All Things Are Full of Gods, page 31
Frankly, I don’t even see how mechanical computation could be said to resemble mental agency. It seems obvious to me that computers function as efficiently as they do precisely by virtue of the absence of anything like mental states. Much of their power lies in maintaining discrete operations that need never achieve—indeed, must never achieve—unity except at the representational level: that is, at the level of a conscious person looking at images on a screen or something of the sort. At the level of the processing of data, there are merely segmented files related to one another only occasionally, by way of algorithms, in as many separated parts as possible. Even in their physical design, computers are austerely modular systems with very few complicated connections. Neither do computers produce streams of consecutive reasoning; their processing speed isn’t a more efficient form of thinking. By the same token, a chess-player considering his or her strategy—imagining, calculating, deliberating risks, contemplating possible future configurations—isn’t engaged in a less efficient form of computation. He or she is doing something qualitatively completely different—and, for that matter, infinitely richer. Actual thought requires a kind of phenomenal reflexivity, a kind of awareness of thinking that refers everything back to the knowing subject; when one is thinking, one experiences that thinking as a phenomenal flow of impressions, sensuous or imaginative. All of that’s absent from the diverse, segregated, unconscious functions by which the algorithms in a chess-playing program eliminate statistically less successful courses of action and then, over the course of countless, mindless reiterations of all possible moves running in several discrete simultaneous operations, produce a representation—and a representation only, and then only when presented to us—of the move least likely to lead to a tactical disadvantage. There’s no intelligence there, no consciousness, no choice—no move actually made by the computer. If it “wins,” it does so precisely by not playing chess and by not contemplating the board.
HEPHAISTOS: And yet, from all that welter of intrinsically meaningless physical competences and forces, meaning still emerges.
HERMES: Hardly. Rather, upon all that welter of intrinsically meaningless physical competences and forces, meaning is adventitiously imposed—as a formal cause.
II
Functionalism, Computationalism
HERMES: [To Psyche:] May I take over again? As you know, the issue of language is one especially dear to my heart.
PSYCHE: Please. I’d be grateful.
HERMES: My thanks. So, then, let me try to define the logic of functionalism as I understand it. It’s the claim that the brain is what Daniel Dennett calls a “syntactic engine,” which more or less as a phylogenic fortuity or hypertrophy was selected by life’s Darwinian logic, and which over the evolutionary epochs has come to function as a “semantic engine.” That is, the brain’s a machine platform, something like a Turing device, one that began its existence as a physiological organ for translating stimuli into responses but that now runs a more sophisticated program for translating “inputs” into “outputs.” In Dennett’s version of the story, this happened in part as a result of the brain’s gradual colonization by “memes”; but, whatever the case, supposedly this program has come to incorporate in its functions a number of brain-states that, in order to serve as useful interfaces between us and all those computational processes, appear to us to be conscious or intentional states with meanings, though in essence they’re nothing of the sort. The governing maxim of functionalism would be something like “Once the proper syntax is established in the neurophysiology of the brain, the semantics of thought will take care of themselves.” That is, once the syntactic engine begins running its impersonal algorithms, the semantic engine will eventually emerge or supervene.
EROS: Sorry, can we pause to refresh our memories? I’ve more or less forgotten what “memes” are supposed to be—assuming I ever bothered to find out.
HERMES: Oh, it’s a term coined by Richard Dawkins, indicating a kind of cultural analogue of genes. A meme is supposedly a unit of shared behavior or thought, like a style of clothing or an architectural fashion or a tune or an idea or a turn of phrase, that’s transmitted from one person to another by imitation and variation. More to the point, memes supposedly replicate themselves in some sense, colonizing the ecology of the brain, adapting, surviving, and displacing less robust memetic populations, and thereby creating and determining the contents of consciousness. It’s a largely worthless concept. I mean, as an ironic metaphor it’s not a bad way of describing the genealogy of popular culture, but as a serious proposal regarding how intentionality is formed it’s vapid. Not that I don’t see its appeal for those devoted to a computational model of mind. If it made sense, it would certainly constitute a convenient conceptual bridge between the uncomprehending physiological competences of brains and the apparent intrinsic intentionality of minds; but, alas and alack, it doesn’t make sense. The basic idea—the notion, that is, of some pre-conscious form of intentional content that proliferates on its own, and that’s situated in the brain the way DNA is situated in the cells of the body, contributing “information” to consciousness in the way that genetic codes contribute “information” to organisms—is impossible to state cogently. Genetic materials are propagated physically because they’re physical realities. “Memes,” however, if such things exist, are composed entirely of intentional content. They couldn’t be “selected” by nature, in the way the units of biological evolution are said to be, but would literally have to be chosen, if perhaps a little passively, by conscious minds. They’d be objects of intentionality, as well as intentional artifacts, but they certainly couldn’t explain intentionality; the existence of a cause can’t be explained by the existence of its own contingent effects.
HEPHAISTOS: For what it’s worth, the versions of functionalism I’d be disposed to defend wouldn’t include “memes,” which I too see as an inane notion.
HERMES: But you’d still agree, I assume, that for any functionalist model of thought—and most especially any computationalist model—one would have to accept this model of a basic syntactic engine or system and then a secondary semantic engine or system that runs parallel to it, in dependency upon it? I mean, you’d accept this idea that semantics supervenes upon syntax, not only in computers but also in brains? And you’d say it’s this parallel stratification of functions that mechanically yields thought and meaning—that thought and meaning are, so to speak, parasitic upon the brain’s computational functions?
HEPHAISTOS: A fair précis of my position.
HERMES: Well, that’s something of a problem, then. I see that you want to show how thought can arise from a purely mechanical substrate; but I also see you can’t accomplish this by reference to mechanical computation. If you really want to suggest a model for how a system of physical “switches” or operations can generate a syntax of functions, which in turn can generate a semantics of thought, which in turn can produce the reality or illusion of consciousness—well, then you’ve got your argument totally backward. None of that’s what a computer actually does, and certainly none of it’s what a brain would do if it were only a computer. If that’s functionalism, then it really is, as Psyche says, a collection of vacuous metaphors. Worse, they’re metaphors based on a phenomenon that doesn’t and couldn’t actually exist. Software isn’t the composite result of uncomprehending competences. Its functions are imposed from above, and are dependent on a higher order of semeiotic content, which is itself dependent upon a fully present teleological and intentional structure of meaning, already containing both syntax and semantics. In any intentional structure, meaning is the ontological ground, not the causal result, and the ontological ground of meaning is the intrinsic intentionality of a mind. It’s all an entirely top-down causal hierarchy: from mind down to a fully formed semeiotic system, then further down to coding, then further down still, all the way to those physical switches whose operations the coding determines. The theory you’re advancing is worthless. To repeat a point we keep returning to, a physicalist reduction of a phenomenon to purely material forces explains nothing if one can’t then reconstruct that phenomenon from principles wholly inherent in that material basis, without invoking any higher causes; and this no computational picture of thought can ever do. Intentional contents—symbols—don’t emerge from the physical processes in a computer; they inform those processes, and arrange them according to a rational rather than physical order. It is they that create the system’s ensemble of competences, not the reverse. Looking from the opposite direction, hoping to discover a causal order rising up from below, one finds instead only the untraversable abyss that separates the intentional nullity of matter—matter as mechanistically conceived, at any rate—from the intentional plenitude of mind. Rather than meaning supervening secondarily upon mechanical functions, those functions are subvenient, so to speak, to meaning. Simply said, computation is ontologically dependent on intentionality and consciousness, so it can’t possibly be their foundation.
HEPHAISTOS: As I think I admitted . . . oh, two days ago, it’s true that computer software has a human programmer, whereas the mind was programmed by evolutionary history. You’re confusing my structural analogy with a genetic analogy.
HERMES: Have you listened to nothing I’ve said? The issue of the dependency of computer functions on higher causes isn’t just a genetic matter of where the software comes from. It’s a structural issue through and through. In a computer, nothing like meaning or intelligibility ever arises from the coding. Look, you’re familiar with John Searle’s “Chinese Room” argument?
HEPHAISTOS: I am.
EROS: And I once more, sad to say, am not.
HERMES: Well, the Chinese Room argument is simply meant to show that even a computer capable of passing the Turing test—that is, a computer able to convince even a diligently skeptical human interlocutor that it’s in fact a conscious agent—would never actually possess subjective awareness or understanding of its own operations . . . at least, not on the basis of its computational processes alone. It’s a simple thought-experiment: Say we designed a program that allowed for an entirely convincing “conversation” in Chinese between a computer and a native speaker of Chinese; would this be the same thing as the computer in some vague sense “knowing” Chinese? To answer this, imagine that instead of a computer there were simply a room in which we’ve installed a man who knows not a word of Chinese, but who’s been equipped with an absolutely infallible set of instructions that allow him, on receiving “input” written in Chinese ideograms through a slot in a door, to produce perfectly appropriate “output,” also in Chinese ideograms, even though he has no idea what any of it—the input or the output—means to readers of Chinese. Manifestly, that man would have no grasp at all of the semantic content he’s supposedly processing. And that, says Searle, means that there’s no intentionality and hence no thinking going on in the translation process, either in the computer or in the Chinese Room, purely on the basis of those syntactic protocols and operational processes.4
EROS: And its importance here is . . . ?
HERMES: Well, for functionalists, thought is essentially a kind of computation, and computation, according to this scheme, is the manipulation of symbols purely in terms of their syntactic, rather than semantic, properties. Semantic content comes after. But how can this be, Searle prompts us to ask, when the program is only mechanical and syntactic while thought is also mental and semantic? How does the former become the latter?
HEPHAISTOS: Don’t neglect, while you’re at it, to mention that it’s an argument that’s been subjected to some fairly excoriating critique in the past.
HERMES: And none of it successful, really.
HEPHAISTOS: Oh, really? You’ve complained of Dennett discreetly introducing intentional and mental language into his description of a robot’s computational functions, only then to use that language to suggest that what we think of as mental states might be generated by such functions. I take your point. Rhetorical sleight of hand is to be avoided. But note here how Searle has introduced the figure of an already complete conscious self into his little fable, but sealed it off from the systems in which that self is embedded. As many have argued, however, the whole point of functionalism is that it’s the system as a whole that understands—that in fact the processing of those symbols by way of their syntactic properties is what yields or accommodates the supervenient semantic stratum of meaning.
HERMES: Which, for reasons already laid out in detail, is absurd. That’s simply not what computers or brains do. Systems don’t and can’t understand anything; only agents can. As for Searle starting with the image of a completely competent mental agent at the center of the system, that’s precisely the point: even if the processing of those ideograms were performed by an unquestionably conscious mental agent rather than by a programmed machine, there still wouldn’t be any semantic knowledge involved at any point in the process, and there’d be no occasion or logic dictating that such knowledge could arise from or supervene upon the process. This isn’t an instance of the “fallacy of composition” either. It’s true that you can’t deny that a system as a whole has consciousness simply on the grounds that one part of the system lacks that capacity. But the man in the Chinese room isn’t part of the system; he is the system, and one totally competent to understand semantic information. He’s the recipient of “input” and the producer of “output”; even the rules he obeys exist only in his mind, even if they’re written down for him somewhere, since the physical medium of ink traces on paper isn’t where that information has a functional existence either. Searle really didn’t need the conceit of the room at all. The point of the story is simply that a system can manipulate syntactic correspondences and equivalences forever and ever, but this will never produce semantics. That said, though, I’d agree that the Chinese Room argument isn’t wholly adequate—unless I’m much mistaken, Searle himself has concluded as much—but only because it still concedes too much to functionalism. In actual fact, computation is devoid not only of semantics, but of syntax as well. This is a problem in most information-theory, as it happens: the failure to distinguish between syntax in the proper sense—that is, the organization of semantic functions in a coherent order—and the mere physical vehicle of information—the physical traces that convey letters, numbers . . . the “bits,” that is. A program doesn’t run on symbols and meanings, true; but neither does it run on syntactical symbolic relations. It runs simply on bare juxtapositions of tokens of code that, from the “perspective” of the computer (so to speak), have neither semantic nor syntactic features of any kind. There are no grammatical functions in a computer—no verbs, nouns, or predicates, no indicative, subjunctive, or optative moods, no imperatives or interrogatives, no conditional forms. Thought and language are semeiotic processes, yes, but computation is, as we’ve said, nothing even remotely like such processes.
In the end, neither computers nor brains are either syntactic or semantic engines; there are no such things as syntactic or semantic engines; syntax and semantics exist only as intentional structures, inalienably, in a hermeneutical rather than physical space, and do so only as inseparable aspects of a single semeiotic system. Syntax can’t exist prior to or apart from semantics, and neither exists except in the intentional activity of a mind. We can talk of syntax in the abstract, but what we’re talking about then is always only an artificial distillate of the complete language of signs that generated it; it can’t produce a semantics because it’s ontologically dependent upon semantics, just as semantics is ontologically dependent on it; it functions only within the system of signs it shapes. Now, of course, to phrase the matter in terms of the “structuralist” school of linguistics, the signifier within any given sign—the actual word in its concrete written or spoken form—may be variable and somewhat fortuitous, but the signified within that same sign—the concept or meaning it indicates—is largely invariable within any given syntax. And, to phrase things in the terms of “poststructuralism,” signs possess their meanings only within and by virtue of all other signs. Or, as C. S. Peirce liked to say, signs are interpreted always by other signs, in an infinite web of signification. But what’s important is that all of it—all language, all semeiotic functioning—exists only by virtue of the prior reality of intentional mind.
As for the operative structure of computer coding, moreover, even the distilled or abstracted syntax upon which that coding relies has no actual existence within a computer. To imagine it does, as dear Psyche likes to say, is rather like mistaking the ink and paper and glue in a bound volume for the contents of its text. Meaning—syntactical and semantical alike—exists in the minds of those who write the code for computer programs and of those who use that software, but not for a single moment does any of it appear within the computer. The software itself, for one thing, possesses no semeiotic unities of any kind; it merely simulates those unities in representations that, at the physical level, have no connection to those meanings. And, as we’ve also said, the results of computation that appear on a screen are computational results only for the person reading them. In the machine, they’ve no significance at all. Even the software’s code, as an integrated and unified system, inhabits no physical space in the computer; the machine contains only notations, binary or otherwise, that cause mechanical processes to produce simulations of the representations for which the software was written. So the functionalist conceit that thought arises as a posterior effect of semantics, and that semantics emerges from syntax, while syntax is generated out of a purely physiological system of stimulus and response—this entire nonsense of an emergent, causally bottom-up hierarchy of meaning and agency—could scarcely be more backwards. When one decomposes intentionality and consciousness into their supposed semeiotic constituents, and signs into their syntax, and syntax into physical functions, one isn’t reducing the phenomena of mind to their causal basis; one is dissipating those phenomena into their ever more rarefied, remote, and impotent dependent effects. Meaning isn’t a physical result of lower functions; it exists in minds that can extract patterns from their own operations and employ them to produce instruments—books, abacuses, Turing devices, what have you—of notation and information-processing. But this happens within what must remain forever an entirely top-down and indissoluble hierarchy of dependent relations, unified at its apex by intentional mind, or all at once it will disintegrate. In any intentional structure, the more “eminent” reality of realized meaning is the ontological ground of the operations that the structure creates and sets into motion. Simply said, as with language, so with intentional agency in its every dimension: it can’t arise from its own contingent consequences.
