7 DOCUMENT RESUME ED 069 296 LI 003 962 AUTHOR Larry H. Chandrasekakan, B.; Reeker, TITLE Artificial Intelli ence - A Case for Agnosticism. INSTITUTION Computer and Information Ohio State Univ.., olumbus. Science Researbh C nter. i REPORT NO OSU -C ISRC -TR- 72 -9 PUB DATE Aug 72 30p.;(13 References) NOTE ( EDRS PRICE MF-30.65 HC -$3.29 DESCRIPTORS *Artificial Intelligence; Computer Programs; *Computers; *C puter Science; Information Processing; Linguistics; Machine *Intelligence; Translation ABSTRACT Almost all'of the ublished work on the philosophical question of mechanical intelligen e has argued in favor of one or the other of the polarities of possfb e,positions. It i$ however possible to take a position between these two extremes, those of the True Believer' and the Infidel. While this agnostic position is not a strong position in the sense of hiving a good many logically compelling arguments in its 'favor, it gains its viability by virtue of the weaknesses in the po ar alternatives. Although agnosticism concerning the possibility f artificial intelligende is believed to be a position held by many omputer scientists, there has been no attempt to present this pos tion in the literature. In an attempt to rerdedy the situaonc this eport presents the agnceticl.s position regarding inte gence..(Author/NH) artifikitial ./ Pt L , U.S DEPARTMENT OF HEALTH. EDUCATIONS WELFARE OFFICE OF EDUCATION THIS DOCUMENT HAS SEEN REPRO. DUCED EXACTLY AS RECEIVED FROM THE PERSON OR ORGANIZATION ORIG. .(0SU-CISRC-TR-72-9) MATING IT POINTS OF VIEW OR OPIN. -r IONS STATED DO NOT NECESSARILY REPRESENT OFFICIAL OFFICE OF EDU- 8 CATION POSITION OR POUCY ARTIFICIAL INTELLIGENCE - A CASE FOR AGNOSTICISM hby B. Chandrasekaran and !carry H. Reeker r % The CompUterand Inf4iation Scien0e Research Center The Ohio State University 44210 Columbus ,1:1Aio . . .August 1972 4, FILMED FROM BEST AVAILABLE COPY 1 PREFACE The Computer and Information Science Research tenter of The Ohio State University is an interdisciplinary research orgpnization which consists of the staff, graduate students, and faculty of many University departments and laboratories. This report is. based on research accom-. plished in the Department of Computer and InfOrumtion Science. V r. 4 ; °Or c co& . 0 2 I. ARTIFICIAL INTELLIGENCE - A 'CASE FOR AGNOSTICISM B. Chandtasekaran.and Larry H. Reeker \ t ellineme Charlie Brown: S1o,p wishy-washy. -I don't` like' it Lucy Oh, yeh? O.K., tell me, can Machines think? . . .Charlie Brown: Well, er, let me'see; on the one hand, but ... Lucy Wishy-washyt 1 0 1 1 . . . vq.i '-' ;.!. .ii - - 1 . -Ni . ) i A.preliminariver ion of this paper-Vas preqgnxed at a special , session on "The P ssikilities and Limitation*. of Artificial. intern- 1 .gence", during the 1911 IEEE Systeis, Man, and Cybernetics Conference .., , in Anaheii California, October, 19/1 5'. .1 N \ Agnosticism, whether in the matter of existence of God or in the matter . 1 has an aura of excessive caution - the of computer simulation of the mind a Is there The juxtaposition: very opposite of bold staking out of'positions. Is the mind a machine? is interesting for another reason -' a soy of a God? empirical semi -decida ility that they share. The question about God Could pre- the affirmative by any.given injlividual to whom He sumably.be answered i Likewise, a. person's doubts about ro- chose to provide sufficient evidence.. bots would probably vanish if.his best friend, about whom no suspicion had The point is that it crossed his brow, turned out to be a clever artifact. is possible to conceive of direct, evidence in favor of the existence of GOd and man-machine equivalence, and difficult to imagine such direct evidence which would Compel belief in atheism or in the impossibility of machines ever being built to simulate Min in all his complexity. ither case at. very good in The prospects for direct evidence are not, 4 In the matter of mechanical simulation of the human mind the in- this time. some computer pfograss direct evidence that we presently have consists of: 3 which purport to show some of Vie performance characterist'ics of the'mind, eN arguments attempting tb show the implausibility of a machine ever being de- signed fulfilling said purpdse, arguments setting forth the likelihood of succeeding in the enterprise of robot - building, and a mass ofdata on the human mind itself. . Almost all of the published work'on the philosophical question of the limitations of mechanical intelligence has argued in .favor of one or the other of the polarities of possible positions; It is however possible to take a . positidh betWeen these two extremes; thoie of the True Believer and the.Infidel. 2. J, - of having While this agnostic position is not a strong position in the.sense (this seems to be a a good many logically compelling arguments in its favor of the 'feature of agnosticism in general), it gaint its viability by virtue We believe agnosticism concerning the weaknesses in the polar alternatives. possibility of artificial intelligence to be a position heldby many computer / in'the Yet there has been no attempt to present the position scientists. This we attempt tokremedy below. literature. . corner, possibly about to fail, is causing [The computer room. a light characters are the computer to cast an intermittent Zong shadow' in which standing and talking about "intelligent machines.] I see that they are holding another compu er:chess tourna- 'True Believer: .Why, scoffers once claimed that machines wo bd never p ay a ment: Zy sophisticated decent'game of chess, but we now haVe some remarik programs. I doubt that they'll soon be' international gr d masters; but Infidel: then, that's beside the point any0ay. oni a matter of True Believer: It certainly is! Why, deli gree whether they are good beginning plaT3ers or randMasters. The programs are being improved regularly. excellent end They pZ frames andostandard openings, and their Zook-ahe d capabilities are... . . . Infidel: (interrupting]: EXcuse my interruption; but you still are missing the point: even if they become grhndmdstera, they are II .4 4 wha: should machines being some arzl People' like you are always .talking as intelligent designer. if machines can think, when you don't even know what is meant. by "tiiink." As a matter of fact, it is not necessary to worry True, Believer: Turing's tesi,4 the "Imi- about the definition..of "think." tation Game," is asgood an. operational definition' as you It is going to be mighty difficult for anyone might need. which passes Turing's test has to deny that a machine captured the -essence of 'thinking. .. Infidel: Certainly any one who thinks can do more than play games, be they ches's or "imitation games." As an exercise in pro- gramming fireworks, Turing's test may be something to wotk towards; but as for "capturing the essence of thinking,'" I think it is presumptuous. "think,"1 Unless somebody defines I think it is unrealistic of you people... 'r-- [interrupting]: You challenge a pod 'operational True Believer: definition ir2 favor of some puddle-headed intuitiari: "[Shouting, It's people. Zike you who jeered at and beginning to flush] the steamboat) do there always have to be a bunch of Neandir... Why, 4st Agtiostic [joining the two and interrupting]: flop no; :fellows, Zet us I can understand the. vipw that the imitation not come to blows. call`' game does not embody every kind pf activity that -one.might Turing evaded that problem by sayi "The "thinking." . .st 'can machinesthink?' is takAedningless ori4inal question, Still,, the imitation game is not a. to deserve dilussion." If a machine were to pass -that test, it would trivial test. be.evidence.that it is possible to buiZd into a machine con- s siderable mastery over language use, a capability to make sophisticated inferences and deductions, a certain cunning, etc.., all evidences of intelligence if displayed by a human. - Even Look at language behavior, for one area. True Believer: Yes! a very simple pmgram, must capture certain ELIZ4t5 which is since it can manage tar make some aspects of human language ' people, angry when they work' with it, just as they would get angry at a person whO was answering back imperinently. Agnostic:Bold it! I would prefer to say ",just as they would get angry at a candy machine which was refusing to yield its candy." For my point is well illustrated 4y machines which "use natural . language,fleand that point is thatIELIZA is more closely anal- agous.to a candy machines than to a human ,for even, quite pos- I sibly, to a chimpangee) in its real knowledge of language. think One must insist that what the machine captures' is not some purely superficial. aspect -of some type of behavior taken ,4 to be indicative orthoughi. But Zet us return to the imitation I game. have my own reasons for considei.ing the imitation game .1 . .a poor operational definition. .. 5. . . rkar? While Turing's test provides us with a standard, or cri- Agncs::::c: terion, for a thinking machine, are we any nearer to designing than we would be without the test? intelligent machines What jusea criterion for thinking; what is needed is mare th those of us who are interested in building "intelligent" programs are Lacking is a knowledge of what specific capabilities.we . should build in. 1,1.* V True Belie0Or: to want is a recipe, something like a What y 0 ; generative finition of thinking. Infidel: What if# thinking was just not satisfactorily definable in this sense, in principle? There are some very important mathematical objects that have this pioperty, you know. Take for-instance, the Set of total recursive functions. Such functions are common enough objects; yet if you try to give a recipe for producing them, you can always find one that's not produced by that. recipe.. 'True Believer: What you.are saying is very similar to the hopes some of you, were nurturing a while ago about how GUdel's theorem was 01424 to resulipeCt.what'yoU like to'can sn's dignity" from the clutches Of us mechanists. Sure there are limitations to machines, i . but as Putnam6 and Benacerraf7 and possibly others hay; indicated, there is no evidence that humans do not po ess similar Jimitationa. Granted, there i = no recipe to produce' all of.the 4 I' ip, drions. But you baue.yet to convince metRat total recursive ct .( the attributes of "thought" form a productive set. I agree that it is futile to 'look toOdel's theorem for Agnostic: proof that the mind is not a machine.8 But the point about the difficulty in obtaining a recipe is not so easily dismissed. For example, as Gel himself The problem is an acute design problem. involved as pointed out, very little is) known about the processes . in the human ability to come up with more and more precise distinctions eadied.for instance, in stronger' and stronger as In order to understand the set theoretic axioms of infinity. "mechanisms" involved,,)e says, a major advance in the foundations of mathematics is needed. Surely at this .stage it is unrea- sonable to insist that "thinking" machines are 'just around One might as well claim the corner." that the ultimate advances in the foundations of mathematics is imminent. * We excldde from consideration arguments.%7hich Mind a machine? Is nr similar essentially revealed involve appeal to'religious convictions Among the mulbiplicitof possible posi- knowledge of the nature of Man. tions, the most interesting ate: 10 The mind is not a formal systamf and although one might be '1. table to'program some 'superficial aspects ofrbehavior, the really important and significant aspects of intelligence - the canno t be captured in a formal ones that stamp us. "human" - sy stem. 7. "t