The Extended Mind Author(S): Andy Clark and David Chalmers Source: Analysis, Vol

Home , Andy Clark, David Chalmers

The Extended Mind Author(s): Andy Clark and David Chalmers Source: Analysis, Vol. 58, No. 1 (Jan., 1998), pp. 7-19 Published by: Oxford University Press on behalf of The Analysis Committee Stable URL: http://www.jstor.org/stable/3328150 . Accessed: 05/04/2014 05:33

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 [email protected].

Oxford University Press and The Analysis Committee are collaborating with JSTOR to digitize, preserve and extend access to Analysis.

http://www.jstor.org

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions The extended mind ANDY CLARK& DAVID CHALMERS'

1. Introduction Wheredoes the mindstop andthe restof the worldbegin? The question invitestwo standardreplies. Some accept the boundariesof skinand skull, and say that what is outsidethe body is outsidethe mind. Othersare impressedby argumentssuggesting that the meaningof our words'just ain'tin the head',and hold that this externalismabout meaning carries overinto an externalismabout mind. We propose to pursuea thirdposi- tion. We advocate a very differentsort of externalism:an active externalism,based on the activerole of the environmentin drivingcogni- tiveprocesses.

2. Extendedcognition Considerthree cases of humanproblem-solving: (1) A personsits in frontof a computerscreen which displays images of varioustwo-dimensional geometric shapes and is askedto answerquestions concerningthe potential fit of suchshapes into depicted 'sockets'. To assess fit,the person must mentally rotate the shapes to alignthem with the sockets. (2) A personsits in frontof a similarcomputer screen, but this time can chooseeither to physicallyrotate the imageon the screen,by pressinga rotatebutton, or to mentallyrotate the imageas before.We can also suppose,not unrealistically,that some speed advantageaccrues to the physicalrotation operation. (3) Sometimein the cyberpunkfuture, a personsits in frontof a similar computerscreen. This agent, however, has the benefitof a neuralimplant whichcan performthe rotationoperation as fast as the computerin the previousexample. The agent must still choose which internal resource to use (theimplant or the good old-fashionedmental rotation), as eachresource makesdifferent demands on attentionand other concurrent brain activity. How muchcognition is presentin thesecases? We suggest that all three casesare similar. Case (3) withthe neuralimplant seems clearly to be on a parwith case (1). Andcase (2) withthe rotationbutton displays the same sort of computationalstructure as case (3), distributedacross agent and computerinstead of internalizedwithin the agent. If therotation in case(3) is cognitive,by what right do we countcase (2) as fundamentallydifferent? We cannotsimply point to the skin/skullboundary as justification,since 1 Authors are listed in order of degree of belief in the central thesis ANALYSIS58.1, January1998, pp. 7-19. ? AndyClark and DavidChalmers

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions 8 ANDY CLARK & DAVID CHALMERS the legitimacyof that boundaryis preciselywhat is at issue. But nothing else seems different. The kind of case just describedis by no meansas exotic as it may at first appear. It is not just the presence of advanced external computing resourceswhich raisesthe issue, but ratherthe generaltendency of human reasonersto lean heavilyon environmentalsupports. Thus considerthe use of pen and paper to performlong multiplication(McClelland et al. 1986, Clark 1989), the use of physicalre-arrangements of letter tiles to prompt word recall in Scrabble(Kirsh 1995), the use of instrumentssuch as the nautical slide rule (Hutchins 1995), and the general paraphernaliaof language, books, diagrams,and culture. In all these cases the individual brain performssome operations,while others are delegatedto manipula- tions of externalmedia. Had our brainsbeen different,this distributionof tasks would doubtlesshave varied. In fact, even the mentalrotation cases describedin scenarios(1) and (2) are real. The cases reflect options available to players of the computer game Tetris. In Tetris, falling geometricshapes must be rapidly directed into an appropriateslot in an emergingstructure. A rotation buttoncan be used. David Kirshand PaulMaglio (1994) calculatethat the physicalrotation of a shape through 90 degrees takes about 100 milliseconds, plus about 200 millisecondsto select the button. To achievethe same result by mentalrotation takes about 1000 milliseconds.Kirsh and Maglio go on to presentcompelling evidence that physicalrotation is used not just to posi- tion a shape ready to fit a slot, but often to help determinewhether the shape and the slot are compatible.The latteruse constitutesa case of what Kirsh and Maglio call an 'epistemicaction'. Epistemic actions alter the world so as to aid and augmentcognitive processes such as recognitionand search. Merely pragmatic actions, by contrast, alter the world because some physical change is desirablefor its own sake (e.g., putting cement into a hole in a dam). Epistemicaction, we suggest,demands spread of epistemiccredit. If, as we confront some task, a part of the world functionsas a process which, were it done in the head, we would have no hesitation in recognizingas part of the cognitive process, then that part of the world is (so we claim) part of the cognitiveprocess. Cognitiveprocesses ain't (all) in the head!

3. Active externalism In these cases, the human organismis linked with an external entity in a two-way interaction,creating a coupledsystem that can be seen as a cognitive system in its own right. All the components in the system play an active causal role, and they jointly govern behaviour in the same sort of way that cognition usually does. If we remove the external component the system's

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions THE EXTENDED MIND 9 behaviouralcompetence will drop,just as it wouldif we removedpart of its brain.Our thesis is thatthis sort of coupledprocess counts equally well as a cognitiveprocess, whether or not it is whollyin the head. Thisexternalism differs from the standard variety advocated by Putnam (1975)and Burge(1979). When I believethat wateris wet and my twin believesthat twin wateris wet, the externalfeatures responsible for the differencein our beliefsare distaland historical,at the otherend of a lengthycausal chain. Features of the presentare not relevant:if I happen to be surroundedby XYZ rightnow (maybeI have teleportedto Twin Earth),my beliefsstill concernstandard water, because of my history.In these cases, the relevantexternal features are passive.Because of their distalnature, they play no rolein drivingthe cognitive process in thehere- and-now.This is reflectedby the fact that the actions performed by me and my twinare physically indistinguishable, despite our external differences. In the caseswe describe,by contrast,the relevantexternal features are active,playing a crucialrole in thehere-and-now. Because they are coupled withthe humanorganism, they have a directimpact on the organismand on its behaviour.In thesecases, the relevantparts of the worldare in the loop, not danglingat the otherend of a longcausal chain. Concentrating on thissort of couplingleads us to anactive externalism, as opposedto the passiveexternalism of Putnamand Burge. Manyhave complained that even if Putnamand Burge are rightabout the externalityof content,it is not clearthat these external aspects play a causalor explanatoryrole in the generationof action.In counterfactual caseswhere internal structure is heldconstant but theseexternal features arechanged, behaviour looks just the same;so internalstructure seems to be doingthe crucialwork. We will not adjudicatethat issue here, but we note that activeexternalism is not threatenedby any suchproblem. The externalfeatures in a coupledsystem play an ineliminablerole - if we retaininternal structure but changethe externalfeatures, behaviour may changecompletely. The external features here are just as causallyrelevant as typicalinternal features of the brain.2 By embracingan activeexternalism, we allowa morenatural explanation of all sorts of actions.Once can explainmy choice of words in Scrabble,for example,as the outcomeof an extendedcognitive process 2 Muchof the appealof externalismin the philosophyof mindmay stem from the intuitiveappeal of activeexternalism. Externalists often makeanalogies involving externalfeatures in coupledsystems, and appealto the arbitrarinessof boundaries betweenbrain and environment.But theseintuitions sit uneasilywith the letterof standardexternalism. In most of the Putnam/Burgecases, the immediateenviron- mentis irrelevant;only the historicalenvironment counts. Debate has focusedon the questionof whethermind must be in the head,but a morerelevant question in assess- ing theseexamples might be: is mindin the present?

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions IO ANDY CLARK & DAVID CHALMERS involvingthe rearrangement of tiles on mytray. Of course,one could always tryto explainmy actionin termsof internalprocesses and a longseries of 'inputs'and 'actions', but this explanation would be needlesslycomplex. If an isomorphicprocess were going on in thehead, we wouldfeel no urgeto characterizeit in thiscumbersome way. In a veryreal sense, the re-arrange- mentof tileson the trayis not partof action;it is partof thought. Theview we advocatehere is reflectedby a growingbody of researchin cognitivescience. In areasas diverseas the theoryof situatedcognition (Suchman1987), studiesof real-world-robotics(Beer 1989), dynamical approachesto childdevelopment (Thelen and Smith1994), andresearch on thecognitive properties of collectivesof agents(Hutchins 1995), cognition is oftentaken to be continuouswith processesin the environment.3 Thus,in seeingcognition as extendedone is not merelymaking a termino- logicaldecision; it makesa significantdifference to the methodologyof scientificinvestigation. In effect,explanatory methods that might once havebeen thought appropriate only for the analysis of 'inner'processes are now beingadapted for the studyof theouter, and there is promisethat our understandingof cognitionwill becomericher for it. Somefind this sortof externalismunpalatable. One reason may be that manyidentify the cognitive with the conscious, and it seemsfar from plau- siblethat consciousnessextends outside the headin thesecases. But not everycognitive process, at leaston standardusage, is a consciousprocess. It is widelyaccepted that all sorts of processesbeyond the bordersof consciousnessplay a crucialrole in cognitiveprocessing: in the retrievalof memories,linguistic processes, and skill acquisition,for example.So the merefact that external processes are external where consciousness is internal is no reasonto denythat those processes are cognitive. Moreinterestingly, one might argue that what keeps real cognition processesin the headis the requirementthat cognitiveprocesses be portable. Here,we aremoved by a visionof whatmight be calledthe NakedMind: a packageof resourcesand operationswe can alwaysbring to bearon a cognitivetask, regardless of thelocal environment. On this view, the trouble withcoupled systems is thatthey are too easilydecoupled. The true cognitiveprocesses are those that lie at theconstant core of the system;anything elseis an add-onextra. Thereis somethingto this objection.The brain(or brainand body) comprisesa packageof basic,portable, cognitive resources that is of inter- est in its own right.These resources may incorporate bodily actions into cognitiveprocesses, as whenwe use our fingersas workingmemory in a trickycalculation, but they will not encompassthe more contingent aspects

3 Philosophicalviews of a similarspirit can be foundin Haugeland1995, McClamrock 1995, Varelaet al. 1991, andWilson 1994.

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions THE EXTENDED MIND II of ourexternal environment, such as a pocketcalculator. Still, mere contin- gencyof couplingdoes not ruleout cognitivestatus. In the distantfuture we maybe ableto plugvarious modules into our brainto helpus out:a modulefor extrashort-term memory when we needit, forexample. When a moduleis pluggedin, theprocesses involving it arejust as cognitiveas if theyhad been there all along. Evenif one wereto makethe portabilitycriterion pivotal, active exter- nalismwould not be undermined.Counting on our fingershas already been let in the door,for example,and it is easy to push thingsfurther. Thinkof the old imageof the engineerwith a sliderule hanging from his beltwherever he goes.What if peoplealways carried a pocketcalculator, or hadthem implanted? The real moral of the portabilityintuition is that for coupledsystems to be relevantto the core of cognition,reliable couplingis required.It happensthat most reliablecoupling takes place withinthe brain,but there can easily be reliable coupling with the environ- mentas well. If the resourcesof my calculatoror my Filofaxare always therewhen I needthem, then they are coupled with me as reliablyas we need.In effect,they are partof the basicpackage of cognitiveresources that I bringto bear on the everydayworld. These systemscannot be impugnedsimply on the basisof the dangerof discretedamage, loss, or malfunction,or becauseof anyoccasional decoupling: the biologicalbrain is in similardanger, and occasionallyloses capacitiestemporarily in episodesof sleep,intoxication, and emotion. If the relevantcapacities are generallythere when they are required, this is couplingenough. Moreover,it may be that the biologicalbrain has in fact evolvedand maturedin ways whichfactor in the reliablepresence of a manipulable externalenvironment. It certainlyseems that evolutionhas favouredon- boardcapacities which are especially geared to parasitizingthe localenvi- ronmentso as to reducememory load, and even to transformthe nature of the computationalproblems themselves. Our visual systems have evolved to relyon theirenvironment in variousways: they exploit contingent facts aboutthe structureof naturalscenes (e.g. Ullman and Richards 1984), for example,and they take advantage of thecomputational short cuts afforded by bodilymotion and locomotion(e.g. Blakeand Yuille,1992). Perhaps thereare other cases where evolution has found it advantageousto exploit the possibilityof the environmentbeing in the cognitiveloop. If so, then externalcoupling is partof the trulybasic package of cognitiveresources thatwe bringto bearon the world. Anotherexample may be language,which appears to be a centralmeans bywhich cognitive processes are extended into the world. Think of a group of peoplebrainstorming around a table,or a philosopherwho thinksbest by writing, developing her ideas as she goes. It may be that language

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions 12 ANDY CLARK & DAVID CHALMERS evolved,in part, to enablesuch extensionsof our cognitiveresources withinactively coupled systems. Withinthe lifetimeof an organism,too, individuallearning may have moulded the brain in ways that rely on cognitive extensionsthat surroundedus as we learned.Language is againa centralexample here, as arethe various physical and computational artifacts that are routinely used as cognitiveextensions by childrenin schoolsand by traineesin numerous professions.In suchcases the braindevelops in a way that complements the externalstructures, and learns to playits rolewithin a unified,densely coupledsystem. Once we recognizethat the crucial role of theenvironment in constrainingthe evolutionand developmentof cognition,we see that extendedcognition is a corecognitive process, not an add-onextra.

4. Fromcognition to mind So farwe havespoken largely about 'cognitive processing', and argued for its extensioninto the environment.Some might think that the conclusion has beenbought too cheaply.Perhaps some processing takes place in the environment,but what of mind?Everything we havesaid so faris compatible with the view that trulymental states - experiences,beliefs, desires, emotions,and so on - are all determinedby statesof the brain.Perhaps whatis trulymental is internal,after all? Wepropose to takethings a stepfurther. While some mental states, such as experiences,may be determined internally, there are other cases in which externalfactors make a significantcontribution. In particular,we will arguethat beliefs can be constitutedpartly by features of theenvironment, whenthose features play the rightsort of role in drivingcognitive processes.If so, the mindextends into the world. First,consider a normalcase of beliefembedded in memory.Inga hears froma friendthat there is anexhibition at theMuseum of ModernArt, and decidesto go see it. Shethinks for a momentand recalls that the museum is on 53rdStreet, so shewalks to 53rdStreet and goes intothe museum. It seemsclear that Inga believes that the museumis on 53rdStreet, and that she believedthis evenbefore she consultedher memory. It was not previ- ously an occurrentbelief, but then neitherare most of our beliefs.The beliefwas somewherein memory,waiting to be accessed. Now considerOtto. Otto suffersfrom Alzheimer'sdisease, and like manyAlzheimer's patients, he relieson informationin theenvironment to help structurehis life. Otto carriesa notebookaround with him every- wherehe goes.When he learnsnew information, he writesit down.When he needssome old information,he looks it up. For Otto, his notebook plays the role usually played by a biological memory.Today, Otto hears about the exhibitionat the Museumof ModernArt, and decidesto go see

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions THE EXTENDED MIND 13 it. He consultsthe notebook, which says that the museumis on 53rd Street, so he walks to 53rd Streetand goes into the museum. Clearly,Otto walked to 53rd Street because he wanted to go to the museumand he believedthe museumwas on 53rd Street.And just as Inga had her belief even before she consultedher memory,it seems reasonable to say that Otto believed the museum was on 53rd Street even before consultinghis notebook. For in relevantrespects the cases are entirelyanal- ogous: the notebook plays for Otto the same role that memory plays for Inga. The informationin the notebook functionsjust like the information constituting an ordinary non-occurrentbelief; it just happens that this informationlies beyond the skin. The alternativeis to say that Otto has no belief about the matteruntil he consultshis notebook;at best, he believesthat the museumis located at the addressin the notebook. But if we follow Otto aroundfor a while, we will see how unnaturalthis way of speakingis. Otto is constantlyusing his notebook as a matter of course. It is centralto his actions in all sorts of contexts,in the way that an ordinarymemory is centralin an ordinarylife. The same information might come up again and again, perhaps being slightlymodified on occasion, beforeretreating into the recessesof his arti- ficial memory.To say that the beliefsdisappear when the notebook is filed away seems to miss the big picture in just the same way as saying that Inga'sbeliefs disappear as soon as she is longerconscious of them. In both casesthe informationis reliablythere when needed,available to consciousness and availableto guide action, in just the way that we expect a belief to be. Certainly,insofar as beliefsand desiresare characterizedby theirexplan- atoryroles, Otto's and Inga'scases seem to be on a par:the essentialcausal dynamicsof the two cases mirroreach other precisely.We are happy to explain Inga'saction in terms of her occurrentdesire to go to the museum and her standingbelief that the museumis on 53rd street, and we should be happy to explain Otto's action in the same way. The alternativeis to explain Otto's action in terms of his occurrentdesire to go to the museum, his standingbelief that the Museumis on the location written in the notebook, and the accessible fact that the notebook says the Museum is on 53rd Street;but this complicatesthe explanationunnecessarily. If we must resortto explainingOtto's action this way, then we must also do so for the countless other actions in which his notebook is involved;in each of the explanations, there will be an extra term involving the notebook. We submit that to explain things this way is to take one step too many. It is pointlesslycomplex, in the same way that it would be pointlesslycomplex to explain Inga's actions in terms of beliefs about her memory. The notebook is a constant for Otto, in the same way that memory is a constant for

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions 14 ANDY CLARK& DAVID CHALMERS Inga;to pointto it in everybelief/desire explanation would be redundant. In an explanation,simplicity is power. If this is right,we can evenconstruct the caseof TwinOtto, who is just likeOtto except that a whileago he mistakenlywrote in his notebookthat the Museumof ModernArt was on 51st Street.Today, Twin Otto is a physicalduplicate of Otto from the skin in, but his notebookdiffers. Consequently,Twin Otto is best characterizedas believingthat the museumis on 51st Street,where Otto believes it is on 53rd.In thesecases, a beliefis simplynot in the head. Thismirrors the conclusionof Putnamand Burge,but againthere are importantdifferences. In the Putnam/Burgecases, the externalfeatures constitutingdifferences in beliefare distal and historical, so thattwins in thesecases produce physically indistinguishable behaviour. In thecases we aredescribing, the relevant external features play an activerole in thehere- and-now,and have a directimpact on behaviour.Where Otto walksto 53rdStreet, Twin Otto walks to 51st.There is no questionof explanatory irrelevancefor this sort of externalbelief content; it is introducedprecisely becauseof the centralexplanatory role that it plays.Like the Putnamand Burgecases, these cases involve differences in referenceand truth-conditions,but they also involve differences in the dynamicsof cognition.4 Themoral is thatwhen it comesto belief,there is nothingsacred about skulland skin. What makes some information count as a beliefis the role it plays,and there is no reasonwhy the relevantrole can be playedonly frominside the body. Somewill resistthis conclusion.An opponentmight put herfoot down andinsist that as she usesthe term'belief', or perhapseven according to standardusage, Otto simply does not qualifyas believingthat the museum is on 53rdStreet. We do not intendto debatewhat is standardusage; our broaderpoint is that the notionof beliefought to be used so that Otto qualifiesas havingthe beliefin question.In all importantrespects, Otto's caseis similarto a standardcase of (non-occurrent)belief. The differences betweenOtto's case and Inga'sare striking,but they are superficial.By usingthe 'belief'notion in a widerway, it picksout somethingmore akin to a naturalkind. The notionbecomes deeper and moreunified, and is moreuseful in explanation. To providesubstantial resistance, an opponenthas to showthat Otto's

4 In the terminologyof Chalmers(forthcoming): the twins in the Putnamand Burge cases differonly in theirrelational content (secondary intension), but Otto and his twincan be seento differin theirnotional content (primary intension), which is the sort of contentthat governscognition. Notional content is generallyinternal to a cognitivesystem, but in thiscase the cognitivesystem is itselfeffectively extended to includethe notebook.

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions THE EXTENDED MIND 15 and Inga'scases differin some importantand relevantrespect. But in what deep respectare the cases different?To make the case solely on the grounds that informationis in the head in one case but not in the other would be to beg the question. If this differenceis relevantto a differencein belief, it is surely not primitivelyrelevant. To justify the different treatment, we must find some more basic underlyingdifference between the two. It might be suggestedthat the cases are relevantlydifferent in that Inga has more reliableaccess to the information.After all, someone might take away Otto's notebook at any time, but Inga'smemory is safer. It is not implausiblethat constancy is relevant:indeed, the fact that Otto always uses his notebook played some role in our justifyingits cognitivestatus. If Otto were consulting a guidebook as a one-off, we would be much less likelyto ascribehim a standingbelief. But in the originalcase, Otto'saccess to the notebook is very reliable- not perfectlyreliable, to be sure, but then neitheris Inga'saccess to her memory.A surgeonmight tamper with her brain, or more mundanely,she might have too much to drink. The mere possibilityof such tamperingis not enough to deny her the belief. One might worry that Otto's access to his notebook in fact comes and goes. He showers without the notebook, for example, and he cannot read it when it is dark. Surelyhis beliefcannot come and go so easily?We could get aroundthis problem by redescribingthe situation, but in any case an occasionaltemporary disconnection does not threatenour claim. Afterall, when Inga is asleep, or when she is intoxicated, we do not say that her belief disappears.What reallycounts is that the informationis easily available when the subjectneeds it, and this constraintis satisfiedequally in the two cases. If Otto's notebook were often unavailableto him at times when the informationin it would be useful, there might be a problem, as the information would not be able to play the action-guidingrole that is centralto belief;but if it is easily availablein most relevantsituations, the belief is not endangered. Perhapsa differenceis that Inga has better access to the information than Otto does? Inga's'central' processes and her memoryprobably have a relativelyhigh-bandwidth link betweenthem, comparedto the low-grade connectionbetween Otto and his notebook. Butthis alone does not make a differencebetween believing and not believing.Consider Inga's museum- going friendLucy, whose biologicalmemory has only a low-gradelink to her central systems, due to nonstandardbiology or past misadventures. Processingin Lucy'scase might be less efficient,but as long as the relevant informationis accessible,Lucy clearly believes that the museumis on 53rd Street.If the connectionwas too indirect- if Lucy had to strugglehard to retrievethe informationwith mixed results,or a psychotherapist'said were needed - we might become more reluctant to ascribe the belief, but such

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions 16 ANDY CLARK & DAVID CHALMERS casesare well beyondOtto's situation, in whichthe informationis easily accessible. Anothersuggestion could be that Otto has accessto the relevantinfor- mationonly by perception,whereas Inga has more directaccess - by introspection,perhaps. In some ways, however, to putthings this way is to begthe question.After all, we arein effectadvocating a pointof viewon whichOtto's internal processes and his notebook constitute a singlecogni- tive system.From the standpointof this system,the flow of information betweennotebook and brain is notperceptual at all;it doesnot involvethe impactof somethingoutside the system. It is moreakin to informationflow withinthe brain.The only deep way in whichthe access is perceptualis that in Otto'scase, there is a distinctlyperceptual phenomenology associated withthe retrieval of theinformation, whereas in Inga'scase there is not.But why shouldthe natureof an associatedphenomenology make a difference to thestatus of a belief?Inga's memory may have some associated phenomenology,but it is stilla belief.The phenomenology is not visual,to be sure. Butfor visualphenomenology consider the Terminator,from the Arnold Schwarzeneggermovie of the samename. When he recallssome information from memory,it is 'displayed'before him in his visual field (presumablyhe is consciousof it, as thereare frequent shots depicting his pointof view).The fact that standing memories are recalled in thisunusual waysurely makes little difference to theirstatus as standingbeliefs. Thesevarious small differences between Otto's and Inga's cases are all shallowdifferences. To focuson themwould be to missthe way in which forOtto, notebook entries play just the sort of rolethat beliefs play in guid- ingmost people's lives. Perhapsthe intuitionthat Otto's is not a truebelief comes from a resid- ual feelingthat the only truebeliefs are occurrentbeliefs. If we takethis feelingseriously, Inga's belief will be ruled out too, as willmany beliefs that we attributein everydaylife. This would be anextreme view, but it maybe the most consistentway to deny Otto'sbelief. Upon even a slightlyless extremeview - the viewthat a beliefmust be availablefor consciousness, for example- Otto'snotebook entry seems to qualifyjust as well as Inga's memory.Once dispositional beliefs are let in thedoor, it is difficultto resist the conclusionthat Otto's notebook has all the relevantdispositions.

5. Beyondthe outerlimits If the thesisis accepted,how far shouldwe go? All sortsof puzzlecases springto mind.What of the amnesicvillagers in One HundredYears of Solitude, who forget the names for everythingand so hang labels every- where?Does the informationin my Filofax count as part of my memory? If Otto's notebook has been tamperedwith, does he believe the newly-

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions THE EXTENDED MIND 17 installedinformation? Do I believethe contents of the page in front of me beforeI read it? Is my cognitivestate somehow spreadacross the Internet? We do not think that there are categoricalanswers to all of these ques- tions, and we will not give them. But to help understandwhat is involved in ascriptionsof extended belief, we can at least examine the featuresof our centralcase that make the notion so clearlyapplicable there. First,the notebook is a constantin Otto's life - in cases wherethe informationin the notebook would be relevant,he will rarelytake action without consulting it. Second, the informationin the notebook is directly availablewithout difficulty.Third, upon retrievinginformation from the notebook he auto- matically endorses it. Fourth, the informationin the notebook has been consciously endorsed at some point in the past, and indeed is there as a consequenceof this endorsement.5The status of the fourth feature as a criterionfor belief is arguable (perhapsone can acquire beliefs through subliminalperception, or throughmemory tampering?), but the firstthree featurescertainly play a crucialrole. Insofaras increasinglyexotic puzzle cases lack these features,the appli- cability of the notion of 'belief' graduallyfalls of. If I rarelytake relevant action without consulting my Filofax, for example, its status within my cognitivesystem will resemblethat of the notebookin Otto's.But if I often act withoutconsultation - for example,if I sometimesanswer relevant ques- tions with 'I don't know' - then informationin it counts less clearlyas part of my beliefsystem. The Internetis likelyto fail on multiplecounts, unless I am unusuallycomputer-reliant, facile with the technology, and trusting, but informationin certainfiles on my computermay qualify.In intermedi- ate cases, the questionof whethera beliefis presentmay be indeterminate, or the answermay dependon the varyingstandards that are at play in various contexts in which the questionmight be asked. But any indeterminacy here does not mean that in the centralcases, the answeris not clear. What about socially extended cognition? Could my mental states be partly constitutedby the states of other thinkers?We see no reason why not, in principle.In an unusuallyinterdependent couple, it is entirelypossi- ble that one partner'sbeliefs will play the same sort of role for the other as the notebook plays for Otto. What is centralis a high degreeof trust, reli- ance, and accessibility.In other social relationshipsthese criteriamay not be so clearly fulfilled, but they might neverthelessbe fulfilled in specific domains.For example, the waiter at my favouriterestaurant might act as s The constancyand past-endorsementcriteria may suggestthat history is partly constitutiveof belief.One mightreact to this by removingany historicalcomponent (givinga purelydispositional reading of the constancycriterion and eliminatingthe past-endorsementcriterion, for example),or one mightallow sucha componentas long as the mainburden is carriedby featuresof the present.

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions 18 ANDY CLARK & DAVID CHALMERS a repositoryof my beliefs about my favouritemeals (this might even be construedas a case of extendeddesire). In other cases, one's beliefsmight be embodiedin one's secretary,one's accountant,or one's collaborator. In each of these cases, the majorburden of the couplingbetween agents is carriedby language.Without language, we might be much more akin to discreteCartesian 'inner' minds, in which high-levelcognition relies largely on internalresources. But the adventof languagehas allowed us to spread this burdeninto the world. Language,thus construed,is not a mirrorof our innerstates but a complementto them. It servesas a tool whose role is to extend cognition in ways that on-boarddevices cannot. Indeed,it may be that the intellectualexplosion in recentevolutionary time is due as much to this linguistically-enabledextension of cognition as to any independent developmentin our innercognitive resources. What, finally,of the self? Does the extended mind imply an extended self? It seems so. Most of us already accept that the self outstrips the boundariesof consciousness;my dispositionalbeliefs, for example,constitute in some deep sense part of who I am. If so, then these boundariesmay also fall beyondthe skin. The informationin Otto'snotebook, for example, is a centralpart of his identityas a cognitiveagent. What this comesto is that Otto himselfis best regardedas an extendedsystem, a couplingof biological organismand externalresources. To consistentlyresist this conclusion,we would have to shrink the self into a mere bundle of occurrent states, severelythreatening its deep psychologicalcontinuity. Far better to take the broaderview, and see agentsthemselves as spreadinto the world. As with any reconceptionof ourselves,this view will have significant consequences.There are obvious consequencesfor philosophicalviews of the mind and for the methodology of researchin cognitive science, but there will also be effects in the moral and social domains. It may be, for example, that in some cases interferingwith someone'senvironment will have the same moral significanceas interferingwith their person. And if the view is taken seriously,certain forms of social activitymight be recon- ceived as less akin to communicationand action, and as more akin to thought. In any case, once the hegemonyof skin and skull is usurped,we may be able to see ourselvesmore truly as creaturesof the world.

Washington University St. Louis, MO 63130, USA [email protected]

University of California Santa Cruz, CA 95064, USA [email protected]

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions THE EXTENDEDMIND 19

References Beer,R. 1989. Intelligence as Adaptive Behavior.New York:Academic Press. Blake, A. and A. Yuille, eds. 1992. Active Vision. Cambridge,MA: MIT Press. Burge,T. 1979. Individualismand the mental. Midwest Studiesin Philosophy 4: 73-122. Chalmers, D. J. (forthcoming). The components of content. Clark, A. 1989. Microcognition. MIT Press. Haugeland, J. 1995. Mind embodied and embedded. In Mind and Cognition, eds Y. Houng and J. Ho. Taipei: Academia Sinica. Hutchins, E. 1995. Cognition in the Wild. Cambridge,MA: MIT Press. Kirsh, D. 1995. The intelligent use of space. ArtificialIntelligence 73: 31-68. Kirsh, D. and P. Maglio. 1994. On distinguishing epistemic from pragmatic action. Cognitive Science 18: 513-49. McClamrock, R. 1995. Existential Cognition. Chicago: Universityof Chicago Press. McClelland, J. L, D. E. Rumelhart, and G. E. Hinton 1986. The appeal of parallel distributed processing. In Parallel Distributed Processing, Volume 2, eds. McClel- land and Rumelhart. Cambridge,MA: MIT Press. Putnam, H. 1975. The meaning of 'meaning'. In Language, Mind, and Knowledge, ed. K. Gunderson. Minneapolis: University of Minnesota Press. Suchman, L. 1987. Plans and Situated Actions. Cambridge: Cambridge University Press. Thelen, E. and L. Smith. 1994. A Dynamic Systems Approach to the Development of Cognition and Action. Ullman, S. and W. Richards. 1984. Image Understanding.Norwood, NJ: Ablex. Varela, F., E. Thompson and E. Rosch. 1991. The Embodied Mind. Cambridge, MA: MIT Press. Wilson, R. 1994. Wide computationalism. Mind 103: 351-72.

This content downloaded from 131.155.215.185 on Sat, 5 Apr 2014 05:33:44 AM All use subject to JSTOR Terms and Conditions