Reply to Commentaries on "Time and the Observer: the Where and When of Consciousness in the Brain," for Behavioral and Brain Sciences, 15, 183-247, 1992
We wish to thank the commentators for their largely constructive criticism. It is gratifying to discover that some of them had been thinking--and in some cases publishing--ideas along similar lines. We claimed that Cartesian materialism, the view of the brain with some deep center where "it all comes together" for consciousness, often seduces even those who explicitly reject it. As Neisser (1976), an early critic of the view, has said: "It is currently a very popular notion, and with good reason. It represents a theoretical coup: not only are the facts of attention apparently explained, but psychology's most elusive target is finally nailed down to a box in a flow chart." (p.103) Even now, Damasio remarks, it "informs virtually all research on mind and brain, explicitly or implicitly." Indeed, serial information processing models generally run this risk (Kinsbourne, 1985). The commentaries provide a wealth of confirming instances of the seductive power of this idea. Our sternest critics (Block, Farah, Libet, and Treisman) adopt fairly standard Cartesian positions; more interesting are those commentators who take themselves to be mainly in agreement with us, but who express reservations or offer support with arguments that betray a continuing allegiance to one or another tenet of the view we sought to discredit.
The issues are extraordinarily slippery, and reading the commentaries has been a daunting experience. If only we had thought of putting it this way, rather than that way--it would have forestalled one all-too-reasonable objection or another. We first present, then, a refined summary of our central view, much improved, we think, by its annealing in reaction to the commentaries. Then we turn to the major themes expressed in the commentaries, and finally we respond to additional important issues raised.
1. Multiple Drafts: an Improved Summary (Roskies and Wood, Lycan, Glymour et al., Jeannerod, Libet, Lloyd, Baars and Fehling, Aronson et al., Block, Clark, Teghtsoonian, Treisman, Velmans, Shepard,
2. Is there a Fact of the Matter? (Lycan, van Gulick, Glymour et al.,Antony, Aronson et al., Lloyd, Velmans, Treisman, Clark, Editorial Commentary, Rollins)
3. When does "Filling In" Happen? (Shepard, Roskies and Wood)
4. Consciousness as a "System" with a Function (Antony, Teghtsoonian, Baars and Fehling, Velmans, Farah, Lycan, Reingold, Bridgman, Lloyd, Block, Van Gulick)
5. Libet and Treisman Tilt at Philosophers (Libet, Treisman, Aronson et al., Van Gulick, Hurley)
6. Other Objections (Wasserman, Gregson, Warren, Block, Van Gulick, Glymour et al., Shepard, Aronson et al., Lloyd,
7. Supporting Arguments (Damasio, Hurley, McDermott, Reingold, Rosenthal, Young)
1. Multiple Drafts: An Improved Summary
All the work that was dimly imagined to be done in the Cartesian Theater has to be done somewhere, and no doubt it is distributed around in the brain. This work is largely a matter of responding to the "given" by taking it--by responding to it with one interpretive judgment or another. This corner must be turned somehow by any model of observation. On the traditional view, all the taking is deferred until the raw given, the raw materials of stimulation, have been processed in various ways. Once each bit is "finished" it can enter consciousness and be appreciated for the first time. As C. S. Sherrington (1934) put it:
The mental action lies buried in the brain, and in that part most deeply recessed from outside world that is furthest from input and output.
In our model, this single unified taking is broken up in cerebral space and real time. We suggest that the judgmental tasks are fragmented into many distributed moments of micro-taking (Kinsbourne, 1988, Damasio, 1989). The novelty in what we attempt here lies in how we develop the implications of this fragmentation. We have not merely broken the Cartesian Theater into thousands of mini-cinemas (Roskies and Wood, Lycan, Glymour et al.). Here is where the intuitive distinction between conscious taking and unconscious taking tends to beguile the theorist. It seems as if we are stuck with only three alternatives:
A. Each of these distributed micro-takings is an episode of unconscious judgment, and the consciousness of the taken element must be deferred to some later process (the Stalinesque show trial in a Cartesian Theater). But then how long must each scene wait, pending potential revision, before the curtain rises on it?
B. Each of these distributed micro-takings is an episode of conscious judgment (multiple mini-cinemas). But then why don't we all have either a kaleidoscopic and jumbled "stream of consciousness" or (if these distributed micro-takings are not "co-conscious") a case of "multiple selves"? Is our retrospective sense of unified, coherent consciousness just the artifact of an Orwellian historian's tampering with memory? As several commentators (Baars and Fehling, Libet, Aronson et al.) ask, how can the manifest coherence, seriality, or unity of conscious experience be explained?
C. Some of the distributed micro-takings are conscious and the rest are not (e.g., Block's "Cartesian modularism" with a spatially distributed "module"). The problem then becomes: what special property distinguishes those that are conscious, and how do we clock the onset of their activity (Clark)? And, of course, since distributed micro-takings may occur slightly "out of order," what "mechanism" serves to unify the scattered micro-takings that are conscious (Teghtsoonian), and in each case, does it operate before or after the onset of consciousness (i.e., which phenomena are Orwellian and which are Stalinesque)?
Our view is that there is a yet a fourth alternative:
D&K. The creation of conscious experience is not a batch process but a continuous process. The micro-takings have to interact. A micro-taking, as a sort of judgment or decision, can't just be inscribed in the brain in isolation; it has to have its consequences--for guiding action and modulating further micro-judgments made "in its light", creating larger fragments of what we called narrative. In the target article we were silent on the mechanisms of interaction, but we did not mean to rule out processes that amount to partial construction of intermediate cases as an effect of a micro-taking, rather than as its basis, as a way of ensuring the appropriate further influences (see our discussion of Shepard and Roskies and Wood, below). This interaction of micro-takings, however it is accomplished in particular cases, has the effect that a modicum of coherence is maintained, with discrepant elements dropping out of contention, and without the assistance of a Master Judge. Since there is no Master Judge, there is no further process of being-appreciated-in-consciousness, so the question of exactly when a particular element was consciously (as opposed to unconsciously) taken admits no non-arbitrary answer.
Jeannerod provides an elegant demonstration of one prediction of this model when he and his colleagues demonstrate disunity of the self when examined in terms of timings of a fraction of a second. The time when the subject initiates a reaction to a stimulus depends on which effector--in this case hand or voice--is used, a finding incompatible with a single locus of decision in the Cartesian Theater (see also Marcel, in press). Whether they have really shown a dissociation between a preconscious (hand) response and a response based on a subjective experience (voice) is open to question. The amount of reprogramming needed to redirect a motor response already in formation would appear to be minimal and the reprogramming is accomplished under circumstances of very high stimulus-response compatibility. The vocal response, the relation of which to the stimulus is arbitrary, is less compatible. Possibly, had the task called for an imitative vocal response, subject to occasional minor modification, and an arbitrary manual response to such a change, the opposite difference would have been found: far briefer vocal than manual latency. There is no way of knowing whether the vocal response preceded, coincided with, or followed awareness of the stimulus. Indeed, there is nothing to know; all that can be said is that the two response types are directed by separate "drafts".
Some of the commentators (Libet, Treisman, Velmans) took us to be defending B, the Orwellian view. Patently we were not; we were instead demonstrating that it was the mirror image of A, the Stalinesque view, and that neither of them could evade the charge of unfalsifiability. Glymour et al. mistakenly suppose that our model "includes the notion of consciousness as an observer and interpreter of some 'draft'." This is precisely not our view; interpretation is not reserved for some one draft; all drafts are products of (not just candidates for) interpretation, independently of consciousness. At least one commentator (Lloyd) took us to be defending C, and criticized us for failing to provide the distinguishing mark of the conscious takings. Other commentators did understand that we were defending something like D&K, but took it to be too radical, metaphysically. Since this is an almost overpoweringly plausible view, we will respond to it first, before going on to other objections.
2. Is there a Fact of the Matter?
The commentators generally agree with us that (1) the time of representing should not be confused with the time represented, and (2) there is no privileged place within the brain "where it all comes together." They do not all agree with us, however, that it follows from (1 and 2) that the Orwellian/Stalinesque distinction must break down at some scale of temporal resolution, leaving no fact of the matter about whether one is remembering mis-experiences or mis-remembering experiences. Here, some claim, we have gone overboard, lapsing into "verificationism" (Lycan, van Gulick, Glymour et al.) or "eliminativism" (Antony, Glymour et al., Aronson et al.) or "anti-realism" (Lloyd, Van Gulick), or some other gratuitously radical position (Velmans, Treisman). This is curious, for we consider our position to be unproblematically "realist" and materialist: conscious experiences are real events occurring in the real time and space of the brain, and hence they are clockable and locatable within the appropriate limits of precision for real phenomena of their type. (For an extended defense of this version of realism, see Dennett, 1991a.) Certain sorts of questions one might think it appropriate to ask about them, however, have no answers, because these questions presuppose inappropriate--unmotivatable--temporal and spatial boundaries that are more fine-grained than the phenomena admit.
In the same spirit we are also realists about the British Empire--it really and truly existed, in the physical space and time of this planet--but, again, we think that certain sorts of questions about the British Empire have no answers, simply because the British Empire was nothing over and above the various institutions, bureaucracies and individuals that composed it. The question "Exactly when did the British Empire become informed of the truce in the War of 1812?" cannot be answered. The most that can be said is "Sometime between December 24, 1814 and mid-January, 1815." The signing of the truce was one official, intentional act of the Empire, but the later participation by the British forces in the Battle of New Orleans was another, and it was an act performed under the assumption that no truce had been signed. Even if we can give precise times for the various moments at which various officials of the Empire became informed, no one of these moments can be singled out--except arbitrarily--as the time the Empire itself was informed. Similarly, since You are nothing over and above the various subagencies and processes in your nervous system that compose you, the following sort of question is always a trap: "exactly when did I (as opposed to various parts of my brain) become informed (aware, conscious) of some event?" Conscious experience, in our view, is a succession of states constituted by various processes occurring in the brain, and not something over and above these processes that is caused by them.
The idea is still very compelling, however, that "realism" about consciousness guarantees that certain questions have answers (even if they are currently unknowable). Embedded in the target article were grounds for rejecting this position, but they did not stand forth for some of the commentators. Here, thanks to the light thrown by the commentaries, is a more succinct version of the basic argument.
The only difference between the Orwellian and Stalinesque treatment of any phenomenon is whether or not the editing or adjustment or tampering occurs before or after a presumed moment of onset of consciousness for the contents in question. The distinction can survive only if the debut into consciousness for some content is at least as accurately timable as the events of micro-taking (the binding, revising, interpreting, etc.) whose order relative to the onset of consciousness defines the two positions. If the onset of consciousness is not so sharply marked, the difference between pre-presentation Stalinesque revision and post-presentation Orwellian revision may disappear, and be restorable only by arbitrary fiat.
As "realists" about consciousness, we believe that there has to be something--some property K--that distinguishes conscious events from nonconscious events. Consider the following candidate for property K: a contentful event becomes conscious if and when it becomes part of a temporarily dominant activity in cerebral cortex (Kinsbourne, 1988, and in preparation). This is deliberately general and undetailed, and it lacks any suggestion of a threshold. How long must participation in this dominance last, and how intense or exclusive does this dominance need to be, for an element to be conscious? There is no suggestion of a principled answer. Such a definition of property K meets the minimal demands of "realism," but threatens the presumed distinction between Orwellian and Stalinesque revisions. Suppose some contentful element briefly flourishes in such a dominant pattern, but fades before leaving a salient, reportable trace on memory (a plausible example would be the representation of the first stimulus in a case of metacontrast). Would this support an Orwellian or a Stalinesque model? If the element participated for "long enough" it would be "in consciousness" even if it never was properly remembered (Orwell), but if it faded "too early" it would never quite make it into the privileged category, even if it left some traces in memory (Stalin). But how long is long enough? There is no way of saying (contra Libet, see below). No discontinuity divides the cases in two.
But perhaps we are overlooking some source of discontinuity; perhaps there is a sharp (sharp enough) break after all, so that the third alternative, view C, can be maintained. Clark claims as much: we might "still discover that there is some functional property . . . which is both necessary for a content to become consciously known and which yields an absolute temporal order of experiences. . . . I cannot see that this possibility is ruled out by anything Dennett and Kinsbourne tell us."
Suppose, then, that what makes some contentful brain events conscious is a property K that has a rather clearcut onset. On such a view, contentful events, like plants, have rather long histories; they are unconsciously sown, develop, briefly bloom (rather suddenly acquiring some salient property K), and then fade into longterm memory or oblivion (losing property K). A single contentful event, let us suppose, can have a temporal subpart that is conscious, marked by the onset and offset of property K. To identify the subset of conscious events, just identify property K and motivate its identification. In order for any such claim to be taken seriously, some reason(s) must be given for singling out this property K, whatever it is, as the mark of consciousness (and hence the time of onset of K as the time of onset of consciousness). It will not do just to announce without further explanation that when events acquire property K, unlike their unconscious kin, they glow in the dark, as it were.
It is this independent motivation, we claim, that cannot be provided. There is no further functional or neurophysiological property K, we claim, over and above the properties that account for the various "bindings" and effects on memory, speech and other behavior, and those properties cannot distinguish between Orwellian and Stalinesque models. Once discrimination and control get distributed around to many sub-agencies, operating on different schedules, all the accomplishments of consciousness occur, one way or another, at one time or another, but no grounds remain for deeming one version of these events the "actual conscious experience."
The Editorial Commentary provides a minimal test case: a "simple, punctate 'pink-elephant-now' thought." Surely its onset in consciousness is unproblematically clockable! As the Editor says, "all we want to know is when the actually experienced NOW actually occurred." Let's consider the details. However "simple" and "punctate" it appears to be, the thought has to be composed, for it consists in (at least) three content components, PINK, ELEPHANT, and--a tricky one--NOW, which must be generated and "bound" together somehow (the thought is of an elephant that is pink, and pink now, not just the thought of ELEPHANT, and PINK, and NOW). The first two components may be generated in different regions, at slightly different times, and could presumably even acquire property K at different times. Or presumably they could get bound together but fail to acquire property K, and fade away without ever being experienced, an unconscious pink-elephant-now content. This raises the question: does the content PINK become conscious before or after it is bound to ELEPHANT? That is, does the "binding" happen before property K is acquired by each element (the Stalinesque theory), or do the content elements acquire property K seriatim, perhaps in the order that they are generated in the brain, with the binding taking place only after all the elements are conscious (the Orwellian theory)? Or does the question not make sense (our theory)? An analysis of the neural invariants recorded over hundreds of variations of pink-elephant-now thoughts may show precisely what conditions ensure that such a thought occurs at roughly such-and-such a time, and may even show considerable variation in time of binding relative to such other temporal landmarks as onset of PINK activity and onset of motor response formation activity, but the one landmark that matters has no independent anchor: onset of K. The subject cannot provide any clues unobtainable by an analysis of the neural activity, because ex hypothesi the different instances are all the same subjectively--simple and punctate.
But what about "the actually experienced NOW"? The content NOW is tricky because, unlike PINK and ELEPHANT, it is indexical: it refers to its own vehicle of representation, and hence bridges the gap between temporal properties represented and temporal properties of representings. But this fact creates a problem instead of solving one: The "concurrent neural event that corresponded to NOW" (the tokening of NOW, in philosophers' jargon) is presumably spread over a pre-conscious forming period and a post-conscious remembered period (like a postcard on which you write "Please visit me NOW" and later mail to a friend, on whose desk it sits for weeks). Unless one can independently identify the onset of consciousness, the discovery of a "NOW" token in the brain hopelessly underspecifies a time of intended "utterance". The actually experienced NOW is presumably not the preparatory, forming-but-not-yet-conscious NOW, or the reverberating-in-memory NOW, but the blooming-with-property-K NOW. How the Editor proposes to identify the thresholds that mark the transition of the tokening of NOW between these three states remains a mystery, since no functional difference has been shown to depend on these transitions, and all the subjective evidence concerns only the temporal properties represented in the thought (e.g., the content: NOW is/was simultaneous with PINK-ELEPHANT), not the temporal properties of the representing. So even in the case of such a simple "punctate" thought, in which no questions arise about its veridicality relative to any external world it is portraying, when the Stalinesque/Orwellian distinction looms, nothing could settle it.
It is unclear to us why the Editor chose the epithet "Reaganesque" for his example, but it strikes us in any case as peculiarly apt for making our point, not his. Recall the question that rang through the land during the Iran-Contra hearings: What did the President know and when did he know it? This question mattered only because it was presumed that the President was actually in charge, actually a responsible decision-maker. But (according to widespread opinion) while the Great Communicator was an excellent public relations officer, he was otherwise a mere figurehead, whose authority was more a convenient fiction than the wellspring of decision. If this is so, there were surely many times when it didn't make any political difference whether Reagan actually said the things that his staff had decided it would be good for him to say--or to have said. He could "fill in" the words if he wished, but if he didn't, no harm would be done. In the case of Reagan, there actually always was a (negligible) fact of the matter, but unless you think there is a Reagan homunculus in you, whose activities are for some reason constitutive of consciousness, there is no reason to believe in the analogous fact of the matter for "Reaganesque" events in your brain. In systems in which the authority and decision-making are distributed, as they were in the Reagan White House, and are in the brain, there is a sort of Virtual President, but there is no fact of the matter about exactly where and when a Virtual President announces to himself "pink elephant now".
The onset of property K (whether it is gradual or sharp) is a temporal property of the representing, not a temporal property represented, and, as all agree, we must be careful not to confuse these. There is a danger of confusion precisely because there is bound to be some sort of regular relation between them: it is no accident that representations of sequences of events will tend to be represented by sequences of representations of those events, that events represented as simultaneous will tend to be simultaneously represented, and so forth. A similar point applies to spatial properties: the representation of x surrounded by a field of y will tend to be accomplished by a representation of x surrounded by a representation of y, and so forth, but the space in the brain in which the representation is accomplished is not the space represented, however regular the correlation between spatial properties represented and spatial properties of the representation. Rollins makes two points in this regard that we accept: (1) temporal isomorphism may be required for some temporal tasks. (2) the appeal to isomorphism need not commit what he calls the canonical fallacy. Sometimes the freedom to order events shrinks for all practical purposes to zero, and isomorphism returns. It is easier to understand this point if we remind ourselves of its spatial analogue: even though there is no general reason why the brain should always represent spatially contiguous regions of the world (or the body, or the retinal image of the world) by spatially contiguous vehicles of representation, there are circumstances in which the exigencies of engineering make this a very good, well nigh inevitable, solution to the design problem. But even when this is the case, we must be careful not to view the spatial isomorphism as a determining feature of the representational content, rather than a limiting condition on the freedom of the vehicles. We must also not make the mistake of "restoring" the isomorphism in some dimly imagined later process. (The back-to-back semi-circles of excitation that occur on the occipital cortex when one is visually stimulated by a large circular ring don't have to be re-united into a circular representation somewhere in order to be the brain's way of representing a circle. The same moral applies to time.)
3. When does "Filling In" Happen?
Shepard, while largely in agreement with us, claims we have gone too far in disparaging "filling in." We agree with Shepard that the target article permitted the inference to be drawn that the brain never bothers doing something like filling in, and that this would be an overstatement. His work has shown clearly that there are phenomena in which there are analog or roughly continuous processes that do amount to a sort of filling in: as he puts it, these processes "selectively activate (or 'fill in') intermediate states corresponding, in concrete detail, to what would be successive intermediate states in the external world." Roskies and Wood make the same point citing further phenomena in which a sort of filling has been shown to occur. Their discussion also provides oft-repeated warnings--they appreciate the need for eternal vigilance--against the dangers of overinterpretation: "One cannot, of course, conclude that these particular neuronal effects are the only or the most important basis for the subjective experience of color-filling." Indeed one must be careful not to jump to the conclusion that a particular instance of neuronal filling in (or, as the connectionists would say, vector completion) should be viewed as the basis, as opposed to an effect, of subjective experience. Vector completion--e.g., in remembering--is not filling in of the sort we were criticizing. Vector completion isn't "paint" used to render the evidence for some ulterior conclusion; vector completion is the conclusion. No more "giving" is needed because no more "taking" is going to occur. Such a process has the nice property of merging the distinction between presentation and judgment, turning the corner from given to taken without turning the corner sharply. Notice how we can accommodate Shepard's phenomena, for instance, without lapsing back into the Cartesian Theater. Suppose, in his illusory motion experiment, a visual subsystem in the subject's brain has arrived at the micro-judgment that the vertical bar moved clockwise. But of course not just clockwise, but at a particular rate and during a particular window of (represented) time. Suppose this elaborated micro-judgment is represented by an analog process, which in turn provides the grounds for a second micro-judgment, provoked by the precisely timed probe spot flashed along the illusory trajectory of the moving bar, to the effect that the spot flashed before or after the bar had crossed that point. Each bit of judgment becomes the basis for further judgment, blurring the distinction between given and taken. The taking that there was motion happened before the taking that the spot was such and such--it had to, logically--but it also provided the basis for the judgment that in mid-course the dot was in such and such a location.
Such processes do provide, as Roskies and Wood claim, "a candidate explanation for observers' reported subjective experience," which does not mean: they provide an explanation of the construction of the "given" which is then subsequently taken by some further process of subjective experiencing. In their discussion of apparent motion, they propose a testable hypothesis:
the perception of apparent motion might begin with an initially nondirectional priming from the first flashed stimulus. If another stimulus with similar characteristics should appear within the primed region before the activation has decayed, a motion signal [our emphasis] might be generated, either by potentiating the intervening circuit, or by sending a signal which is interpreted by other areas as coding motion.
The point we would make--and we gather Shepard, Roskies and Wood would agree--is that a motion signal is not to be confused with a motion picture, even when it is accomplished by a process (which is not itself accessible to consciousness) that generates intermediate states. (For more on "filling in" and evidence of analog or "roughly continuous" processes in perception, see Dennett, 1991b.)
4. Consciousness as a "System" with a Function
One of philosophy's virtues is that it sometimes makes explicit the assumptions that are more covertly driving the imaginations of theorists in other disciplines, exposing them for assessment. A fine case is provided by the commentaries: the assumption that might be formulated in the following slogan:
Every real thing must be in its own box.
According to Antony, the most charitable interpretation of our position is "eliminativism": "consciousness does not exist." He eventually arrives at the correct reading of our view: "conscious experiences are temporally located, but the time scale appropriate for the measurement of neurophysiological events (e.g., milliseconds) is too fine-grained to determine nonarbitary temporal boundaries for experiences." From this, however, he draws the ominous--and mistaken--conclusion that "conscious experiences have no role in the functional organization of the brain." (See Velmans for a similar conclusion, and also his BBS target article, 1991) The conclusion we would draw instead is that conscious experiences have no role in the functional organization of the brain in virtue of meeting some criterion of consciousness. Becoming a conscious experience does not clearly endow an event with potencies it previously lacked. Antony says it follows from this that "conscious experiences cannot be identified with functional states or processes," but this is true only in the sense that conscious experiences cannot be identified as a type with any particular functional type. That is a far cry from "eliminativism." The events that constitute conscious experiences do play functional roles that are datable down to the millisecond, but when or whether they play those roles is independent of when or whether they make the cut into the elite circle of conscious events.
The assumption that consciousness cannot be real unless there is a consciousness subsystem can be seen playing a slightly less visible role in the commentaries of Teghtsoonian, Baars and Fehling, Velmans, and Farah. (Bridgman clearly expresses the case against the assumption.)
Teghtsoonian argues for a "central monitor" on the grounds that cross-modal comparison is a demonstrably real phenomenon. (Contrary to what he supposes, we do not claim to offer any arguments, new or old, against dualism. We assume that dualism is a dead horse, but that Cartesian materialism is alive and well, and worth combatting.) We agree with him that very many experiments, by S. S. Stevens and others, demonstrate conclusively that people can do cross-modal comparisons, and that these have to involve getting all the information into the same physical system. This "single monitor" is called the brain. Teghtsoonian takes the brain's capacity to serve all these comparative purposes as evidence for the existence of monitoring "systems" and then goes on to say that presumably similar monitoring systems exist for other attributes. Depending on how we understand "system" this is either trivially true, or entirely unlikely. People can compare the experience of falling in love with the experience of contracting a disease, or going into battle, but this gives no grounds for thinking that there is a monitor system (a "module" as some would say) specifically made for this comparison task. We can construct comparison "machinery" on the fly, ad lib, as we need it--thanks to the richly interconnected brain that provides the underlying hardware. The risk we run if we start diagramming these feats with the "boxological" flow charts of functionalism is that we will assume that the boundaries of the boxes we draw have a saliency and integrity that projects beyond the cases that inspired us to draw them. (Our only misgivings about Lycan's "inner sense" version of "UnCartesian materialism" are with his eagerness to reify all the episodes of corner-turning from given to taken into the acts of "internal scanners", which then encourages him to assert that "An internal monitor is an attention mechanism, that can presumably be directed upon representational subsystems and stages of same." The difference in emphasis is slight, but the effects multiply. Before you know it, you are apt to have a warehouse of Crick-style searchlights illuminating--for their operators?--various attention-grabbing scenes. We urge on him Bridgman's concluding point: "Consciousness is not a monitor of mental life, but a result of mental operations separated from the immediate sensorimotor world." )
In the same vein, our only quarrel with Baars and Fehling is with their assumption that they cannot be realists about consciousness unless they consider their Global Workspace to be a "system" of limited capacity (and every item is either inside or outside the system), instead of saying that the brain has a limited capacity to perform the feats described on the left of their Table 1, and that its performance of them can be seen as constituting the establishment of something rather like a Global Workspace (see also Bridgman, who expresses the same view, which is elaborated in Dennett, 1991b). Responding to Reingold's well-put request, we would view Baars and Fehling's Table 1 not as marking distinct qualitative marks of consciousness, but as perspicuously describing two poles (there are others) between which quantitative (but not easily quantified!) variation establishes the intuitive opposition between unconscious and conscious.
The difference between Velmans' "integrationist" model and our Multiple Drafts model is very succinctly stated: we agree with him that there are distributed processes of integration that tend, normally, in the fullness of time, to achieve a relatively integrated version of the world--the stream of consciousness that emerges in subjects' introspective and retrospective protocols; we just insist that there is no bridge across the stream! Velmans claims that on his view "within any given time window, information is integrated into a single conscious stream, and that unless there is evidence to the contrary, subjects' reports about what they experience are accurate." If one were to expand this time window of consciousness to include every instant between initial stimulation and ultimate report, one would rule out all pre-experiential and post-experiential editing: for instance, the subject would be held to be just as conscious of a raw retinal image as of any subsequent interpretations of it. If one tries to shrink the window down so that all adjustments of content can be located as pre-experiential or post-experiential, one must draw in the bridge across the stream. What counts as "evidence to the contrary" depends on where one draws these boundaries (consider the insanely radical Orwellian hypothesis that we are conscious of the dizzying swim of imagery on our moving retinas in spite of what we subsequently say--we just continually forget it). Velmans notes, correctly, that the Orwellian view is in danger of being unfalsifiable. Indeed, and the same danger faces its Stalinesque twin.
Farah provides a nice demonstration of a familiar sort of theoretical blindness: since you can (if you insist on it) redescribe any phenomena without mentioning consciousness, you can convince yourself that models of those phenomena that do mention consciousness tell us nothing about it. The target article is about consciousness because the phenomena in question are invisible to any research program that does not interpret subjects' protocols as accounts of conscious experience. If you treat the subjects' verbal behavior as so much emitted noise, the questions about subjective sequence, for instance, just cannot arise. Farah says that we "have not, so far, shed any light on the mechanisms of consciousness." What does she think these are, if not the very mechanisms of which we treat? If she means that there are additional mechanisms of consciousness, she has simply revealed her own unregenerate Cartesianism, materialist or dualist. If Farah is merely insisting that it is open to the theorist to claim that a subject who performs speech acts apparently descriptive of conscious experiences might be a zombie just unconsciously mouthing the words, we would agree, but we didn't think it necessary to direct our argument to those who still believe in zombies.
More explicit exploitation of the assumption that if it's real, it must be a system, can be found in the commentaries of the philosophers, Lloyd, Block, and van Gulick. It beguiles Lloyd when he misattributes view C to us, claiming that our theory is not a theory of consciousness at all because "some parts of the multiple parallel streams are conscious and some are not." Rather, we claim that there is no crisp way of telling exactly which parts of the multiple parallel streams are conscious. Any one of the streams sometimes contributes to awareness and sometimes not. No one stream is necessarily conscious by its very nature (see Kinsbourne in preparation for an extended discussion). And that is our theory of consciousness. It is a bad day for "realism" when only phenomena with hard-edged necessary and sufficient conditions can be considered real. More pointedly, when Lloyd goes on to declare that our "anti-realist" arguments "collapse in the face of the reality of conscious experience, as undeniable for us as it was for Descartes," he pledges his allegiance to one of the most persistent illusions of Cartesian materialism: how it seems to be fixes how the seeming must be. Consciousness seems to be all or nothing, but as undeniable as that fact is, it does not license the conclusion that consciousness is all or nothing, that either there is a box in which consciousness, and nothing but consciousness, exists, or consciousness does not exist at all.
Block proposes that the seeming is accomplished in a "Cartesian Module," the one functional place in the brain that provides us with "phenomenal consciousness." He claims that we mistakenly assume without argument that there no such thing as phenomenal consciousness. Well, what is it? Block supposes that we all just know: "our fundamental access to phenomenal consciousness derives from our acquaintance with it." Whatever phenomenal consciousness is, Block tells us it is not any of the sorts of consciousness captured by such functional properties as "inferential promiscuity" (essentially, the property represented on the left of Table 1 in Baars and Fehling) or accompanying second-order thoughts (see Rosenthal). Block sees an opening for what he thinks will be a non-arbitrary way of sorting brain events into the conscious and unconscious: "What if some of the brain representations of an event . . . are phenomenally conscious whereas others are not?" Indeed, what if? Shall we find out which these are by asking subjects? If so, all the arguments of the target article apply, for the quandaries arise precisely because of the problems encountered in interpreting the protocols or introspective claims of subjects. Does Block then hold, perhaps, that it is only possible to determine correlations of phenomenal consciousness with brain events by auto-cerebroscopy? Even here we will run into problems, according to Block's own account, for even the most careful and persistent autocerebroscopist is stuck relying on his own judgments of what seems to be happening, but "to assume that the subjective is exhausted by judgments is to beg the question against phenomenal consciousness, which, if it exists, is not just a matter of judgments." So a subject studying himself with an autocerebroscope (and his native introspective talents) will not be able to avoid begging the question against the existence of phenomenal consciousness.
It may be folly for us to attempt to say anything more about a putative phenomenon so sublimely inaccessible to investigation as phenomenal consciousness, but since both Block and Van Gulick venture to propose models of it, a comment is in order. The main attraction of Stalinesque theories, apparently, is that they agree with Descartes that the Show Must Go On, and it is worth noting that both models are Stalinesque. This is explicitly stated in the case of Van Gulick's GIPS model. It is less obvious in the case of Block's Cartesian Modularism, the view that "all conscious events occur in a single system," but it follows from the supposition that events occurring in a single system will be the effects of events occurring in other systems--the idea is that after some unconscious discriminatory machinery does its work, it sends some effect to the Consciousness Module so that its results can be properly bathed in phenomenal properties. We are in no position to assert that any model of "phenomenal" consciousness would have to be Stalinesque, but it looks like a good bet.
Consider Van Gulick's GIPS model. Like Block, he claims that his model "draws a clear distinction between phenomenal and nonphenomenal representation," but he similarly does not venture any account of what this distinction is. He claims that neurological evidence could be used to secure it, however, and in his account of metacontrast supposes that we could confirm or disconfirm his Stalinesque interpretation by relying on "what we independently [our emphasis] know to be the neural correlate of phenomenal representation of the relevant stimulus." But this cannot just be asserted--as if we could tell by looking at the neurological evidence that some of the neural events glowed in the dark. Our point is that when we turn to the only imaginable source of "independent" evidence for the distinction--the judgments of subjects, as expressed in their protocols (or our own judgments, if we use the autocerebroscope)--we find the Janus-faced evidence of Orwell/Stalin. Moreover, when we acknowledge, as Van Gulick insists, that such judgments cannot be viewed as infallible (he mistakenly supposes that our attack on the "strawman" of Cartesian materialism depends on this), we must admit the need for some Archimedean point that can impeach a particular judgment on neurological grounds, throwing us back on the requirement of motivating, independently, the identification of some neurological feature (some property K) as the mark of "phenomenal" representation. Can this be done? The target article claims that it cannot, drawing for support on a general argument that makes no specific mention of "phenomenal" consciousness. Block has given us no reason to think that phenomenal consciousness (whatever that is) would be more rather than less readily anchored to independent neural correlates, and as he notes, there are plenty of detailed arguments directed against the concept of phenomenal consciousness in Dennett, 1991b.
5. Libet and Treisman Tilt at Philosophers
Libet and Treisman seem to be under the impression that we are both philosophers, a compliment we accept in the spirit it was given, adding that we are not only philosophers, and regretting that this misapprehension on their part has led them to the most egregious misreadings of the target article.
Libet makes it clear that he is not a dualist. We are happy to have given him the opportunity to disengage himself more sharply than before from his dualist supporters. We also accept his clarification of what we had viewed as an equivocation between a radical and mild version of "backwards referral." We note, however, that in the very sentence in which he claims to have been clear about the distinction, he uses a dangerously equivocal phrase; he claims to have demonstrated "the subjective timing of sensory experience as appearing [our emphasis] before the time of the adequate neural representation for the experience." In the absence of his declaration, one might well be tempted to read this the way the dualists did: as the claim that a sensory experience actually appears (blooms in consciousness) before the time of the adequate neural representation, rather than what he apparently means: it only appears to appear in consciousness before that time.
Libet makes it clear at the outset that his primary motive for preferring a Stalinesque model is that it seems to him to explain "the actual experience of a single narrative"--something our view, in his opinion, cannot handle. (Aronson et al. and Van Gulick also seem to think that it is easier to explain the observed unity of consciousness by positing a single Stalinesque theater, but as Hurley notes, "the unity of the collection of vehicles in the theatre provides only the illusion of understanding of the unity of consciousness.") Often Libet simply assumes the Stalinesque interpretation of his experiments. When he does address the Orwellian alternative (which is not, as he thinks, our model), he misunderstands it. For instance, he thinks he can eliminate the Orwellian alternative to his Stalinesque account of metacontrast by claiming that Orwellian "intensification" of memory is more ad hoc than "obliteration", ignoring the fact that the Orwellian alternative was introduced by us as one in which memory "improvements" were inserted. And in his description of his study of somatosensory thalamus stimulation (Libet et al., 1991) he mixes the Stalinesque and Orwellian interpretations of his results without noticing: "Clearly, subjects had a memory of a brief stimulus, which they later correctly identified [in a forced choice guess], but no reportable [our emphasis] awareness." That is just what the Orwellian would say, insisting that this showed that subjects were (briefly) conscious of the stimulus in spite of the fact that it was not subsequently reportable.
Libet's response to the technical criticisms we mentioned is misplaced. We chose to direct attention to his interesting work in spite of the widespread opinion that they "involved very few subjects, were inadequately controlled, and have not been replicated." Referees and readers of early drafts of the target article had objected to our choice of his phenomena on these grounds (see also Treisman), so we felt obliged to acknowledge that opinion. In fact we share it, and are quite qualified, as is Churchland (see especially 1981b), to express criticisms about both his methods and interpretations. We do view the target article as supporting his conviction that there should be further attempts to replicate this difficult work.
Libet points out that there were three factors in addition to the one we cited for his estimate of the time of "neuronal adequacy" for consciousness, but the other three are more obviously question-begging in the context of fending off Orwellian alternative models. His discussion in 3.2 and 3.3 repeats his misunderstandings of the Orwellian alternative (and the idea that it is our alternative) mentioned above. For instance, he fails to notice that (1) the Taylor and McCloskey (1990) results are neutral between an Orwellian and Stalinesque treatment, (2) his claim that "the experimental evidence is precisely that the initiating neural events produce conscious intention but only after a delay of about 350msec" is itself question-begging, and (3) in his analysis of the Grey Walter experiment, he is not entitled to take the subjects' reports at face value as an indication of the order in which the crucial events actually happened in the brain.
Treisman misreads the target article, and then directs his lengthy discussion against a fantasy of his own construction. Most of his points are simply versions of points we make ourselves, or thought too obvious to need rehearsal, and his assertions about the implications of our view are almost all far wide of the mark. Perhaps we are to blame, but others did not misread it as radically as he has. For instance, he says there is no paradox; that was our point. The paradox arises only when one interprets the events through a mistaken view, the Cartesian Theater. He reads the questions we put in the mouth of a naive mechanist as questions we ourselves set out to answer, when in fact we set out to discredit those questions much as he does (we are not Orwellians, a fact that has escaped him).
He claims that we make two major errors. First, we confuse two different sense of "represent". He cites no instances, and in fact we use the term always in a sense which carries a connotation of content that is potentially usable by the system. Second, we suppose that "subjective sequence is "determined in consciousness". This idea that we "fail to distinguish between neural processing on the one hand, and hypothetical processes proceeding 'in consciousness' on the other" is a wild misreading, since it is precisely our point that there is no such arena in which anything could be determined by any hypothetical processes. Of course it is neural processing that determines subjective sequence. Our point is that once this has happened, there is no further phenomenon: there is no "running in proper sequence" or "projection" of the contents whose order has been fixed by those neural processes.
Once Treisman's views are sorted out, they emerge not, as he thinks, as an alternative to ours, but as a truncated or underspecified version of ours. For instance, he agrees with us that, thanks to the invariance of what he calls relation U, the subjective sequence need not mirror the objective sequence of events in the brain. And he also says that "We can safely conclude that the extent of the delay before the neural mechanisms reach a perceptual decision on an input may vary with the context in which the subject operates and the nature of the task." Of course. But he does not go on to address the question of whether this implies that consciousness of the relevant events is in all regards delayed until all such perceptual decisions have been made. Does he think, for instance, that there is no consciousness at all of the second tap in a cutaneous rabbit test until the brain has had a chance to arrive at the perceptual decision that the tap is displaced? Until he actually confronts such questions, we won't know whether he is in complete agreement with us, or is seduced by either an Orwellian or a Stalinesque vision.
Wasserman (see also Libet) points out that the problems of temporal smear--ranging up to several hundred milliseconds, are not due directly, as we suggested, to the slow speed of axonal transmission. It is rather the competition between larger patterns of neuronal activity that take time to be resolved--but the fact remains that the speed limit for such resolution is ultimately a function of the basic cycle times for individual neural elements. While we also agree with Wasserman that the ultimate basis for any neural integration must be the summation of graded potentials inside single cells, single neurons are too "stupid" to be responsible, on their own, for whole micro-takings, so we must advance to the level of larger patterns of activity to find appropriate machinery for accomplishing such judgments. We ventured no details on how to model these processes, but Gregson provides an elegant account of hypothesized dynamics of neural systems that predicts the existence in principle of temporal anomalies of consciousness, and his analysis may contribute to an explanation of how the multiple draft machinery works. In particular, it provides a response to Wasserman (who argues that integration in the last analysis must take place by summation, on the mini-theatrical stage of the individual neuron), by suggesting coupled neural dynamics as an integrative principle. Incidentally, his point that three or more coupled processes sometimes become chaotic appears neatly applicable to Norman's (1967) observation that when subjects identify rapidly successive sequences of visual or auditory stimuli, they often believe themselves to have experienced the last two stimuli in reverse order. Clearly, according to Gregson, such an occurrence does not necessitate belief in sequence concretely located in the brain.
Wasserman usefully elaborates the details of the Kolers and von Grünau experiment with color phi, but we would hold that he is mistaken in claiming that their choice of dependent variable, having the subjects do a matching task instead of giving an oral response, altered the situation in such a way as to avoid the Orwell/Stalin quandary. Given the instructions, when a subject leaves the knob in a particular orientation, this amounts to uttering a predefined speech act, a more refined and accurate speech act than would be verbally achievable, but still the product of judgmental processes that allow room for the quandary to arise.
Warren is in more agreement with us than he realizes. He slightly misreads our remark about how, as a matter of logic, certain discriminations must amount to temporal discriminations. We agree with him that "patterns are recognized first, and then properties of these patterns inferred"--when they are, which is seldom the case for some such properties. Consider the same point in a different domain. If subjects can distinguish between tachistoscopic presentations of the words "back" and "buck", this must be, as a matter of logic, a case of being able to distinguish two stimuli that differ only in one property: the presence or absence of the tiny bit of black that distinguishes the "a" from the "u" in the typeface used. But of course the subject doesn't make this discrimination by first looking for that bit of black and then drawing an inference about the identity of the word.
Block begs to differ from our interpretation of the quotations of Harnad, Mellor, and Churchland, and invites the reader to look again. We second the invitation. In his discussion of the Mellor quote, he says "the issue here is not one of judgments, but of experience of relations among experiences." We take Van Gulick to attempt the same defense of Mellor. This seems to mean that the experience of having the experience of A-as-happening-before-B must be the experience of having the experience of A before having the experience of B, but Mellor offers no argument for this (deeming it self-evident, we presume), and, as we argued in the target article, there are reasons for abandoning this traditional prejudice. Glymour et al. produce a passage that succinctly illustrates the fundamentally tempting confusion: "Suppose a temporal sequence composed of experiences E1, E2, E3, in that order. Either the content is experienced in that order too [our emphasis], or that temporal sequence is not experienced at all." To sustain their claim, the italicized phrase must be read to carry no implications at all about whether the content is experienced as in that order, but this is hardly a familiar reading. Suppose, for instance, the temporal sequence in question is a dream, and suppose it is actually dreamt backwards (E1 is the shocking conclusion, E2 is the midgame, E3 is the opening scene). Now if dreams are experiences, then either the content is experienced in the order conclusion-midgame-opening or it is not experienced at all, but it must not be supposed to follow that such a dream would be experienced as anomalous in any way (Dennett, 1976). If the dreamer found the whole experience anomalous, this would not be because of the order in which the content was experienced, but because, for one reason or another, the content was experienced as being in the anomalous order.
Shepard faults us for claiming that no one draft is more "correct than another." Here he overlooks our intended meaning, which was not, as he supposes, that no one draft is apt to be a more veridical account of what happened in the environment and in the control structure of the agent (we agree with his remarks about this), but simply that no subsequent introspective report can claim to be a veridical account of "what happened in consciousness"--as opposed to an account of what happened in the brain one way or another.
Aronson et al. offer figures that do not accurately represent the claims made in the target article. Their figure 1a shows Orwellian editing to be pre-conscious, and figure 1b apparently models view C, not the multiple drafts view (the same misreading found in Lloyd). We are unclear about how to interpret their preferred diagram 1c, but the lines from "local discriminations" to the "functional theater" makes it appear to be some version of the Cartesian Theater after all.
7. Supporting Arguments
We want to express our appreciation to Damasio, Hurley, McDermott, Reingold, Rosenthal and Young for commentaries that support our position with considerations drawn from very different quarters.
Damasio, Reingold and Young all confirm our conviction that, in spite of disclaimers, allegiance to the imagery of the Cartesian Theater is a widespread bad habit of thought, not a strawman as some commentaries suggest.
Damasio cites neuroanatomical grounds for resisting this temptation, which is a valuable antidote (though not, of course, a refutation) to Libet's claim that "the existence of such a single locus is not yet experimentally excluded." Damasio says, correctly, that a satisfactory model of consciousness should indicate how the dis-integrated fragments operate to produce the integrated self, and we have tried to respond to this demand in sections 1 and 2 above. We agree that somatic states play an ineliminable role in any robust integration of consciousness, but the issues raised by the specifics of such a proposal are beyond the scope of our discussion.
Damasio also alludes to Sherrington's idea, recently refurbished by von der Malsburg (1987) and others, that--one might say--timing is binding. We agree that there may well be something deeply right about this idea, but if so, two implications should be noted: (1) The timing that binds is not a photo-finish but a phasic relation between concurrent events whose neural consequences have a considerable temporal spread, and (2) insofar as simultaneity of phase behavior plays the role of pure binding, it necessarily becomes unavailable to play the role of a basis for a judgment of simultaneity. (For instance, the successive notes of a melody are bound, but certainly are not represented as occurring simultaneously, nor does the use of delay lines to create a simultaneous representation of the successive notes appear to be a workable mechanism for binding. For an analysis of some of the relevant issues, see Port, 1990, and Port and Anderson 1990)
Young uses the McGurk effect to support the Multiple Drafts model, by showing how unreliable introspection is as a source of knowledge about how what seems to happen actually happens.
Rosenthal and Hurley both note that a particularly tempting traditional Cartesian assumption is that whatever consciousness is, it is an intrinsic property of those mental states that are conscious. Both then show, by somewhat different routes, how this assumption suffices to make the Cartesian theater seem inevitable. Giving up consciousness as an intrinsic property, Rosenthal recommends that we adopt the term transitive consciousness for what we have called property K: the property the acquisition of which makes a mental state conscious. At first this appears to open up what might be called a forced choice for the theorist: Orwellian or Stalinesque? But he shows that any answer will be, as we claim, arbitrary. Hurley arrives at the same verdict. These commentaries provide valuable support for our position, by showing how its most "radical" conclusions are consistent with a traditional metaphysical position that relinquishes only one dubious feature of the Cartesian heritage: the idea of consciousness as an intrinsic property.
Hurley also anticipates and rebuts the objection that our view is "verificationist," and in particular notes that "To object that the 'observer' to whom manifestation is being required is oneself at a slightly later time is to assume the temporal atomism about consciousness that is [our] target."
McDermott speculates about the underlying moral anxieties that may well provide a hidden agenda in support of Cartesian materialism. As he says, in a forthright rejection of this theme, "if people are valuable, it is not because they are imperishable souls connected to bodies only for a brief sojourn." There is more on this important topic in Dennett 1991b.
Dennett, D. C., 1976, "Are Dreams Experiences?" Phil.Review, pp.151-71.
Dennett, D., C., 1991a, "Real Patterns" J.Phil., 87, pp. 27-51.
Dennett, D. C., 1991b, Consciousness Explained, Boston: Little Brown
Kinsbourne, M., 1985, "Parallel processing explains modular informational encapsulation," Behavioral and Brain Sciences 8. p.23.
Kinsbourne, M., 1988, "Integrated Field Theory of Consciousness," in A. J. Marcel and E. Bisiach, eds. The Concept of Consciousness in Contemporary Science, Oxford: Oxford Univ. Press.
Kinsbourne, M., "The Distributed Brain Basis of Consciousness."
Neisser, U., 1976, Cognition and Reality, San Francisco: Freeman
Norman, D. A., 1967, "Temporal Confusions and Limited Capacity Processors," Acta Psychologica, 27, pp.293-297.
Port, R., 1990, "Representation and Recognition of Temporal Patterns," Connection Science, 2, pp.151-76.
Port, R., and Anderson, S., 1990, "Dynamic Network Models for Audition," Indiana University Institute for the Study of Human Capabilities, Technical Report ISHC-TR01-RP-90.
Sherrington, C. S., 1934, The Brain and its Mechanism, London