The Architectural Nonchalance of Commonsense Psychology

RADU J. BOGDAN
1993 Mind & Language  
Eliminativism assumes that commonsense psychology describes and explains the mind in terms of the internal design and operation of the mind. If this assumption is invalidated, so is eliminativism. The same conditional is true of intentional realism. Elsewhere (Bogdan 1991) I have argued against this 'folktheory-theory' assumption by showing that commonsense psychology is not an empirical prototheory of the mind but a biosocially motivated practice of coding, utilizing, and sharing information
more » ... om and about conspecifics. Here, without presupposing a specific analysis of commonsense psychology, I want to challenge a key implication of the 'folk-theory-theory' assumption to the effect that commonsense psychology is committed to a definite architecture of the mind. It is this architectural commitment that invites the elimination of commonsense psychology by connectionism (Stich 1983, ch. 11; Stich 1988; Ramsey, Stich, Garon 1990) or its vindication by the language of thought or syntax-over symbols view, which I will call syntactic computationalism (Davies 1992 ; see also Fodor 1975 Fodor , 1981 Fodor , 1987 Pylyshyn 1984) . Since the study of mental architecture is scientific, why bother with commonsense psychology? 2 Why are its alleged architectural commitments important as a basis for elimination or vindication? Three reasons spring to mind. One is that both eliminativism and realism use commonsense psychology as a stick to prop their scientific candidate and beat on its opponent. For eliminativism, in particular, if connectionism is right about mental architecture, not only is commonsense psychology wrong and eliminable, but so may be its scientific sponsor, syntactic computationalism. This is because commonsense psychology, which has been around for much longer than syntactic computationalism, is often viewed by friend and foe alike as a source of inspiration for syntactic computationalism. 1 Another reason is that commonsense psychology appears to provide an apriori conception of our cognitive selves, as thinkers, which emphasizes a formal rationality that can be implemented only by the architectures envisaged by syntactic computationalism. Eliminating commonsense psychology on architectural grounds is thought to explode this apriori conception that eliminativists find wrong and counterproductive. Finally, there is the ability of commonsense psychology to explain and predict cognitions and actions successfully. The thought naturally occurs that the success of this ability must 1 For intentional realists like Fodor and Pylyshyn the commonsense framework is the inspiration for the cognitive-scientific framework: see Fodor 1975 Fodor , ch. (1990 explicitly talk about "one sort of psychological model that exploits and builds upon the posits of folk psychology" (p. 500), and later hold that "in the psychological literature there is no shortage of models for human belief or memory which follow the lead of commonsense psychology in supposing that propositional modularity is true" (p. 506). There are many other examples of the cognitive science-common sense connection in the philosophical and foundational literature. 3 have an architectural grounding since its explanations and predictions are causal. I find none of these reasons for the architectural significance of commonsense psychology compelling. The last two will be taken up later on, as we speculate that commonsense psychology is likely to view a thinker functionally, as a problem solver, rather than architecturally, as a symbol manipulator; and that its explanatory success may rest on its architectural nonchalance rather than commitment. The first reason, having to do with using commonsense psychology in the battle between connectionism and syntactic computationalism, is a bogus that had better be disposed of right away. In spite of the well documented influence of commonsense psychology on the philosophy of syntactic computationalism, the major inspirations of the latter come from other directions. I am thinking of Chomsky's theory of grammar, early psychological models of concept formation, memory, and thinking, computer simulations of cognition, computer vision, and the reality of computers. As the professional literature testifies, it is in terms of such phenomena that the merits of connectionism and syntactic computationalism are judged. Significantly, after handwaiving in the direction of commonsense psychology, the archintentional realist, Jerry Fodor, gives his most forceful arguments for syntactic cognition in terms of compositionality, computation, systematicity, and productivity, and illustrates them with structural descriptions of visual scenes and parsing trees (Fodor 1975, chs. 3 and 4; 1987, Appendix). These are hardly the sorts of features that commonsense psychology cares about, and are precisely the sorts of features that syntactic computationalists 4 think connectionism cannot explain (Fodor and Pylyshyn 1988) . Similarly, when Ramsey, Stich, Garon, or Davies, discuss connectionism pro and con, the examples are memory retrieval, formation of tenses, and phonetic learning -again, hardly the stuff about which commonsense gets excited or excessively opinionated. This matter being clarified, the question we turn to is whether commonsense psychology has architectural commitments. I think it hasn't, which is why I want to show that the architectural elimination (or vindication) of commonsense psychology is a misguided project. My plan is this. I examine in the next section the notion of propositional modularity proposed by Stich (1983), and developed by Ramsey, Stich, and Garon (1990) , suggest that it allows different architectural readings, and then force it into a dilemma by arguing that, on an implementational reading, propositional modularity may be relevant to commonsense psychology but has no architectural bite, while on a stronger reading, it has the opposite effects, bite but no relevance. Then, in the following three sections, I examine the evidence (in the form of apriori notions, introspection, and attributions of propositional attitudes) invoked for the architectural commitments of commonsense psychology, and find it inadequate. I conclude with a speculation about how the architectural nonchalance of commonsense psychology bears on its explanations. To simplify the terminology, I describe as archphilosophers all those who think that commonsense psychology has architectural commitments, whether they are eliminativists, realists, or of other stripes. My argument, here, is not that they are wrong (which they are) but rather that they have not made a convincing 5 case for their position. Degrees of Architectural Involvement Propositional modularity is the notion that mental states are (i) functionally discrete, (ii) semantically interpretable, and (iii) play a causal role in cognition and behavior in virtue of (i) and (ii). Stich (1983, pp. 237-242) and Ramsey, Stich and Garon (1990, pp. 504, 506) characterize this notion in the commonsense terms of beliefs, memories, and other propositional attitudes, and take it to have architectural import. For our discussion, it suffices to say that the notion of cognitive architecture captures the resources that are systematically responsible for the form and operation of representations. The resources include basic components of representations (e.g., symbols or neural nets) and basic programs that handle representations. For syntactic computationalism, these are resources that store, retrieve, and manipulate symbol structures according to their form. This indicates a certain "style of processing" (Fodor and Pylyshyn, 1988, pp. 5, 10) in which programs articulate simple symbols into complex structures, and combinatorially manipulate such structures according to their composition. Connectionism contrasts this "style of processing" with its own, based on a nonsyntactic and nonsymbolic architecture of activation networks. The notion of propositional modularity allows several degrees of architectural involvement. The most minimal is that of an architecture that merely implements propositional modularity. This is the implementational sense 6 in which elementary particles implement thought or paper transactions implement money exchanges. Since in its talk of beliefs and desires, commonsense psychology is said to posit states that are functionally discrete, semantically valued, and causally effective, it must be committed to some unspecified architecture that implements propositional modularity. What the implementational commitment does not entail is that the architecture itself is propositionally modular. To get that entailment, we need an argument for a stronger architectural involvement. At this point, we should note in anticipation, a dilemma is beginning to take shape. On the one horn, the minimally implementational degree of architectural involvement may be the closest the notion of propositional modularity comes in approximating the architectural commitments of commonsense psychology. Unfortunately, this is too shallow and trivial a commitment to be invalidated by connectionism (or vindicated by syntactic computationalism). On the other horn, any stronger commitment might allow such an invalidation (or vindication) but at the price of misrepresenting commonsense psychology. Eliminativism and realism need a stronger architectural involvement on the part of commonsense psychology if their favorite cognitive-scientific paradigm is to yield elimination or vindication. According to Ramsey, Stich and Garon (1990) , commonsense psychology appears to have such a stronger involvement insofar as it entails that beliefs and desires are acquired or lost one at a time, or that the same belief is involved in various inferences, or that out of several possible beliefs only one was causally active in generating an action, or the like. These entailments point to properties that only a propositionally 7 modular architecture could have, and to processes that only such an architecture could run. In addition, commonsense psychology needs a stronger architectural involvement to satisfy the conditions on propositional modularity. Specifically, the third condition, on causation, could not be met unless the functional units were causally efficacious in virtue of their semantic values. That causal efficacy requires appropriate architectural resources. How strong should this new involvement be? The line taken by Stich and his collaborators points to a rather weak involvement that bears on mental sentences as units, and not on their internal form. This cannot be good enough for the archphilosopher. Left unanalyzed, mental sentences have no psychological value, not only because their formation and transformation are left in the dark, but also because, as a result, we have no understanding of the causal mechanics by which mental sentences play the functional roles required by propositional modularity. Mental sentences are neither semantically nor causally potent merely in virtue of having truth values. Semantically, they represent because their parts do; causally, mental sentences could not represent, and could not generate other sentences, unless they are manufactured and recombined out of parts. Psychologically, it is precisely the internal composition of a mental sentence that makes its semantic value causally efficacious. 2 This is an axiom of syntactic computationalism. I am not 2 Ramsey, Stich, Garon (1990, p. 506) write that the propositionally modular psychological models inspired by commonsense psychology view "beliefs or memories as an interconnected collection of functionally discrete, semantically interpretable states which interact in systematic ways." I submit that no plausible account of the interconnectedness and systematic interaction of beliefs and memories can avoid appeal to the internal composition of sentences. Stich himself admits that the "unstructured format" of sentential 8 saying that this is the right axiom, or that commonsense psychology is committed to a syntactic architecture. I am saying only that a sentential reading of propositional modularity leaves the route from semantics to causation gappy if not mysterious, thus preventing commonsense psychology from coherently satisfying all the conditions on propositional modularity. Two further facts militate against a weak sentential involvement. One is that the ordinary attributions of propositional attitudes are notoriously sensitive, both semantically and causally, to the internal constituents of the attributed contents. We know that beliefs fail to retain their semantic value and causal efficacy if, when equivalent parts of their contents are substituted, the equivalence is not recognized by the believer. The archphilosopher would want the architectural commitment of commonsense psychology to reflect this intrasentential sensitivity. Another fact is that connectionism is interesting generally, and as an elimination ploy, in particular, precisely to the extent to which it is an alternative to a syntax-over-symbols architecture, and much less interesting if merely suggesting nonsyntactic mechanisms for hopping from one sentence unit to another. An intrasentential reading is therefore needed for the alleged architectural commitments of commonsense psychology to meet coherently all the models has a hard time locating and tracking the right information routes in memory recall and other tasks (Stich 1983, p. 239), after having agreed with language of thought theorists that propositional (i.e., semantic and nonmental) theories of belief misrepresent commonsense psychology (op.cit., p. 29) I am not sure that is such a misrepresentation, but the fact is that a mental sentence theory comes close to have the same psychological irrelevance as a nonmental propositional theory of attitudes. 9 requirements of propositional modularity, and thus be subject to elimination (or vindication). This is the line taken by Davies (1992) with respect to connectionist eliminativism, and I think it is the right way to go. Yet I hasten to add that the conclusions of this paper also apply to the weaker reading of propositional modularity. I call this stronger reading syntactic modularity. The question, then, is whether commonsense psychology has a strong architectural commitment to syntactic modularity, as archphilosophers (should) claim. If so, what sort of commonsense psychology do they have in mind, and what is their evidence for the claim? These are the questions we turn to next. The Evidence As understood here, commonsense psychology is not what ordinary folk think about the mind, whatever that may be. As I see it, commonsense psychology is a universal competence, possibly innate and surely matured at an early age, which is constituted by a repertory of concepts and attribution schemas that are employed naturally and unreflectively to interpret and predict cognitive, emotional, and behavioral conditions (Bogdan 1991). People exercise this competence spontaneously and successfully, irrespective of what they happen to think about the mind. People often have the silliest views about the mind (as the history of philosophy and folklore abundantly testifies) and yet succeed in interpreting their conspecifics by using commonsense psychology. In a word, commonsense psychology is one thing, false consciousness about the mind or about commonsense psychology itself another thing. Neither 10 form of false consciousness is good evidence for the nature and operation of commonsense psychology, and neither plays any role in our story. The argument here is that archphilosophy has not shown that the commonsense concepts and attributions entail or evidence a commitment to syntactic modularity. The argument is not whether commonsense psychology is right or wrong in its architectural commitments, but whether archphilosophy shows that it has them. (I deny that it shows that.) The argument is concerned only with the evidence that motivates the archphilosopher to infer to the alleged architectural commitments of commonsense psychology. (I deny that the evidence motivates such an inference.) Again, the argument is not about what ordinary folk think of, or infer from, that evidence but rather about how the archphilosophers reconstruct the evidence to substantiate their charge. The archphilosophical position holds that the functionally discrete, syntactically organized, and semantically interpretable mental states with causal roles in cognition are the propositional attitudes of commonsense psychology. It also holds that the commonsense conception reflects architectural commitments to syntactic modularity. How do archphilosophers know these truths? Not scientifically. We do not know scientifically how commonsense psychology works. Aside from very recent studies in animal and developmental psychology, no systematic inquiry has been mounted to find out the rationale, nature, and operation of commonsense psychology. But there is no need for such an inquiry, the archphilosopher would say, smiling knowingly. For we have not only apriori and introspective evidence, but also the evidence of the attributions of propositional attitudes, to conclude that commonsense 11 psychology is committed architecturally to syntactic modularity. These three sources of archphilosophical evidence are worth scrutiny. THE APRIORI EVIDENCE. Commonsense psychology must be committed to syntactic modularity in an apriori sense because this is what the commonsense concept of thinker means (Davies 1992). To be a thinker is to instantiate a syntactic architecture. My reply is that even if we grant that the commonsense concept of thinker could be determined apriori, it does not follow that the concept entails syntactic modularity. For a softening up, consider this possibility. We could conceive of a thinking architecture with these properties: (i) it is not syntactic; (ii) it executes most of its cognitive tasks in nonsyntactic formats; yet (iii) for a definite class of tasks (e.g., logical, linguistic) it emulates syntactic modularity by "creating artifacts in the form of physical representations that we can manipulate to get answers to very difficult and abstract problems" (Rumelhart, Smolensky et al, 1986, p. 44). This (connectionist) position does not have to be right, only conceivable. Since it is, thinkerhood does not entail a syntactic architecture. Still, you would retort, this is not commonsense conceivability. In letter it is not, but in spirit it may be. As we will note shortly, in discussing introspection, we are internally aware of free and unorganized perceptions and ideas that get trimmed, firmed up, and syntactically regimented once they are formulated linguistically. This realization allows us intuitively to conceive of a narrow and disciplined architecture, on the top of a much larger and undisciplined one, getting to work intermittently when certain types of cognitive tasks must be carried out. 12 A more direct objection to the apriori argument is the following. It is not clear what we mean by an architecturally syntactic thinker. Is a syntactic architecture a necessary and sufficient condition, or only necessary? It is not logical niceties but levels of explanation that matter here. From an apriori stance, a thinker may be a means-ends reasoner or problem solver rather than a syntactic modulator. If a person or animal appears to deliberate and choose, commonsensically we are inclined to count her a thinker, without knowing or bothering about how she does it. We take a similar stance toward intelligence as a measure of how well or quickly problems are solved rather than as a description of the cognitive means employed. Pressed with examples and theory, the commonsense psychologist might admit that syntactic modularity is a way of implementing means-ends thinking. Syntactic modularity would then be a necessary but not sufficient condition for being a thinker. In that case, from the standpoint of commonsense psychology, logic and language appear as syntactic implementations of thinking. We think (i.e., solve problems, pursue goals) by means of logic and language programs, for definite tasks. As with other implementations (e.g., atomic forces ensuring the solidity of objects), commonsense is wisely agnostic about how the mental logic and language work architecturally. The implementational sense of architectural involvement once again advertises its plausibility. INTROSPECTION. What about introspection, the inner phenomenal prop of commonsense psychology? When treated commonsensically, doesn't it show that thinking is syntactically modular? Let us see. What exactly are we 13 introspecting from the stance of commonsense psychology? What does the latter enable us to grasp and conceptualize from the passing show of our inner mental life? Certainly neither programs nor underlying mechanisms. These cannot be introspected; and even if they could, we wouldn't get them right anyway. As psychologists have known for decades, introspection is biased, unrealistic, and unreliable. As far as introspecting thinking is concerned, I see two alternatives. One is a passive, kibbitzing introspection, a diffuse awareness of what we are doing mentally. This mode of introspection often fails to deliver the "structured thought" required by syntactic modularity. Spontaneous thinking is a typical example. I do not know about you, but I introspect and commonsensically identify little structure when thinking spontaneously. I introspect (mostly) free, anarchic, dynamic, and diffuse associations of barely formulated and unfocused ideas, assaulted and intermittently displaced by other sensory and memory inputs. If, Heaven forbids, my naive introspection of my spontaneous thinking were generalized as commonsense psychological evidence of functional architecture, connectionism would win hands down. The other alternative is our introspection of linguistically regimented outputs and of socially disciplined forms of argument and attribution. We so introspect when we talk to ourselves, or think aloud, or express our thoughts publicly, or engage in attributions, including self-attributions, of propositional attitudes. Am I not aware and conceptualizing commonsensically (right now, as I develop this very thought, talking sotto voce to myself) that I think in forms that are syntactically modular? I might be (OK! I am) but only of the evidence of 14 the outputs of the operation of my mental architecture. Informed by commonsense concepts, my introspection does not reach the architecture itself or its modus operandi, nor do those concepts give me any reason to infer from the introspected data (disciplined outputs) to the syntactic modularity of the architecture. As we noted already that, the output evidence is compatible with a different (connectionist) architecture or else uninformative about architecture. ATTRIBUTIONS. Yet surely, the archphilosopher would retort, the strongest case is made by propositional attitudes and their attributions. When I attribute to you the thought that cats are more fun and less noisy than barking dogs, isn't my commonsense attribution assuming the presence in your head of data structures made of grammatical and logical parts, functionally discrete, semantic, and, when attitudinized, causally efficacious in virtue of these features? Isn't this a commitment to an architecture responsible for such features? When we take one to have acquired a belief or lost a desire, isn't our commonsense conceptualization assuming that semantic and functionally discrete units are causally active or inactive, respectively, in virtue of their architectural instantiation or lack of it? The archphilosophers can make their case only if they have a solution to (what I call) the gap problem. There is a gap to be bridged between the attributions of propositional attitudes, on the one hand, and the data structures and processes, and their architectures in the head, on the other. The solution to the gap problem must consist in a principled and systematic mapping of 15 propositional attitudes into syntactic architectures. The gap problem is at the heart of the debate about the elimination (or vindication) of commonsense psychology, so it deserves some discussion. In the next section we examine the general dimensions of the gap problem, and then consider why it may have no solution. The Gap Revealed One solution to the gap problem would be simply to assume that there is a mapping from propositional attitudes to architectures. Some archphilosophers may have bought this solution unconsciously. Since we are interested in arguments, not fictions, we move on. One serious archphilosophical argument might go as follows. A reason why people can be said to have propositional attitudes in their heads is that the commonsense psychological discourse so portrays them. The attribution language portrays people as possessing beliefs and desires, and also portrays beliefs and desires as doing causal work. This is not enough. The mental form of the attitudes must be also established, and that form must be syntactically modular. But we know what the mental form is (the argument continues) because the attribution discourse displays it. The attribution discourse does display that-clauses which describe contents, and attitudes which describe the roles of the contents. The question
doi:10.1111/j.1468-0017.1993.tb00279.x fatcat:mhlr3t66erga3cv2scbay3gu6m