Draft: version 0,94 (b) Not for distribution, citation, or … Comments welcome
The Foundations of Computing
Brian Cantwell Smith*
Computer and Cognitive Science Departments Indiana University, Bloomington, in 47405 usa
Will computers ever be conscious? Is it appro- 1. Empirical: It must do justice to computa- priate—correct, illuminating, ethical—to un- tional practice (e.g., be capable of explaining derstand people in computational terms? Will Microsoft Word: the program, its construc- quantum, dna, or nanocomputers radically alter tion, maintenance, and use); our conception of computation? How will com- 2. Conceptual: It must discharge all intellectual puting affect science, the arts, intellectual his- debts (e.g., to semantics), so that we can un- tory? derstand what it says, where it comes from, I have never known how to answer these what it costs; and questions, because I have never been sure what computation is. More than twenty-five years 3. Cognitive: It must provide a tenable founda- ago, this uncertainty led me to undertake a tion for the computational theory of long-term investigation of the foundations of mind—the thesis, sometimes known as “cog- computer science and artificial intelligence. nitivism,” that underlies artificial intelligence That study is now largely complete. My aim in and cognitive science. this paper is to summarise some of the results.1 The first, empirical, requirement, of doing jus- tice to practice, helps to keep the analysis 1. Project grounded in real-world examples. It is hum- bling, too, since the computer revolution so reli- The overall goal has been to develop a compre- ably adapts, expands, dodges expectations, and hensive theory of computation. Since the outset, in general outstrips our theoretical grasp. But I have assumed that such an account must meet the criterion’s primary advantage is to provide a three criteria: vantage point from which to question the legit- imacy of all extant theoretical perspectives. For I 1This paper is largely excerpted from, and is intended to take it as a tenet that what Silicon Valley treats serve as an introduction to, a series of books that collec- as computational, is computational; to deny that tively report, in detail, on the study identified in the would be considered sufficient grounds for re- opening paragraphs. The study of computing is presented jection. But no such a priori commitment is in The Age of Significance: Volumes I–VI (Smith, forthcom- ing); the metaphysical territory to which that study leads, given to any story about computation—includ- in On the Origin of Objects (Smith 1996). ing the widely-held Turing-theoretic conception
Copyright © 1996 Brian Cantwell Smith *Brian Cantwell Smith
Page 2 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G
claiming that people are formal symbol ma- 4. Digital state machines (DSM): the idea of nipulators, for example, Fodor would naturally an automata with a finite disjoint set of inter- be wrong if computers were formal symbol ma- nally homogeneous machine states—as paro- nipulators and people were not. But he would died in the “clunk, clunk, clunk” gait of a also be wrong, even though cognitivism itself 1950’s cartoon robot; might still be true, if computers were not formal 5. Information processing (IP): what is in- symbol manipulators, either. volved in storing, manipulating, displaying, In sum, cognitive science is, like computer and otherwise trafficking in information, science, hostage to what I call the constitutive whatever that might be; and project: formulating a true and satisfying theory 6. Physical symbol systems (PSS): the idea, of computing that honors these three criteria. made famous by Newell and Simon, that, Needless to say, no one of them is easy to meet. somehow or other, computers interact with (and perhaps also are made of) symbols in a 2. Six Construals of Computation way that depends on their mutual physical embodiment. Some might argue that we already know what computation is. That in turn breaks into two By far the most important step in getting to the questions: (i) is there a story—an account that heart of the foundational question, I believe, is people think answers the question of what com- to recognize that these construals are all concep- puters are; and (ii) is that story right? tually distinct. In part because of their great fa- With regards to the first question, the an- miliarity (we have long since lost our inno- swer is not no, but it is not a simple yes, either. cence), and in part because “real” computers More than one idea is at play in current theo- seem to exemplify more than one of them—in- retic discourse. Over the years I have found it cluding those often-imagined but seldom-seen convenient to distinguish six construals of com- Turing machines, complete with controllers, putation, each requiring its own analysis: read-write heads, and long tapes—it is some- times uncritically thought that all six can be 1. Formal symbol manipulation (FSM): the viewed as rough synonyms, as if they were dif- idea, derivative from a century’s work in for- ferent ways of getting at the same thing. Indeed, mal logic & metamathematics, of a machine this conflationary tendency is rampant in the manipulating symbolic or (at least poten- literature, much of which moves around among tially) meaningful expressions without regard them as if doing so were intellectually free. But to their interpretation or semantic content; that is a mistake. The supposition that any two 2. Effective computability (EC): what can be of these construals amount to the same thing, done, and how hard it is to do it, mechani- let alone all six, is simply false. cally, as it were, by an abstract analogue of a Thus the formal symbol manipulation con- “mere machine”; strual (fsm) is explicitly characterized in terms 3. Execution of an algorithm (ALG): what is of a semantic or intentional aspect of computa- involved, and what behavior is thereby pro- tion, if for no other reason than that without duced, in following a set of rules or instruc- some such intentional character there would be tions, such as when making dessert; no warrant in calling it symbol manipulation.
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 3 T H E F O U N D A T I O N S O F C O M P U T I N G
The digital state machine construal (dsm), in particular surface behavior is the essential locus contrast, makes no such reference to intentional of being computational, with questions about properties. If a Lincoln-log contraption were how that is achieved left unspecified and un- digital but not symbolic, and a system manipu- cared about (effective computability is like this, lating continuous symbols were formal but not with algorithm execution somewhere in the digital, they would be differentially counted as middle). computational by the two construals. Not only Not only must the six construal be differen- do fsm and dsm mean different things, in other tiated one from another; additional distinctions words; they have overlapping but distinct exten- must be made within each one. Thus the idea of sions. information processing (ip)—by far the most The second and third construal—effective likely characterization of computation to appear computability (ec) and algorithm execution in the Wall Street Journal, and the idea responsi- (alg)—similarly differ on the crucial issue of se- ble for such popular slogans as ‘the Information mantics. Whereas the effective computability Age’ and ‘the information highway’—needs to construal seems free of intentional connotation, be broken down, in turn, into at least three sub- the idea of algorithm execution, as I have char- readings, depending on how ‘information’ is un- acterized it, seems not only to involve rules or derstood: (i) as a lay notion, perhaps dating recipes, which presumably do mean something, from the nineteenth-century, of something like but also to require some sort of understanding an abstract publicly-accessible commodity, car- on the part of the agent producing the behavior. rying a certain degree of authority; (ii) so-called Semantics is not the only open issue. It is “information theory,” a semantics-free notion similarly unclear whether the notions of “ma- that originated with Shannon & Weaver (1949), chine” and “taking an effective step” internal to spread out through much of cybernetics and the ec construal make fundamental reference to communication theory, is implicated in Kol- causal powers, material realization, or other mogorov, Chaitin, and similar complexity mea- physical properties, or whether, as most current sures, and has more recently been tied to no- theoretical discussions suggest, effective com- tions of energy and, particularly, entropy; and putability should be taken as an abstract mathe- (iii) the semantical notion of information advo- matical notion. This is no small question. If we cated by Dretske (1981), Barwise & Perry (1983), do not yet understand the mind/body problem for Halpern (1987), and others, that in contrast to machines, how can we expect computational the second deals explicitly with semantic con- metaphors to help us in the case of people? tent and veridicality. There are other differences among the con- Clarifying all these issues, bringing the sali- struals. They differ on whether they inherently ent assumptions to the fore, showing where they focus on internal structure or external input/ agree and where they differ, tracing the roles output, for example—i.e., on whether (i) they they have played in the last forty years—ques- treat computation as fundamentally a way of tions like this must be part of any foundational being structured or constituted, so that surface reconstruction. But in a sense these issues are all behavior is derivative (the formal symbol ma- secondary. For none have the bite of the second nipulation and digital state machine construals question raised at the beginning of the section: are of this type); or whether (ii) the having of a of whether any of the enumerated accounts is
Page 4 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G
right. Naturally, one has to say just what this carry information about or signify other states in means—has to answer the question, that is, the world (or at least can be taken by people to “Right of what?,” in order to avoid the superfi- do so). This semantical or intentional character cial response: “Of course such and such an of computation is betrayed by such phrases as analysis is right; that’s how computation is de- symbol manipulation, information processing, fined!” This is where the empirical criterion programming languages, knowledge representa- takes hold. More seriously, I am prepared to ar- tion, data bases, etc. Furthermore, and this is gue for a much more radical conclusion: that, important to understand, it is the intentionality when subjected to the empirical demands of of the computational that motivates the cogni- practice and the conceptual demands of cogni- tivist thesis. The only compelling reason to sup- tive science, all six construals fail—for deep, over- pose that we (or minds or intelligence) might be lapping, but distinct, reasons. computers stems from the fact that we, too, deal with representations, symbols, meaning, infor- 3. Diagnosis I: General mation, and the like. For someone with cognitivist leanings, What is the problem? Why do these theories all therefore—as opposed, say, to an eliminativist fail? materialist, or to some types of connectionist— The answers come at many levels. In the it is natural to expect that a comprehensive the- next section, I discuss some construal-specific ory of computation will have to focus on its se- problems. But a general thing can be said first. mantical aspects. This raises problems enough. Throughout, the most celebrated difficulties Consider just the issue of representation. In or- have to do with semantics. It is widely (if tac- der to meet the first criterion, of empirical ade- itly) recognized that computation is in one way quacy, a successful candidate will have to make or another a symbolic or representational or in- sense of the myriad kinds of representation that formation-based or semantical—i.e., as philoso- saturate practical systems—from bit maps and phers would say, an intentional—phenomenon.4 images to knowledge representations and data Somehow or other, though in ways we do not bases; from caches to backup tapes; from low- yet understand, the states of a computer can level finite-element models used in simulation model or simulate or represent or stand for or to high-level analytic descriptions supporting reasoning and inference, from text to graphics 4Although the term ‘intentional’ is philosophical, there to audio to video to virtual reality. As well as be- are many philosophers, to say nothing of some computer ing vast in scope, it will also have to combine and cognitive scientists, who would deny that computa- decisive theoretical bite with exquisite resolu- tion is an intentional phenomenon. Reasons vary, but the most common goes something like this: (i) that computa- tion, in order to distinguish: models from im- tion is both syntactic and formal, where ‘formal’ means plementations; analyses from simulations; and ‘independent of semantics’; and (ii) that intentionality has virtual machines at one level of abstraction from fundamentally to do with semantics; and therefore (iii) virtual machines at another level of abstraction, that computation is thereby not intentional. I believe this is wrong, both empirically (that computation is purely in terms of which the former may be imple- syntactic) and conceptually (that being syntactic is a way mented. of not being intentional); I also disagree that being inten- In order to meet the second, conceptual, cri- tional has only to do with semantics, which the denial re- terion, moreover, any account of this profusion quires.
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 5 T H E F O U N D A T I O N S O F C O M P U T I N G
tively foundational construals of computation Intentionality Computation are implicitly defined in terms of just such a Computation Intentionality background theory of semantics, but do not ex- plain what semantics is, and thus fail the second (a) (b) conceptual criterion. This group includes the Figure 1 — Explanatory dependence first formal symbol manipulation construal so favored in the cognitive sciences, in spite of its of representational practice must be grounded superficial formulation as being “independent of on, or at least defined in terms of, a theory of semantics.”7 Other construals, such as those semantics or content, partly in order for the that view computation as the behavior of dis- concomitant psychological theory to avoid va- cuity or circularity, and partly so that even the dependence” theory (1987), and Millikan’s “teleoseman- tics” or “biosemantics” (1984, 1989). For comparison computational part of the theory meet a mini- among these alternatives see e.g. Fodor (1984) and Mil- mal kind of naturalistic criterion: that we un- likan (1990). derstand how computation is part of the natural 7Because formal symbol manipulation is usually defined world. This is made all the more difficult by the as “manipulation of symbols independent of their inter- pretation,” some people believe that the formal symbol fact that the word ‘semantics’ is used in an in- manipulation construal of computation does not rest on a credible variety of different senses across the theory of semantics. But that is simply an elementary, range of the intentional sciences. Indeed, in my though apparently common, conceptual mistake. As dis- experience it is virtually impossible, from any cussed further in §4, the “independence of semantics” postulated as essential to the formal symbol construal is one location within that range, to understand independence at the level of the phenomenon; it is a the full significance of the term, so disparate is claim about how symbol manipulation works. Or so at that practice in toto.4 least I believe, based on many years of investigating what 5 practitioners are actually committed to (whether it is Genuine theories of content, moreover—of true—i.e., holds of computation-in-the-wild—is a sepa- what it is that makes a given symbol or structure rate issue). The intuition is simple enough: that semantic or patch of the world be about or oriented to- properties, such as referring to the Sphinx or being true, wards some other entity or structure or patch— are not of the right sort to do effective work. So they can- are notoriously hard to come by.6 Some puta- not be the sort of property in virtue of the manifestation of which computers run. At issue in the present discus- sion, in contrast, is independence at the level of the theory 5In computer science, to take a salient example, the term (or, perhaps, to put it less epistemically and more onto- “the semantics of a,” where a is an expression or construct logically, independence at the level of the types). And here in a programming language, means approximately the fol- the formal symbol manipulation construal is as dependent lowing: the topological (as opposed to geometrical) tem- on semantics as it is possible to be: it is defined in terms of poral profile of the behavior to which execution of this it. And defining yourself in terms of something is not a program fragment gives rise. By ‘topological’ I mean that way to be independent of it, as the parent of any teenager the overall temporal order of events is dictated, but that knows. Symbols must have a semantics, in other words their absolute or metric time-structure (e.g., exactly how (have an actual interpretation, or be interpretable, or fast the program runs) is not. As a result, a program can whatever), in order for there to be something substantive usually be sped up, either by adjusting the code or run- for their formal manipulation to proceed independently ning it on a faster processor, without, as is said, “changing of. Without a semantic character to be kept crucially in the semantics.” the wings, the formal symbol manipulation construal 6Best known are Dretske’s semantic theory of informa- would collapse in vacuity—would degenerate into some- tion (1981), which has more generally given rise to what is thing like “the manipulation of structure” or “stuff ma- known as “indicator semantics”; Fodor’s “asymmetrical- nipulation” — i.e., materialism.
Page 6 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G
crete automata—and also, I will argue below, 1. What the construal says or comes to—what even if this is far from immediately evident, the claim it makes about what it is to be a com- recursion-theoretic one that describes such be- puter; havior as the calculation of effective func- 2. Where it derives from, historically; tions—fail to deal with computation’s semanti- 3. Why it has been held; cal aspect at all, in spite of sometimes using se- mantical vocabulary, and so fail the first em- 4. What’s right about it—what insights it gets pirical criterion. In the end, one is inexorably at; driven to the conclusion represented in Figure 1 5. What is wrong with it, conceptually, empiri- (b): that, in spite of the advance press, especially cally, and explanatorily; from cognitivist quarters, computer science, far 6. Why it must ultimately be replaced; and from supplying answers to the fundamental in- 7. tentional mysteries, as suggested in Figure 1 (a), What about it should nevertheless be re- must, like cognitive science, await the devel- tained in a “successor” more adequate theory. 8 opment of a satisfying theory of intentionality. 4.a. Formal Symbol Manipulation
4. Diagnosis II: Specific I sometimes call the fsm construal antisemanti- cal, to capture its underlying idea that computa- So none of the six construals provide an account tion is the “manipulation of symbols indepen- of semantics. Since I take computation to be se- dent of their semantics.” On analysis, it turns mantic, that means they fail as theories of com- out to be motivated by two entirely different, putation, as well. And that is just the beginning ultimately incompatible, intuitions. The first of the problems. They also fail for detailed motivation is at the level of the theory, and is structural reasons—different reasons per con- reminiscent of a reductionist desire for a “se- strual, but reasons that add up, overall, to a re- mantics-free” account. It takes the fsm thesis as markably coherent overall picture. a claim that computation can be described or In this section I summarize just a few of the analysed in a semantics-free way. If that were problems, to convey a flavor of what is going true, or so the argument goes, that would go on. In each case, to put this in context, the aim some distance towards naturalizing intentional- is to explicate: ity (as Haugeland, primary adherent of this in- terpretation, says, it would help “make the world safe for semantics”). 8As suggested in the preceding footnote, philosophers are less likely than computer scientists to expect a theory of There is a second motivating intuition, dif- computation to be, or to supply, a theory of intention- ferent in character, that holds at the level of the ality. I.e., they would not expect the metatheoretic struc- phenomenon. Here the ideas is simply the fa- ture to be as expected by most computer scientists and ar- miliar observation that intentional phenomena, tificial intelligence researchers—namely, as indicated in Figure 1 (a), with a theory of intentionality resting on a such as reasoning, hoping, or dreaming, carry theory of computation. But that does not mean they on in relative independence of their subject mat- would necessarily agree with Figure 1 (b). Many philoso- ters or referents. Reference and truth, it is recog- phers seem to think that a theory of computation can be nized, are just not the sorts of properties that independent of a theory of intentionality. Clearly, I do not believe this is correct. can play a causal role in engendering behav-
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 7 T H E F O U N D A T I O N S O F C O M P U T I N G
ior—essentially because they involve some sort The fundamental idea underlying the formal of relational coordination with things that are symbol manipulation thesis, that is, is that a too far away to make a difference. This relational barrier of this double allegedly aligned sort can characteristic of intentionality—something I call be drawn around a computer, separating a pris- semantic disconnection—is such a deep aspect of tine inner world of symbols—a private kingdom the phenomenon that it is hard to imagine it be- of thought, as it were—thought both to work ing false. Without it, fantasy lives would be (ontologically) and to be analysable (theoretic- metaphysically banned; you would not be able ally) in isolation, without distracting influence to think about continental drift without bring- from the messy, unpredictable exterior. ing the tectonic plates along with you. Not surprisingly, the traditional motivating For discussion, I label the two readings of examples motivating the fsm construal, such as the formal symbol manipulation construal con- theorem proving, meet this complex condition. ceptual and ontological, respectively. The onto- First, they involve internal symbols designating logical reading is natural, familiar, and based on external situations, thereby satisfying align- a deep insight. But it is too narrow. Many coun- ment: data bases representing employee salaries, ter-examples can be cited against it. Space does differential equations modeling the perihelion of not permit rehearsing them here.9 Instead, to Mercury, first order axioms designating Platonic get to the heart of the matter, it helps to high- numbers or purely abstract sets. Second, espe- light a distinction between two kinds of “boun- cially in the paradigmatic examples of formal ax- dary” thought to be relevant or essential—in- iomatizations of arithmetic and proof systems of deed, often assumed a priori—in the analysis of first-order logic (and, even more especially, computers and other intentional systems: when those systems are understood in classical, especially model-theoretic, guise), the system is 1. Physical: A physical boundary between the system and its surrounding environment— assumed to exhibit the requisite lack of interac- i.e., between “inside” and “outside”; and tion between the syntactic proof system and the model-theoretic interpretation, satisfying isola- 2. Semantic: A semantic boundary between tion. In conjunction, the two assumptions al- symbols and their referents. low the familiar two-part picture of a formal sys- In terms of these two distinctions, the ontologi- tem to be held: a locally contained syntactic sys- cal reading of the formal symbol manipulation tem, on the one hand, consisting of symbols or construal can be understood as presuming the formulae in close causal intimacy with a proof- following two theses: theoretic inference regimen; and a remote realm of numbers or sets or “ur-elements,” in which 1. Alignment: That the physical and semantic the symbols or formulae are interpreted, on the boundaries line up, with all the symbols in- other. It is because the formality condition relies side, all the referents outside; and on both theses that the classical picture takes 2. Isolation: That this allegedly aligned bound- computation to consist exclusively of sym- ary is a barrier or gulf across which various bol–symbol transformations, carried on entirely forms of dependence (causal, logical, expla- within the confines of a machine. natory) do not reach. The first—and easier—challenge to the anti-
semantical thesis comes when one retains the 9See The Age of Significance: Volume II.
Page 8 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G
first alignment assumption, of coincident vironmental interaction within computational boundaries, but relaxes the second isolation practice, it is not at all clear that this would be claim, of no interaction. This is the classical possible, without violating the condition of em- realm of input/output, home of the familiar no- pirical adequacy embraced at the outset. But it tion of a transducer. And it is here that one en- does not matter, ultimately, because the critique counters the most familiar challenges to the fsm is only half way done. construal (such as the “robotic” reply to Searle’s More devastating to the fsm construal are (1980) Chinese room argument, and Harnad’s examples that challenge the alignment thesis. (1991) “Total Turing Test” as a measure of in- It turns out, on analysis, that far from lining up telligence). Thus imagine a traditional percep- on top of each other, real-world computer sys- tion system—for example one that on encoun- tems’ physical and semantic boundaries cross- ter with a mountain lion constructs a symbolic cut, in rich and productive interplay. It is not representation of the form mountain-lion-043. just that computers are involved in an engaged, There is interaction (and dependence) from ex- participatory way with external subject matters, ternal world to internal representation. By the in other words, as suggested by some recent “si- same token, an actuator system, such as one that tuated” theorists. They are participatorily en- would allow a robot to respond to a symbol of gaged in the world as a whole—in a world that the form cross-the-street by moving from indiscriminately includes themselves, their own one side of the road to the other, violates the internal states and processes. This integrated independence assumption in the other direc- participatory involvement, blind to any a priori tion, from internal representation to external subject-world distinction, and concomitantly world. intentionally directed towards both internally Note, in spite of this interaction, and the and externally exemplified states of affairs, is not consequent violation of isolation, that align- only architecturally essential, but is also critical, ment is still preserved, in both cases: the trans- when the time comes, in establishing and ducer is imagined to mediate between an inter- grounding a system’s intentional capacities. nal symbol and an external referent. Yet the vio- From a purely structural point of view, four lation of isolation alone is enough to defeat types of case are required to demonstrate this the formality condition. This is why transducers non-alignment: (i) where a symbol and referent and computation are widely recognized to be are both internal; (ii) where a symbol is internal uneasy bedfellows, at least when formality is at and its referent external; (iii) where symbol and issue. It is also why, if one rests the critique at referent are both external; and (iv) where sym- this point, defenders of the antisemantical con- bol is external and referent internal. The first is strual are tempted to wonder, given that the op- exemplified in cases of quotation, meta-struc- erations of transducers violate formality, whether tural designation, window systems, e-mail, com- they should perhaps be counted as not being pilers, loaders, network routers, and at least ar- computational.10 Given the increasing role of en- guably all programs (as opposed, say, to data- bases). The second, of internal symbols with 10Thus Devitt (1991) restricts the computational thesis to external referents, can be considered as some- what he calls “thought-thought” (t-t) transactions; for thing of a theoretical (though not necessarily him output (t–o) and input (i–t) count as non-computa- practical) default, as for example when one re- tional.
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 9 T H E F O U N D A T I O N S O F C O M P U T I N G
members the sun’s setting over a lake. The third 1. Physical transducers, for operations or and fourth are neither more nor less than a de- modules that cross between the inside and scription of ordinary written text, public writ- outside of a system; ing, etc.—to say nothing of pictures, sketches, 2. Semantic transducers, for operations or conversations, and the whole panoply of other modules that mediate between symbols and forms of external representation. Relative to any their referents. particular system, they are distinguished by whether the subject matters of those external And it is this bifurcation, finally, that irrevoca- representations are similarly external, or are in- bly defeats the antisemantical claim. For the ternal. The familiar red skull-and-cross-bones only remotely plausible notion of transducer, in signifying radioactivity is external to both man practice, is the physical one. That is what we and machine, and also denotes something exter- think of when we imagine vision, touch, smell, nal to man and machine, and thus belongs to articulation, wheels, muscles, and the like: sys- the third category. To a computer or person tems that mediate between the internals of a involved, on the other hand, an account of how system and the “outside” world. Transducers, they work (psychoanalysis of person or machine, that is, at least in informal imagination of prac- as it were) is an example of the fourth. titioners, are for connecting systems to their 11 By itself, violating alignment is not enough (physical) environments. What poses a chal- to defeat formality. What it does accomplish, lenge to the formal (antisemantical) symbol ma- however, is to radically undermine isolation’s nipulation construal of computation, on the plausibility. In particular, the antisemantical other hand, is the semantic transducer: those as- thesis was challenged not only because these ex- pects of a system that involve trading between amples show that the physical and semantic an occurrent state of affairs, on the one hand, boundaries cross-cut, thereby undermining the and a representation of it, on the other. Antise- alignment assumption, but because they illus- mantics is challenged as much by disquotation trated the presence, indeed the prevalence, of as by driving around. As a result, the only way to retain the onto- effective traffic across both boundaries—between and among all the various categories in ques- logical version of the fsm construal is to disallow (i.e., count as non-computational) the opera- tion—thereby negating isolation. And this, in turn, shows up, for what it is, tions of semantic transducers. But that is absurd! the common suggestion that transducers, be- It makes it clear, ultimately, that distinguishing cause of violating the antisemantical thesis, that subset of computation that satisfies the on- should be ruled “out of court”—i.e., taken not tological version of the antisemantical claim is to be computational (à la Devitt (1991)). It not only unmotivated, solving the problem by should be clear that this maneuver is ill-advised; fiat (making it uninteresting), but is a spectacu- even a bit of a cop-out. For consider what a larly infeasible way to draw and quarter any ac- proponent of such a move must face up to, 11This statement must be understood within the context when confronted with boundary non-align- of cognitive science and the philosophy of mind. It is ment. The notion of a transducer must be split in telling that the term ‘transducer’ is used completely differ- two. I.e., someone interested in transducers ently in engineering and biology (its natural home): to would have to distinguish: signify mechanisms that mediate changes in medium, not that cross the inside/outside or symbol/referent boundary.
Page 10 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G tual, real-life, system. For no one who has ever 4.b. Effective Computability built a computational system has ever found any reason to bracket reference-crossing operations, Although different in detail, the arguments or to treat them as a distinct type. Not only against the other five construals are similar in that; think of how many different kinds of ex- style. In each case, I have tried to develop a amples of semantic transducers one can imag- staged series of counterexamples, not simply in ine: counting, array indexing, e-mail, disquota- order to show the construal false, but to serve as tion, error-correction circuits, linkers, loaders, strong enough intuition pumps on which to simple instructions, data base access routines, base a positive alternative. Space precludes pointers, reflection principles in logic, index op- spelling out details, except for a few words erations into matrices, most Lisp primitives, and about effective computability—the idea that the like. Furthermore, to define a species of underwrites recursion theory, complexity the- transducer in this semantical way, and then to ory, and the official (mathematical) “theory of remove them from consideration as not being computation.” Note, for starters, that whereas genuinely computational, would make compu- the first formal symbol manipulation construal tation (minus the transducers) antisemantical is predominant in artificial intelligence, cogni- tautologically. It would no longer be an inter- tive science, and philosophy of mind, it is this esting claim on the world that computation was second, effective computability construal, in antisemantical—an insight into how things are. contrast, that underlies most theoretical and Instead, the word ‘computation’ would simply practical computer science. be shorthand for antisemantical symbol mani- Fundamentally, it is widely agreed, the the- pulation. The question would be whether any- ory of effective computability focuses on “what thing interesting was in this named class—and, can be done by a mechanism.” But two concep- in particular, whether this conception of com- tual problems have clouded its proper apprecia- putation captured the essential regularities un- tion. First, in spite of its subject matter, it is al- derlying practice. And we have already seen the most always characterized abstractly, as if it were answer to that: it is no. In sum, introducing a a branch of mathematics. Second, it is imagined notion of a semantical transducer solves the to be a theory defined over (for example) the problem tautologically, cuts the subject matter numbers. Specifically, the marks on the tape of at an unnatural joint, and fails to reconstruct the paradigmatic Turing machine are viewed as practice. That is quite a lot to have against it. representations—representations, in general, of Furthermore, to up the ante on the whole numbers, functions, or other Turing machines. investigation, not only are these cases of “se- In contrast to the received view, I argue two mantic transduction” all perfectly well-behaved; things. First, I claim that the theory of effective they even seem, intuitively, to be as “formal” as computability is fundamentally a theory about any other kind of operation. If that is so, then the physical nature of patches of the world. In those systems either are not formal, after all, or underlying character it is no more “mathemati- else the word ‘formal’ has never meant independ- cal” than anything else in physics—even if we ence of syntax and semantics in the way that the use mathematical structures to that physical re- fsm claim construes it. Either way, the ontologi- ality. Second— and this is sure to be conten- cal construal does not survive. tious—I argue that recursion theory is funda-
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 11 T H E F O U N D A T I O N S O F C O M P U T I N G
mentally a theory of marks. More specifically: inexorably to the following very strong conclu- sion: that what goes by the name “theory of The representation relation, alleged to go from computation” fails not because it makes false marks to numbers, in fact runs the other way, claims about computation, but because it is not from numbers to marks! I.e., the truth is 180° a theory of computation at all.12 off what we have all been led to believe. In sum: I recommend redrawing the intel- All sorts of evidence are cited in defense of this lectual map. What has been called a “theory of non-standard claim. For example: computation” is in fact a general theory of the 1. Unless one understands it this way, one can physical world—specifically, a theory of how solve the halting problem; hard it is, and what is required, for patches of the world in one physical configuration to 2. An analysis of history, through Turing’s pa- change into other physical configuration. It ap- per and subsequent work, especially includ- plies to all physical entities, not just to comput- ing the development of the universal Turing ers. It is no more mathematical than the rest of machine, shows how and why the represen- physics. And thus it should be joined with tation relation was inadvertently turned up- physics—because in a sense it is physics. It is a side down; spectacular achievement of which the century 3. The analysis makes sense of a number of oth- should be proud: a thoroughly-modern mathe- erwise-inexplicable practices, including, matical theory of causality. among other examples: (i) the use of the word “semantics” in practicing computer sci- 5. Method ence to signify the behavior engendered by running a program, (ii) the rising popularity Similarly strong conclusions can be arrived at by of Girard’s linear logic, and (iii) the close as- pursuing each of the other construals. Rather sociation between theoretical computer sci- than go into them here, I instead want to say a ence and constructive mathematics. word about method. Specifically, about formal- It follows from this analysis that all use of se- ity. For there is a theme underlying all six of mantical vocabulary in the “official” theory of these critiques: that part of what has blinded us computation is meta-theoretic. As a result, the to the true nature of computation has to do theory is not a theory of intentional phenomena— with the often pretheoretic assumption that that is, is not a theory that deals with them as computers are formal. intentional phenomena. Whereas the formal In one way or another, no matter what con- symbol manipulation construal fails to meet its strual they pledge allegiance to, just about every- own criterion, in other words, of being “defined one thinks that computers are formal—that independent of semantics,” this second con- they manipulate symbols formally, that pro- strual does meet (at least the conceptual reading 12Even if it is derivatively intentional. I do not believe of) that condition. Ironically, however, it is in that computation’s intentionality is inherently derivative, achieving that success that the recursion-theo- as it happens, but even those who do think that must ad- retic tradition ultimately fails. For computation, mit that it is an intentional phenomenon of some sort. as was said above, and as I am prepared to ar- For derivative does not mean fake. And if “derivatively in- tentional” is taken not to be a constraint at all, then one is gue, is an intentional phenomenon. Which leads forced to say just what does characterize computation.
Page 12 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G grams specify formal procedures, that data struc- They are alike in foisting recalcitrant theo- tures are a kind of formalism, that computa- retical issues onto center stage. Consider explic- tional phenomena are uniquely suited for anal- itness, for example, of the sort that might ex- ysis by formal methods. In fact the computer is plain such a sentence as “for theoretical pur- often viewed as the crowning achievement of an poses we should lay out our tacit assumptions in entire “formal tradition”—an intellectual orien- a formal representation.” Not only has explicit- tation, reaching back through Galileo to Plato, ness—and its partner in crime, implicit- that has been epitomized in this century in the ness—stubbornly resisted theoretical analysis, logic and metamathematics of Frege, Russell, but both notions are parasitic on something else Whitehead, Carnap, Turing, etc. we do not understand: general representation.14 This history would suggest that formality is Or consider “a-contextual.” Where is an overall an essential aspect of computation. But since the theory of context in terms of which to under- outset, I have not believed that this is necessarily stand what it would be to say of something (a right. For one thing, it has never been clear logical representation, say) that it was not con- what the allegiance to formality is an allegiance textually dependent? to. It is not as if “formal” is a technical or the- Considerations like this suggest that particu- ory-internal predicate, after all. People may be- lar readings of formality can be most helpfully lieve that developing an idea means formalizing pursued within the context of the general theo- it, and that programming languages are formal retical edifices that have been constructed (more languages, and that theorem provers operate on or less explicitly) in their terms. Five are particu- formal axioms—but few write formal(X) in larly important: their daily equations. Moreover, a raft of differ- 1. The antisemantical reading mentioned above: ent meanings and connotations lie just below the idea that a symbolic structure (represen- the surface. Far from hurting, this apparent am- tation, language, symbol system, etc.) is for- biguity has helped to cement popular consensus. mal just in case it is manipulated independent Freed of the need to be strictly defined, formal- of its semantics. Paradigmatic cases include so- ity has been able to serve as a lightning rod for a called formal logic, in which it is assumed cluster of ontological assumptions, methodolog- that a theorem (such as mortal(socrates)) is ical commitments, and social and historical bi- derived by an automatic inference regimen ases. Because it is tacit, goes deep, has historical roots, and permeates practice, formality has replies that the term has very different connotations in been an ideal foil, over the years, with which to different fields. Some mathematicians and logicians, for example, take it to be pejorative, in contrast to the major- investigate computation. ity of theoretical computer scientists, for whom it has Almost a dozen different readings of “for- almost the opposite connotation. mal” can be gleaned from informal usage: pre- 14On its own, an eggplant cannot exactly be either formal cise, abstract, mathematical, a-contextual, digital, or explicit, at least not in its ordinary culinary role, since in that role it is not a representation at all. In fact the only 13 explicit, syntactic, non-semantic, etc. way to make sense of calling something non-representa- tional explicit is as short-hand for saying that it is ex- 13At one stage I asked people what they thought “formal” plicitly represented (e.g., calling eggplant an explicit in- meant—not just computer scientists, but also mathemati- gredient of moussaka as a way of saying that the recipe for cians, physicists, sociologists, etc. It was clear from the moussaka mentions eggplant explicitly).
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 13 T H E F O U N D A T I O N S O F C O M P U T I N G
without regard to the reference, truth, or (or is correctly classified in terms of) the given even meaning of any of its premises. concept. And finally, the fourth view—that to 2. A closely-allied grammatical or syntactic read- be formal has something to do with being math- ing, illustrated in such a sentence as “infer- ematical, or at least with being mathematically ence rules are defined in terms of the formal characterisable—occupies something of an on- properties of expressions.” Note that whereas tological middle-realm between the subject- the antisemantical reading is negatively char- matter orientation of the first three and the acterized, this one has a positive sense. methodological orientation of the fifth. 3. A reading meaning something like determi- The ultimate moral for computer and cogni- nate or well-defined—i.e., as ruling out all tive science is similar to the claim made earlier ambiguity and vagueness. This construal about the six construals: not one of these readings turns out to be related to a variant of the of ‘formal’ correctly applies to the computational computationally familiar notion of digitality case. It can never be absolutely proved that com- or discreteness. putation is not formal, of course, given that the notion of formality is not determinately tied 4. A construal of “formal” as essentially equiva- down. But I am prepared to argue that no stan- lent to mathematical. dard construal of formality, including any of 5. A reading that cross-cuts the other four: for- those just enumerated, is both (i) substantive mality as applied to analyses or methods, per- and (ii) true of extant computational practice. haps with a derivative ontological implication Some readings reduce to vacuity, or to no more that some subject matters—such as computa- than physical realisability; others break down in tion, perhaps?—are uniquely suited to such internal contradiction; others survive the test of analytic techniques. being substantial, but are demonstrably false, The first two are often treated as conceptually even of current systems. One is inescapably led equivalent, but to do that is to assume that a to the following conclusion, in all its historical system’s syntactic and semantic properties are irony: that the computer, darling child of the necessarily disjoint—which is almost certainly formal tradition, outstrips the bounds of the false. The relationship between the third (deter- very tradition that gave rise to it. minate) reading and digitality does not have so much to do with what Haugeland (1982) calls 6. The Ontological Wall “first order digitality”: the ordinary assumption that a system’s states can be partitioned into a Where does this all leave us? Over time, investi- determinate set, such as that its future behavior gations of the sort described above, and consid- or essence stems solely from membership in one eration of the conclusions reached in them, have element of that set, without any ambiguity or convinced me that none of the reigning theories matter of degree. Rather, vagueness and indefi- of computation, nor the reigning methodologi- niteness (as opposed to simple continuity) are cal attitude to computation, will ever lead to an excluded by a second-order form of digitality— analysis strong enough to meet the two criteria digitality at the level of concepts, in the sense of laid down at the outset. More specifically, I was there being a binary “yes/no” fact of the matter led to the following position, which in various about whether any given situation falls under ways I held for almost twenty years: I was
Page 14 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G
1. In awe of the depth, texture, scope, pluck, luck. The two issues are inextricably entan- and impact of computational practice; gled—a fact of obstinate theoretical and meta- theoretical consequence. 2. Critical of the inadequate state of the current A methodological consequence will illustrate theoretical art; the problem. Especially within the analytic tra- 3. Convinced that a formal methodological dition (by which I mean to include not just ana- stance stood in the way of getting to the lytic philosophy, e.g. of language and mind, but heart of the computational question; and most of modern science as well, complete with 4. Sure in my belief that what was needed, its formal/mathematical methods), it is tradi- above all else, was a (situated, embodied, em- tional to analyse semantical or intentional sys- bedded, indexical, critical, reflexive, all sorts tems, such as computers or people, under the of other things—it changed some, over the following presupposition: (i) that one can parse years) theory of representation and semantics. or register the relevant theoretical situation in advance into a set of objects, properties, types, In line with this metatheoretic attitude, as will relations, equivalence classes, and so one (e.g., already have been indicated, I kept semantical into people, heads, sentences, real-world refer- and representational issues in primary theoreti- ents, etc.)—as if this were theoretically innocu- cal focus. Since, as indicated in the last section, ous—and then (ii), with that ontological parse the official “theory of computation,” derived in hand, go on to proclaim this or that or the from recursion and complexity theory, pays no other thing as an empirically justified result. attention to such intentional problems, to strike Thus for example one might describe a mail-de- even this much of a semantical stance was to livering robot by first describing an environ- part company with the center of gravity of the ment of offices, hallways, people, staircases, lit- received theoretical tradition. ter, and the like, through which it is supposed You might think that this would be conclu- to navigate, and then, taking this char- sion enough. And yet, in spite of the importance acterization of its context as given, ask how or and magnitude of these intentional difficulties, whether the creature represents routes, say, or and in spite of the detailed conclusions sug- offices, or the location of mail delivery stations. gested above, I have gradually come to believe If one adopts a reflexively critical point of something much more sobering: that the most view, however, as I have systematically tried to serious problems standing in the way of devel- do, one is inexorably led to the following con- oping an adequate theory of computation are as clusion: that, in that allegedly innocent pretheo- much ontological as they are semantical. It is not retical “set-up” stage, one is liable, even if un- that the semantic problems go away; they re- wittingly, to project so many presuppositions, main as challenging as ever. It is just that they biases, blindnesses, and advance clues about the are joined—on center stage, as it were—by even “answer,” and in general so thoroughly prefigure more demanding problems of ontology. the target situation, without either apparent or Except that to say “joined” is misleading, as genuine justification, that one cannot, or at least if it were a matter of simple addition—i.e., as if should not, take any of the subsequent “analysis” now there were two problems on the table, seriously. And that is problematic, in turn, not where before there had been just one. No such just because it rejects standard analyses, but be-
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 15 T H E F O U N D A T I O N S O F C O M P U T I N G cause it seems to shut all inquiry down. What nitive scientists, and artificial intelligence re- else can one do, after all? How can one not searchers are not the only ones running up parse the situation in advance (since it will against the demands of ontology. Similar con- hardly do to merely whistle and walk away)? clusions are being reported from many other And if, undaunted, one were to go ahead and quarters. The words are different, and the per- parse it anyway, what kind of story could possi- spectives complementary, but the underlying bly serve as a justification? It seems that any phenomena are the same. conceivable form of defense would devolve into Perhaps the most obvious fellow travelers another instance of the same problem. are literary critics, anthropologists, and other In sum, the experience is less one of facing social theorists, vexed by what analytic cate- an ontological challenge than of running up gories to use in understanding people or cultures against an ontological wall. Perhaps not quite of that, by such writers’ own admission, compre- slamming into it, at least in my own case; rec- hend and constitute the world using concepts ognition dawned slowly. But neither is the en- alien to the theorists’ own. What makes the counter exactly gentle. It is difficult to exag- problem particularly obvious, in these cases, is gerate the sense of frustration that can come, the potential for conceptual clash between theo- once the conceptual fog begins to clear, from rist’s and subject’s world view—a clash that can seeing one’s theoretical progress blocked by easily seem paralysing. One’s own categories are what seems for all the world to be an insur- hard to justify, and reek of imperialism; it is at mountable metaphysical obstacle. best presumptuous, and at worst impossible, to Like many prior claims, such as that all ex- try to adopt the categories of one’s subjects; and tant theories of computation are inadequate to it is manifestly impossible to work with no con- reconstruct practice, this last claim, that theoret- cepts at all. So it is unclear how, or even ical progress is stymied for lack of an adequate whether, to proceed. theory of ontology, is a strong statement, in But conceptual clash, at least outright con- need of correspondingly strong defense. In my ceptual clash, is not the only form in which the judgment, to make it perfectly plain, despite the ontological problem presents. Consider the bur- progress that has been made so far, and despite geoning interest in “complex systems” coa- the recommended adjustments reached in the lescing in a somewhat renegade subdiscipline at course of the six specific analyses enumerated the intersection of dynamics, theoretical biol- above, we are not going to get to the heart of ogy, and artificial life. This community debates computation, representation, cognition, infor- the “emergence of organisation,” the units on mation, semantics, or intentionality, until the which selection operates, the structure of self- ontological wall is scaled, penetrated, disman- organizing systems, the smoothness or rough- tled, or in some other way defused. ness of fitness landscapes, and the like. In spite One reaction to the wall might be depres- of being disciplinarily constituting, however, sion. Fortunately, the prospects are not so bleak. these discussions are conducted in the absence For starters, there is some solace in company. It of adequate theories of what organization is, of is perfectly evident, once one raises one’s head what a “unit” consist in, of how “entities” arise from the specifically computational situation (as opposed to how they survive), of how it is and looks around, that computer scientists, cog- determined what predicates should figure in
Page 16 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G characterizing a fitness landscape as rough or excellent stead as a supply of concrete cases of smooth, etc. The ontological lack is to some ex- middling complexity—what in computer sci- tent recognized in increasingly vocal calls for ence is called an appropriate “validation “theories of organization” (a theory of organiza- suite”—against which to test specific metaphys- tion is metaphysics with a business plan). But ical hypotheses. “Middling” in the sense of nei- the calls have not yet been answered. ther being so simple as to invite caricature, nor Ontological problems have also plagued so complex as to defy comprehension. Crucially, physics for years, at least since foundational is- too, they are examples with which we are as sues of interpretation were thrown into relief by much practically as theoretically familiar (we the developments of relativity and quantum me- build systems better than we understand them). chanics (including the perplexing wave-particle duality, and the distinction between “classical” 7. Summary and “quantum” world views). They face con- nectionist psychologists, who, proud of having Thus the ante is upped one more time. Not developed architectures that do not rely on the only must an adequate account (any account manipulation of formal symbol structures en- that meets the three criteria with which we coding high-level concepts, and thus of having started) include a theory of semantics; it must thereby rejected propositional content, are nev- also include a theory of ontology. It is not just ertheless at a loss as to say what their architec- intentionality that is at stake, in other words; so tures do represent. And then of course there are is metaphysics. But still we are not done. For on communities that tackle ontological questions top of these two very strong conclusions lies one directly: not just philosophy, but poetry and art, final one—if anything even stronger: where attempts to get in, around, and under ob- Computation is not a (determinate, jects a have been pursued for centuries. autonomous) subject matter. So there are fellow-travelers. But no one, so In spite of everything I said about a “theory of far as I know, has developed an alternative onto- computing,” that is, and in spite of everything I logical/metaphysical proposal in sufficient detail myself thought for twenty years, I no longer be- and depth to serve as a practical foundational lieve that there is a distinct ontological category for a revitalized practice. Unlike at least some of computing or computation, one that will be arguments for realism or irrealism, and also un- the subject matter of a deep and explanatory like some treatises pro or con this or that phi- and intellectually satisfying theory. Close and losophy of science, the task is not a meta- sustained analysis, that is, suggests that the metaphysical one—of merely arguing for or things that the things that Silicon Valley calls against a way of proceeding, if one were to pro- computers, the things that perforce are comput- ceed. Rather, the concrete demand is for a de- ers, do not form a coherent intellectually delim- tailed, worked-out account—an account, as one ited class. Computers turn out in the end to be says, that “goes the distance.” And for this pur- rather like cars: objects of inestimable social and pose, with respect to the job of developing an political and economic and personal impor- alternative metaphysics, the computational tance, but not in and of themselves, qua them- realm has unparalleled advantage. Midway be- selves, the focus of enduring scientific or intel- tween matter and mind, computation stands in lectual inquiry.
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 17 T H E F O U N D A T I O N S O F C O M P U T I N G
Needless to say, this is another extremely the claim is that such theory as there is—and I strong claim—one over which some readers may take it that there remains a good chance of such be tempted to rise up in arms. For if I am right, a thing, as much as in any domain of human ac- it implies that there will never be a satisfying tivity—will not be a theory of computation or and intellectually productive “theory of comput- computing. It will not be a theory of computa- ing” of the sort I initially set out to find. So all tion because computers per se, as I have said, do the previous conclusions must be revised. It is not constitute a distinct, delineated subject mat- not just that a theory of computation will not ter. Rather, what computers are, I now believe, supply a theory of semantics, for example, as and what the considerable and impressive body Newell has suggested. Or that it will not replace of practice associated with them amounts to, is a theory of semantics. Or even that it will not neither more nor less than the full-fledged social depend or rest on a theory of semantics, as inti- construction15 and development of intentional arti- mated in Figure 1 (b). It will do none of these facts. That means that the range of experience things because there will be no theory of compu- and skills and theories and results that have been tation at all. developed within computer science—astound- Given the weight that has been rested on the ingly complex and far-reaching, if still inade- notion of computation—not just by me, or by quately articulated—is best understood as prac- computer science, or even by cognitive science, tical, synthetic, raw material for no less than full but by a large fraction of the surrounding intel- theories of causation, semantics, and ontol- lectual landscape—this might seem like a nega- ogy—i.e., for metaphysics full bore. tive conclusion. Indeed, you might conclude Where does that leave things? Substantively, that I (and no doubt others as well) have spent it leads inexorably to the conclusion that meta- these last twenty-five years in vain. But in fact physics, ontology, and intentionality are the there is no need to worry. For I firmly believe only integral intellectual subject matters in the almost exactly the opposite: vicinity of either computer or cognitive science. The superficially negative conclusion (that com- Methodologically, it means that our experience puting is not an autonomous subject matter) with constructing computational (i.e., inten- makes the 20th-century arrival of computation tional) systems may open a window onto some- onto the intellectual scene a far more interesting thing to which we would not otherwise have and important phenomenon than it would oth- any access: the chance to witness, with our own erwise have been. eyes, how intentional capacities can arise in a “merely” physical mechanism. It is sobering, in On reflection, in fact, it is clear that the fact that retrospect, to realize that the fact that computers neither computing nor computation will sustain are computational has placed a major theoretical the development of a distinct theory is by far block in the way of our understanding how im- the most exciting conclusion that the computer portant they are! They are computational, of and cognitive sciences could possibly hope for. course; that much is tautological. But only when Why so? Because I am not saying that com- we let go of the conceit that that fact is theoreti- putation in the wild is intrinsically a-theoreti- cally important will we finally be able to see, cal—and thus that there will be no theory of without distraction—and thereby, perhaps, at these machines, at all, when day is done. Rather, least partially to understand—how a structured
Page 18 Copyright © 1996 Brian Cantwell Smith T H E F O U N D A T I O N S O F C O M P U T I N G lump of clay can sit up and think. hold a candle to computer or cognitive—or And so that, in these last five years, has been rather, as we should now call it, in order to rec- my project: to take, from the ashes of computa- ognise that we are dealing with something on tional critique, enough positive morals to serve the scale of natural science—epistemic or in- as the inspiration for, basis of, and test of, an tentional science. entirely new metaphysics. A story of objects; a Hawking (1988) is wrong. It is we, not the story of reference; a story of history; a story of natural scientists, who must develop a theory of society. For sheer ambition, physics does not everything.
References
Barwise, Jon & Perry, John (1983): Situations and At- pp. 213–225. titudes. Cambridge, Mass.: MIT Press. Hawking, Stephen W. (1988) A Brief History of Time, Devout, Michael (1991), “Why Fodor Can’t Have It Toronto: Bantam Books. Both Ways,” in Loewer, Barry and Rey, Georges Hutchins, Ed (1995), Cognition in the Wild, Cam- (eds.), Meaning in Mind: Fodor and His Critics, bridge, Mass: MIT Press. Oxford: Basil Blackwell, pp. 95–118. Millikan, Ruth (1984), Language, Thought, and Other Dretske, Fred I. (1981): Knowledge and the Flow of In- Biological Categories, Cambridge, Mass.: MIT formation, Cambridge, Mass.: MIT Press. Press. Dreyfus, Herbert (1993), What Computers Still Can’t ——— (1989), “Biosemantics”, Journal of Philosophy Do Cambridge, Mass.: MIT Press. lxxxvi:6, June 1989 pp. 281–297. Fodor, Jerry (1980), “Methodological Solipsism Con- ——— (1990), “Compare and Contrast Dretske, sidered as a Research Strategy in Cognitive Psy- Fodor, and Millikan on Teleosemantics,” Philo- chology,” Behavioral and Brain Sciences, Vol. 3, No. sophical Topics 18:2, Fall 1990, pp. 151–161. 1; March 1980, 63–73. Reprinted in Fodor, Jerry: Newell, Alan and Simon, Herbert A. (1976): “Com- Representations, Cambridge, ma.: MIT Press, 1981. puter Science as Empirical Inquiry: Symbols and ——— (1984), “Semantics, Wisconsin Style,” Syn- Search,” Communications of the Association for these, 59, pp. 231–250. Computing Machinery, vol. 19 (March 1976), No. 3, ——— (1987), Psychosemantics, Cambridge, Mass.: pp. 113–126. Reprinted in John Haugeland, ed., MIT Press. Mind Design, Cambridge, Mass: MIT Press (1981), Halpern, Joe (1987): “Using Reasoning about Knowl- pp. 35–66. edge to Analyze Distributed Systems,” Annual Re- Searle, John (1980): “Minds, Brains, and Programs,” view of Computer Science, Vol. 2, pp. 37–68. Behavioral and Brain Sciences, Vol. 3, No. 3, Sept. Harnad, Stevan (1991), “Other Bodies, Other Minds: 1980, pp. 417–458. A Machine Reincarnation of an Old Philosophical Shannon, Claude E. & Weaver, Warren (1949), The Problem,” Minds and Machines 1, pp. 43–54. Mathematical Theory of Communication, Urbana, Haugeland, John (1982), “Analog and Analog,” Phi- Illinois: The University of Illinois Press. losophical Topics (Spring 1981); reprinted in J. I. Smith, Brian Cantwell (1996), On the Origin of Ob- Biro & Robert W. Shahan, eds., Mind, Brain, and jects, Cambridge, Mass: MIT Press. Function: Essays in the Philosophy of Mind, Norman, ——— (forthcoming), The Age of Significance: Vol- Oklahoma: University of Oklahoma Press (1982), umes i–vii.
——— end of file —————— ðð
Version of 2:43:51 p.m. on Saturday, August 31, 1996 Page 19