I
Can the mind change the world?
NED BLOCK
Hilary Putnam origi nated the idea that mental states are computational states. At first (Putnam, 1960), his view was that although mental states are not identical wit h computational states (or " logical states." as he then called them), there are useful analogies between them. Later (Putnam, 1967). he argued in favor of the identity on the grounds that it was more plausible to suppose mental states are function al states (as he then called them) than that they are behavioral or physical states. This doctrine - funct ionalism - has dominated the philosophy of mind for over twenty years. Shortly after proposing functionalism. Putnam re jected it again (J 973), and he has maintained this position ever since (Putnam. 1988). Putnam was my teacher during both my undergraduate and gradu·ate days, and I fear I have absorbed his ambivalence toward functionalism. My teacher has had a habit of changing his mind, but never has he done so within a single essay, and so in this chapter I have surpassed him. My chapter starts out as an argument for functionalism, but it ends up suggesting an argument against it. The issue is whether we ca n avoid epiphenomenalism, which I here understand as the doctrine that what we think or want has no causal relevance to what we do. I propose functionalism as a way of warding off arguments for epiphcnomenalism, but then I argue that functionalism may bring epiphenomenalism in its wake. The orientation of the chapter is toward the sciences of the mind, and their relation to intentional content, that is, what is shared hy the
I am graleful for support IO the America n Counci l or Learned Sociclies and lo the Nalional Science Foundation. granl numhcr DTR88 12559. I am also grateful to Jerry Fodor, Paul Horwich, Frank Jackson, Gabriel Segal. and Marianne Talbot for !heir commenls on an earlier draft; to Thomas Nagel, who was the commentaror on the paper on which lhis chapler is based at a conference al Columbia University in December 1988; to Simon Blacktiurn, who was !he commentalor on the paper at Oxford in February 1989; to Martin Davies. Dorothy Edgington. and Oarry Smith for discussion or 1he issues while I was at Birkbeck College. University of London; and to the audiences al a number of philosophy colloquia. .. · · :;
138 NFD ULOCK
hclief that grass grows and the desire that grass grows, the that grass model. T grow.~ that buth sraces are directed toward. The question at hand is qucstioni whether the scie nces of the mind prc..:cludc int entional content from useful pt. causal relevance to behavior. One argument that the intentional con J think ti tents of our beliefs, thoughts, and the like have no effects on our premise behavior could he put this way: The processors in the head are not The r. sensitive to content, so how could content have any effect on the Turing r outputs or changes of state of the system of processors? And if content such as , can't affect the operation of this system of processors, how could it play sensitive any role in producing behavior? This argument seems formidable Uu1 if 1h whether one thinks of the processors as neural devices reacting lO a gate. h neural inputs or, instead, from the cognitive science point of view, as gates? S computational devices processing representations. ' Jn this chapter, I se ntat io1 confin e myself to the prohlcm as it arises in the cognitive science influcnc• I approach that is dominated by the computer model of the mind. I soning a assume a very specific picture of cognitive science and its relation to the computl commonsense conception of intentional content, namely, the view ac mind (c cording to which there is an internal system of representation from assumrn I whose meanings our intentional contents derive (Fodor, 1975; Pylys hyn, My r '. 1984 ). One of my reasons for couching the discussion in terms of this view is that although those who adopt this view are motil'ated by the I. t· aim of showing how our commonsense beliefs about cont ent (including I I. l our belief in content's causal efficacy) are vi ndicated by the computer II model of the mind, the problem of thl' epiphenomenalism of content 111. s arises within this view in an extremely simple and straightforward (and (.' poignant) way. The viewpoint assumed throughout the chapter is that of r a supporter of the computer model in cognit ive science who also would IV . I like to believe that the contents of our thoughts arc indeed ca usally g relevant to what we do. A suht I The problem I have in mind might be put in terms of The Paradox of relcvarn the Causal Efficacy of Content, namely, th c: thar grass ml>Je l. This s<.:eon c.J daim is by for the most conc roversial, hul I won't be at hand is questioning it here. My reasons are that I think it is true. that I see no intent from usdul purpose to dividing meaning and content in this cn11text. and that Jtion al con I think the best bets fo r resolving the paradox are to que:-. ti on the third :cts on our premise amt whether the reasoning that leads to the paradox is right. !ad arc not T he reasoning behind the paradox goes so mething li ke this: Any feet on the;: Turing machine can be constructed fro m simple primit ive p rnces~l) r s cl if content such ;.is 1111<1 ga tes. or gates, and the like. (See Minsky, llJC17 .) CJates are :ould it play sen!'> itive to the sy ntactic fo rms of representations, not tl1eir meanings. fo rmiJable But if the meaning 1>f a rcprc:!se nt ation cannot in liuence tl1c behavior of reacting to a ga te. how coulJ iL influence th t.: hehuvior of a computer - a system of of view. as gates? Since intentional content reduces lO meanings of internal rep re ; chapter, I sentations, and si nce meanings of internal repre!>t! nt ati ons ca nnot live scie nce inllut.:nce behavior, content cannot influence hehavior t.:it her. The rea the minJ. I soning assumes that at least as far as our thinking is con('erneJ, we are lation to the computers. This idea - which is simply the computer model of the he view ac- mind (o ur cognit ive mind, that is) - may he wrong, but I will be ·tation from assum ing it m explore wh ere it leads. '5; Pylyshyn, My plan fo r the chapter involves erms of this •ared by the I. Expl is not causally relevant to the behavior is the last letter of the name of the city in which the belief was formed. Note that the point is not that beliefs, thoughts, desires, and the like (mental states or events that have content) are causes, for example, of behavior. (I assume that they arc.) Rather, the point is that when mental events have effects, they typically have those effects (rather than different effects) because the mental events have the contents that they have, rather than some other contents. Typically, if the beliefs, thoughts, and so forth, had had contents that were appropriately different, they would have had quite different effects. For example, had I believed that everywhere except the United States is a dangerous place, then I wouldn't have left the country. My metaphysical stance in this chapter is one in which mental events are the causes of behavior, and their contents are properties of those Of cou events that may or may not be causally relevant to the events' effects. I remembe1 shall say that property P of event c is causally relevant to effect e and computer: that c causes e in virtue of P more or less interchangeably. Also, I shall hering mi put the claim that content is causally relevant to something by saying that it ca1 that cont ent is causally efficacious, and not epiphenomena!. Of course, storing a my use of 'epiphenomenalism' is importantly different from the tradi doctrine . tional one. I'm not raising any possibility of content being a property of premise. a distinct mental substance, events in which are caused by events in the thought t brain even though events in the mental substance never cause anything. internal r And epiphenomenalism in my sense does not entail that content is itself propositir caused hy an underlying physical state that also causes behavior. The thinking t reason I use this old word ('epiphenomenalism') for this newer problem the head. is that the problem I raise is the modern heir of the old problem. Talk of The second premise - that content reduces to meanings of internal talk. The representations (symbols in the head) - is much more tendentious situation, than the first, and is certainly not part of commonsense wisdom about the syster the mind. But it is a straightforward way of making commonsense differs frc realism about content compatible with the view that the machinery of tionalism ~ the mind is one of computations on internal representations. According cont rover: to the computer model, the mind (or its cognitive aspect) can be tional. thought of as a system of processors that take representations as inputs, The th transform them in various ways, and then send them to other proces syntactic sors, as in computers. One can think of the representations in such a their mca system as being in certain computational relations to the whole system. and gate i These computational relations are determined hy the ways the system inputs yic would treat the representation given various different states of its I. Is 1 component processors. A sample computational relation is that of 11. Do storing a representation. 111. [)(' . Can rhc mind change rite ~vorld ? l ~ l i e name of 1d the like :ample, of hat wh en 1ther than . that they thoughts, rent, they ieved that e, then I Figu re I tal events ; of those Of course, computers tha t you can buy don't currently think or · effects. I remember. But if computers can be programmed to think, or if we are 'ect e and computers, this story can be extended to thinking things. Thus, remem so, I shall bering might consist in the storing of a representation in such a way by saying that it can later be accessed. Remembering that grass grows would be )f course, storing a representation that means that grass grows. This is the the tradi doct rine of reduction of content to meaning stated in the second ropcrty of premise. More generally, us is argued in Fodor (1975), to have the nts in the thought that p is to be in a certain computational relation to an anything. internal representation tha t means that p, and likewise for other nt is itself propositional attitudes. The slogan - and it is only a slogan - is that tvior. The thinking that grass grows is having ' Grass grows· in the thought box in : problem the head. Iem. Talk of the thought box in the head is (thinly) disguised functional f internal talk. The sent ences that are in th e thought box share a computational ndentious situation, a role within the system - in other words, a function within om about the system. T his functional theory of what a thought is (and how it monsense differs from a desire) should be flrmly distinguished from other func :hinery of tionalisms to be mentioned in this paper, especially the much more \ ccording controversial it.lea that the meaning of a representation is itself func ) can be tional. as inputs, The third premise is that internal processors are sensitive to the ;r proces synt ac.: tic form of the internal representations th at they process, not in such a their meanings. Consider the and gate of Figure 1. What makes it an le system. and gate is that it emits a 'I ' if and only if both inputs are 'l's'; all other ie system inputs yield a '()' as outpu t. The a11d gate ies of its s that of I. Is not sensitive to I I. Does not react to llJ. Docs not detect .. ~:~: . . ' · · ..,.Z,;~~···~;i'_.~~mBDDBJEmmm•m------• • ""'I .,-- - ~ • ~r~~ ••.., --:: ,· , . ' • ,'{~~ ~-·.- '":""i•41 ·, ..~~ •.• '- ;- .. r, '•~ ' ' , , I • ;' • : I' I • ' ·~ 142 N E I> BLO C K whether the 'l's' represent truth or the number 1 or nothing at all. Rather, the and gate is sensitive to, reacts to, and detects only whether the inputs are both ' l's' or not. Thus it is sensitive to the syntax. not the meaning of its inputs, and likewise for the primitive processors postu lated by cognitive science accounts of the mind. (The primitive proces sors of a system are the ones for which there is no explanation within cognitive science of how they work; their operation can he explain ed only in terms of a lower level branch of science, physiology, in the case of human pr!mitive processors, electronics in the case of standard computer primitives.) Note that the sense of 'syntax' I am using here (somewhat mi slead hanJ. an ingly) means fom1 class. 'I' and 'O' arc different syntactic objects in this th e other sense. It is important to he aware that synt ax in this sense of the term is cog ni1 ive another functional notion. English orthography is also functional, al putc111ial though this may be obscured by the rigidification of function hy conven shall see tion. For %xampl %, you wi ll hav% littl % trouhl% figuring out what ··going r 1%tt %r of th % alphab%t is th % on % to whi ch th% unusual symbol in worries ; this s%nt%nc% should b% tr%at%d as functionally %quiva1 %nt. section i ~ Consider an input-output system whose input and output registers to syntax are bi-stable, and take on values of either 7 volts or 4 volts. Suppose soluliPn. that if both input registers arc at 4 volts, th en the output is 4 volts, and every other input yields the 7-volt output. Then (1) the system is an and gate, (2) the 4-volt va lue counts as a ' l ' for this and gate, and the 7-volt 2. . va lue counts as a 'O '. A differently constructed and gate might he one The tit le for which 7 volts counts as a ' I '. ' 1' is conventionally assigned to states example. of computer registers using this type of consideration. The functional hanc.lk o roles of the bi-stable slates of registers simultaneously determine our only knm identifications of the devices in the system (e.g., as adders and gates) represent and our identifications of the states of the registers as symbols. So it is decimal having a certain functional role that makes a stale satisfy a sym acric follow ing description, in the sense of syntax used here. Note that it woul d be a mistake to say that 4 volts in the first gate (J I () mentioned has the same meaning as 7-vo lt s in the second gate. We I , 0 don't know what th e meanings of the ' l 's' in either gate are until we U -t I see other aspects of their function . These ' l's' could be used to mean [ I one, or true, or green; the input-output function does not choose The fir among these and other possibilities. The aspects of function relevant to Ii.isl is t rn syntax are different from, though overlapping with, the aspects of Herc i ~ function relevant to meaning. ·o· in thi ~ In the next section, I will explain the picture common in cognitive ga te. 1"11c science of sensible or" rational" relations among contents deriving from inpu h a1 th e correlation between rational relations among contents on the one illus t1 ;1tet Cun the J11ind change tht! H'orld? 143 hing at all. l 1ly whet her l ca.x, not the I ;sors postu tive proces H ion within e explained in the case .>f standard Figurt; 2 1at mislcad hand, and rroct:'>sing rt:latio11s among syntactic objects in the brain on •jects in chis the uthi.:r. This idea will illu 'o !ratc hoth the strength and wc ;1kness of the · the term is cognitive scie nce riccure I will he assuming. The strcng! h lies in its lCtio nal, al - potential fur a mechanica l account uf thought and reason, but as we 1 by conven- shall see, it Joes so in a v•ay that rdegate;;s syntactic properties to 1g out what "going proxy" f11r semantic propaties. and that opens it up to the ii symbol in worries about cpiphcnomt:nalism discus:>ed here. My ain 1 in the ne ...;t al % nc. section is to mut iva1c prcmis<.: 3 - that internal proces:>or'> are sensitive ·ut registers to syntax, 1101 rnc;rning - ~inc e it will take the heJt in rny proposed ts. Suppose soluti1 1n. 4 volts, and m is an and 1d the 7-volt 2. The brain as a synlactic engine driving a semanlir engine ight he one The title you just read ca n be understood by atten tion to a simrle ed to states example, a cornnwn type ur computer adder stripped d11wn so as to ! functional ham.lie only one-digit addenda. To understand !he example, you nec:!d termine our only know the following simpk facts ahout hi na ry notation: 0 and l are ; and gates) reprc"cntec.J alike in binary and decimal. but the binary 1ranslation of Jots. So it is decimal ·2· is · tll'. The adder pictured in Figure 2 \\ill solve the a syntactic following four prnhkms: he first gate 0 1 0 = 0 .d gate. We I t U = I are until we 0 t I ;ed to mean ~ I = Ill not choose The first thrcl' equations ;m:: true in both binary and dc1 ·imal, but the 1 relevan t to last is 11 uc only in binary. aspects of I krc is huw 1lic ;1cJdcr wur ks. The two digits to be acJdnJ. a ' I' amJ a ·o· in this case, ;1rc co1111ectcd both to an und gate anti J.11 exclusive-or in cognitive gate. The lattt:r gate i'o a .. dillcrencc detector ... i.e., it ouq uts a · 1· if its eriving from inputs arc difkn:nt. anu 11 '()' if thl.!y arc che same. In the case on the one illuslratcd in FiFt11 L· 2. til e 1·1dusi1 ·1·-or ga te secs a ditfcrc ice, and so ic 144 NE D BLOCK The fi rst maps the : : onto the 8J The ke two fun ct 0 I. A II . Sy· Figure J II J. u ~ The symt out puts a ' l ' to the rightmost box of the answer register. The and gate meanings outputs a 'O', and so the device computes the answer. The e.xclusive-or useful re gate does the "work" in the first three problems, and is needed only for function carrying, which comes in only in the last problem as illustrated in descri bed Figure 3. Both inputs are ·J's', and so the and gate outputs a ' l ' to the these two leftmost hox of the answer register. The other gate makes the rightmost usual sen hox a 'O', and so we have the answer. how it i ~ Seeing the adder as a syntactic engine driving a semantic engine numbers. requires noting two funct ions: one maps numbers onto other numbers, semantic and the other maps symbols onto other symhols. The latter function is of symbo concerned with th e numerals as symbols - without attention to their The idea meanings. Here is the symbol function: nature ha tions amc '(r, ·o· -+ ·o· of the sy 'O', , 1~ ') ' "syntactic ' I '. 'U' - ·r and 1·s ti - ' l(J' • 1· , 'I' - meaning - proces· This symbol function is mirrore c.I by a function that maps the numbers a correla 1 represent ed by the numerals on th e left on to th e numbers represented their fon by the numerals on the right. This function will thus map nurnhers ont o drive our numbers. We ca n speak of this function th at maps numhers onto Now t1 numbers as the semantic function. since it is concerned with the driving a meanings of the symhols, not the symbols themselves. (It is important intention not to confuse the notion of a semantic function in this sense with a our cogn• function that maps symbols onto what they refer to; J shall discuss this the synta latter function shortly.) Here is the semantic function (in decimal ca usal we notation - you must choose some notation to express any fu ncti on): o. 0 0 0, 1 --t l I, 0 -+ 1 In this s I , I -t 2 making ti Can rlre 111i11d change rhe world? The firs t function maps symbols ont o symbols; the second function maps the numbers referred to by the argu ments of the first function onto the numbers referred to by the va lues of the first function. The key idea behind the adder is that of a correlation between these ) two functions. The. Jcsigncr has joined together I. A meaningful notntion (binary notation) 11. Symbolic manipulations in that notation Ill. Useful relalinns among the meanings of the symbols The symboli c.: manirulations correspond to useful relations among the and gate meanings of !he sy mbols - namely, the relations of addition. The xclusi1Je-or useful relations n111ong the meanings arc captured by the semantic :d only fo r function above, and the corresponding symholic relations are the ones stratecJ in described in the symbolic function above. It is the con-elation between ' '1' to the these two functions (which establishes a semantic function in the more rightmost usual !'ense of a function from words to their referents) that explains how it is that a device that manipulates symbols manages to add tic engine numbers. Now the idea of the brain as a syntactic engine driving a numbers, semantic engine is just a generali zation of this picture to a wider class :unction is of symbolic activities. namely. the symbolic activities of human thought. •n to their The idea is that we have symbolic structures in our brains. and that nature has seen to it that there are correlations between causal interac tions nmong these structures and sensible relations among the meanings of the symbolic structures. The primitive processors " know" only the " syntactic" form of the symhols they process (e.g., what strings of O's and 1·s they see), and not \\.·hat the symbols mean. Nonetheless, these meaning-hlind rrirnitive processors control processes that " make sense" - processes nf decision. problem solving, and the like. In short, there is e numbers a correlation between the meanings of our internal representations and !presented their forms. And this explains how it is that our syntactic engine can nbers onto drive our semantic engine . 2 1hers onto Now the picture just sketched of the brain as a syntactic engine with the driving a semantic engine n:vea ls how it is that a mechanistic theory of important intentionality can invite the charge of epiphenomenalism. It seems that nse with a our cognitive processes exploit a correlatio11 between the semantic and Jiscuss this the syntactic. The syn tactic.: properties of the representations do the in decimal causal work, and the semantic properties come along for the ride. unction): 3. The appeal to laws In this section, I will exnmine a putative solution, that is, a way of making the cogniti\t: science ric.:ture (premises 2 and 3) compatihle with 146 NED llLOC!-: the causal relevance of content (to the behavioral and other effects of contentful mental states, of course - I'll be leaving the prepositional phrase out often, just speaking, elliptically, of the causal relevance of content). Actually, I shall start by briefly mentioning a rcJuctionist proposal just to set it to one side. If content properties could be identified with, say, neurophysiological properties, then there would be no opening for epiphenomenalism. If content properties are simply identical to neurophysiological properties, then the causal efficacy of the neural would guarantee the causal efficacy of content. Whatever the merits of physiological reductionism, it is not available to the cognitive science point of view assumed here. According to cognitive science, the essence of the mental is computational, and any computational state is without I " multiply realizable" by physiological or electronic states that are not ·•fork ·· c; identical with one anothe r, and so content ca nnot be identified with any Herc i one of them:l fire to a Note that in rejectin g this putative solution, I am not rejecting a enough "physicalistic" point of view. If all the nomologically possible things explosirn that can have computational properties are physical things, then the enough I computational point of view, embracing this idea, is itself physicalistic. will cxpl· Even if the computational properties that characterize mentality are in Nov,· t. this sense physical, that c..loes not make them physiological or electronic and clcc (or syntactic, for that matter), and so physicalism in this sense does not elec tron ~ lead to any suggestion that processors in the head can Jetect content or electrica. meaning. cient for There is anotha putative solution 10 which I will devote more rnnducti allention, one that appeals to a nomnlogical view of causaiion. The idea ' I take 1 is that there arc non-strict psyc hological laws involving content, and 11utin11 u that law - even non-strict law - makes for causa l relevance. Fodor I the n(mJ I ( l 987h, forthcoming) argues that intent ion al laws provide non-strict crcast: d intentional sutllcient conditions for behavior, and that is what makes explo:-,io1 content causally relevant to behavior. It will pay us to examine a simple rising d version of this nomist perspective: Fis causally relevant to an effect e if relevant the instantiation of F is nomologically sutticient fo r e - even if the that allo nomological sut1iciency holds only ceteris parihus. lo explo' The trouble with this simple version of the nomist approach is The o familiar: there can be correlation without ca usation - even nomologi avoid tit cal correlation of F, with G without a causal relevance relation between cases h c.l F and G. And nomological correlation can involve nomological suffi cases. a1 ciency. Suppose C (for cause) is nomolng ica lly sufticient for and uncon t erm~ nf to (for troversially causally relevant E e fl cct) . Suppost: X is nomologi way !l> g cally correlated with C bt:cause X and C share a causal source, and so about wl X is not causally relevant to C. Then, X is nomologically sufficient for E he one I (·an rht 1J1i11d clumi;e rh e wvr/J? 1-l 7 r effects of cpositional .: levance of ·eJuctionist s could be e would be C;iusa.1 are simply efficacy of hacever the Figure -I 1e cognitive ~cicnce, the inal state is without being cau ~; tlly rclc.:' ant to E. See Figu re ..f, which pictures this hat are not "fork .. case. ed with any Herc is an example . Consider a setup in which a metal roJ connects a fire to a bomb. So long as tit<.: thermal conductivity of the rod is low, not rejecting a enough heat is tr;1nsferred from the fire to the bomb to cause an sible things explosion. But if the thermal conductivity of the rod is increased ,s, then the enough (say, by altering it..; compDsi tion ). then the heat from the fire 1h ysicali st ic. will ex plot.le l he humh. -i tality are in Now· then:.: is a l:1w - th e Wiedemann-Franz Law - linking thermal Jr electronic and electrical i.:ont.luctivity under normal conditions. (The same free .se Joes not electrons carry hu1h charge and heat.) Hence for this setup, ri sing t content or elec1rical conJuc1i' ity , logcther with other things being equal, is suffi cient fo r an t.:xplosion . Hence by the nomist criterion, rising electrical cvotc more conductivity is c au~ a lly rekvanl ll > the explosion. in. The idea I take it that 10 1lic cxtcnl thac we have a prctheoretical grasp on !he ont ent. a nd notion uf causal n.:lcvancc , I his consequence l>f the simple version of 1nce. Fodor 1he nom ist criterion is simply wrong. The eleclrical Cl>!lductivity in e non-strict crease does not (;1usc 1hc expl os ion; certainly it does not cause !he what makes explosion i11 1·irt11c· uf hein.t: a 1ise i11 c:lertrical c:o11cluc1ii·ity. Rather, the ine a simple rising electrical condue1ivity is an inactive concomitant of the causally .n effect e if relevant risi ng thermal conductivity. It is the ri sing thermal conductivity even if the that allows more heat tu be conducted 10 the bomh, causing the bomb to explode:. approach is The obviuus numist response \vould he to reformul!ltc the thesis to n nomologi avoid these fo rk c:1ses. The idea would he to formulate what the fork ion between cases have in wmmon that uistinguishcs tht.:m from genuine causal logical suffi cases. and use this result 1u produce a definition of causal relevance in . and uncon terms of .. non-fork'' numnlogical sufficiency. I think this is a promising is nomologi W'JY 111 go ab(l ut cliaracteriLing causal rdcvancc, hut I have two doubts urce, and ·so about whether the resulting nomist conception of causal relevance will ;hcient for E he one 1hat saves us frnm cpiphcnomenalism. 148 NED BLOCK First, recall that the problem of epiphcnomenalism arose to begin law is causa with because the cognilive science point of view held that nature had no progress produced a correlation between syntax and semantics. only the former being directly causally related to behavior. What reason is there to think that an independently motivated definition of " non-fork'' nomo logical sufficiency would count semantics as any more causally relevanl to behavior than risi ng electrical conductivity is to the explosion in the We have jus example considered earlier? Indeed, the correlation between semantics of nomologi and syntax via a common cause is a good example of a fork. nalism. I k11< Second, -the issue of the epiphenomenalism of the semantic may re le\·a nee. 11 simply resurface in a different form in an independently motivated section, I wi ' account of causal relevance in terms of nomologicity. Horwich (I 987, - the Coun Chapter 8) frames a definition of causation in terms of the idea of a vancc is con basic law. A ca use is linked to its effects hy chains of clircct nomological way of avoic. determination, and <..lircct nomological determination is determination Su pposc v i via basic law. "A direct cause of some effect is an essential part of an a {'I'. 'ff} ru antecedent condition whose intrinsic description entails, via basic laws a ' O' ou tput, of nature, that the effect will occur" (ibid.). Cause c causes effect e if just !he sam and '()' hat.I there is a set of events, r 1 ••• e,, linked by basic laws, and if c = e 1, and l e = e,,. The account concerns event causation, not the causal relevance these symho of properties. But assuming that it could be supplemented to form an been using account of causal relevance using the notion of a basic law. we can see rrimit ivc f)f • I that although it shifts the terms of the disc ussion of epiphenomenalism, object (say, ; I another strir i it does not solve the problem. Is content epiphenomena!? That depends l on whether the laws !hat determine behavior in terms of content are ca us ed the s: philmophica •' basic. But one's decisions on whether these laws are basic - and. indeed, what basicness is - will be (and should he) conditioned hy the meanin ~ one's decisions on what is causally relevant to what. conclusion i ~ Of course the fact that the nomist conception of causal relevance will nal, so arc t I not solve this pruhlcm all by itself is not a reason to reject the nomist I want to 1 conception. tions arc thl Fodor (forthcoming) says !hat if there is a causal law that F instantia for cpiphcrn· tions cause G instantiations (cetcris pan"b11.r ), then F is causally relevant The f uncti to the G instantiations that arc cause<..! by F instantiations. The appeal thinking, pl a to causation in this sufficient condition is harmless for Fodor's pur bines with a poses, but not for mine. For how are we supposed to know whether the hctween inp iI laws that relate content to behavior are camal laws? For example, arise from ti consider the (very low 4uality) law to the effect that instantiations of fun ction a I ist ~ wanting G and believing that A is required for G cause in stantiations of body. others A, ceteris paribllS. To know whether the content properties arc causally these imping relevant to the action. we must know if the law is causal. But if Fodor's ists take fun suffici ent condition is all we have to work with, the issue of whether the partly int ern < 1111 tlw 111111cl c:lrc111gt.' the 1rorlcl:1 l 4l) ' arose to begin law is causal just is t lic epiphcnorn<:nalism issue we are discussing, so int nnture had no progrc: 1.,s has been 111 adc. l nly the fo rmer I on is there to n-fork" nomo ~. A proposal usally relevant requires no commitment on 1his matter. More speciflcu lly. nothing I say here !> hould be construed as an endorsement of" narrow content." coming th1 Empedocl1 A functional role theory of meani ng is a theory nf what meaning is that yields an account of what it is about representations 1hat gfres . me amng a them their me anings, not a ·' semantics" in the sense of a theory of functional particular constructions in particular languages. Thus, it would not be functional the job of a function al role theory of meaning to explain why it is that Perhaps 'The temperature is 70°', and 'The temperature is rising' do not entail incl udes a; '70° is rising'. Theories of the sort that would Jeal wi th such questions here wher often address th emse lves to what meaning is, but they do not yield an hence it s · account o(what gives representations th eir meanings. Usually, for their causal cha purposes C:t ny "surrogate " of meaning - anything th at acts like mean coming th i ing in th e way tlrnt set theoretic entities " act like'' numhcrs - will do. fri end arr (See Hlm:k, 1986, 1987, for more on this distinction.) A functional role fl eeing wo th eory can he thought of as a " use" theory of meaning, wh ere the uses avo id a nee. are at least partly in si de th e head. sentence i One motivation fur such an account can perhaps he seen more cl early been diffe1 if one reflects on a case in which one learns the meanings of words Jn sum, without anything approximating an eliminative definition for them. tation wou Consider, for example, th e learning of a new scientific theOI)' wi th its been ditte1 new theoretical terms. e.g., 'force', 'mass', ' moment um ', 'energy'. These different f1 new terms are not definable in everyday language (or anythi ng like an Counterfac "ohsc rvation language"), th ough they arc definable in terms of one In t er m ~ another. T he s1u dc11t learns these terms by coming to understand how can be se n to use them in thought , in experiment and observati on, and in solving Are the mi problems on quizzes. If meaning is functional role, th en it is easy to see inputs'.' wh)' learning new terms is acquiring their use . It is a plus for a th eory of Of Ct) Ur ~ what meaning is if ii al so tells us what it is to know and learn meanings. Thinking < That is all I will say here to motivate the func.:til>n al role account. 5 processo r · Now on to th e Counterfactual Argument. represt: ntn You will recall that the Counterfactual Argument said that it ap involve di t peared that a given syntactic object would have caused the same output represc.: nt a even if it had meant something quite different from what it actually would ha v1 meaning. 1 means or even if it had meant n"o thing al a ll , so th e sy ntactic form of a representation scrl'l'llS uff it s meaning from having any causal rele for a gem va nce." proces~ur ~ Suppose, for example, that the scntcncl: "I here is danger coming this compkx r way' is in the "bclid hox" (with its normal meaning). causing one to wou ld han meaning. l fl ee. According to the Countcrfactual Argumc..:nt , the sentence would possibili ty . have caused the fl eeing even if it had meant that my long-lose fri end This pl11 approaches, or !hat Empedocles le aped. Bui if funct ional rule seman tonom1n1s tics is correct. then ii is not at all guaranteed that 'There.: is dange r 1980). Oh ~ Can the 111i11cl change the ll'or/il? 151 lly. nnthing I say coming this way' would ha\L' l·ausctl fleeing even ii it had meanr that iw cont ent. " Empedode:-. leaped. For if it h<.1 Then the fact that the frog's fly reprcsentt.:r has been ca rrying informa tion about flies has causally concributed to giving this represi:ntation the functional role that it has - being produced by retinal movement flashes and guiding zapping. More generally, the line of 1hought is that The plot ~u what our representations have been indicating about the world has had ism and sla) an influence - via evolution - on their having the functional roles the victory < that they have in our hcaJs. that funct io I do not wish to gu into th is reasoning in any detail , consider can win a h objections to it, or talk ahout the extrapolation from frogs to people. I the end be want to point out only thac even if the reasoning is entirely correct in its crucial ca~e own terms, jt does not show that the informational content of a section is tll nrn ha vi: lh1 representation is part of what is causally relevant to (in th e sense of a ''triggering" cause) the behavioral output th at th<: representation causes, tentative. and so Dretskc's proposal is nnt a solution to the problem heing In hrid. 1 considered in this paper. consist in ti Suppose that the frog has a fly-word ('FLY' ), the informational properties) cont ent of which has been causally relevant tu the estahlishment of lhc input ~ and intcrnul functional role of ' FLY', incluJing the guiding of the frog's non-fu11cti111 tongut: zapping behavior. ls lhis informational Cl>rllenl thert.:by causally nut the tunt relevant to the! prouuclion of, i.e .. involveu in th e causation of. any To get al that of a fu 1 particular zapping? This is the cpiphcnomenalism i ~sut: of th is paper (applied tu this case). The answer is obviously not. X can have cau!)a lly which I me: or other (sa· promoted the pattern of Y ~ Z without in any way trigge ring the one anoche. curre nc token of z. Fm X can have promoted Y --+ Z without now causi ng Y or enabling Y to cause Z. The informal ion al content of prope rt ics 11 ' FLY' does not causally contribute lo the appearance of this loKcn of cnies.) Con ~ ·FL y · in the frog 's head. Thal is done hy the fl y that caused it. And ignoring the once 'FLY' lws apreared in lhc frog, th e informational conc cnt does provokes th not enable or aid ·FLY' in proJucing a Lapping. To dramatize the anger. The ' point, suppose that thi.: current ·FLY' token is a mis reprcsi.: ntat ion live, of havi1 caused, say. by a 8 -B. This fly token nonethdess indicates Hies (a some propt· 1 misrer>resentation of a fl y is still a rcprcscnta1ion of a fly). The history docs the pre li ve n e~!-. cau of correlation of ·FLY' tokens with flies has contributed tO the func bull is coo tional rok of· FLY' tokens in the frog, but once that role i~ se t, the past correlation is irrelevant to the process hy wh ich the B-B now produces provoke the the current ·fly· wken, which in turn prnduces the zapping that pops Another l ' the 13-8 into the frog's gut. We can tell the whole mechanistic story property. t Ii· first-oruer <.:I abouc this causal prncc~s without saying anything ahouc how th e mecha x is dormiti' I nisms that s ub~erve it arose. And it is this former question ch at this (when x is it paper is about: is cont enl part of the causal proce s~ by wh ich behavior is produced? your nuticin1 I falling aske1 (an r/((' 111111d «liu11gc rite 11 c>rltl :' 155 'ing informa sentation the 6. Functional properties d movement The plot so tar is: fun ctionalism nh.:ds arguments for cpiphenomenal ought is that ism and slays them. II wo uld be nice to stop here. but alas, the story of ··orld has had the victory of fun<.:tionalism m·cr epiphcnomenalism is fiction. I argued ctional roles that functionalism docs defeat the Countcrfoctual Argument, hut you ca n win a battle and still lose the war. Fum:tionalism loses the war in ai l, consider the end because functional prnpcrtics are causally inert in certain . to peoplt!. I crucial e<.1-.e :-.. Or rather. I fear that i.1 11 this is !rue. The point of chis correct in its section is w raise a sh:cptical doubt. The issues iJre complex, and l do :ontent of a not have the "P<1ce It\ C'i plnrc thc111 :1dcquatcly. Su my claims must bt: 1e se nse uf a tentative. ation causes, In brief. mv puinl is this. h1ncti1H1 al propaties are properties that l)hlem being consist in the having tif sumc properties nr other (say non-functional properties) that ha\'c ccrtain causal relations to one another ami to nformational inputs and outputs. In the production of those outputs, it is the 1ment of the no11-fu11ctio11 al pmpcrtiL: S that :ire s1;111dardly the causally relevant ones, .>f the frog's not the funct ional 1mipcnics. reby causally To get at the poinl. let's consider a slightly more general noiion than .tion of, any that of a functional property. the mition tJf a second-order property, by if this paper which I mc:rn :.i property tha t consists in the having of some properties 1ave ca usally or other (say lirst-orJn properties) that have certain causal relations to iggering the one another. (The gri..:atcr gener;tlity here is jus1 that second-order .vithout now properties needn't i11nilve inputs :.inJ outputs, as with function;d prop I content of erties.) ConsiJcr the bullfighter's cape. The myth (which wc will iJccepr, .his token of ignoring the inconvenient color-blindness of bulls) is that i1s red color Jscd it. And provokes !he hull ; that is. redness is causa ll y relevant to the bull's ·ontcnt does anger. The c .1pe also has the "l:CtH1d-urdcr propc:rty of hcing provoca amatizc the tive, of having some pniperty 1>r other that provt)kcs the bull, of having presentation so me property or other that is L·ausally relevant to the bull's anger. But ates flies (a does the prolDC(ltirencss llf till.' care provoke the hull? Is the provoca The history tiveness causally rele\·:1 111 to the bull's anger? Ir would seem not. The to the func bu ll is tno stupid for that. The provocativeness of the cape mighr set, the past provoke the ASPCA. hut not the bull. Jw produces Another example: c1111sider dormitivity construed as a second-order ng th~1t pops property, the p o~st:ssit>ll uf some property or other (for example. a anistic swry fi rst-order chemical prnperty) that is causally relevant to slcc.p. That is, 1 the mecha x is dormitive = x ha :- some plllperty that is causally relevant w sleep on that this (when x is ingested). II a dnrmitive pill is slipped inlO your food without ich behavior your noticing. the property of the pill that is causally relevant to your falling aslct:p is a (p1 L·s 11111ahl~ lir!->t-order) chemical property. not, it 156 NED Ill <>CK would seem. the otions will act Second-order properties involve having some properties or other, and 1m, ;mother in though it is often hclpf ul to think of the propenies quantified over as mitiviry of the first-order. actually they can he any properties at all. There is a general to your falling procedure (see Lewis. 1970) for defining a second-order property, given a theory in which the property plays a role so long ;1s the theory allows hen it ca n be some sort of a distinction between theoretical and observational terms. ·ss of the cape (In the case of a psychological theory. the distinctilln would he cashed hennmcnon in as theoretical = mental. and observational = input ,'output.) If the the ho effect. If a oretical terms arc 'T1' .•. 'T,, '. we can write the theory as T(T1 ••• T,,). its dormitivity, leaving out all mention ol' the Dbservational entities. Then we can • •• • •• ]. ification of the define 'T ,' as follows: .r ha s T 1 == EF1 EFn [T(F1 f) & x has F1 1itivity can be So far, the case I've mac.Jc for the limited causal inertness of second uitivc pill that order properties is based entirely on examples. But if the co unterintu 1ity requires its itive consequences of this causal inertness are as had as I will claim they rnitivc pill, and arc, a natural re sponse will be simply lo live with rejecting my way of and close my taking such ex<1rnplcs. That is, if I am convincing Liter when I say why up, nnc of my the causal inertness of the second-order commits us to a view of the It of taking my special sciences I hat is hard to sv..;allow, the rcason;1ble response would and sues me. he that we should just suppose that dormitivity is causally relevant to the pills were sleep, and provocativeness does affect the bull. So I wi ll try to get at the principles that underlie our reaction to the examples. I can think of ilways causally two. 1ed.H The only First. let us return to thl.'. nomist conception of causal relevance. On ies seem to be that conception, second-order properties are not ~ausal l y relevant to . . rng recognizes the effects in terms of which they arc defined because they are not cases where a nomologically sufficient fur those effects. Consider <.lormitivity and 1d that effect is sleep. The relation between the two is more like the relation between property by an being a widow and having had a hushand than th;1t between, say, heat ntence of this and expansion. If a pill is dormitivc in the following sense: x is Jard cases, the dormitive iff x has some prl)perty that causally guorantees (this is where this definition differs from the one offered earlier) sle ep if x is ingested inert in some - and I take the pill, it follows that I sleep. The L1ct that dormitivity is ms of a nomist sufficient for sleep is perfectly intelligible in terms of this logical ! nomologically relation. What reason is there to suppose that there -must also be a keeps trying to nomological relation between dormitivity and sleep? rcnignition of Now. I am very much not saying that a logival relation between Jsc) dormitivity properties pn·cludes a nomological relation. This is as much a fallacy al relevance to for properties as fur Davidsonian token events. Suppose dormitivity is is nomologically my aunt's favorite property, and sleep is my uncle ·s, and that my uncle :c to cancer but tracks changes in my aunt's favorite property, changing his own so that 1tandard causal his is always entailed by hers. Then dormitivity and sleep will be hath nomologically and logically related. Logically related under one set of nt.) 158 Nl·.D 111 .0CK descriptions, nomologically under another. My point is not that a logical In other w. relation precludes a nomological relation, hut rather that the logical explosion \· relation between uormitivity and sleep tells us perfrctly well why conductivil 1 dormitivity involves sleep. There would have to be some 5pecial reason ing an epip to postulate a nomological relation as well, and since the stOI)' about my The pi ct aunt and uncle wasn't true, I don '1 sec any such special reason. The analogy ju! point that this example is me;rnt to make is that it would he amazing if relations n there was always some special reason why a second-order properly was different, a nomologically related to the effe cts in terms of which it is defined. (limited) This consideration is based on the fa ct that sect)nd-order properties er used to pr< are defined in terms of effects. A second argument is based on a chemi cal pr different feature of sc.;concJ -ordcr prnpenics, the quantification involved tivil y illtistr in them. nalism' an J Suprmsing that provm:ativcncss provokes the hull would be supposing gu men l a ls1 a strange ·sort of ovc rd c.; tc.;rmin a ti on of the bull's anger. Of course ove rdetcrm overdetermination docs sometimes happen. (The placebo effect is an coincitkncl' example.) But to suppose that it always happens would be to suppose a is always a bizarre systematic uvcn.ktcrmination. Wheneve r wt: have a first -o rder coinciJencl' causal relation we can always defi ne a second-order property on the causally efl i model of provocativeness. • th e prnpnty rcla!cd to 1>11 5. These ot h< a gcars-and compurcr irn tronic implc causally rclc' Figure 5 such imrlcm The up<; hc desire for an ice cream cunc is causally releva nt to my !!ning to the ice what cu1111w cream shop. 11 implement al At this poi111 , I shall change gears ahruptly. /\lthn11gh . as l'\'c said, sl ril ngc con\ the case I've matle for the (limited) incflicacy of the second-order comrarahly property is far from conclusive. I think I have said- enough for it 10 be tional scicnt taken seriously. What I now propose to uo is simply ns.mme that ( 'onsid er second-order properties are not always causnlly relevant lo the effects plausibly f 111 in terms of which thl'y arc defined. and go on to discuss 1hc conse tancc and quences of this idea. actual char;1 The skeptical point of this section depends on !he fact !hat the dominant 0 1 mechanisms that manipulate genes and the mechanisms that manipu of the di~l 1 late the internal rep rese ntation ~ in the hrain are not inte ll igent, and so there arc n cannot recognize second-order properties. Hence these are just the prorerty of kinds tlf cases in which it would seem that the scconu-nrder properties chemical pr arc inefficacious with respect to the effects in terms nf which they arc via the prm defined. Likewise for other special sciences whose properties nrc rlausi tional prop bly second-order. proc cs~ors Uy way of an example. consider a simple finit e automawn, !he one tics such ; specified hy th e machine tahk of Figure 5. What the table requires of p rocessor~ any ma chine that it ckscribcs is that it have two states. S 1 anc.J 5 2 , such rrorcrt ies that wh en th e machine is in S 1 a11d sees a 'I' it cries 'oc.lc.I' and goes into which thcv 111 52 : and when the machine is in S~ and secs a ' I', it cries 'even· and goes gene . hack in to S,. And when the machine is in S 1 / S ~ and sees no input In su m. ('-' symholizcs the null input). it cries 'cven·; ·odd' (respectively). inconsistcn staying in the same state. Thu<; the machine tells us whether it has seen I. Speri. an odc.J or even numhcr of· rs· (though 1101 entirely sunT..,sfully. since it scicrn is under the impression thar 0 is an even numher). 2. Speci Now suppose we have a machine on the desk in front of us that ~- runcl satisfie s Figure 5 the prnpcrty 1ha1 n111-;i~f'.. 111 h:l\·i11g twn other properties that are reli1tcd tu 1rnc a1111th u a11d Ill inputs and outputs as spccifk<..1 in Figure 5. These other propcrtici; would he mechanical properties in the case o f a gears-and-pulleys impkmL'ntaliPn. electronic properties in a usual compute r i111plcmcntatio11, ;111<..I so t>n. If we arc Jealing with an elec tronic implementation. then it is the electronic properties that are causally re le va nt to the cry of 'oud·. not the property of possessing some such im plementation proper!~. The up<\ hot is lhat comp11tt11io11al properties lu11 ·c 110 causal relel'lmce fO 1g to the ice wlim cumpurcrJ clo. these c.:fkcts hcin!! entirely the product of the implementations of the cn111putatio11al prorertics. This is indeed a as I've said, strange cnnclusio n. and if I i llll right about second-order properties. ccoml-ordc r <.:t1111p<1r:1hly st1 ang.e C(l nl'lu ~i11 11 s l'lltild he 1t:acl1cd about olhn func for it to he tional scic11n:s. 1.rn11ne that Consider the rrorc r1 y of !ici ng a brown-hair gene. This property is ) the effects plausihly functionally characteriz:.ibk in terms of its patterns of inheri ; the consc- tance and interaction \\ith uther genetic properties in determining actual characteristics. for e:w 111ple. in the fact thal brown-hair ge nes are lct that the dominant Dvc r hlonc.J-hair genes. Such second-order properties are part ha t manipu of the distinctive conccptu;tl apparatus of Mendelian genetics. But gcnt, and so the re arc 1H> ge netic 111ecl1 ;111isms that detect c.) r arc sensitive lo the are just the property of being a hruwn-hair gene. Genetic mechanisms detect bio ·r prnpcrties chemical prnpcrt ies. anti the propert y of being a brown-hair gene arises ich thev arc via the processing hascd c>n hioc.:hernical properties. Similarly, computa s arc plausi- tional properties such as S 1 men tinned above arise via the activity of processors thal detect sy ntact ic properties. not computational proper i C.Hl, I he one ties such as S1. And se nwntic properties arise via the activity of ; requires of processors lhat c.ktcct syn tact ic properties. All these second-order :.1n d S z. such properties arc epiphenomena! with respect Lo the effects in terms of nd goes into which t ltcv arc dcfinccJ. such as l:irnwn hair in the case of the hrown-hair en· and goes gene. 111 ~es no input Jn sum. we want to 111;1i11tai11 all of the fu ll owing. but they arc ·espcct ive ly ), inconsiste nt : :r it has seen I. Special science rrnper1ie'> are c;1u <\a lly rek,·;111t to the effects those fully. si nce it sc.:ienc.:es predict ;111J explain. 2. Special science prcirc r!iL'S arc often functional. 1t of us th at J. Fune.: I iunal proper! ics arc sla nt.lard !~· causally irrelevant !o the ; it sees a ' 1', e ffe cts they are defined in tcrms (lf. state S 1 (of a iroduction of J have just argued for J. JI I am right. then we have what for many of provocative us is a uilemma: we rnus1 ei ther abandon functionalism or accept >rope rty, it is epiphcnomc nalism. But am I right about Y? Recall that the case for J 162 Nl· I> 1.11.<> CK depended mai nly on the nomological approach to causation, and per causal-expla r haps the way out is to reject that approach. Before accepting that of my fallin! conclusion, however. it would he best to examine another way out. second-order appeal to du 7. Because Dormitivity causally relc' The claim that second-order properties are standardly causally irrele appeal to a 1 vant to th e effects in terms of which they are defined seems to fty in the ti on of sleep face of a va ri ety of commonsense and scientific facts, for' example, true chemical pm uses of the ro ll nwi ng: second-order I slept because I took a sleeping pill. in the posse~ My muscle tension evaporated bccau::.c I took u muscle relaxant. cause it invol The vase broke because il was fragile. property ca11 J have blue eyes hccause I have two bluc-cyt: genes causally dlic fi rst-order or l shall argue that such facts only seem incompatible wit h the (stan for the stand dard) causal inefficacy of the second order because of two ambiguities tha t easily escape notice. Lrpla11 atio11 and ctw.rn1io11 Func tional le.: One con'itru. One ambiguity is that ht:twccn explanatory and causal usc5. of ·because'. dormitiv1 1y ""'" It is easy to be unclear ahout \.\hcther a ·oecame' statement is true in one equally j an explanatory or causal sense of ' hccau!->c '. If th e dispositional/ func This is th<: n< tional terms in th e sent ences lisled above are understood to refer to dormitivc:: vir, second·order properties, then the sentences are only true on an ex construal is l planatory, not ;1 causal, reacJi11g of· because· - if the line of tlwught of lion that pid the last seet ion is correct. former const1 Audiences who have heard this chapter read as a paper have ofc en to cont ext. · 1 objected that al though thc.!re is a clear distinction between ex planation sa me SeL·o 1H.I and causation. th e dist i net ion het ween the n wsal-expla 11 1.1to1y rele1'w1ce ditfcrc111 lir ~l and ca usal relc•1'a11ce of properties that I rely on here is a distinction potions. For without a difference. Herc is the issue: The kind of explanat ion in diffaenl fror volved here is ca usal explanation. which is one of a number of species Similar!), ·I he of explanation. So the kind of explanatory relevance of properties at lotteries (Lev issue is causal-explanatory relevance. And what is the distinction be To sum u1 tween causal-L·xplanatory relevance of properties and causal rdevance rekva nt Lu !> I of properties? the 'hecaust'. Here is a way of seeing that there is a difference and what ii is: notice ascribed to tl that we can ca usa lly explain a case of sleep via appeal 10 the sccond David Le~ order property of dormi tivity. I fell asleep because I took a pill that had Smart's itka the property of ha ving ~omc prnperty or nt ht:r that c;wse!'> skcp. Th i~ is Functional l e Can rlie 111111cl clwng!' rlie 11·or/11:' lhJ i, and per causal-explanatory because it ruks out alternative causal explanations ~pting thal of my falling asleep - for l!xample, that I was gra~ing pa~ers. But 1y oul. second-order properties being (standardly) causally 1nefficac1ous, the appeal to dorrnitivity is not an appeal to a causally relev~nt prope~ty. Dnrmitivity is causal-explanatoril y relevant tn sle ep without being causally relevant to sleep. Why? T he appeal to dnrmiiivity im·oh·es an ;ally irrele appeal lo a property that is genuinely causall y relevant to th: produc o tty in the tion of sleep, namdy, the unnamct.I property (pn.:sumahly a lirst-order 1mpk. true chemical property) whose existl!nce is mentioned in the analysis of the sccoml-ort.ll.!r propl!rt y, thl.! onl.! that the tlmmitivity ol the pill consists in tht: possl.!ssinn of. Dormi1ivi1y has rnusal-explanawry relevance be cause it involves a causully dlicaciuus properly. A causally inefficacious 1l. property c<.111 nonetheless be causal-explanawry if it "brings in"' a causally ellicacious property. Uut the causally rdevanl property is the first-order one , not the second-order one (assuming that my argument the: (stan- fo r the standard causal inertness of the second-order was right). 1mbiguities .·lmhig11i1y of f111 1c1io11al 1em1s Functional terms arc ambigullUS. Ll.!t 's lakl! 'dorrn1tivi1y' as an example. One construal is thl! seconu-orucr one much di sc11"isl'd here already - 1 · · bccau!'>c • dormitivity -"" pu.uess111g a prup('n)• ilwt causes sleep. The other construal, t is true in one equally justil ic..:<.J in ordinary usage, is rhc propeny rhat rnuses sl!'ep. )nal/ func This is the natural reading uf ·uormitive virtue' in the ;1sscrt i1>n that the to rdcr to durmitive virtue of Secon;.il is causally responsible for sleep. This latter on an cx construal is one in which ·Jormitivity' is analyzed as a t.letinite descrip thought of tion that picks out the firs1 -orda property that is quantified ove r in the former cunstrual. The two cunstruals arc vay dilh:rcnt in their relation have often to context. · Dormitivity', on the formt:r construal. always picks out the xplanation same seconu-nrc..l er property. But on tht: latter conscrual, il picks out v relei•w1ce dilferent fir st-order properties in the case of diffcrt:nl types of slcepi11g dist i net ion potions. For the prnpercy thac ca uses sleep in 1he c11se of Seconal is tnation in- difforent from the property that ca uses sleep in the case uf Valium. of spc:cies Similarly, '1he winning number" picks out Jilferc..:nc numhers in dilferenl .)perties at lotteries {Lewis's example). .n ction be To sum up: In the second-order sense. uorrnilivity is not causally • relevance relevant tu skt:p; in the first-order ~c nse. it is. And the plausibility of the 'because' claims mentioned al the outset of thi' section can be it is: notice ascribed to the first-orJc:r interpreration. 11 ne second- David Lewis has long advocated a form of functionalism based on 111 thal had Smart's it.lea of topic neulral analyses of mental 1er111s (Lewis, 1972). ·er. This is Functional lerms arc defined by Lt.:wi s in th l.! faller ol the two manntrs IM NFI> 111.0C"I\. • mentioned above - in terms of definite descriptions. Many functional out a psycl ists adopt a functiunal state identity thesis, one that identifies mental out a Mc; states with second-order properties. Lewis, hy contrast, says that mental property. states are fimc:tional/y specifi<'d brain st at es. Many fun ctionalists take Pickwickia pain to be a second-order state, but Lewis takes pain to be the the concCJ first-order hrain state that plays the role characterized hy the second part of thL order state. Where some functionalists identify mental states with states tinnal pro that consist in causal roles, Lewis l<1kcs mental states to be the entities nothing ar that . have the causal roles. The issue is aboul how to regiment the tus of clc<.: language of the mental (and other 1hcorctical langu relevant to sleep," "Content is causally relevant to behavior," and the state 5 1 h likc.12 If we accept Lewis's reading, we can preserve the talk that would of" Even! he appropriate if special science properties were causally efficacious. So tional stat · it seems th;-it we can avoid epiphenomenalism. and not have to give up the causa functionalism. And there is no need to reject a nomological theory of \arly, sim · causal relevance. psycho log. Unfortunately. the Lewis solution (as I shall call it) is only cosmetic. believing To construe the functional terms of a special science in Lewis's manner of the me is to construe them as picking out " lower level'' properties, properties Constn of an implemefllation science. 'Believing that grass is green' on the make the Lewis construal picks out a physiological property. (In us, that is; in the distinct iv( case of a hypothetical intelligent computer, it would pick out an way of r• electronic property.) And in a certain sense, 'beli eving that grass is Lewis's n green' does not pick out a psychological property: that is, it does not But 'havi pick out a property that is part of the distinctive conceptual apparatus ·s,· pi ck~ of psychology. On Lewis's construal, 'brown-hair gene' (a term used by tion stat< Mendelian geneticists long before the advent of molecular biology) Computa picks out a piece of D!'IA, the concept of which would have been utterly sciences. alien to Mendel. Consider the computational state S 1 of Figure 5. I said to them it was a computational state. and indeed the computational definition enccs. T was given in the text. But on Lewis's construal. 'S 1' picks out an sciences electronic state or a mechanical state. or some other implementation conceptt: state, depending on the way the automaton is constructed. 'S 1' would be spcci (·an the mind clwnge the world? 16.) Many functional out a psychological property. and likewise 'brown-hair gene· does pick itfentifies mental out a Mendelian property and ·5 ' does pick out a computational . says that mental 1 property. But these: consl ru;.ils o f "psychological rroperty'', etc., are unctiona lists take Pickwickian. Psychological properties, on this construal, are not part of pain to be the the conceptual apparatus of psychology; Mendelian properties are not ti hy !he sccond p:irt of the conccrtual apparatus of Mendelian genetics; and computa statcs with states tional propert ies are properties that a computer scientist may know lo he the entities nothing about. properties that arc really part of the conceptual appara to regiment the tus of electronics or mechanics or hydraulics. es). My point is this: You tfo nol vindicate the causal e ffic acy of the ng of functional properties of a special science by construing it s terms as referring to the ·ake and eating it properties of other sciences, impleme ntation sciences. The Lewis con itivity is causally strual docs preserve a way of talking that we want. We can speak of the · havior." and the state S being causally relevant to the production of the machine's cry e talk that would 1 of "Even!" But the cost is that we no longer take S to be a computa !ly efficacious. So 1 tional state (except in a Pickwickian sense). So a Lewis vindication of t have to give up the causal efficacy of the computational is certainly Pickwickian. Simi logical lhcory of larly, since believing that grass is green is a physiological, not a psychological, state, the preservation of talk of the causal efficacy of is only cosmetic. believing that grass is green is n(l real vindication of the causal efficacy 1 Lewis's manner of the mental. ~rtics. properties Construing terms of the special sciences in Lewis's manner does not s green' on the make the second-order properties I have been talking about any less us, that is; in the distinctive of their special sciences. Rather, it merely forces a different dd pick out an way of referring to them. 'Believing that grass grows', construed in ng that grass is Lewis's manne r, picks out a physiol ogical state, not a functional state. it is. it does not But ' having tile hcl id that grass grows' picks out the functional state. ~ ptual apparatus 'S ' picks out an electronic or mechunical or some other implementa (a lerm used by 1 tion state. hut 'being in S 1' picks out a genuine computational state. >lecular biology) Computational states are what are distinctive about the computational 1ave been utterly sciences. and Lewis's way of talking does nol prevent us from referring f Figure 5. I said to them or make them any less distinctive of the computational sci tional definition ences. The same holds fo r the functional properties of all the special , · picks out an sciences for which functional properties are part of the distinctive implementation conceptual apparatus. These properties remain distinctive of these ~ d . 'S ,' would be special sciences even if we construe terms a la Lewis. and these start• that has the properties remain causally im:rt (that is, causally inert with respect tu nor pick out a effects in terms of which they arc defined). In sum, moving to the Lewis way of talking does not solve the :>erty picked out epiphcnomenalism problem, but only provides a way of talking that .rcen · does pick allows us to avoid facing it. The basic picture pro\·ided by the Lewis way I 00 NI.I> llLOC'K • of talking is th e same as lhe one I have been trying to avoid: special 2 The: idi.:a d Sec: abo [) sciences whose distinc:live properties arc causally inert but thac can be hrst appc:a1 used to pick ou t causally cl!icacious properties of lowt.:r-lcvel sciences. ca u~al 1n1c: T his brings me to a second objection to che Lewis soluti on. Second among 1h1.: if ii encou order properties in the sense discussed here art! properties that consist ~ymbob, at in the having of properties wi th such and such relat ions to one another. meaning ~ ;1 As I mentioned earl ie r, the properties quantified over need not be itlcntifying nc:urolog1c; first-order. In the case of our menrnl properties construed as second 3 This has lo order properties, th e quantified properties are physiological. Phy~ iologi oc:en dialk cal properties are biologicul properties, and biological properties ge ner ~1.:c Kim ( I' ( l980a) a1g ally are prime· candidates fo r functional analysis. Thus ·believing that and Blac l-. b grass grows' construed in lhe Lewis manner, si nce it picks oul a 4 Ahernativc: physiological property, may pick ow another jimcrio11ally imlfritl11M~d tl.!mpc::ratu1 explode .... i property. Perhaps biological terms themselves pick out first-order prop parl will co erties, but perhaps nol. Perhaps only the most fundamental of physical IO bt U llCOI properties are ge nuinely firs t-order. In that case. even mu ch of physics the: explo ~ i 1 ClllHJ Uc:l i\'11 ~ would be epiphcnomt.: nal! O r perhaps not evt:n the most fundamental t.:1rndi1io 11 i ~ physical properties arc first-order. (Maybe there isn't uny mllst fun N I JI Ill fl! . h: illc 11 1i1y r cquiro.:~ hc.-11 1!! i:;i11,cd by Sall) . l.d'o1r ;111d I <•ewer may led 1ha1 lhc " ·•r 011 1 i' lt1 lalk t1C1I ahnul lhc pr11d11r1i11 11 of B1m·c. h11 1 ral ht•t o l Bruce lil..t· t·venh lypc'>. ' nor rok ens. lhc \\:ty 1111t llt' . ly p l''· '!! ;irE!llnll'lll Bibliogra phy cc ns 11lf 1he even 1( the Ollll:k. Neu. I 1JXlla . Rcc1cfin1:s 111 /'l11lo.w11lt1· of P.frc hnlo.1..'1 ·. I ; ,f, I . Cambridge. I l IJHfil for a Mass.: l larvard University Pn: .... -.. 1 11 11 toge t ht: r l!J80h. " What Is r u11r1irn1alis11i'. " In Block (198()). flt JI/II 1/o~ t n i/ l 98rl . .. Advertisement for a Semantics l<1 r Psyc hology." In I'. A French ct a l. , es i11 which ulfkient for eds .. Mid11 t'.1·1 Stwlic.1 in l'htl11w11h.1'. Vol. X . t-.linm:apnlis: University of nc;try ra ther !\f innesola Pres-., 6 15-7H . I 1187 ... Fu net ion al Role anu Truth Conditions ... /'111c t'<'d111g' of tlw / fm 1melinJ1 t: frog has a Sout·rr . .... upplcnientary \'Olu me. h i , I :'7-81. dcre11us o n Churchlc111d. P. S. . 198(1. Nl'11mp/11/11wph,1'. Cambridge. Mass.: MIT Press. . II 1101 a11ply ic Sn· 1hc Dchmelt, I Jans. 1989. "Triton . .. . electron . ... cosn11111 . ... : An infin ite rcgrcs siun'? " flm1 ·t•<'cl111g1 of'"" National / ln11lem,v of Sci<•11n·s. ~6. 8618. ispo,il io11;1l 1990. "Expnir11c11ts on the structure of an imJiviLlual clc1rn:ntary p;1rticle," rope rl ies - Scfrna. 247, 5J9-·t'i. : t:ial case. Dennett . Daniel, 1981. " Three Kimi-; of l11ten1ional P-;yc hology." In R . Healy, .1irned that anti Pcllit ed .. R£'c/11c rion. Time ancl Rcallfv. C rn1hridge: Camoridgc University Press. .,., in term'\ Dretskc. rrccl. 1988. F.xplt1111111~ /11-ltul'im · Uc·a.wnv"' a ll'orlt! of Ca11.'ic.:1. Cam t rrn:: n I ~ used bridge. Mass. : f\ t!T Press. ali11 n argu· Enc, Berent. 198.1. " In Defence of the Identity Theory." Jm1111al uf Philo.w phy. go further. 811. 279 - QH. propt:rt ics i caleu glass 198(), "Essentia l ism with lnJi, id ual Essences: Causation, Kinds. Super ;.ire many vcnience, and Restricted ldc111ities." In P. A. French cl al., eds .. .\fidwest i nne ase in Swdies 111 l'hilowphy_ Vol. X I. tin111.:apolis: University of Minnesota Press, he sp ecific 40J-2h. molecular o ns ;.imong Fodor. Jerry, 1975. The !.a11.1(11a~" of 1'/w11xl11 . Camhridg<.: . Mass.: I larvard re perhaps Univl!rsity Press. ir pn,iti1111 1980. "Mctl10dologic;1' Solipsism Considered HS a Rt:sl!n rch Strategy in Cog .. Bur what ni1ive Psyrhology.'· 8t·l1111 ·ioml nncl 11mi11 Sc:ienas, .1. 417 -24. 'llN. 19<>0.) l987a. ' "'-'"·'11111.'11111111ics . CamhridJ!e. Mass.: MIT Prcso; . II}" I Oll.\t11fy I locate all I 987b ... t-.faking Mind Mati e r More." This is Fodm's reply to l.cp11rc and tween two Loewe r ( f ll87). The abstract appeared in Jvunwl of l'li il11sophy. 84. M2. va ncc thal In press. " Making Mind Matte r Mon:." Philo.wphical Top.in. 1 t>f c:.iusal 1980. R<'ha/'· 1re es grow Haugcland. John. " Programs. Caus;il Powers. and l11! <.:r1tio nali1y." . rrusts in ioral anti Bmin Sciena.s, .1. 4J2-3. eing faster 1981 . ,\fine/ l ks1xn. Camhridge. Mass.: f\- llT Press. ·r multiply Horwich. Paul. 19~7 . .- tsy11111wmef 111 Timi'. Cambridge. f\ 1a s-;.: M t:r Pre-;s. I program- Jackson. Frnnk. and Philip Pe ttit. 1988. ·-r-unctiurwli-;m anti Uroau Content. '' ,\f111tl. 97 . .187. 38 1- 400. 1 ~ anything 1p Kitcher Kim . J aegv.·1111. l Kilcher, Patricia. 1980. " l fow lo Rec.Jun: a Funclillnal Psychology." Pltifosophy of Science, 47, 134-40. 1982. "Genes. HccJuc1ion ;1ncJ f u11ctio11;.il P~yd10logy. " l'hilow11l11· of Sc1t·11u» F 49, 633-6. Kilcher, Philip. 1982. " Gene~ ." 8n11sh Jounwl for the P/11/u.wphy vf S(lmce. 33. 337-59. LePore, E., ancJ B. Loewer. 1987. "Mind Mailers," Jounwl oj Philv.1Uph.r. 84. 630-4 1. Lewis, David, 1970 . .. How lo Ddln\.'. Tht:on.:tical Terms." Jo11n111/ of I'ltilosoph)', 67, 427- 46. 1972. "Psychophysical and Thcoret ic;.il Id ent i fi cat ions ... .· I1111rn/cHi1111 Jo11mal of l'hi/osoph.r. 50, 249-58. Reprinted in Block ( 1980). Minsky, Marvi n, 1967. Co111pwurio11 . Englewood Cliffs, N.J .: Prtnticc-l lall. Newe ll , Ahin, 1980. ''Physical Symhol Syslcms," Cog11iti1 ·e Science>, 4, 1.15-83. Putnam, llilal)1• 1960. "Minds and Machines." In S. llook. ed .. l>i111 n1.1 io11s of Mind. New York: New York University Pr<.:ss. 1967. "Psychological Prcdica1es." In \\'. H . Capiian and D . D . Merrill, eds., Sc.:ic ntili c.: re Art, Mind, and Rl'iigwn. Pittsburgh: University of Pillsburgh Press. Rc1i1led istic rrmluc "The Na1urc of Mental S1a1cs" in Pu tnam\ colle cted pap<.:rs. Reprinted in causal ~I rue Block (I 980). ac.loplion o 1973. 1975. " Philosoph y and Our Mental Life ." Moul, La111:1w1:c · and Reality: (fo_: 01 ii fl LJ I Philo.\Opl11rnl Pupers. Vol. 2. Camhridgc: Cambridge Univcrsi1 y Press. 1 causal ~11 uc Hcprirllcd in Block ( 1980), and in somewhat different form. in 1laugcland (1981). Originally published in Cognition , 2 (1 973), wilh a scclion on IQ the ca~c lo thal has hccn omitted from all of the reprinted vcrsions. appare111 ly , 1988. l kprese111a1io11 a11tl Rcaliry. Cambridge. Mass.: MIT Pre~s . involve 1 ek Pylyshyn, Zenon, 1984. Co111p111a1io11 and Cogmtio11: !S.W<'S in the Fo111ula1w11s of turcs; tilt.: c• Cng11 i1i1 ·r Science. Camhridge, Mass.: MIT Press. tu he thCOl) R id1ardson, Rohcrt, 1979. "J7unctionalism ancJ R c du ctio ni ~rn , " l'hilowplty of J lJ8Ha, I lJ8' SL"i~nce. 46. 533 / 558. Putnam l 9i Schiffer, Stephen. 1987. Rc11111n11rs of ,\-feunin1: . Cambridge. Ma ss.: MIT Press. that appeal Searle, John, 1980. "Scarlc's Reply LO Critics of· Mind ~. Brains and Programs'," Belial'iorul and Brain Sciences, 3, 450-7. Sosa, Ernest, 1984. ·'Mind-Body lntt:raction and Supervenicnt Causa1ion.·· In P. A. French c t al.. eds .. Midwcs1 Slluiies in I'hilo.w11hy. Vol. IX . Minneapo lis: U11ivcr:..11y of Minnesula Press. 271-82. A variL'l)' 1 thcory-dt:p• i ~ti c and p . . variou~ sci edge liu t conccpb a linguist re c lhl!Ol)-illd l