Is the Brain a Digital Computer? Author(s): John R. Searle Reviewed work(s): Source: Proceedings and Addresses of the American Philosophical Association, Vol. 64, No. 3 (Nov., 1990), pp. 21-37 Published by: American Philosophical Association Stable URL: http://www.jstor.org/stable/3130074 . Accessed: 19/08/2012 03:34 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. . American Philosophical Association is collaborating with JSTOR to digitize, preserve and extend access to Proceedings and Addresses of the American Philosophical Association. http://www.jstor.org IS THE BRAINA DIGITALCOMPUTER? JohnR. Searle ofCalifornia/Berkeley University Address theSixty-fourth Presidential delivered AnnualPacific Division ofthe before Meeting Association inLosAngeles, March30, 1990. American California, Philosophical I. Introduction, Strong AI, WeakAI andCognitivism. Therearedifferent a Presidential totheAPA;theoneI have Address waystopresent chosenissimply toreport onworkthatI amdoingright I am now,on workinprogress. someofmyfurther intothecomputational modelofthe goingto present explorations mind.[1] Thebasicideaofthecomputer modelofthemindisthatthemindistheprogram and thebrainthehardware ofa computational A sloganoneoften seesis "themind system. is to thebrainas theprogram is to thehardware." [2] Letus beginourinvestigation ofthisclaimbydistinguishing threequestions: 1. Is thebraina digital computer? 2. Is theminda computer program? 3. Can theoperations ofthebrainbe simulated on a digital computer? I willbe addressing answered in the 1 andnot2 or3. I think2 can be decisively Sinceprograms aredefined or and since minds have negative. purely formallysyntactically an intrinsic mentalcontent, it follows that the itself cannot immediately program by constitute themind.The formal oftheprogram doesnotbyitself the syntax guarantee of mental contents. I showed this a decade the in Chinese Room presence ago Argument meforexample, couldrunthestepsoftheprogram forsome (Searle,1980). A computer, mental as such without a word of Chinese. Chinese, capacity, understanding understanding Theargument rests on thesimple thatsyntax isnotthesameas,noris itby logicaltruth itself sufficient semantics. So theanswer to thesecondquestion isobviously "No". for, The answerto 3. seemsto me equallyobviously at on least a natural "Yes", Thatis,naturally thequestion means: Is there somedescription interpretation. interpreted, ofthebrainsuchthatunderthatdescription could do a simulation of you computational theoperations ofthebrain.Butsinceaccording toChurch's that can be thesis, anything as a setofstepscan be simulated on a digital givena precise enoughcharacterization itfollows that the has an affirmative answer. The computer, trivially question operations ofthebraincan be simulated on a digital in thesamesenseinwhichweather computer thebehavior oftheNewYorkstockmarket orthepattern ofairline over systems, flights LatinAmerica can. So ourquestion isnot,"Istheminda program?" Theanswer tothat 21 22 APA PROCEEDINGS, VOL. 64,NO.3 to thatis,"Yes". The The answer is,"No". Noris it,"Canthebrainbe simulated?" I am of "Is the brain a And for is, purposes thisdiscussion question digital computer?" that as to: "Are brain computational?" processes taking question equivalent ifquestion2 wouldlosemuchof itsinterest thinkthatthisquestion One might unless the mind is a program, a negative answer.Thatis,onemight that receives suppose Butthatisnotreally thebrainisa computer. tothequestion whether thereisnointerest of arenotconstitutive thecase. Evenforthosewhoagreethatprograms bythemselves to there is more the Granted that an there is still mental question: phenomena, important be the itmight ofthedigital mindthanthesyntactical nonetheless, computer; operations are statesand mentalprocesses case thatmentalstatesare at leastcomputational mental of states. formal these over the structure This, operating computational processes ofpeople. takenbya fairly in fact,seemstometheposition largenumber likethis:At thattheviewisfully I amnotsaying clear,buttheideaissomething thereareso tospeak,"sentences aresyntactical; brainprocesses somelevelofdescription in the butperhaps or Chinese, in English in thehead". Theseneednotbe sentences a have like of sentences, (Fodor,1975). Now, any syntactical they "Language Thought" from ofsyntax can be separated andtheproblem ormeaning, anda semantics structure in the is:Howdo thesesentences ofsemantics The problem ofsemantics. theproblem of the headget theirmeanings?But thatquestioncan be discussed independently that to answer A sentences? work in these brain does the How typical processing question: latterquestionis: The brainworksas a digitalcomputer computational performing inthehead. structure ofsentences overthesyntactical operations a mind I calltheviewthatallthereistohaving tokeeptheterminology straight, Just can mental brain the view that a ishaving program, (and AI, processes) processes Strong a brain is the that Weak the view and besimulated computer, digital computationally, AI, Cognitivism. whatmotivates andI hadbetter ThispaperisaboutCognitivism, sayat thebeginning and Nicholls or Kuffler brain about the books read it. Ifyou (1976)) (1983) (sayShepherd oninthebrain.Ifyouthenturntobooksabout ofwhatisgoing picture yougeta certain ofthe ofthelogicalstructure 1989)yougeta picture (sayBoolosandJeffrey, computation about books then turn to Ifyou ofcomputation. science, (sayPylyshyn, cognitive theory is reallythesameas whatthe 1985)theytellyouthatwhatthebrainbooksdescribe to thisdoesnotsmellright booksweredescribing. Philosophically speaking, computability of sense to follow an of at the at least meandI havelearned, my beginning investigation, smell. II. The PrimalStory as I canwhyCognitivism tostateas strongly I wanttobeginthediscussion bytrying ofhuman the relation is a about There intelligence hasseemed story intuitively appealing. classicpaper(1950),andI believeitis thatgoesbackat leasttoTuring's tocomputation view.I willcallitthePrimal oftheCognitivist Story: thefoundation thesis(sometimes inmathematical We beginwithtworesults logic,theChurch-Turing theChurch-Turing calledChurch'sthesis)and Turing'stheorem.Forour purposes, thereis someTuringmachinethatcan implement thesisstatesthatforanyalgorithm PRESIDENTIALADDRESSES 23 thatalgorithm. theorem isa Universal Machine which Turing's saysthatthere Turing can simulate Machine. Now if we we have these two the anyTuring put together result thata Universal Machine can whatever. implement Turing anyalgorithm Butnow,whatmadethisresult so exciting? Whatmadeitsendshivers upanddownthe a of whole of workers in artificial is thefollowing spines generation young intelligence the brain is a Universal Machine. thought: Suppose Turing forsupposing thebrainmight bea Universal Well,arethere anygoodreasons Turing Machine?Letus continue withthePrimal Story. It is clearthatat leastsomehumanmental arealgorithmic. abilities Forexample, I canconsciously dolongdivision the of an for bygoing through steps algorithm solving It is furthermore a consequence oftheChurch-Turing thesis longdivision problems. andTuring's theorem thatanything a humancando algorithmically can be doneon a Universal Machine.I canimplement, forexample, theverysamealgorithm Turing thatI use forlongdivision on a digital In a case,as described such computer. by both the human and the mechanical are I, Turing(1950), computer, computer thesamealgorithm; I amdoingitconsciously, themechanical implementing computer Now it seems reasonable to there alsobe a wholelot nonconsciously. suppose might ofmental oninmybrain which arealsocomputational. processes going nonconsciously Andifso,wecouldfindouthowthebrainworks theseveryprocesses bysimulating ona digital as wegota computer simulation oftheprocesses fordoing Just computer. sowecouldgeta computer simulation oftheprocesses forunderstanding longdivision, visualperception, etc. language, categorization, "Butwhataboutsemantics? After arepurely Hereanother set all,programs syntactical." oflogico-mathematical results comesintoplayin thePrimal Story. The development ofproof showedthatwithin certain wellknownlimits the theory semantic relations between can be entirely mirrored propositions bythesyntactic relations between thesentences thatexpress thosepropositions. Nowsuppose that mental contents intheheadareexpressed inthehead,thenallwewould syntactically needtoaccountformental wouldbe computational the between processes processes inthehead. Ifwegettheproof elements thesemantics will syntactical theory right takecareofitself; andthatiswhatcomputers do:theyimplement theproof theory. We thushavea welldefined research We trytodiscover theprograms program. being in thebrainbyprogramming to implement thesameprograms. implemented computers We do thisin turnbygetting themechanical tomatchtheperformance ofthe computer humancomputer thepsychologists tolook (i.e.topasstheTuring Test)andthengetting forevidence thattheinternal arethesameinthetwotypes ofcomputer. processes in Now whatfollows I wouldlikethereader tokeepthisPrimal inmind--notice Story of theprogram especially Turing'scontrastbetweentheconsciousimplementation bythe humancomputer and thenonconscious ofprograms, whether implementation bythebrain or by the mechanicalcomputer; noticefurthermore the idea thatwe mightjustdiscover 24 APA PROCEEDINGS, VOL. 64,NO.3 in nature, theverysameprograms thatwe putintoourmechanical running programs computers. onefinds certain common Ifonelooksat thebooksandarticles Cognitivism supporting often assumed it is thatthe but nonetheless often First, unstated, assumptions, pervasive. form of is some dualism. to the view the brain is a alternative that digital computer only ofimmortal Cartesian Theideais thatunless intheexistence souls,youmust youbelieve believethatthebrainis a computer.Indeed,it oftenseemsto be assumedthatthe ourmentalstatesand thebrainis a physical mechanism whether determining question the are the same whether thebrainis a digital speaking, computer question.Rhetorically thatunlesshe acceptstheideathatthebrainis intothinking ideais to bullythereader the views.Recently he is committed tosomeweird somekindofcomputer, antiscientific von notbe an old fashioned fieldhas openedup a bitto allowthatthebrainmight kindofparallel a moresophisticated butrather Neumann processing digital computer, style is to risklosing that the brain is to Still, deny computational computational equipment. inthescientific community. yourmembership arecomputational whether brainprocesses is it also assumed thatthequestion Second, in thesame is be factual It to settled is justa plainempirical investigation by question. is or whether a as whether the heart that such greenleavesdo pump way questions or is no room for fact. as of There were settled matters logicchopping photosynthesis fact. Indeed I hard scientific about matters of we are since talking analysis, conceptual an of this the title doubt that work in this field would think who paperposes manypeople isnomore a digital at all. "Isthebrainreally computer?" question appropriate philosophic at neuro-muscular "Is neurotransmitter the a philosophical than really junctions question acetylcholene?" seem andDreyfus, suchas Penrose toCognitivism, Evenpeoplewhoareunsympathetic aboutthe issue.Theydo notseemto be worried to treatit as a straightforward factual ButI ampuzzled be thattheyaredoubting. whatsortofclaimitmight bythe question a its of could constitute sort fact about the brain What being computer? question: even is thehasteand sometimes feature of thisliterature another Third, stylistic are over. What are the withwhich foundational carelessness exactly the questions glossed is discussed? What ofbrains thatarebeing andphysiological features anatomical exactly to connect? to these two a digital Andhowaretheanswers supposed questions computer? aboutO's and in thesebooksand articlesis to makea fewremarks The usual procedure andthengetonwiththemore oftheChurch-Turing thesis, summary l's, givea popular in readingthis andfailures.To mysurprise achievements suchas computer exciting things hiatus.On theone thatthere seemstobe a peculiar literature I havefound philosophical resultsranging fromTuring'stheorem hand,we have a veryelegantset ofmathematical On theother torecursive function thesis toChurch's hand,wehavean impressive theory. set of electronicdeviceswhichwe use everyday. Since we have such advanced musthave we assumethatsomehowsomebody and such goodelectronics, mathematics to theelectronics.But themathematics workofconnecting done thebasicphilosophical we are in a peculiarsituation as faras I can tellthatis not thecase. On thecontrary, on such absolutely the wherethereis littletheoretical among practitioners agreement is a symbol? What a is What fundamental as, exactly exactly digitalcomputer? questions are two conditions what Under Whatexactlyis a computational exactly physical process? same the program? systems implementing PRESIDENTIALADDRESSES 25 ofComputation m. The Definition Sincethere isnouniversal onthefundamental I believe itisbest agreement questions, togobackto thesources, backto theoriginal Alan definitions Turing. givenby toTuring, a Turing machine cancarry outcertain According elementary operations: It can rewrite a 0 on itstapeas a 1,itcanrewrite a 1 on itstapeas a 0, itcanshift the to the or it can shift the to the controlled It is 1 left, 1 tape square tape square right. by a program ofinstruction andeachinstruction a condition andan actionto be specifies carried outifthecondition is satisfied. Thatis thestandard ofcomputation, itis at leasta bit definition but,takenliterally, If home are most to misleading.youopenupyour you computer unlikely findanyO'sand 1'sorevena tape. Butthisdoesnotreally matter forthedefinition. To findoutifan is a it turns out that we do not have object really digital computer, actually tolookforO's andl's, etc.;rather wejusthavetolookforsomething thatwecouldtreat as orcount as orcouldbeusedtofunction as O'sandl's. Furthermore, tomakethematter morepuzzling, itturns outthatthismachine couldbemadeoutofjustaboutanything. AsJohnson-Laird "It could be made out of like and levers old an fashioned mechanical says, calculator; cogs itcouldbe madeoutofa hydraulic whichwaterflows; itcouldbe made system through outoftransistors etchedintoa silicon whichan electrical flows.It current chipthrough couldevenbe carried outbythebrain.Eachofthesemachines usesa different medium to represent of the or absence of binary symbols--the water,the positions cogs, presence levelofthevoltage andperhaps nerveimpulses" Laird,1988,p. 39). (Johnson Similar remarks aremadebymostofthepeoplewhowrite onthistopic.Forexample, Ned Block(Block,1990),showshowwecanhaveelectrical gateswherethe1'sandO's areassigned tovoltage levelsof4 voltsand7 voltsrespectively. So wemight think that weshouldgoandlookforvoltage levels.ButBlocktellsusthat1 isonly"conventionally" toa certain level.Thesituation morepuzzling whenheinforms us assigned voltage grows further thatwedidnotneedtouseelectricity at all butwecouldhaveusedan elaborate ofcatsandmiceandcheeseandmakeourgatesinsucha waythatthecatwill system at theleashandpullopena gatewhichwecan alsotreatas ifitwerean 0 or 1. strain The point,as Blockis anxiousto insist, is "theirrelevance of hardware realization to Thesegatesworkindifferent computational description. waysbuttheyarenonetheless computationally equivalent" (p.260). Inthesamevein,Pylyshyn saysthata computational sequence couldberealized ofpigeons trained topeckas a Turing machine!" by"a group 1985,p. 57) (Pylyshn, Butnowifwearetrying totakeseriously theideathatthebrainisa digital computer, wegettheuncomfortable result thatwecouldmakea system thatdoesjustwhatthebrain doesoutofpretty muchanything. on thisview,youcanmake Computationally speaking, a "brain" thatfunctions andmineoutofcatsandmiceandcheeseorlevers justlikeyours orwaterpipesorpigeons oranything elseprovided thetwosystems are,inBlock'ssense, You would need an awfullotofcats,or pigeons or "computationally equivalent". just orwhatever itmightbe. The proponents ofCognitivism waterpipes, reportthisresultwith sheerand unconcealeddelight.But I thinktheyoughtto be worriedaboutit,and I am goingto tryto showthatit is justthetipof a wholeicebergofproblems. 26 APA PROCEEDINGS, VOL. 64, NO.3 IV. FirstDifficulty: Syntaxis notIntrinsicto Physics ofmultiple ofcomputationalism notworried Whyarethedefenders bytheimplications thatthe it is functional accounts The answer is that of think realizability? they typical same functionadmitsof multiplerealizations.In thisrespect,computersare just like can be made of brass or steel, so and thermostats.Justas carburetors carburetors indefinite hardware materials. can be made of an of range computers are definedin and thermostats The classesof carburetors But thereis a difference: ofcertainphysical effects.That is why,forexample,nobodysays termsof theproduction out of pigeons. But the class of computersis defined you can make carburetors is a of O's and l's. The multiplerealizability the in terms of assignment syntactically in different can be same achieved the effect not the fact that of consequence physical is arepurelysyntactical.The physics butthattherelevantproperties substances, physical of and state of and of as the irrelevant insofar it admits O's assignments except l's transitions betweenthem. whichmightbe disastrous: But thishas twoconsequences would seem to imply 1. The same principlethat impliesmultiplerealizability of is definedin termsof the assignment If computation universalrealizability. whatever because be a would then anyobject digitalcomputer, everything syntax in made to it. You could describeanything could have syntactical ascriptions termsofO's and l's. to physics. The ascriptionof syntactical 2. Worse yet,syntaxis not intrinsic who treatscertainphysical to an relative is agentor observer properties always as phenomena syntactical. be disastrous? Now whyexactlywouldtheseconsequences howit producesmental the brain to know how works,specifically Well, we wanted that thebrainis a digital be told that to answer question phenomena.And itwouldnot and solar in thesenseinwhichstomach,liver,heart, system, thestateofKansas computer we that we had was are all digitalcomputers.The model mightdiscoversomefactabout is a it show that would theoperationof thebrainwhich computer.We wantedto know in a way were brains which in iftherewas notsomesense intrinsically digitalcomputers hearts or thatgreenleaves intrinsically pumpblood. intrinsically photosynthesis perform word the or us a of It is not matter "pump"to hearts assigning arbitrarily"conventionally" And whatwe were fact of the matter. is an actual There to leaves. or "photosynthesis" that would makethem about brains of the matter fact a that in "Is there way askingis, brains are digital to be that not answer It does told, yes, question digitalcomputers?" a is because everything digitalcomputer. computers ofcomputation, definition On thestandardtextbook 1. For any object thereis some description of thatobjectsuch that underthat is a the digitalcomputer. description object PRESIDENTIALADDRESSES 27 there 2. Foranyprogram issomesufficiently issome objectsuchthatthere complex of the under which it is the Thus for implementingprogram. description object back is now the wall behind the Wordstar my right implementing example becausethereis somepattern of moleculemovements whichis program, the with structure of Wordstar. formal But if the wall is isomorphic implementing Wordstar thenifitisa bigenough wallitisimplementing anyprogram, including inthebrain. anyprogram implemented I thinkthemainreasonthattheproponents do notsee thatmultiple or universal is is a that do not it as a see of much a realizability problem they consequence deeper thatthe"syntax" isnotthenameofa physical likemassorgravity. feature, point, namely On thecontrary andeven"semantic as ifsuch theytalkof"syntactical engines" engines" talkwerelikethatofgasoline or diesel as if it a could be engines engines, just plainmatter offactthatthebrainoranything elseis a syntactical engine. I think itisprobably toblocktheresult ofuniversal possible realizability bytightening ofcomputation. weoughttorespect thefactthatprogrammers upourdefinition Certainly andengineers itas a quirkofTuring's definitions andnotas a realfeature regard original ofcomputation. works Brian Vinod Smith, Goel,andJohnBataliall Unpublished by that a more realistic definition of will suchfeatures as the suggest computation emphasize causal relations and of states,programmabilitycontrollabilitythe amongprogram and situatedness in therealworld.Butthesefurther restrictions on the mechanism, definition ofcomputation areno helpin thepresent discussion becausethereallydeep is thatsyntax isessentially an observer relative notion.Themultiple problem realizability ofcomputationally in different mediawasnotjusta signthat equivalent processes physical theprocesses wereabstract, butthattheywerenotintrinsic to thesystem at all. They on an from outside. We were for some facts of thematter depended interpretation looking whichwouldmakebrainprocesses but the we have defined computational; given way there never could be such facts of the matter. We on the one computation, can't, any that a is if we can a to it and hand,say then anything digital computer assign syntax intrinsic to itsphysical whether or nota supposethereis a factualquestion operation natural suchas thebrainis a digital system computer. Andiftheword"syntax" seemspuzzling, thesamepointcan be statedwithout it. Thatis,someone claim that the notions of"syntax" and"symbols" arejusta manner might ofspeaking and thatwhatwe arereallyinterested in is theexistence ofsystems with discrete andstatetransitions between them.On thisviewwe don't physical phenomena needO'sandI's; theyarejusta convenient shorthand. thismoveis really But,I believe, no help. A physical stateof a system is a computational stateonlyrelative to the tothatstateofsomecomputational orinterpretation. Thesame assignment role,function, ariseswithout suchas computation, and problem algorithm O'sand 1's becausenotions do not name features of statesarenot intrinsic program physical systems. Computational discovered within thephysics, to thephysics. theyareassigned Thisisa different from RoomArgument andI should haveseen argument theChinese it tenyearsago butI didnot. The ChineseRoomArgument showedthatsemantics is not to syntax.I am nowmakingtheseparateand different intrinsic pointthatsyntaxis not intrinsic to physics.Forthepurposes oftheoriginalargument I was simplyassuming that thesyntactical characterization ofthecomputer wasunproblematic. Butthatis a mistake. 28 APA PROCEEDINGS,VOL. 64,NO.3 thatsomething a digitalcomputer Thereis no wayyoucoulddiscover is intrinsically to an observer becausethecharacterization ofitas a digital is alwaysrelative computer to thepurely features ofthesystem.As a syntactical whoassigns physical interpretation thatthethesis ofThought thishastheconsequence hypothesis, appliedto theLanguage unknown thatthereare,intrinsically, is incoherent. Thereis nowayyoucoulddiscover tosomeagentoruser isa sentence inyourheadbecausesomething sentences onlyrelative modelgenerally, the who usesit as a sentence.As appliedto thecomputational ofa physical is a characterization ofa process as computational characterization system an doesnotidentify as computational oftheprocess from andtheidentification outside; characterization. relative itisessentially an observer ofthephysics, intrinsic feature on limits therearea priori I amnotsaying Thispointhastobe understood precisely. a pattern ofevents innature.We couldnodoubtdiscover wecoulddiscover thepatterns on this of thevi program to the implementation in mybrainthatwas isomorphic is to say as a is to that But process say somethingfunctioning computational computer. It the is events a of more than that occurring. requires pattern physical something we some of a might Analogously, by agent. interpretation assignment computational innatureobjectswhichhadthesamesortofshapeas chairsandwhichcould discover be usedas chairs;butwe couldnotdiscover therefore objectsin naturewhichwere themorusedthemas who to relative some as chairs, agents regarded except functioning chairs. V. SecondDifficulty: The Homunculus Fallacy is Endemicto Cognitivism Thishas isnotpartofphysics. at a problem. So far,weseemtohavearrived Syntax is then is defined thatifcomputation theconsequence syntactically nothing intrinsically Is thereanywayoutofthis ofitsphysical a digital properties. solelyinvirtue computer butitisout in taken is a and it there science, is, Yes, cognitive waystandardly problem? in the seen the works I have of the Most and into fire. ofthefrying computational pan on thehomunculus somevariation ofthemindcommit fallacy.The ideaalways theory with.A is to treatthebrainas ifthereweresomeagentinsideit usingit to compute a from as of vision the task describes who Marr is David case (1982) proceeding typical of the visualarrayon theretinato a three-dimensional two-dimensional description the is: Who is reading worldas outputof thevisualsystem.The difficulty external the works on standard and in other Marr's looks it book, Indeed, throughout description? its treat to order in the inside a homunculus to invoke have as we if system subject, as computational. operations genuinely with a problem, is notreally because, feelthatthehomunculus fallacy Manywriters Since is this: The idea be can the homunculus feel that Dennett "discharged". (1978),they intoprogressively canbe analyzed ofthecomputer thecomputational simpler operations itseemsthatthe wereachsimple untileventually "1-0"patterns, units, "yes-no", flip-flop, withprogressively homunculican be discharged stupiderhomunculi,until higher-level at thatinvolvesno realhomunculus we reachthebottomlevelofa simpleflip-flop finally willeliminatethehomunculi. all. The idea,in short,is thatrecursive decomposition It tookme a longtimeto figureout whatthesepeopleweredrivingat, so in case someoneelse is similarly puzzledI willexplainan examplein detail:Supposethatwe have PRESIDENTIAL ADDRESSES 29 a computer thatmultiplies sixtimeseightto getforty-eight. Now we ask "Howdoesit do it?" Well,the answermightbe thatit addssix to itselfseventimes.[3] But ifyouask "Howdoesit add six to itselfseventimes?", theanswermightbe that,first, it converts all and ofthenumerals intobinary foroperating notation, second,itappliesa simplealgorithm on binarynotationuntilfinally we reachthebottomlevelat whichtheonlyinstructions are of the form,"Printa zero,erase a one." So, forexample,at the top level our homunculus six timeseightto get forty-eight". intelligent says"I knowhow to multiply But at thenextlower-level is he replacedbya stupider who says"I do not homunculus but I can do addition."Belowhimare some actuallyknowhow to do multiplication, oneswhosay'We do notactuallyknowhowto do additionormultiplication, but stupider we knowhowto convertdecimalto binary."Belowtheseare stupider oneswhosay"We do not knowanything about anyof thisstuff, but we knowhow to operateon binary whojustsay "Zeroone, symbols."At thebottomlevelare a wholebunchof homunculi zeroone". All of the higherlevelsreduceto thisbottomlevel. Only thebottomlevel reallyexists;thetoplevelsare all justas-if. Variousauthors(e.g.Haugeland(1981),Block(1990)) describe thisfeature whenthey a semanticengine. But we stillmust say thatthesystemis a syntactical enginedriving facethequestionwe had before:What factsintrinsic to thesystemmakeit syntactical? Whatfactsaboutthebottomlevelor anyotherlevelmaketheseoperations intozerosand ones? Without a homunculus thatstandsoutsidetherecursive we do noteven decomposition, havea syntaxtooperatewith.The attemptto eliminatethehomunculus fallacythrough recursivedecomposition to the fails,because the onlyway to get the syntaxintrinsic is to put a homunculus in thephysics. physics Thereis a fascinating featureto all of this. Cognitivists concedethatthe cheerfully levelsofcomputation, 6 times 8" are observer thereis nothing higher relative; e.g."multiply there that to it is all in the really corresponds directly multiplication; eye of the But theywantto stopthisconcessionat the lowerlevels. The homunculus/beholder. electroniccircuit,theyadmit,does not reallymultiply 6X8 as such,but it reallydoes and and these so to But O's manipulate manipulations, speak,add up to multiplication. l's to concedethatthehigherlevelsofcomputation arenotintrinsic to thephysics is already to concedethatthelowerlevelsarenotintrinsic either.So thehomunculus fallacyis still withus. Forrealcomputers ofthekindyoubuyin thestore,thereis no homunculus problem, each useris thehomunculusin question. But ifwe are to supposethatthe brainis a we are stillfacedwiththe question"Andwho is the user?" Typical digitalcomputer, homunculus sciencearesuchas thefollowing: "Howdoesthevisual questionsin cognitive from how does it distance fromsize of systemcomputeshape shading; computeobject retinalimage?"A parallelquestionwouldbe,"Howdo nailscomputethedistancetheyare to travelin theboardfromtheimpactofthehammerand thedensity ofthewood?"And theansweris thesamein bothsortsofcase: Ifwe are talkingabouthowthesystem works neither nailsnorvisualsystems We as outside homunculi intrinsically computeanything. and it is oftenusefulto do so. But you do not mightdescribethemcomputationally, understand hammering by supposingthatnails are somehowintrinsically implementing and you do not understand visionby supposingthe systemis hammering algorithms implementing, e.g.,theshapefromshadingalgorithm. 30 APA PROCEEDINGS, VOL. 64, NO.3 VI. ThirdDifficulty: SyntaxHas No Causal Powers Certainsortsof explanationsin the naturalsciencesspecifymechanismswhich ofthephenomenato be explained.This is especially function causallyin theproduction of disease,the accountof commonin thebiologicalsciences. Thinkof thegermtheory and even the Darwiniantheoryof of inherited the DNA traits, theory photosynthesis, and in each case the is specified, naturalselection. In each case a causal mechanism Now ifyougo back of the mechanism. of the an output specification gives explanation andlook at thePrimalStoryit seemsclearthatthisis thesortofexplanation by promised Cognitivism.The mechanisms bywhichbrainprocessesproducecognitionare supposed thecausesof we willhave specified theprograms to be computational, and byspecifying do notneed is that we often of this research One remarked, program, cognition. beauty Brain in to order to knowthe detailsof brainfunctioning processes explaincognition. buttheprogram ofthecognitive level programs, onlythehardware implementation provide aregiven.On thestandardaccount,as statedby is wheretherealcognitive explanations Newell forexample,thereare threelevels of explanation,hardware,program,and (Newell calls this last level, the knowledgelevel), and the special intentionality level. of cognitive scienceis madeat theprogram contribution aboutthiswhole thenthereis something ButifwhatI havesaidso faris correct, fishy theorywas at least project. I used to believethatas a causal accountthecognitivist's a versionofit thatis coherenteven to false;but I nowam havingdifficulty formulating thesisat all. The thesisis thatthereis a whole thepointwhereit couldbe an empirical in the brain,O's and 1's flashing lot of symbolsbeingmanipulated throughthe brainat to even the mostpowerful naked but to the invisible not and eye only lightning speed is thatthe and it is thesewhichcause cognition.But thedifficulty electronmicroscope, O'sand 1's as suchhave no causal powersat all becausetheydo notevenexistexceptin hasno causalpowersotherthanthose theeyesofthebeholder.The implemented program has no real existence,no ontology, mediumbecause the program of the implementing medium. that the of Physically speakingthereis no such thingas beyond implementing a separate"program level". of the You can see thisif you go back to the PrimalStoryand remindyourself and Turing'shumancomputer.In Turing's difference betweenthemechanicalcomputer andit is functioning levelintrinsic to thesystem humancomputer therereallyis a program is consciously human is the This because to at level to convert that output. input causally this and a certain the rules for explainshis causally computation, doing following the same to mechanical the we when But perform computer program performance. to relative now is a of us, the the interpretation computational computation, assignment to the causationintrinsic outsidehomunculi.And thereis no longera levelofintentional rules,and thisfactexplainshis following system.The humancomputeris consciously any rulesat all. It is following behavior,but the mechanicalcomputeris not literally rules,and so forpractical,commercial designedto behaveexactlyas ifit werefollowing like the tellsus thatthebrainfunctions purposesit does not matter.Now Cognitivism commercialcomputerand this causes cognition. But withouta homunculus,both haveno causalpowers andthepatterns andbrainhaveonlypatterns commercial computer media. So it seemsthereis no wayCognitivism in additionto thoseoftheimplementing couldgivea causal accountofcognition. PRESIDENTIAL ADDRESSES 31 Howeverthereis a puzzleformyview. Anyonewho workswithcomputers even do that often in fact causal knows we that to the casually give appeal explanations can that when I hit For we this I such and such results say key got program. example, becausethemachineis implementing thevi program and nottheemacsprogram; and this lookslikean ordinary causalexplanation.So thepuzzleis,howdo we reconcilethefact that syntax,as such, has no causal powerswith the fact that we do give causal would these sortsof explanationsthat appeal to programs?And, more pressingly, an model for will explanations Cognitivism, theyrescueCognitivism? provide appropriate Couldwe forexamplerescuetheanalogywiththermostats out thatjustas the bypointing notion"thermostat" in causal of to the figures explanations independently anyreference of its so the notion whileequally physics implementation, "program", mightbe explanatory of thephysics. independent To explorethispuzzlelet us tryto makethecase forCognitivism the byextending PrimalStoryto showhowtheCognitivist workin actualresearch investigative procedures is to program a commercial so thatit simulates practice.The idea,typically, computer somecognitive suchas visionorlanguage.Then,ifwegeta goodsimulation, one capacity, that givesus at least Turingequivalence,we hypothesize that the braincomputeris thesameprogram as thecommercial and to testthehypothesis we look running computer, forindirectpsychological evidence,such as reactiontimes. So it seems that we can in exactlythe causallyexplainthebehaviorof thebraincomputer bycitingtheprogram samesensein whichwe can explainthebehavior ofthecommercial computer.Now what iswrongwiththat?Doesn'titsoundlikea perfectly scientific research legitimate program? We knowthatthecommercial conversion of inputto outputis explainedby computer's a program, and in the brainwe discoverthe same program, hence we have a causal explanation. Two thingsoughtto worry us immediately aboutthisproject.First,we wouldnever foranyfunction ofthebrainwherewe actuallyunderstood acceptthismodeofexplanation howit workedat theneurobiological level. Second,we wouldnotacceptit forothersorts ofsystem thatwe can simulatecomputationally. To illustrate thefirst point,considerfor et example,thefamousaccountof"WhattheFrog'sEyeTells theFrog'sBrain"(Lettvin, al. 1959 in McCulloch,1965). The accountisgivenentirely in termsoftheanatomyand of thefrog'snervoussystem.A typicalpassage,chosenat random,goes like physiology this: 1. SustainedContrastDetectors An unmyelinated axonof thisgroupdoesnotrespondwhenthegeneralillumination is turnedon or off. If thesharpedgeof an objecteitherlighteror darkerthanthe and continues backgroundmovesinto its fieldand stops,it dischargespromptly no matterwhattheshapeoftheedgeor whethertheobjectis smalleror discharging, field.(p. 239). largerthanthereceptive I have neverheardanyonesaythatall thisis justthehardware and implementation, thattheyshouldhave figured out whichprogram the frogwas implementing. I do not doubtthatyou could do a computersimulation of the frog's"bugdetectors".Perhaps 32 APA PROCEEDINGS, VOL. 64, NO.3 how the frog'svisual someonehas done it. But we all knowthatonce you understand the irrelevant. level" is works, systemactually "computational just of othersortsof systems.I am, To illustrate thesecondpoint,considersimulations forexample,typingthesewordson a machinethatsimulatesthe behaviorof an old fashionedmechanicaltypewriter. [4] As simulations go, the wordprocessing program betterthananyAl program I knowofsimulatesthebrain. But no simulatesa typewriter sane person thinks:"At long last we understandhow typewriters work,they are in generalthat not the case word It is of processing programs." simply implementations of the causal simulations phenomenasimulated. provide explanations computational simulations So whatis goingon? We do notin generalsupposethatcomputational in place of or in additionto neurobiological of brainprocessesgiveus anyexplanations accountsof how the brainactuallyworks. And we do not in generaltake "X is a relation. That is, we do not simulationof Y" to name a symmetrical computational the typewriter thattherefore a the simulates that because typewriter computer suppose simulatesa simulatesa computer.We do notsupposethatbecausea weatherprogram is the hurricane of of the behavior that the causal hurricane, bythe provided explanation unknown brain where to these should we make an principles exception program.So why the for there are Are anygood grounds making exception?And processes concerned? an is explanationthatcitesa formalprogram? whatkindof a causal explanation from Here,I believe,is thesolutiontoourpuzzle.Once youremovethehomunculus from the outside thesystem, youare leftonlywitha patternofeventsto whichsomeone Nowtheonlysenseinwhichthespecification couldattacha computational interpretation. of the patternby itselfprovidesa causal explanationis thatifyou knowthata certain forthepattern. existsin a system youknowthatsomecauseorotheris responsible pattern if you So you can, forexample,predictlaterstagesfromearlierstages. Furthermore, youcan byan outsidehomunculus, alreadyknowthatthesystemhas beenprogrammed of thehomunculus.You can to theintentionality thatmakereference giveexplanations say,e.g., thismachinebehavesthe way it does becauseit is runningvi. That is like and doesnotcontainany thatthisbookbeginswitha bitabouthappyfamilies explaining becauseit is Tolstoy'sAnna Kareninaand not longpassagesabouta bunchof brothers, The Brothers Karamazov.Butyoucannotexplaina physicalsystemsuchas Dostoevsky's a patternwhichit shareswithits computational or a brainby identifying a typewriter doesnotexplainhowthesystemactually of thepattern becausetheexistence simulation, is at muchtoo higha level thepattern worksas a physical system.In thecase ofcognition to explainsuch concretemental(and therefore of abstraction physical)eventsas the of a sentence. or theunderstanding of a visualperception occurrence work andhurricanes Now,I thinkitis obviousthatwecannotexplainhowtypewriters simulations.Whyis theysharewiththeircomputational bypointingto formalpatterns it notobviousin thecase of thebrain? Here we come to thesecondpartof oursolutionto thepuzzle. In makingthecase thatthebrainmightbe implementing we weretacitly forCognitivism algorithms supposing forcognition,in the same sense that Turing'shumancomputerand his mechanical whichwe have seen thatassumption But it is precisely algorithms. implement computer an whathappenswhena systemimplements to be mistaken.To see this,ask yourself the thestepsof consciously goesthrough algorithm.In thehumancomputerthesystem so theprocessis bothcausalandlogical;logical,becausethealgorithm provides algorithm, ADDRESSES PRESIDENTIAL 33 a setofrulesforderiving theoutput from theinputsymbols; causal,becausethe symbols a the conscious effort to in thecaseofthe is agent making go through steps.Similarly an mechanical the whole includes outside andwiththe homunculus, computer, system homunculus thesystem is bothcausaland logical;logical,becausethe homunculus an interpretation to theprocesses and causal,becausethe of themachine; provides hardware of themachine causesit to go through theprocesses.Buttheseconditions cannotbemetbythebrute, blindnonconscious ofthebrain. operations neurophysiological In thebraincomputer thereis no conscious intentional of the implementation algorithm as there isinthemechanical becausethatrequires anoutside homunculus either, computer toattacha human but there can't be as there nonconscious computer, any implementation is inthecomputational to the we events. The most could findin interpretation physical thebrainis a pattern ofevents whichis formally similar to theimplemented in program themechanical but that as has no causal to call its own computer, pattern, such, powers andhenceexplains nothing. In sum,thefactthattheattribution ofsyntax identifies no further causalpowers is fataltotheclaimthatprograms causal of To the provide explanationscognition. explore ofthis, letusremind ofwhatCognitivist ourselves look consequences explanations actually like.Explanations suchas Chomsky's account ofthesyntax ofnatural or Marr's languages accountofvisionproceed a setofrulesaccording towhicha symbolic bystating inputis converted intoa symbolic In a for case, example,single output. Chomsky's inputsymbol, intoanyoneofa potentially infinite ofsentences number S, isconverted bytherepeated of a set of syntactical rules. In Marr'scase, representations of a two application dimensional visualarray areconverted intothree dimensional of theworld "descriptions" in accordance withcertain Marr's distinctions between the algorithms. tripartite computational solution ofthetaskandthehardware ofthe task,thealgorithmic implementation has Newell's a become famous as statement of the (like algorithm distinctions) general oftheexplanation. pattern Ifyoutaketheseexplanations as I do,it is bestto think ofthemas saying naively, thatit isjustas ifa manaloneina roomweregoingthrough a setofstepsoffollowing rulestogenerate sentences or3D descriptions, as thecasemight be. Butnow,let English us askwhatfactsin therealworldaresupposed to correspond to theseexplanations as to the brain. In for we are not to think that case, example, applied Chomsky's supposed theagentconsciously a setofrepeated of rules;norarewe goesthrough applications to think that he is his theset ofrules. supposed unconsciously thinking waythrough Rather therulesare"computational" andthebrainiscarrying outthecomputations. But whatdoesthatmean?Well,wearesupposed to thinkthatit isjustlikea commercial Thesortofthing thatcorresponds totheascription ofthesamesetofrulesto computer. a commercial is to to the ofthoserulesto the computersupposed correspond ascription brain.Butwehaveseenthatinthecommercial the isalways observer computer ascription theascription is maderelativeto a homunculus relative, whoassignscomputational tothehardware states.Without thehomunculus there isnocomputation, interpretations circuit.So howdo we getcomputation intothebrainwithout justan electronic a homunculus?As faras I know,neither norMarreveraddressed thequestionor Chomsky even thoughttherewas such a question. But withouta homunculusthereis no of the programstates. There is just a physical explanatory powerto the postulation 34 APA PROCEEDINGS, VOL. 64, NO.3 causal levelsof the brain,withitsvariousreal physicaland physical/mental mechanism, description. VII. FourthDifficulty:The BrainDoes Not Do Information Processing to whatI thinkis,in someways,thecentralissuein all In thissectionI turnfinally of this,the issue of information processing.Many people in the "cognitivescience" and theywill is simplyirrelevant feel that of mydiscussion will much scientific paradigm as follows: it argueagainst betweenthebrainand all of theseothersystems Thereis a difference youhave been in thecase of simulation a this difference and explainswhy computational describing, whereasin thecase of thebraina computais a meresimulation, theothersystems and not merelymodelingthe functional tionalsimulationis actuallyduplicating is an reason is that thebrain,unliketheseothersystems, of the brain. The properties in the brain about And this fact is, words, your processing information system. "intrinsic".It is just a fact about biologythat the brain functionsto process and sincewe can also processthe same information information, computationally, role a different brain have of models altogetherfrom processes computational well research is a defined there So the weather. models for of, example, computational information brain the which the "Are processes by computational procedures question: thesameas theprocedures processthesameinformation? bywhichcomputers What I just imaginedan opponentsayingembodiesone of the worstmistakesin are used science.The mistakeis tosupposethatin thesensein whichcomputers cognitive To see thatthatis a mistake, brainsalso processinformation. to processinformation, withwhatgoeson in thebrain. In thecase of the whatgoeson in thecomputer contrast in a formthatcan be processedby encodes some information outside an agent computer, of the realization of thecomputer.That is,he or she providesa syntactical thecircuitry thatthe computercan implement information in, forexample,different voltagelevels. a seriesof electricalstagesthattheoutsideagentcan The computerthengoes through has no ofcourse,thehardware eventhough, and both interpret syntactically semantically not does the And beholder. the all in the of It is or semantics: intrinsic physics eye syntax an the can to it that matterprovided output Finally, implement algorithm. get only you as symbols can interpret whichan observer is producedin theformofphysical phenomena witha syntaxand a semantics. But nowcontrastthatwiththebrain. In thecase of thebrain,noneof therelevant relative(thoughof course,likeanything, theycan neurobiological processesare observer oftheneurophysirelativepointofview)and thespecificity froman observer be described clear,let us go throughan example. ologymattersdesperately.To makethisdifference modelof visionwill standard me. A toward car a I see computational coming Suppose about the visual arrayon myretinaand eventuallyprintout the take in information sentence,"Thereis a car comingtowardme". But thatis notwhathappensin theactual reactionsare set seriesofelectro-chemical biology.In thebiologya concreteand specific cellsofmyretina,and thisentire up bytheassaultofthephotonson thephotoreceptor PRESIDENTIALADDRESSES 35 ina concrete visualexperience. results Thebiological isnotthat eventually reality process ofa bunchofwords orsymbols the visual rather it is a matter by beingproduced system, of a concrete conscious visual this visual event; very specific experience.Now,that visualeventis as specific concrete andas concrete as a hurricane or thedigestion ofa meal. We can,withthecomputer, do an information of model that event or processing ofitsproduction, as wecando an information model of the weather, processing digestion or anyotherphenomenon, butthephenomena themselves arenotthereby information systems. processing In short, thesenseofinformation thatis usedin cognitive is at science, processing muchtoohigha levelofabstraction tocapture theconcrete of intrinsic biological reality The"information" inthebrainisalways tosomemodality orother. intentionality. specific It is specific to thought, or vision,or hearing, or touch,forexample.The levelof information whichis described in thecognitive sciencecomputational models processing ofcognition, on theotherhand,is simply a matter ofgetting a setofsymbols as output toa setofsymbols inresponse as input. We areblinded to thisdifference "I see a car bythefactthatthesamesentence, toward can be used to record both the visual and me", coming intentionality theoutput ofthecomputational modelofvision.Butthisshouldnotobscure from us thefactthat thevisualexperience is a concrete eventandis produced in thebrainbyspecific electrochemical To confuse theseeventsandprocesses withformal biological processes. symbol is to confuse thereality withthemodel.The upshotofthispartofthe manipulation discussion is thatinthesenseof"information" usedincognitive itis simply false science, tosaythatthebrainis an information device. processing VI. Summary oftheArgument Thisbrief hasa simple andI willlayitout: argument logicalstructure textbook isdefined interms 1. On thestandard definition, computation syntactically ofsymbol manipulation. 2. Butsyntax andsymbols arenotdefined in terms ofphysics.Thoughsymbol tokens arealways and"samesymbol" arenotdefined in tokens, physical "symbol" terms ofphysical features. inshort, is notintrinsic tophysics. Syntax, 3. Thishastheconsequence thatcomputation isnotdiscovered inthephysics, itis toit. Certain areassigned orusedorprogrammed assigned physical phenomena orinterpreted andsymbols areobserver relative. syntactically. Syntax thatyoucouldnotdiscover 4. It follows thatthe brainor anything else was a intrinsicallydigitalcomputer, although youcouldassigna computational interpretation toitas youcouldtoanything else.Thepointisnotthattheclaim "The brainis a digitalcomputer" is false. Ratherit does notgetup to thelevel of falsehood.It does not have a clearsense. You willhave misunderstood my accountifyou thinkthatI am arguingthatit is simplyfalsethatthe brainis a Is as ill defined digitalcomputer.The question"Is thebraina digitalcomputer?" 36 APA PROCEEDINGS, VOL. 64,NO.3 "Isit as thequestions "Isitan abacus?", "Isit a book?", "Isit a setofsymbols?", a setofmathematical formulae?" thanothers. 5. Somephysical facilitate thecomputational usemuchbetter systems and use them. In suchcaseswe are the That is whywe build,program, inthesystem thephysics inbothsyntactic andsemantic homunculus interpreting terms. different we thengivedo notcitecausalproperties 6. Butthecausalexplanations ofthehomunculus. thephysics oftheimplementation andtheintentionality from thehomunculus 7. Thestandard, tacit, fallacy. though wayoutofthisistocommit of and to models The homunculus is endemic cognition fallacy computational beremoved the recursive cannot standard They by decomposition arguments. are to a different addressed question. thatthe brainis doing 8. We cannotavoidthe foregoing results bysupposing are intrinsic far as its The as "information brain, operations processing". and It is a doesnoinformation concerned, organ its biological processing. specific In the forms of cause intentionality. neurobiological processes specific specific and sometimes there are brain, theycause intrinsically, neurobiological processes consciousness. Butthatis theendofthestory. [5] Footnotes seeSearle(1980)andSearle(1984). [1] Forearlier explorations ofbooksandarticles anddefended in a largenumber This view is announced [2] andThought more or less the same which to have of title, e.g.Computers many appear et and and eds. al, (Sharples 1988),The (Feigenbaum Feldman, 1963),Computers Thought and and the Mind 1985), 1988),ComputationCognition (Pylyshyn, (Johnson-Laird, Computer and of course, Modelof the Mind"(Block,1990, forthcoming), "The Computer andIntelligence" 1950). (Turing, Machinery "Computing times.Butthat sometimes eight [3] People saythatitwouldhavetoaddsixtoitself six addedto itself because is times Six addedto itself is badarithmetic. fifty-four, eight is made. howoftenthismistake zerotimesis stillsix. It is amazing Batali. wassuggested byJohn [4] Theexample oftheissues ofpeoplefordiscussions number a to am indebted I remarkably large [5] areduetoJohnBatali,Vinod inthispaper.I cannotthankthemall,butspecialthanks Goel,IvanHavel,KirkLudwig, DagmarSearleandKlausStrelau. References ModeloftheMind". "TheComputer Block,Ned (1990),forthcoming andLogic.Cambridge C. (1989). Computability Richard Boolos,GeorgeS. andJeffrey, Press, Cambridge. University PRESIDENTIALADDRESSES 37 onMindandPsychology. DanielC. (1978). Brainstorms: MIT Dennett, Philosophical Essays Mass. Press, Cambridge, andThought. E.A. andFeldman, McGraw-Hill J.eds.(1963). Computers Feigenbaum, New and York San Francisco. Company, NewYork. ThomasY. Crowell, Fodor, J.(1975). TheLanguage ofThought. Mass. Mind MIT ed. Press, (1981). John, Design. Cambridge, Haugeland, P.N. and the Mind. Harvard The Press, (1988). Johnson-Laird, Computer University Mass. Cambridge, Kuffler, JohnG. (1976). FromNeuronto Brain. Sinauer StephenW. and Nicholls, Mass. Associates, Sunderland, Lettvin, H.R.,McCulloch, W.S.,and Pitts,W.H. (1959). "Whatthe J.Y.,Maturana, the Brain". Tells 47, ofRadioEngineers, Frog'sEye Proceedings oftheInstitute Frog's in 230-255. 1940-1951, McCulloch, 1965,pp. reprinted San Francisco. andCompany, Marr,David(1982). Vision.W.H.Freeman Warren The Embodiments Mind. MIT Mass. McCulloch, S. (1965). Press, of Cambridge, Z. and MIT Mass. Pylyshyn,(1985). Computation Cognition. Press, Cambridge, R. (1980). "Minds, andPrograms", Brains TheBehavioral andBrainSciences. Searle,John 417-424. 3, pp. R. (1984). Minds, andScience.Harvard Brains Searle,John Press, University Cambridge, Mass. D. (1988). Computers M.,Hogg,D., Hutchinson, C., Torrance, S., andYoung, Sharples, andThought. MIT Press, Mass. and London. Cambridge, GordonM. (1983). Neurobiology. OxfordUniversity Press,NewYorkand Shepherd, Oxford. Alan(1950). "Computing andIntelligence." Mind,59,433-460. Turing, Machinery