
Brattle Research Corporation, Artificial Intelligence and Fifth

Generation Computer Tectinologies

Computers can be used to collect information about objects, events, or

processes; and, of course, computers can organize large amounts of
information more efficiently than people can. People, however,

**
Brattle Research Corporation, Artificial Intelligence and Fifth Generation
Computer Technologies{hostor\), p. 5.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE?

instinctively do something that has been very difficult to program into

a computer: we discover relationships between things; we sense
qualities and spot patterns that explain how various items relate to
each other.
Newspaper photographs are nothing more than collections of
minute dots (Figure l-d); yet without any conscious effort, we
discover the patterns that reveal faces and other objects in those
photos. Similarly, one of the ways that we make sense of the world is

by recognizing the relationships and patterns that help give meaning

to the objects and events that we encounter.
If computers are to become more intelligent, they must be able

to make the same kinds of associations between the qualities of

objects, events, and processes that come so naturally to people.

Al TECHNOLOGIES
The debate about the essence of artificial intelligence is not one that
will be resolved quickly.

Figure 1-6.
The Texas Instruments
Professional Computer
Pictured as a Collection
of Dots

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE''

As a relatively new science,

AI continues to experience rapid
changes in focus and .scope.
primary tool, the computer, has
Its

continually and dramatically increased in power, and de<reased in

cost, througliout the history of AI. These advances in computer
technology often are accompanied by correspondingly dramatic
advances in AI research.
."
"At the Moment . .

A I researchers tend lo One seemingly insignificant phrase in Elaine Rich's definition of

explore new frontiers in artificial intelligence presented at the beginning of this chapter is the
computer science.
notion that AI studies what people do better than computers "at the
Therefore, some of these
researchers move on to
moment."
new problems as soon as With this definition, Rich has identified yet another area of
current problems essen- contentionamong those who would define artificial intelligence:
tially are solved. when an AI technology is developed to the extent that humans no
longer outperform computers, some AI researchers no longer consider
if AI studies only what
that technology to be part of AI. After all,
people do better than computers "at the moment," then AI ceases to
study a problem when computer performance reaches or exceeds the
level of human performance.
AI research has traditionally been at the cutting edge of
computer science. AI continually redefines what computers can do
and pushes existing computer technology to its limits. When a
particular AI problem is substantially solved, AI researchers typically
move on to newer, more challenging problems. The former problem is
"no longer interesting," they may say, or it has become "simply a
'

matter of engineering.
This results in a very fluid definition of artificial intelligence as
the scope of the field continually shifts in new directions. One
difficulty with the "at the moment" concept is that it renders it
'

impossible to ever develop an AI product. If a technology advances

'

' '

sufficiently to allow for product development, that technology can no

longer be classified as artificial intelligence.

Tlie AI Evolution
Thus far, area.s of AI With all the factors that contribute to imprecision in the definition of
research have been evol\ AI, one useful way to understand AI has been to inspect what it is that
ing continually. However
AI researchers do. In other words, it has been useful to define AI not
as more people identify
research taking place in a
by using a formal definition, but by identifying the areas of research
particular area as AI, that that are being conducted by people who consider themselves (and/or
area will tend to remain a who are considered by others) to be working with artificial
part of AI. This could intelligence. Because so many people in AI continually seek new
result in a more static
directions in search of more challenging problems, defining the field
definition of artificial
intelligence.
by examirung the research has contributed to a continually evolving
definition of AI.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE?

Roger Schank, head of the AI Lab at Yale University, stresses

that to understand AI, you must understand that it is in a constant
state of change. "AI tends to be an elusive subject," he notes.
"Artificial intelligence is best understood as an eixdutian rather than a
''
revolution.
Although the boundaries of AI will certainly continue to
expand, it is reasonable to assume that the definition of AI may be in
the process of becoming more static. Awareness of AI among the
general public is increasing dramatically. As fields of research become
firmly identified with AI in the public perception, it is likely that those
fields will continue to be classified as AI even after the focus of AI
research moves on to greener pastures.
Areas of AI Research
There are identifiable areas of research that generally, if not
universally, are included in discussions of artificial intelligence. Here
are brief introductions to several of the AI technologies that are
explored in this book.

Expert Systems
Currently, the most well- An expert system is a computer program designed to act as an expert
known area of AI researcli ina particular domain (area of expertise). Also known as a knowledge-
is expert systems, where
based system, an expert system typically includes a sizable knowledge
programs include expen
base, consisting of facts about the domain and heuristics (rules) for
level knowledge of a par
ticular field in order to applying those facts {Figure 1-7).
assist experts in that field

Figure 1-7.
The Components of the
Knowledge Base of an
Expert System
DOMAIN OF EXPERT SYSTEM
 WHAT IS
ARTIFICIAL INTELLIGENCE?

Expert systems currently are designed to assist experts, not to

replace them. They have proven to be useful in diverse areas such as
medical diagnosis, chemical analysis, geological exploration, and
computer system configuration.
Since the expert systems field promises a great deal of
practical application and commercial potential in the near future, it
has begun to attract an enormous amount of attention. Expert systems
technology is poised to become the first AI technology to have a
widespread impact on business and industry.

Natural [Link] Processing

The two aroa-s of natural The utility of computers Ls often limited by communication
language [Link], difficulties. The effective use of a computer traditionally has involved
understanding and
the [Link] of a programming language or a set of commands that you
generation, are intended
must use to communicate with the computer. The goal of natural
to .simplify our communi
cation with a computer. language processing is to enable people and computers to
communicate in a "natural" (human) language, such as English,
rather than in a computer language.
The field of natural language processing is divided into the
two sub-fields of:
Natural language understanding, which investigates methods of
allowing the computer to comprehend instructions given in
ordinary English so that computers can understand people more
easily; and
Natural language generation, which strives to have computers
produce ordinary English language so that people can understand
computers more easily.

Speech Recognition
Through speech recogni- The focus of natural language processing is to enable computers to
tion research, we some communicate interactively with English words and sentences that are
day may be able to com- typed on paper or displayed on a screen. However, the primary
municate with a compute
interactive method of communication used by humans is not reading
by speaking naturally in-
stead of typing. and writing; it is speech.
The goal of speech recognition research is to allow computers
to understand human speech so that they can hear our voices and
recognize the words we are speaking. Speech recognition research
seeks to advance the goal of natural language [Link] by
simplifying the process of interactive communication between people
and computers.
Computer Vision
Providing computers with It isa simple task to attach a camera to a computer so that the
the ability to understand computer can receive visual images. It has proven to be a far more
their surroundings the
difficult task, however, to interpret those images so that the computer
is

goal of computer vision

research.
can understand exactly what it is seeing.

16 UNDERSTANDING AFITIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE?

People generally use vision as their primary means of sensing

their environment; we generally see more than we hear, feel, smell, or
taste. The goal of computer vision research is to give computers this
same powerful facility for understanding their surroundings.
Currently, one of the primary uses of computer vision is in the area of
robotics {Figure 1-8).

Robotics
.\lthougli we currently A robot is an electro-mechanical device that can be programmed to
have robots that can per perform manual tasks {Fiffu re 1-8). The Robotic Industries Association
form pre-programmed
formally defines a robot as "a reprogrammable multi-functional
activities, AI robotics re
searchers want to design
manipulator designed to move material, parts, tools, or specialized
robots that can change devices through variable programmed motions for the performance of
'•>
their actions based on a variety of tasks. "
their rnvironmrnt

Figure 1-8.
An Inspection Robot That
Combines a Robot Arm
and a Computer Vision
System to Inspect Parts
and Assemblies

'"William B. Gevarter, Intelligent AfacftinfsCEnglewood Cliffs, NJ; Prentice-Hall,

1985), p. 1.59.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE?

Not all of robotics is considered to be part of AI. A robot that

performs only the actions that it has been pre-proj;ranimed to perform
is considered to be a dumb' robot, possessing no more intelligence
'
'

than, say, a dishwasher. An "intelligent" robot includes some kind of

sensory apparatus, such as a camera, that allows it to respond to
changes in its environment, rather than just to follow instructions
"mindlessly."

Intelligent Computer- Assisted Instruction

Intelligent computer- Computer-assisted instruction (CAI) has been in use for many years,
assisted instniction pro- bringing the power of the computer to bear on the educational
grams customize instruc
tion by using learning
process. Now AI methods are being applied to the development of
techniques that are appm intelligent computer-assisted instruction (ICAI) systems in an attempt
priate for a particular to create computerized "tutors' that shape their teaching techniques
'

student. to fit the learning patterns of individual students.

Automatic Programming
To simplify and shorten In simple terms, programming is the process of telling the computer
the programming process exactly what you want it to do. Developing a computer program
tools are l)ejng developed
to write computer pro-
frequently requires a great deal of time. A program or .system (a group
grams automatically. of interrelated programs) must be designed, written, tested, debugged
(cleansed of errors), and evaluated, all as part of the program
development process.
The goal of automatic programming is to create special
programs that act as intelligent "tools" to assist programmers and
expedite each phase of the programming process. The ultimate aim of
automatic programming is a computer [Link] that could develop
programs by itself, in response to and in accordance with the
specifications of a program developer.

Planning and Decision Support

Making a computer When you have a goal, either you rely on luck and providence to
capable of aiding in achieve that goal or you design and implement a ptow. The realization
business planning is one il

of a complex goal may require the construction of a formal and

i

the goals of planning and

decision .supp<irt research detailed plan.
The development of plans in business, for example, may
necessitate the collection and evaluation of significant amounts of
information. Intelligent planning programs are designed to provide
active assistance in the planning process and are expected to be
particularly helpful to managers with decision-making responsibilities.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
AFITIFICIAL INTELLIGENCE'

DOES IT MATTER?
Most iH-opk" working in Is the goal of artificial intelligence to create machines that actually are
the AI field arc not con- intelligent or to create machines that simulate intelligence? Must the
cerned that a single,
techniques of AI programming be modeled after human cognition, or
generally-accepted defini
are the means of creating intelligent machines irrelevant? Which areas
tion of AI currently does
not exist. of research can be labeled AI, and which ones cannot be so labeled?
The presence of questions like these is a direct result of the
lack of an accepted definition of artificial intelligence. This raises
another question: how important is it to develop a concise, generally-
accepted definition of artificial intelligence?
While it may be awkward to discuss something that we have
not defined precisely, our human intelligence allows us to overcome
the ambiguity and grasp the essential concepts that underlie AI. In
fact, some people in the field downplay the importance of arriving at a
precise definition of AI.
When the editors of the journal Artificial Intelligence asked a
series of questions to several AI luminaries, they included questions
such as " Is AI a single discipline? " and " What distinguishes it from
other fields?" None of the respondents mention any discomfort with
the lack of precision in the definition of AI, and several stress its lack
of importance.
Saul Amarel of Rutgers University, for example, says that he
"would not be overly concerned at present with the 'nature of AI' and
definitions of the discipline. " " Jerome Feldman of the University of
Rochester is even more direct; he does not "think it matters at all
whether AI is a discipline or where its boundaries might lie."'- In
Figure 1-9, Randy Davis expresses his opinion on the importance of
developing a definitive definition of artificial intelligence.

WHAT HAVE WE LEARNED?

1 There is not universal agreement among AI researchers about
exactly what constitutes artificial intelligence.
2. Various definitions of AI focus on different aspects of this branch of
computer science, including intelligent behavior, symbolic
processing, heuristics, and pattern matching.
3. Cognitive science investigates the mechanics of human behavior
and provides valuable insights to AI researchers.
4. One common way of programming a computer to exhibit intelligent
behavior is to model, or simulate, human behavior.
5. Areas of AI research include expert systems, natural language
processing, speech recognition, computer vision, robotics,
intelligent [Link] instruction (ICAI), automatic
programming, and planning and decision support.
'
' Daniel G. Bobrow and Patrick J. Hayes, eds., "Artificial Intelligence— Where Are
V/eV , Arltfirial Intelligence2h{March 1985); 375-415.
'2 Ibid.

UNDERSTANDING ARTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE'

Figure 1-9.
An Interview with Randy
Oavison "The
Importance of Defining Q: For a program to be considered Al, must it solve problems using the same
Artificial Intelligence" technique that a person would use?
DAVIS: No. There are two different agendas you could have. One is
at least
closer to what cognitive scientists are trying to do, which is to understand how

people solve problems. The other is more oriented toward what we call machine
intelligence, which is that you want to solve the problem in any reasonable
fashion, and you are not concerned about whether models people exactly. it

People fall at all points of the spectrum. It's not an "elther-or" issue it's —
"and." Yes. you can understand how people work; and yes. you can develop
ideas and techniques which solve difficult problems in ways that are less like
people; and yes, all of the above.
In fact, I get a little impatient with a question like "Do you want to do X or
Y?" The answer is not "or." The answer is "Yes, all of the above."
Q: Some Al people seem to be quick to point al someone else's work and say,
"That's not Al "
Are you suggesting a more inclusive policy — saying that Al
covers many different things?
DAVIS: There are many legitimate concerns; one need not choose one or the
other.
The map is populated by all sorts of approaches which have all sorts of
goals. Where you want to draw the line and call one "Al" and another "not
Al" — that is also a question that I will admit to some impatience on. Because I

don't know what you get out of it.

Suppose draw the line somewhere. So? Other than name calling, what
I

have accomplished?
I

Q: So is not an important question?

it

DAVIS: don't think is particularly important. The only time starts to get
I it it

important is when starts to have pragmatic consequences.

it

The most obvious example of that is in marketing because, of course, Al —

has become a hot. new "buzzword." Somebody says "This is an Al product,"
and the only reason I will enter into an argument and say "No, it isn't" is when I

think somebody is being lied to. Somebody is trying to sell something

uninteresting that is probably an old idea wrapped up in a reprogram and given a
new label so somebody will buy it again.
The only reason to be concerned about the label is if somebody is trying to

take unfair advantage of the label.

WHAT'S NEXT?
This chapter has discussed the general concepts of artificial
intelligence and introduced you to several areas of Al technology. The
next chapter explores the history of AT and introduces you to some of
the key people in the field.

UNDERSTANDiNG ARTIFICiAL INTELLIGEI^CE

WHAT IS
ARTIFICIAL INTELLIGENCE?

Quiz for Chapter 1

1 V\ hich of the fi)llowmg have 9. A key element of AI is a'an

computers traditionally done better , which is a " rule of
than people!" thumb."
a. Storing information a. heuristic
b. Responding flexibly b. cognition
c. Computing numerically C. algorithm
d. All of the above d. digiton
e. a and c above 10. An Al technique that allows
2. Which of the following have people computers to understand
traditionally done better than associations and relationships
computers? between objects and events is
a. Recognizing relative importance called:
b. Finding similarities a. cognitive science.
C. Resolving ambiguity b. heuristic processing.
d. All of the above C. relative .symbolism.
e. a and c above d. pattern matching.
3. The field that investigates the 11. A computer program that contains
mechanics of human intelligence is: expertise in a particular domain is

a. artificial science. called an:

b. cognitive science. a. automatic processor.
c. psychology. b. intelligent planner.
d. sociology. C. expert system.
d. operational -symbolizer.
4. One method of programming a
computer to exhibit human 12. The knowledge base of an expert
intelligence is called modeling or: system includes both facts and:
a. simulation. a. theories.
b. cognitization. b. heuristics.
C. psychic amelioration. C. algorithms.
d. duplication. d. analyses.

5. A bidirectional feedback loop links 1 3. The area of Al that investigates

computer modeling with: methods of facilitating
a. pattern matching. communication between people
b. heuristic processing. and computers is:

c. human intelligence. a. natural language processing.

d. cognitive science. b. decision support.
C. symbolic processing.
6. Comparing the human mind to a
d. robotics.
"meat machine" isa
view of the workings of the mind. 1 4. Natural language processing is
a. cognitive divided into the two subfields of:
b. behaviorist a. time and motion.
C. mechanistic b. algorithmic and heuristic.
d. relativistic C. symbolic and numeric.
d. understanding auid generation.
7. The brain contains about
cells. 15. The primaryinteractive method of
a. one thousand communication used by humans is:
b. one million a. reading.
C. one billion b. writing.
d. 100 billion C. speaking.
d. seeing.
8. One definitionof Al focuses on
problem-solving methods that 16. The primary method that people
process: use to sense their environment is:

a. numbers. a. reading.
b. symbols. b. writing.
C. actions. C. speaking.
d. algorithms. d. seeing.

UNDERSTANDING AFTTIFICIAL INTELLIGENCE

 WHAT IS
ARTIFICIAL INTELLIGENCE''

17. An mli'llinciu ii)t)<>t:

a. ii's|Kiii(ls to changes in its

i'M\iiDnment.
b. lolldws Instructions mindlessly.
C. possi'sses no more intelligence
than a dishwasher.
d. does all of the above.
18. Shaping teaching techniques to fit
the learning patterns of individual
students is the goal of:
a. automatic programming.
b. decision support.
C. intelligent computer-assisted
instruction.
d. expert systems.
19. S(X'cial programs that assist
programmers are called;
a. symbolic programmers.
b. heuristic processors.
C. Intelligent programming tools.
d. program recognizers.
20. Intelligent planning programs may
be of special value to managers
with responsibilities.
a. programming
b. customer service
c. ijersonnel administration
d. decisionmaking

UNDERSTANDING ARTIFICIAL INTELLIGENCE