A Primer on Artificial Intelligence. PUB DATE 75

DOCUMENT RESUME

ED 114 116 IR 002 751

AUTHOP Leal, Ralph A. . . TITLE /A Primer on Artificial Intelligence. PUB DATE 75 ,. , NOTE '33p. 1

' :-- FDRS PRICE MF-$0.76 HC-$1a95 Plus Postage- pESCPIPTORS *Artificial Intelligence; *Bionics; tOdputer Programs;' *Cyb4inetics;'Information The'ory; IntelligenC Factors; Machine Translation; *Progr,aming Languages. . r ABSTPACT A survey of litetature on, recent advances in the fidld of artificial intelligence provides, a comprehensive introduction to this field for the non-technical'reader. Important areas covered are:, (1) definitions, (2)t'ie brain and thinking, (3) heuristic sahrch, and (4) pr'ograminig languages used in the research of artificial intelligence. Some examples of hrtificial intelligence are described. Suggestions for additional reading and a 28-item bibliography are'appended. (DS)

4,r

**********************************************************************

* . Documents acquired by ERIC include manylinforeal unpublished * materials not available from other sources. ERIC makes every effort * * to obtain the best copy, available. Nevertheless, items of marginal" * reproducibility' are often encountered and this-affects the,qualit.y

* of the microfiche and hardc'opy reproductions ERIC makes available . *,via the ERIC .Document Peproduction Service (EDPS). EDRS is not * responsible for the quality of the original document. Reproductions *, supplied bey EDPS are the best that-can/be madd from the original. *, ***************4******************************************************* s.4) Title: A Primer' on Artificial Intelligence r7q

.Author: ,-Ralph A. Leal r-q State University of New York- College at Old Westbury

C:) . Abstract: This survey of recent ad4iances in the field offartificial 11-11 intelligene is'aimed at the non-technical reader. Its purpose is to present a concise and comprehensive introduction to the field.. Areas covered' include definitions, the brain and thinking, heUristic searchand an overview of pogramming languages in artificial intelligence research. The appendices include a classical article by Dr. Vannevar Bush entitled "As We May Think" and biblio- graphical suggestions for additional reading.,

Keywords: Artificial Intelligence; AI; Huristic Search; Programming Languages in Artificial Intelligence; Vannevar Bush.,

42 pages, appendices, bibliographies

U S OEFIARTMENT OF HEALTH, 1$01.1CATION LWELFARE NATIONAL INSTITUTE OF 4, IOUCATION THISDOCUMENT HAS OUCEO EXACTLY SEEN REPRO THE PERSON OR AS RECEIVED FROM 4. ATiNG IT POINTSORGANIZATION ORIGIN OF VIEW OR OPINIONS STATED DO NOT SENT OFFICIAL NECESSARILY REPRf NATIONAL INSTITUTE EOUCATION POSITIONOR oppLity OF

A Ar,STRACT

Leal, Ralph A. A PRIr22 CN ARTIFICILL D-TELLI.7,:tNCE. New York, 1974.

Artificial intelligence has been .defined. as 'the totality of if attempts to make and understand machines that are able to perform tasks that, recently, only human beings could perform, and to perform /them with effectiveness and speed comparable toa human.' (Banerji, 1969).

Although the field has nany divisions and subdivisions, the Ast important work can be classifiedinto four areas: (1) Game Playing; (2) Problem Solving; (3) PatternRecognition', and (4) Semantic tnformatiOn Processing.

Attempts to understand. the basic.requiremen;s for the construct- don of an 'intelligent' nachine have led to exploration of/the physical functioning of the brain. The most famous theorem in this area` is the McCulloch-Pitts Theory of Formal Neural Networks.

An undeAtanding of thought processes is also important to the_ artificial intelligence field.gleolai Asomov, of the Institute of Cybernetics of the Academy of Sciences of the USSR; r as propoSeda general model of ahuman bei0 which has three types of programs: one for himself - for self - preservation; one for stock 7:the repro- dUctive instinct, and (11-e-'or species- the programs of social behavior.

The Englilth mathematician-A.' M. Turing, a firm believer in the possibility of creating tificial intelligence, has summarized the objections to that possibility in an article entitled 'Can A Machine Think?' Among the objections we find are: (1) the theological;, (2) the heads-in-the-sand; (3) the mathematical; (4) the argument from consciou?ness; (5) the various disabilities argument; (6) Lady Love- lace's grgument; (7) the argument from continuity in the nervous system; (8) the argument front informality of behavior, and (9) the argument from extra-:sensory perception.

7- Heuristic searchls a mathematical process used to move from an initial state to a desired state using a repertoire of actions when the sequence of steps is not exactly known. Donald MiChie believes that this process offers good payoff Possibilities fdr future research in the AI field,

A number of Programming languages, characterized by their ability to d9 list processing and symbol manipulation.; have'been developed to assist AI researchers. The loci for the creation of these languages have been the Massachusetts Institute of Technology, Stanford Research

- Institute, and the University of Edin",surgh.

Despite the raging debates in favor and against the possibility that 'thinking machines' may te constructed, we must conclude that not all the evidence is yet in, and lean- very tentatively - towards 'the positive side of the argument. TABLE OF CONTEN7S 4

I. v Abstract

r- Introduction III. Definitions

IV. The Brain' and Thinking > V. Can Machints Think ?-

VI. 1-leuristic Search

VII. Programming Languages'in AI Research

VIII.Some Examples of AT ..

IX.* Conclusions

Appendices

A. LC Science Tracer Bullet "Artifidial Intelligence"

(B. Direct brain-to-computer interface on,,:the way

C. "Artificial, Intelligence" - Encyclopedia of Librdry and Information Science :C

"As We May Think."

Bibliography

4' INTRODUCTION

. In 1945,1)r. Vannevar Bush wrotean article in The Atlantic Monthly whi has become a classic in the,field ofinformation science. 4 : 1 In that e, Dr. Bush foresaw a world in which man'squest for knowledge be tailored to individual needs through theuse of ..'association trails' stoked in each person's homelibrary - the Memex. .

In the same 1 reaC'hing boldly into the 'future, hewrote:

In, the outside world, allforms of intelligence, whether of sound or sight, have been reducedto the form of varyingcurrents in an electric circuit in order thatthey may be transmitted. Inside the huinan frame exactly thesame sort of process occurs. Must we always transformto mechanical movements inorder to proceed from one electrical phenomenonto another? 1

The September, 1974 issue of Cariputer Decisionsdescribes research currently underway atthe Stanford Research Institute in thefollowing terms:

Imagine, if you can? a computer installationwith none of the input devices you're used to. No card readers, tape readersr.crt terminalsor tele- typewriters. Instead of all thiscumbersome and inefficient hardware,you've got a special helmet. You will enter data directly from your braininto the computer. 2

We have chosen the above examples to introducethe subject ofour research paper- artificialintelligence- because they are illustra-

tive of the advance'swhich are possible in the, various fields ofscience

* 1..Bush, Vannevar. "As We May Think." TheAtlantic Monthly, July 1945, p. 108. (See Appendix D.)

:4 "DirectBrain-to-Computer Interface On The Way." ComputerDecisions, September 1974, p. 13. (See Appendix B.)

tarce L ; bror ce,s 2

I within the span ofthree decades. What Vannevar Bushreferred.to.as a "... suggestive thought, but it hardly warrantsprediction without losing touch withreality and. immediateness 3 ..." in 1945 is already being tested in a California' aboratoryin 1974. The subjectiof our study has suffered attadksby a number of/ respected members of the scientific community,who have put forth

powerful arguments todemonstrate that the . very concept of artificial .--/-elligenceis blzar? e andunderservingf serious . .,, consideration. In the following pages tie will attempt topresent an unbiased a and compre- hensilite picture'of an evolving field. 7

3. Bush, Vannevar, loc. cit.

11.

G , Alt DEFINITIONS

Donald G. Fink, in his Computersand the Human Mind,

4-Tinesartificial intelligence in the f011owingmanner:

(a) The ability of machinesto organize inforL nation into meaningful patterns; abilityto recognize, store, recall, and3abipulatesuch patterns in playing games, solvingproblems, answering questions,etq., and in Controlling the actions of other mechanisms;Xb) the ability of a machine to adapt to itsenvironment, parti- cularly to respond to patternsof stimulation not explicitly foreseen in its design; (c)the obser- ved performance of such machinesas measured by comparison with, or in competitiona4p3nst,.human, intelligence.4

Ranan B. Banerji, in his Theory of ProblemSolving.- An IP-- Approach to Artificial Intelligence, _ puts forth the following defini-

Lion:

The field commonly. called artificialintelligence may, perhaps, be described as the totalityof ahtemptS to make'and understand machinesthat are able to perform tasks that, until,recently, only 1171;gn beings couldperform and to perform them with effectimpess and speed comcarableto a uman. (Italics his) 5

Philip C. Jackson, Jr., in his Introductionto Artificial Intel-

ligende, defines it in these terms:.

PArtificial intelligence is the ability ofmaChines.t: to do things that people wouldsay require intelligence. 6

p- . ! 1 14 4. Fink', Donald G. Computers and the aMind,-ti. 25.

5. Banerji, Ranan B. Theory of Problem Solving, p. 1.

6. Jackson, Jr., Philip C. Introduction to Artificial Intelligence, p. 1.

4 3

I 4

Hubert L. Dreyfus has written that, although thefield has

many divisionS and subdisions, the mostimportant work can be

classified into four areas:

. 1 1. Game praying / s

2. Problem solving'.

3. Pattern recognition

4. Semantic information processing

, The first series of efforts attemptto 'teach' machines to play

.44 such games as checkers, chess,poker, etc. with a level of proficiency

approaching that of an interested adult.The second set ot efforts

include training in the solution ofmathematical, iogl.Cal, and other

types of problems. The third group,'4:.enerally using supportingtele-

vision equipment, seeks to createthe ability to discriminateamong

similarvisual or auditory patterns. Finally, in-the fourth category

we find efforts at machine itt,P-nslation of foreignlanguages, as well

as programs thatanswer a variety of questions on very clearly-delimi- , ted fields. ,

Professor Frank George, of the Department ofCybernetics at

Brunel University; has defined the basicproblem of artificial

intelligence in these terms:

The basic problem of artificial intelligence is as to whether or.not machines (i.e.arti-

, ) 7. Dreyfus, Hulert IL. What Comouters Can't Do, p. xxxiii.- 1.

ficially conStrutted sytemssuch as digital computers) can be made to"think." The pro-. b/em is partly a semanticone, since the word "machine" and the word, "think"both present definitional ,prOb/ms, andwe could clearly define both terms insuch away that machines could obviously thinkor could obviously not think as the.case may be.8 ..

The next section of our paper will present a briefanalysis of the work which has been done in seeking to link the ,brain,thought processes, and artificial intel7igepcg.

8. George, Frank. "What Is Artificial Intelligence?" Data Processing, September-October 1972,p. 328. THE BRAIN AND THMIKING ..vor

In his very penetrating article "The Brain considered As .

a Thinking Machine," Dr. William Feindel wrote:

the Elliman brain h'as certain critical ad- 'SO vantages over these electronic devices. It is far more compact.An electronic brain having as many as twelve thouSand million thinking units would take up an enormous volume. -Even if miniaturized, at thevery best an electronic brain would'be comparable in size to a very large government grain I. elevator. The human brain uses a' very small amount of electrical energy, something between ten and twenty-five watts.Furthermore, it is portable and comes with a lifetime guarantee! '9

Concern with the bratn and thinking gOes back to-antiquity, but

it is only in recent .times that scientists have begun to attemptto

develop a theory which might e applicable to computers. The most

famous theorem in this area is'called the McCUiloch-Pitts Theoryof- 4 Formal Neural Networks,eand is explained bi Johnvon Neumann in his 7 "The General and Logical Theory of Automata" in, the followingterms:

The functioning of a formal neural network maybe defined by singling out so` of the inputs of the entire system and some of its outputs, and then describing what original stimuli on the former are to cause what timate stimuli on the ptter... Any functioning which'can be defined at all logically, strictly, and unambiguoimly in a finite number of words can also be realized such a formal neural network. 10

alb

9. Feindel, William. "The Brain Considered As/a Thinking Machine." in Memory, Learning, and Language, p. 22.

10. von Neumann, John. "The General and Logical Theory of Automata." in The World of Mathematics, pp. 2089-2090.

7 1U 11 Mortimer Taube quotes Warren Weaver'sexplanation of ,the 4 theorem by statingthat it is theoretically possible to constructa robot which coulddeduce any legitimate conclusion from a finiteset of premises, provided thatthe constructionincluded a set cf,tegene- rative loops. 12 Arbib, who has been strongly influenced b the ideasof Warren

McCUlloch, analyzes theproblem from two distinct'angles: whathe has termed the artificial intelligence approach, andwhat he calls the brain theory approach. He distinguishes betweenthe two approaches

by stating that thefdrmer is based an carrying out operationson data

, which it s passively stored, while the latteremp e parallel activity of a multitude of operations.This second approachis referred to as "... distributed action-oriented computationin layered somatotopically organiledmachines." 13

Nicolai M.,Asomov, of the Institute of Cyberneticsof the Academy-of Sciences of the USSR, is the authorof an'article entitled "Simulation of Thinking Processes," which appeared in'thebook 41. Purposive Systems.14 Asomov believes that simulationor modeling of thinking is necessary for studying humancognition, for creating artificial intelligence, and for developing modelsof human interaction. 4

11. Taube, Mortimer Computersand Cannon Sense,p. 19

'12. Arbib, Michael A. The Metaphorical Brain. New York, Wiley- Interscience, 1972. / X:---

13. Arbib, op. Cit.,p. 5.

14.' Asomov, Nicolai M. "Simulation of ThinkingProcesses." In Purposive Systems. New York, Spartan Books,1968.

1 8 \.

-we, He' proposesa general model of a human being whichcan be represented by an automaton having threetypes of programs; one for himself- for self-preservation;one for stock - the'reproductive instinct,and one forspedies- the programs of social behavior. He goes on to state that six programs determine human behavior. They are:

P . 1. Program of bOdily feeling - longperception and -processing of information from thebody. -' 2. Perception of outerinfluence and of models of recognition meaning and qualitiesof the sur- rounding world.

3. Program of action - imparting of energy andinfor- oration to thesurrounding world.

4. Program of speech.

5. Program of consciousness.

6. Program of creativity and work. 15

Clearly, if anautomaton could be constructed andprogrammed with the above list of attributes,we could categorically state that wehave achieved 'artificialintelligence.' Unfortunately, weare far from having arrlvednatthat point.

In his principle of complication in reproduction,. vonNeumann felt that belowa certain point, if automata'could reproduce,they would only be able make less complicated machines thanthemselves. Above that point, and atleaSt conceptually, it wouldbe possible for automata to treatehigher entitles.16

15. Asomov / 2 op. cit.,) p.41.

16. von Neumann, ok. p. 2098.

12 9

Herbert Anschutz, inan article entitled "Prospects for ,the

Development of the Psychocyberneticsof Intelligent Behavior," also

found in Purposive Systems, writes that'cybetmeticistimake use of

three basic disciplines:'informationtheory, the theory of control,

and automaton theory, and that knowledge'about structure can be . t largely replaced by knowledgeabut function, "... *putthe behavior .. r 17 of the blade box overan abstract set of States." 0, Be further states, "... webavefound that. every data oceising,

system, including the human bcain,is subject to a generalized . .

combination of'information theory and the theo-of automata." 18 , ,;./4 . 'We have discerned two principal pidblems corrinectedwiththe

; area of the brain and thoughtprocesses. First, despite intensive

research, little is known of theway in which the brain functions ih

physical terms. Secondr.although it is possible to postulatea

series of requirements for the creationof artificial intelligence,

the bridge between concept and constructionhas not been establishedi.,

t :exceptfor some minimal examples whichwe will examine below:

'In ournextsection we will consider a series ofar

presented by.A.M.'Turing in negatiVeresponse to the qUestion "Can

Machines Think?"

01/1*.+ 17. Anschutz, Herbert, in Puroosive Systems, p. 1136

18. Anschutz, 22. p. 117.

1 3 1

CAN MACHINES THINK ?

A. M. Turing, the.British mathematician ,andlogician, must

be credited with twpsignificant contributions to ourfield, both of 19 which are found in his article "Can A eachine Think?"

The first contribution, which has come to be known as 'Turing's

test,' is a'positive one. The author suggests that one way in which

artificial intelligence could be tested ,world' be through what he calls

'The Imitation Game.' The object pf the game, which is played by

three individuals: a than (a); a woman (b), and an interrogator(c)

who maybe of either sex, is to enable the Interrogator todetermine

the sex of either of the other two players.The rules of the game

call for (a) to try to get (c) to make a wrong identification,and

for (b) to try to get (c) to make a rightidentification. (a) is

free to lie, but (b) must always tell the truth. The players ate

physically separated, So (c),cannot.Anany insights'into thesex

of the players except through their printedrftponses.Turing

suggests that if a machine could be designed andprogrammedtto success--

fully play the role-Of (a), it could be creditedwith, possessing what

he defines as artificial intelligence.

Turing's second contribution, although couchedin,negative terms,

is also a positive one. As a believer in artificial intelligence,

Turing sought to marshallthe,arguments which had been put forth

. , against the possibility of such an occurrence.We have chosen to

summarize these arguMents.

of Mathematics: .19. Turing, 11, "Can a Machine' Think?" in The World pp. 2099-2123. 11

The theological objection. This argument statesthat thinking is a function of man'ssimmortal soul, given to him byGod. Therefore, no animal or inanimateobject can think. . ; The 'heads in the sand' objection. Inasmuch as the . . consequences . , of machines thinking are too dreadful, letus not consider the.possi-1'. bility and fervently hope that theynever will.

THe mathematical objection. There are a number ofresults, of mathematical logic which a can be used to showthatthereare limita- tions to thepowers of discrete-statemachines.Ohe best known of 20 these objections is known as GOedel's theorem, and it shows that

in any sufficiently,powerful logital system _ staipments can be formu- le.-d which can neither be proVed not disprovedwithin the system, unlss it can be shown that. the systeM itself isinconsistent. The argument from consciousness.This argument states thata' machine cannot be said to think until it composesa sonnet or writes a piece of music and is conscious of the fact that it hasdone so. The argUment from various disabilities.The form of thisargu- ment allows for machines to perform a number of'thinking' tasks, but' suggests that there is '1 one =more tasks that a machinecould not

possibly perform.,

Lady Loyelace's argument. The most detailed informationavailable about Babbage's AnalyticalEngine comes froma memoir written by Lady

Lovelace: In this memoir, a classiC in the field, Lady LovelaCestates

20.Nagell'Erfiest and Newman, James, R. "Goedel's Proof."in The . World of Mathematics,pp. 1668-1695. o

12 p . that the machine can do whatever we know how to orderit to perform,

no more, no less. The argument from continuity in the nervous system. This' . argument states that the nervous system isa continuous -, rather than

a discrete-state machine. ,Therefore, no machine could hope to repro--

duce the workings of the nervous system.

The argument from informality of behavior. It is not possible

to predict each and everyone of man's activities. Therefore, it is

inconceivable to-create.a machine which could faithfullyreproduce

man's behavior.

The argument from extra-sensory perception. Since we must acknow-

ledge the existence of such unexplained phenomena as telepathy, clair-

voyance, precognition, and psycho-kinesist we cannot- possibly, expect

machines to account for these things..

Clearly, the arguments presented by Turing constitute a powerful

array of logic against the possibility of artificial intelligence.

The reader is urged to ppek out the reference article to see how

the author demolishes,pach.and every one of the arguments described

above.

1-0 ..}LEURITAOEARCH

Donald Michie is a member of the faculty of the Department ce Machine Intelligence and Perception at the Un4versity of Edinburgh,..-. one of the three leading places in the field of artificial intelligence.

At a meeting of the Mathematical Society of the University in 1970,

Michie explained the development of a new technology, such as artifi- - - rtal intelligence, in terms of three evolutionary stages, each of

_ wilich overlaps or merges with the others over time.

The first stageff technological development is the phase of ad hoc innovation, exemAlfied by the Wright brothers in the field of aeronautics or Watt in the development of stea4power. The second phase is the phase of formalization, that phase at which a new science ,7 is born, thermodynLMics in. the wake of the steam engineers, or aero- dynamics in the wake of the fliers, or information and switching theories in the wake of Marconi and Bell. The final phase is one of stabilization and, systematic development,,whenthe researchers have 21 moved on to newer fields and the practitioners have arrived.

Mollie believed that machine intelligence, as a whole was in headlong transit between phase one and phase two..

In the same article, Michie describes a class of problems which are' defined in the following terms:

Given: an initial state a definition of a desired state a repectoir of actions,

Find: a sequence of actions which willtransform the initial state into a desired state

21. Michie, Donald.. "Heuristic,Search." Thea. Computer Journal, Februarftl971, pp. 96-102.0,

/- 3:4

Michie believes that in thef4ure, heuristic searchstudies, at least within theonte* Ff artificial intelligence,are likely to

,pay increased attentio4tp devisingautomatic methods of imposing

. classifications on problem spaces. The object of thieactivitywould be to subdivide the spaces into regions, and to havea different

operator-selection regime be 22 associated with each region..- _:._

22. Ibid.

18 . PROGRA.MMINGLANGUAGES IN AI RESEARCH

For more than a decade, thelist processing and symbol-maniRulationlanguages -- such as COMIT, (Bobrow' and Raphael 1964) have been the media for almost all AI achieveMents.-Although the effectiveness.of.research with theselangua- ges has improveddramatically due primarily . to greatly expandedmemory sizes and new --interactive debuggingfacilities, the langua- -ges themselves have remaink remarkably. .--

The article from which the above quote was extracted-provides . a very useful summary of some of the computer languages whichhave been utilized in AI research. We have reproduced capsule-descriptions . of these lang4ages for the sake of increasing readerawareness in a .

not-very-coMmon field. .

3":=-1 - ASSET is a programming language based on setrwhichwas developed

at the University ofAberdeen, during 1971. .

ECL is an extensible ;anguagesystem developed at Harvard in

1972 for use with automatic programming. It offers a combinationof agreeable language sources, an interpreter for list ``structureprogram

representation and various levelsof compiler. 1 LISP-70 was developed in 1973, and is based on the LISP system. :-

Although the full system is not yetavailable, a prototype called

IpSP2 has been used in pattern matching and automatic rulemaintenance.

PLANNER and MICRO-PLANNER, a subset,were written in LISP in 1972,

at the MIT AI laboratory.MICRO!PLANNERruns only on theePDP-10

NIT, but transfer to other LTSPsystems for experimental use is being

accomplished.

23. Bobrow, Daniel and Raphael, Bertram. "New Programming Languages for Artificial Intelligence Research."Computing Surveys, September 1974, pp. 153-174. .16

POPLER 1.5 is a programmatic language implementedas an extension of POP-2, a system developed itat the University of Edinburgh for .

research in artificialintelligence..

WCP/INTERLISP are languages which were developeddirectly from

pA. They eliminate theslow utility functions and.lack,ofdebugging

tools of that language. -' (ILTsP was piggy- backed unto.INTERLISP, and both were developed at the Stanford ResearchInstitute.

SA11.is a language project of the Stanford ResearchInstitute AI Project. It is considered to be one of the most stable, debugged,and , heavily usedof the languages in the article.It runs on a PDP -l0

under the DEC 10-50monitor.

SMALLTALK is a language developed in1973 which uses thenotion of classes as an interrogative nechanicm.The language enables4a

user to obtain an answer froman instanceof a class without knowing

whether the data is stored orprocedurally,developed. '- The languages described in the article are characterizedby their ability to do list processing and/or symbol manipulation,and have a been developed to aid AIresearchers in their effortsto enable

computers to perform more andmore complex tasks. SOME EXAMPLES OF AI

This section refers the readerto a number of written 4 desCriptions of experiments inwhich computers were made to play

.games, solve problems, translatesentences, or recognize patterns. , - However, it must be understoodthat the.written word, isa poor subs- - - - - titute for the excitement which pervadesa computer installation when . a proVam designed to accomplish Oneoranother lofthe above tasks is

being tested for the firsttile, or when it finally accomgishtsthe

task it was designed tocarry out.

A general type of problem which AI researchers have attemptedto solve is the situation-space problem. Briefly described, a problemof

this type "... consists ofan initial situation, a set of possible

situations, and a set of possibleactions, together with a specifica- .` ^ tin of how the various situations can be produced from each other by .

different actions, and the specificationof a final, desired situation, 1 or goal.24

One of the best programs for solvingsituation-space problems is

the General Problem Solver (CPS)program of Newell, Shaw, Simon, and

Ernst. The program uses descriptions of objectsand operators -

situations and actions- and using a means-ends analysis technique

(a name given by the authors) proceedsas follows;

1. Evaluate the difference between thecurrent situation and the goal.

2. Find an operator that typicallylowers the type of difference found in step 1.

24. Jackson, op. cit., p. 72. J 18

3. Check to see ifthe operatorfound in step 2 can be applied tothe current situation; it can, then if apply it, elsedetermine a situation required for the applicationof that operator, andestablish it asa new (sub) goal; theygo to step 1. 25

The program hasbeen applied to many different simpleproblems, such as theMissionary-Cannibals Problem, the Three-CoinsProblem,

. and the%Towerof Hanoi Problem. However, when theprogram was given ". . 7 the Seven Bridgesof Konigsberg Problem, shown to.beunsolvable by Biller in1736, it triedagain and again until it gaveup. .

_ - . Another computer program which has achieveda modicum. of fame is the checker- playing program written by Dr. A. L. Samuels,of the . International Business _ Machines Corporation.sSamuels taught the computer how to playcheckers and to continually improveits game until "... In August1962, the computer took on Mr. RobertW. Neely, : 'a former Connecticut checkers champion, and one of thenation's foremost players,' andMr. Mealy lost."26

Other games for which programs haire been written,with varying degrees ofsuccess, are Chess, Nim, Go, Tic-Tac-Toe,and Bridg-It. Many of thesegames are available aOmmercially, and it isworthwhile for the interestedreader who wants to explore the programimplica- tions to acquirea manual version and to seek to plot theprtbabilis-, tic alternatives of the variousmoves.

, Nilsson suggests that problem solving hastwo basic elements:

A 25. Jackson, cEl. cit.,p. 76.

26. Cote, Jr., Alfred The Search for theRobots, p. 199.

...... 19

representation and search. The first.element dealswith the way in which a problem is formulated, and Nilssonstates

The problem of representation iscommon to any problem-solvingapproach, but unfortunately research in artificial intelligence-- has not yet producedany general automatic -_ -method for skillfully formulating problems..27

. *_The second element - search - deals with the stepsused in.:

arriving at the solutionof a given problem. -

The possibility that a machine might be ableto perceive its , , surroundings-has interested Al researchers forsome time. -One of the considerations in constructing a robot capableof surviving in the real world4iS thatit must have the capability to analyzeits envi- ronment and to take the necessary steps to avoidmishaps.- The field of patternperception encompasses visual patterns,sound patterns, symbol patterns, and even reasoning patterns. However, not all:of

these segments havebeen explored.

Among the exercises which computers} isually witha television camera hook-up, have beln able to perform, 'we findone in which the computer is shown a series of aerial photographsof bridges running -over rivers and bridges running over roads, andeach type is ident- ified for the computer. Upon recall, the computerhas scored better than 80% in properly identifying the subtledifferences. One of the classical'efforts in the area of perceptionwas under- ) taken by Frank Rosenblatt, of the Cornell UniversityAeronautical

27. Nilsson, Nils J. Problem-Solving Methods inArtificial Intelligence, p. 8. - K

23 20

Laboratory, wasa machine called the Perceptron.The first Percep- tron was designedwith the ability to recognize theletters of the alphabet. The commercialapplication of the idea in:the formof OCR (optical L , character recognition)will not escape the reader. - Semantic information processing is probablythe area of cial intelligencein which most experimentation has beencarried out. MI his book Semantic , . Information Processing, Marvin Minskyprovides detailed analysesof a number of programs which have beensuccessfully developed in thefield.

The firstprogram discussed in Minky's work isBertram Raphaells. octoral thesis. It describes the development ofa prograM named SIR. emantic InformationRetrieval, in which "... an attempt(was made) to builda memory structure that converts theinformation it receives intoa systematic, efficient representation The system ...works by understanding the statements when they aremade, consoli-

. dating thisunderstanding by adding to or modifyingthe network."28 Daniel G. Bobrow isthe author of a program calledSTUDENT, which "...acc pts as input . " a comfortable butrestricted subsetof English whichan be used to express a wide variety ofalgebra story problems." 29

Space limi ationspreclude a more extensive analysesof the myriad of p6gr which have been written in supportof AI efforts.

23. Minsky, M=in, ed. Semantic InformationProcessing, p. 4. 29. Minsky, 2E. cit,p. 135.

24 . CONCLUSIONS NC'

In his evaluation of computer applications in the field of

artificial intelligence, Donald G. Fink writes

- We find that alm?st all the-computer applications fail to qUalify as "intelligent" under (a) or (b). of the d inition (see p... 3) - useful, yes, - - beyond fondest dreams of Eckert and Mauchly. ,Comput save time and tedium almdSt beyond compreh ion, but intelligent, in the sense of being le to answer'questions not explicitly foresee ? Not proved for the great majority of -- them. 30

Stanley Jakil in his very penetrating analysis of the relation- .,_ . f ...... ° :1-'. ship between B Mind and Computers, states . _ , , .---- . . 77 , 'The stage has been reached where the question U Machines think?" is viewed as a projecti- . ve test of personality. Those answering "yes" described as "self-confident humanists..." ose who refuse to admit the possibility of. ng machines are called "doubters, pessi-

ts... 31 ' - ,

Htbett L. Dreyfus, the most outspoken of the critics ofartifi-

cial intelligente says

my general th6sis will be that the field of artificial intelligence exhibits a recurring pattern: early, dramatic success followed by sudden unexpected difficulties. 32

then, there was HAL, the computer in 2001: A Space Odyssey... 1.0

30. Fink, a.. cit., pp. 225-226.

31. Jaki, Stanley L. 'Brain, Mind and Computers, pp. 255-256.

32. Dreyfus, op. cit., p. xxxiii.

2b. 22

'On a more positive vein, Jackson writel,

... it may be possible for artificial intelligence to greatly reduce theamount of human labor necessary to, operate theeconomy of the 'world Computers and AI research may play an important part in helping toovercome the food, population, housing, and othercrises -that currently grip the earth... 33

The following lines from Locks/ey Hall,by Alfred, Lord

Tennyson, are representative ofthis author's feelings towards his

Subject:

. For I dipt into the.fUture, faras human 'eye could see Saw the Vision of the World, and allthe wonder that would be.

33. Jackson, 22. cit., p. 398.

V 2Ci 'Lt ScienceTracer. Bullet

Reference Section, Science and Technology Division Library of Congress, 10 First Street, S.E., Washington,D.C.20540

ARTIFICIAL INTELLIGENCE, Compiled by B. J.' Krevitt

TB 73-4 ' January 1973

8 SCOPE: Includes material on the techni61 and philosophical aspects of the question: Can a machine think?The inter- disciplinary nature of the topic is reflected by its appearance in the literature of the engineering sciences, mathematics, linguistics, psychology and the biological

sciences. 4 .

An introduction to the topic appears in:

Davis, Ruth M. Artificial intelligence. Jn,Encyciopedia of library and information science. v. 1.--New York, M. Dekker, 1968.* p. 627-632. PZ1006.E57Vertical file*

SUBJECT HEADINGS under which books on artificial intelligence can be located in the LC card catalogs include the following:

ARTIFICIINTELLIGENCE (Highly relevant) CYBERNETICS (Highly relevant) BIONICS.(Relevant) CONSCIOUS AUTOMATA'(Relevant) SELF-ORGANIZING SYSTEMS (Relevant) INFORMATION THEORY-(More general) MACHINE TRANSLATING (Related) PATTERN RECOGNITION (Related) PERCEPTRONS (Related)

BASIC TEXTS:

Dreyfus, HArt L. What computers can't do: a critique of artificial reason. New York,'Harper and Row, 1972. 259 p. Q335.D74 1972*

Findler, N. V., and Bernard Meltzer, eds. Artificial intelligence and heuristic programming. New York, American Elsevier, 1971 37 p. Q335.A787

*Available in the Science Reading Room Collection "* *Available in the Science Reading Room Microform Collection

t 7% ,./...ceigenbaum',.Eaward, and Julian Feldman. Computers and thought. New York, McGraw-Hill, 1,26.3. 535 p. Q335.5.F4*

Jaki, Stanley L. Brain, mind and computers.' Olew York, Herder and Herder, 1969. 267 p. Q335.J3*

Taube, Mortimer', Computers and common sense. New York, Columbia University Pres.s, 1961. 136 p. Q31.0.T33

ADDITIONAL TEXTS:

kiener, Norbert. Cybernetics, or control and communication in' the animal and,the machine. New York, J. Wiley, 1948. 194 p. Q175.W6516*

. The human use of human beings; cybernetics and society. Boston, Houghton Mifflin Co., 1950., 241 p. Q175.W6517* e. Other books including material on engineering cyberneticsare shelved under the following LC call numbers: TA166-168 and TJ212-225.

HANDBOOKS, ENCYCLOPEDIAS and DICTIONARIES which contain informationon artificial intelligence include:

EnCyclopaedia of cybernetics. Translated from the German Lexikon der Kybernetik by G. Gilbertson. Manchester, Manchester U. P:; New York, Barnes and Noble, 1968. Q310.L413 1968*

Meetham, A. R. Encyclopaedia of linguistics, information, and' control. Oxford, New York, Pergamon Press, 1969. 718 p. Q360.M35 1969*.

BIBLIOGRAPHIES:

Alum, Fazlul. Cyberneticsi,,automation, computers, control, ergonomics, information 'theory, and machine translation: 'a subject guide: Provisional ed. London, New Science Publications, 1968. 110 p. 27405.C9A4*

Campbell, Alan. Bionics and biocybernetics. Washington, Aero- space Technology Division, Library of Congress, 1968. 366 p. (ATD'report 68-77-108-4) Z664.23.A2

Dutton,. John M.: and William H. Starbuck.Computer simulation models of human behavior: a history of an intellectual technology. In Institute of Electrical and Electronics Engineers. Systems, Man and Cybernetics Group. IEEE transactions on systems, man and cybernetics, v.SMC-1, An. 1971: 128-171. Q300.143

28 ^-1

-3-,

Gibbs, Katye MI Bionics and related research; a report bibliography. Arlington, Va., Armed Services Technical Information Agency, 190. 177 p.. AD 294 150Vertical file

Sweitzer, Dorothy.I. Biological and artificial intelligence. Pasadena, Jet Propulsion Laboratory, California Instituteof Technology, 1960. 185 p. (Astronautics information literature search, no. 254) 27405.A7S9Vertical file*

. Supplement. Pasadena, Jet Propulsion LaboratOry, California Institute of Technology,1961. 160 p. Z7405.A7S9 Suppl. Vertical file*.

SELECTEDTECHNICAL REPORTS:. I Illinois. University. Biological Computer Laboratory. Accomplish- . went summary, 1971-1972. Urbana, 1972. 275 B. AD 744 009**

Massachusetts Institute of Technology.Artificial Intelligence Laboratory. Description and theoretical analysis (using' schemata) of planner: a language for proving theoremsand. manipulating models in a robot.Cambridge, Mass., 1972. 403 p., AD 744 620**

Stanford University. Computer Science Department. Artificial Intelligence Project. Project technical report. Stanford, Calif., 1971. 1 v. AD 724 867** .

System Development Corporation, Santa Monica, Calif.Problem solving and learning by man-machine teams. Santa Monica, 1971. 229 p. AD 729 070**

STATE-OF-THE-ART REVIEWS and CONFERENCE PROCEEDINGS:

American Society for Cybernetics. Purposive systems; proceedings of the first annual symposium of the American Societyfor Cybernetics. New York, Spartan Books, 1969, c1968. 179.p. Q300.A45

American Society for Cybernetics.Cybernetics and the management of large systems; proceedings of the secondannual symposival of the American Society for Cybernetics. New York, Spartan 'Books, 1969. 264 p. Q295.A45

American Society for Cybernetics.Cybernetics, simulation, and conflict resolution; proceedings of the third annualsymposium of the American Society for Cybernetics. New.York, Spartan Books, 1971. 249 p. HM136.A58 -4-

American Society for Cyhrnetics. Cybernetics, artificial intelligence, and ecology; proceedings of the fourthannual symposium of the American Society for Cybernetics. New York, ' Spartan Books, 1972. 333 la: Q335.A444

International, Joint_Conference on Artificial Intelligence,1st, Washington, D.C., 1969. Proceedings. gedford?'Mass., 1969. 715 p.. Q335.5.155 1969 -

Machine intelligence. Proceedings of the MachineIntelligence. Workshop. 1st, 1967-' New York, American Elsevier,A335.M27 JOURNAL ARTICLES andother literature on artificial intelligenCeare indexe4/primarily in the following abstracting andindexing services: --- Computer Abstracts (1957-)Z6654.C17C64* --- ^ See: Artificial Intelligence (Section 4) and index--

Computer & control Abstracts(19,66-)* See: Artificial Intelligence(Section 61.40).

Computer and InformationSystems (1962-) QA76.146* See: Artificial Intelligence(Section CA2)Nandindex:

Engineering Index (1884-)Z5851.E62*

See: Cybernetics , -

. , Quarterly BibliographyofComputers and Data Processing QA76.Q3* (1971-) See: Artificial Intelligence

Science Citation Index(1961-) Z7401.Z87*- See especially Permuterm Subject Index (1966-)for, entries under ArtificialIntelligence

. . Other indexes, listed here, should be used foran exhaustive search. Only a limitedreturn can be expected for the time spent. Directions are generally given in the front of each issue: -

Computing Reviews (1960-)QA76.C5854* Cybernetics Abstracts (1964-)Q300.R413*

REPORTS and other types of literatureare indexed in the following guides:

Scientific and Technical Aerospace Reports (1963-)TL500.S35* See: Artificial Intelligence Cybernetics r ; . Note: Consult reference librarian for location ofabstracting and indexing services inthe Science ReadingRoom. 30 -5-

government Reports Index (formerly U.S. GovernmentResearch and Development ReportsIndex) (1946-) Z7405.R4q513* See: Artificial Intelligence Cybernetics

Rand Corporation. Selected Rand Abstracts (1963-)AS36.R284* See: Artificial Intelligence Cybernetics /

PRIMARY JOURNALS that often containearticles relevant to artificial intelligence are:-

Artificial intelligenceQ335.A785 Association for ComputingMachinery. Journal QA76.A77 IEEE transactionsOn information theory Q350.I2 Institute of Electrical andElectronics Engineers. IEEE transactions on computersTK7885.A1F2 Institute of Electrical andElectronics Engineers. IEEE transactions on systems science and cybernetics(1965-1970) TA168.122 Institute of Electrical and ElectroniCs Engineers. Systems, Man and Cybernetics Group. IEEE transactionson systems, man and cybernetics (1971-) Q300.143

SELECTED MATERIALS available in the Science ReadingRoom vertical file include:

The age of thinking machines. Business week, no. 2200,Oct. 30, 1971: 90-92+

Hamming, Richard W. The computer and the intellectualfrontier. Computers and automation,v. 21, June 1972: 25-29+

Kugel, Herbert C., Computer intelligence andgeneral software development. Softwar&age, v. 4, Oct./Nov.1970: 14-15. ) Thring, M. R. Robots in home and industry. In 'American Society for Cybernetics. Cybernetics, artificialintelligence, and ecology; proceedings of the kourthannual symposium of the American Society for,Cybernetics. New York, Spartan Books, 1972. p. 253-268.

Turing, A. M. Intelligent machinery. In Machine intelligence5; proceedings of the machine intelligence workshop." NewYdrk, American Elsevier, 1970. p. 3-23. Q335.M27

31 BTBTJOCAPHY

Arbib, Michael A. The Metaphorical Brain. An Introduction to Cyber- netics as Artificial Intelligence and Brain Theory. New York, -Wiley-Interscience, 1972.

Banerji, Ranan B.Theory of Problem Solving. An Approach to Artificial Intelligence. New York, American Elsevier Publishing Company, Inc.,.1969.

Barrow, H. G., et. al. "Tokyo-Edinburgh Dialogue on Robots in Artificial .Intelligence."The Computer Journal, volume 14, February (1971, pp. 91-95.

Bobrow, Daniel and Raphael, Bertram. "New Programming Languages for ArtifiriA1 Intelligence Research."Computing Surveys, volume 6, SepteMber 1974, pp. 253-174.

Bush, Vannevar. "As We May Think."The Atlantic Monthly, volume 176, July 1945, pp. 101 -108.

Cote, Jr., Alfred J. The Search for the Robots. New York, Basic Books, Inc., Publishers, 1967.

Davis, Ruth M. "Artificial Intelligence." in Encyclopedia of Library and Information Science. Volume 1. New York, Marcel Dekker, Inc., 3968.

"Direct Brain-to-Computer Interface on the Way." Computer Decisions, September 1974, p. 13.

Dreyfus, Hubert L. What Computers Can't Do. A Critique of Artificial Reason. New York, Harper & Row, Publishers, 1972.

Feindel, William, ed. Memory, Learning', and Language. Toronto, University of Toronto Press, 1960.

Fink, Donald G. Computers and the !AL-Ian Mind. An Introduction to Arti- ficial Intelligence. Garden City, Anchor Books, 1966.

Frank, George. "What is Artificial Intelligence?"Data Processing, volume 14, September- October 1972, pp. 328-330.

Jackson, Jr., Philip C. Introduction to Artificial Intelligence. New York, Petrocelli Books, 1974.

Jaki, Stanley L. Brain, Mind and Computers. -New York, Herder and Herder, 1969.

32 Krevitt, B. I., comp."Artificial Intelligence." LC Science Tracer Bullet, TB 73-4, January 1973. as Licklider, V. C. R.Libraries of the Future. Cambridge, The M.I.T. Press, 1965.

Michie, Donald. 'Heuristic Search." The Computer Journal, volume 14, February 1971, pp. 96-102.

Minsky, Marvin L. "Artificial Intelligence." Scientific Americas, 1 . September 19.66, pp. 246-252+.

. Semantic Information Processing. Cambridge, the M.I.T. Press, 1968.

Nagel, Ernest and Newman, aames.R. "Goedel's Proof." in The World of Mathematics.' Volume.3, pp. 1668-1695.

Newman, James R., ed.The World of Mathematics.' New York, Simon and Schuster, 1956.

"Nilison, Nils J.Problem7polvir9 Methods in Artificial Intelligence. New York, McGraw -Hill. Book Company, Inc., 1971.

Shannon, Claude E. "A Chess-Playing Machine." in The World'of Mathematics. Volume 4, pp. 2124-2133.

Taube, Mortimer. Computers and Common Sense. New York, McGraw-Hill

Book Company, Inc., 1961. .

Turing, A. M. "Can A Machine Think?" in The World of Mathematics. Volume 4r pp. 2099-2123.

Von Foerster, Heinz, et. al. eds. Purposive Systems.Proceedings of the First Annual Symposium of the American Society for Cybernetics. New York, Spartan Books, 1968.

Von NeuMann, John. The Computer and the Brain. New Haven, Yale University Press, 1958. I

. ',The General and Logical Theory of Automata." in The World ,of Mathematics. Volume 4, pp. 2070-2098.

3,;