Judgments and Drafts Eight Years Later

Andrew Brook

Philosophy and the Cognitive Science Programme

Carleton University

Ottawa, ON, Canada

K1S 5B6

Isolating the task

In Consciousness Explained (1991a; hereafter CE),(1) Dennett's most extended treatment of consciousness to date, he lays out an interpretationist approach to what consciousness seems to be like (heterophenomenology), mounts a series of challenges to the idea that some basic questions about conscious experience have determinate answers, sketches what he calls a Multiple Drafts model of what consciousness is actually like, and urges that conscious experiences are judgments, not (what has been understood by terms like) sensible images. In general, he has two main targets: states or events in which something seems to us to be a certain way - Seemings in his parlance (i.e., what others call qualia or appearances) - and what is seemed to, the Subject (subjects are also meaners and agents).

Now that some years have passed, how does this picture of consciousness look? On the one hand, Dennett's work has vastly expanded the range of options for thinking about conscious experiences and conscious subjects. On the other hand, I suspect that the implications of his picture have been oversold (perhaps more by others than by Dennett himself). The rhetoric of CE is radical in places but I do not sure that the actual implications for commonsense views of Seemings and Subjects are nearly as radical.

For thirty years now. Dennett has worked to give an account of content and of consciousness. Dennett's characteristic modus operandi for dealing with content (intentionality) is as follows (1978f, 1987d). First he argues that a target psychological or behavioral phenomenon can be accounted for in terms of the subjects' reasons via an idealization of the subject's instrumental rationality. This mode of explanation goes with a certain metaphysics of mind. The patterns it reveals are real but they are patterns discernible in behaviour, not in the head (1991b, p. 98). The phenomenon in question then comes out as another such pattern. Adopting this technique is adopting the intentional stance. Dennett mops up any residual resistance with some vigorous, often verificationist counter-arguments. His well-known hostility to qualia (CE, Ch. 12) and to even the bare possibility of zombies (Polger, this volume) are two of the best known of these mopping up operations. ('If it looks like a duck and quacks like a duck, then most probably it is a duck' is the basic move.)

Now, consciousness is a psychological phenomenon. Yet the intentional stance plays little role in CE. Dennett so much as mentions it only about four times and his longest discussion of it is less than one page (pp. 76-7). The first two-thirds of the book are written in an unabashedly realist idiom - and here I mean the old-fashioned realism of conscious experiences as states in the head, not the special realism of 'Real Patterns' (1991b).

So how does the metaphysics of the intentional stance relate to the multiple drafts model of consciousness, etc.? Here is Dennett a few years after CE:

Conscious experiences are real events occurring in the real time and space of the brain [1994a, p.135].


Sensory qualities are nothing other than the dispositional properties of cerebral states to produce certain further effects in the very observers whose states they are [1994b, p. 146].

He has consistently denied this about psychological states such as beliefs and desires. If they are to be mapped onto anything real, it is to real patterns in behaviour, not real events in the brain. Recall the Jacques, Sherlock, Tom, Boris thought-experiment. The four men all come to believe that Jacques shot his uncle dead in Trafalgar Square but they acquire this belief in utterly different ways and even via different languages - Jacques by doing it, Sherlock by investigating the deed, Tom by reading about it in the Guardian, and Boris by reading about it in Pravda. Says Dennett,

If [theorists] insist [that there is nevertheless] a similarly structured object in each head, this is a gratuitous bit of misplaced concreteness, a regrettable lapse in ideology [1987b, p. 55].

How do these two positions hang together?

Well, the intentional stance may not play much of a role in CE but a closely related notion does: heterophenomenology. The intentional stance is a device for arriving at a view of what is moving a subject. Heterophenomenology is a device for arriving at a view, from the point of view of an outsider, of how things seem to that subject. (This used to be called empathy before the term was bent out of shape by ramming it together with sympathy. Thus Dennett, one of the most graceful of writers, had to resort to a horrendous neologism.) It is part of the metaphysics of the intentional stance that there need not be (and almost certainly are not) any states in the subject's brain that are the beliefs, desires, etc., ascribed to her. In apparently the same way, it seems to be part of the metaphysics of what we might call the heterophenomenological stance that there need not be any states in the subject's brain that are or contain the way things seem to her. At any rate, this much is true. Just as the intentional stance is a device for ascribing a pattern of beliefs and motives that fits not just with brain states but also with behaviour, history and environment, heterophenomenology is a device for ascribing a view of how self and world seem to the subject that fits not just with brain states (and information they may contain) but also with the subject's behaviour, history and environment,

So Dennett has a two-part model:

1. A realist part concerning the events and dispositions that are what is really in play when people are conscious, things seem a certain way to them, etc.

2. An interpretationist part concerning how things seem to people.

Can these two parts be reconciled? One wonders. How, for example, can conscious experiences come out as real events, sensory qualities real dispositions, while how things seem to someone might be a mere fiction, no more real than a novel (p. 98)? Surely 'sensory qualities' is simply a name for how something seems to someone. Can sensory qualities both be and not be real events or dispositions in the brain? We need an interpretation that avoids this outcome.

There is a risk of ambiguity in Dennett's treatment of the heterophenomenological stance. What is he denying? Merely that we are as we seem to ourselves? Few will quarrel with that, though his wholesale jettisoning of the categories of folk psychology goes further than we most people (p. 319). (Most of us think that when we seem to be aware of such basic things as beliefs and desires in ourselves, here at least we are probably right.) But Dennett seems to want to go farther. He seems to want to say that when we seem a certain way to ourselves, there need not be any events or pattern of events in us that contains or constitutes that way of seeming (p. 98). (How much further he wants to go in this direction will occupy a good part of this paper.)

The difference here is the difference between a Seeming being accurate and it having a vehicle. If we think again of the novel analogy, it would be the difference between the novel's story being true and the novel being a real object. This is a substantial difference! It may contain a way out of the prima facie contradiction we generated above. More generally, I want to explore exactly what metaphysics of mind the analysis of CE supports. In addition, I want to ask whether CE contains new any arguments for this metaphysics, whatever exactly it is, arguments different from those used in earlier works to defend the metaphysics of the intentional stance.

CE has three parts. Part I is introductory and Part III is mostly aimed at the special preoccupations of philosophers. In Part II Dennett mounts his challenges, aiming to show that conscious experiences are less determinable than we might be inclined to think, and sketches the Multiple Drafts model. Then in the first two chapters of Part III he gives his theory of conscious experiences as judgments. So Part II and the first chapters of Part III are the crucial sections for our purposes.

Traditional notions and the trilevel approach

It will help frame the question of what metaphysics of consciousness is supported by CE and how if we put the issue into a broader context. For the few hundred years that people have had a stable notion of consciousness at all,(2) folk psychology has ascribed a fairly constant group of features to Seemings and to Subjects (not necessarily under these names, of course). Clearly, the metaphysics of the intentional stance rejects some aspects of these notions. The holism and externalism of the intentional stance entail not only that 'meanings just ain't in the head', in Putnam's memorable (1975) phrase, but that content of any sort 'just ain't in the head'. All that's in the head is "syntax" (in the wildly inflated sense of the term 'syntax' used in contemporary philosophy of mind) (1987b). They also entail that indeterminacies are a central feature of the mental. If so, there is nothing that has some of the central features traditionally ascribed to Seemings and Subjects.

This is a kind of eliminativism. If the intentional stance is eliminativist about Seemings and Subjects, what about CE? Is CE as eliminativist as the intentional stance? More so? Put the question a bit differently. To what extent does the book argue that Seemings and Subjects as traditionally conceived do not exist and to what extent does it just want a better model of them? (Modeling something does not challenge its existence.) And whatever the answer to these questions, what does CE offer by way of new arguments for its position?

Let's try to make the eliminativist issue more precise. Start with that mainstay of cognitive science, the triple of task, procedure and implementation.(3) Here is an example. Add two numbers. The task is to add the numbers correctly. The procedure is whatever algorithm is used to perform this task. And the implementation is whatever arrangements in a physical system the running of the algorithm happens to consist in on this occasion.

The three notions go with three, nested kinds of description. A single task will usually be an assembly of dozens or even hundreds of units of procedure, units that may vary widely from instance to instance of the same task, and a single procedure may well be an assembly of dozens or even hundreds of implementation units, units that also may vary widely from instance to instance of the same procedure (Marr, 1982; Dawson,1998).

Consciousness is more than a repertoire of tasks. This is one of the things that makes it cognitive science-resistant. So we need to broaden the first notion. The following loose concept will do for our purposes: Features by which we identify something. Let us call this the F-level (for identifying features).(4) More is almost certainly involved in the middle level of consciousness than algorithmic procedures. Processes of other kinds are also probably involved. So let us call the second level the P-level (for processes of all kinds, not just procedures). Call the third level the I-level (for implementation). We won't have much to say about it.

Now, when Dennett goes after the picture of Seemings and the Subject of the tradition, is he going after the F-level picture, certain P-level theories developed in the tradition to account for these F-level features, or both? If he denies the existence of something central to either F-level picture, then he is an eliminativist about at least that aspect of consciousness. If he rejects only traditional P-level theories of what processes and procedures subserve the F-level features, then he is not an eliminativist, he just wants a better theory. Which is it?

Here is a sketch of the F-level features of Seemings found in folk psychology:

F-level concept of Seemings: Seemings are real states or events or some sort in us in which something appears to be some way, something is like something for us.

In the tradition, most theories of the P-level processes and procedures that realize these F-level features are like would contain the following elements:

P-level theory of Seemings: Seemings are reidentifiable events in the head with clear start and stop points. These events that can be identified and reidentified independently of other similar events around them. They present scenarios to something called "the mind"

and so on (the list is not meant to be exhaustive). As we said, if Dennett rejects significant aspects of the F-level concept, he is an eliminativist about Seemings. If he rejects only the P-level account, he just wants a better P-level theory. Which is it?

Dennett clearly rejects the P-level theory and any theory like it. When he says that we don't have qualia in any way different from what a CADBLIND system could have (p. 375), that "I'm denying [that Descartes' real seemings] exist." (p. 363), he is at least doing that much. However, the fierce rhetoric of CE suggests more than that. The rhetoric suggests that Dennett wants to deny that any states or events have even the F-level properties of a Seeming, and many people have taken him just this way. Here are some of the things that Dennett says:

[The category of] the way things actually, objectively seem to you [is a] bizarre category [p. 132].

There is no such phenomenon as really seeming - over and above the phenomenon of judging in one way or another that something is the case [p. 364].

There is no such thing [as] actual phenomenology [p. 365].

Yet it is actually unclear what these statements amount to. The first statement occurs in an attack on an aspect of a P-level theory, namely, Cartesianism (as he sees it), and the second and third in the context of mounting his own P-level alternative. Dennett may not be going after the F-level picture in any of them. OK, so what about things like this?

[A subject's heterophenomenological world is a] theorist's fiction [p. 98].

Heterophenomenology exists ... as novels and other fictions exist [p. 98].

These do look strongly eliminativist about Seemings, as does the later claim that Seemings are, like centers of gravity, mere abstracta, postulates useful in some explanatory exercise but not corresponding to any material state or process (pp. 95-6). We will try to sort out exactly what kind of eliminativism and how Dennett argues for it (whatever "it" turns out to be) in the next two Sections.

Dennett knows full well that we certainly seem to have Seemings, of course. Indeed, he insists on the point. But this is just autophenomenology, heterophenomenological interpretation applied to self. How things seem to someone may be totally different from how they are. Put a bit paradoxically, the fact that we seem to have Seemings does not entail that we do! (We will pass over the appearance of self-refutation in this sentence, at least for a while.)

About the Subject, the issue of eliminativism is less complicated. The F-level story of the Subject in folk psychology is built around something like list (1):

List 1: F-level concept of the Subject. A Subject

� operates by forming things like beliefs and desires about objects, events, etc.

� marshals various cognitive resources (beliefs, desires, values, concepts, etc.) to generate an intention to do something and then marshals diverse bodily resources to do it

� is aware of a number of things at once in some way that allows us to relate things to other things

� consciously focuses attention on tasks

and so on (again the list is not meant to be exhaustive). List (2) spells out the P-level theory developed by the tradition to explain what is going on at the F-level. It is Dennett's main stalking horse.

List 2: Dennett's account of the traditional P-level theory of the Subject

A system able to do the things on list (1) would have:

� a Cartesian theater

� a Central Meaner, the shared contents of which are highly integrated and 'enter consciousness' as though the mind has a synchronous clock speed

� a class of mental events called, variously, appearances, qualia, and the felt quality of things. These events are viewed as in their own traditional P-level theory sketched above; for example, as having clear start and stop points.

The list of features in list (2) provide a theory or model of the items in list (1).

Dennett is clearly an eliminativist about the items on list (2) and any other like it. But that is merely to reject certain theories of what is needed to realize the features of Subjecthood. It is not to eliminate our F-level concept of the Subject. To be an eliminativist about Subjects as traditionally conceived, one would have to go after the items on list (1), not just list (2). If Dennett accept that Subjects have at least most of the F-level features ascribed to them in our folk psychology and just denies that it takes anything like the items on list (2) to realize these features, he is not an eliminativist about Subjects. He just wants a better theory of them.(5)

Are indeterminacies of consciousness a problem for Seemings?

Seemings first. Dennett's treatment of Seemings has three main parts, all of which were mentioned at the beginning of this paper: the heterophenomenological stance, the series of challenges, and the theory of conscious experiences as judgments. Let's start with the challenges.

Judging by the intentional stance, eliminativism in Dennett about the F-level notion of Seemings could take two forms and it is important to separate them. It could take the form of a claim that there are no states or events in which something seeming like something, or it could take the form of a claim that, though there are states or events that have that property, they do not have some other crucial feature of Seemings of the F-level picture. Here is how the difference plays itself out in Dennett's writings on intentional content (see Viger, this volume). Dennett's more instrumentalist writings take the first route - intentional explanation is merely a device for connecting behaviour to various inputs and the entities it postulates to explain these connections have no more reality than that. Intentional entities are abstracta. More recent writings take the second route - intentional phenomena are real, but they are not in the brain as the traditional F-level picture of them would have it. They are real patterns discernible in behaviour (1991b).

Dennett does not take the first route with Seemings, not entirely at any rate. Something seeming like something has to involve conscious experience in some way, and this closes off the second route, at least partially. As we saw, he says that conscious experiences are "real events ... in the brain". So if he is an F-level eliminativist about Seemings at all, he has to take the second route: insist that these "real events in the brain" lack some central F-level feature ascribed to Seemings by the tradition. They cannot lack location in the brain in the way that intentional patterns are said to do, so what they do lack?

The answer to that question is complicated and far from straightforward. Let's start with a simpler one: Do any of Dennett's challenges to the determinateness of conscious states challenge the traditional F-level picture of Seemings? Here is a summary of the main challenges:

1. There is often no clear sense to be attached to the idea of a specific point at which something becomes conscious (Stalinesque/Orwellian impasses).

2. The time order that contents of consciousness appear to have is often different from the time order in which we encountered the events originally.

3. Often there is no way to distinguish between an organism misrepresenting one thing and correctly representing another thing.(6)

4. We can carve mental events up in ever finer slices, describe them in boundlessly varied descriptions, and embed descriptions of them in higher- and higher-order attitudes without limit.

5. No sense can be made of the notion of an inverted spectrum.

All these claims have considerable interest in their own right. Do they undermine the F-level notion of Seemings? Seemings, recall, are not just something seeming like something - Dennett clearly accepts that that occurs. They are states or events of something seeming like something. It is not clear that that Dennett's challenges threaten such states or events. My reasons for saying this vary from claim to claim, so we need to examine them individually.

1. Stalinesque/Orwellian impasses (S/O).

An S/O impasse is generated when something is misremembered but there is no way to determine whether it was misperceived (the Stalinesque option: history is falsified from the beginning) or perceived correctly but immediately misremembered (the Orwellian option: history is initially recorded correctly but falsified afterwards). An example: The Lady Remembered with Glasses.

A short time after seeing a lady who is not wearing glasses, we remember seeing her with glasses. And the question is, where did the mistake occur? It seems that there should be two possibilities: the lady in question might have been misperceived with glasses (the Stalinesque option) or she might have been perceived correctly but, due to a later revision, immediately misremembered (the Orwellian option). Dennett argues that no possible information could adjudicate between these options, so they are not really options and there is no fact of the matter about when the error began (p. 125).

Why might S/O impasses be a problem for the existence of Seemings? Well, if such elementary impasses as these cannot be resolved, then it would seem that there is nothing that meets the most basic requirements for being a real Seemings: there being a clear point at which one ends and the next one begins, a way of determining when we are encountering the same one again, when we are encountering another one of the same kind, and so on.

Are all S/O impasses merely apparent, a distinction that corresponds to no difference? The answer is not obviously yes. Consider pain and pain tolerance (1978b, pp. 222-4). A person has suffered some mild bang, yet complains he cannot work, the pain is so bad. Does this person feel unusually intense pain for such a mild injury or does he have the usual modest pain but a very low threshold at which pain becomes seriously debilitating?(7) On the other hand, we see a Freud soldiering on even though cancer of the jaw has rotted his whole cheek away and left a great gaping hole. Does he have severe pain but astonishing pain tolerance or does he just happen to feel less pain from this lesion than others would feel? These examples generate S/O impasses but the difference between levels of pain and levels of pain tolerance seems to be perfectly real, certainly in many cases. (Pain creates problems for Dennett in other ways, too, as we will see.)

In fact, it is not so clear that S/O impasses are irresolvable in principle. One option is the "outside in" approach. Start from cases where everything is clear: a foot is mashed to a pulp and the person is screaming; a woman was perceived and remembered correctly for a time but then an error crept in. Next correlate the conscious contents with whatever one can find, perhaps even brain-states (though Dennett thinks it unlikely that we will ever do that). Then apply the correlates to the cases where the short time-scale, etc., generates an impasse. Dennett himself uses this strategy to great effect in 1978b and touches on it in 1988 and a number of times in CE (pp. 85, 96, 326 fn. and 396). His reasons for viewing it as inevitably inconclusive in certain key cases are not clear.(8)

But suppose that some S/O impasses are genuinely irresolvable? Would Seemings then be in trouble? I think not. The process of generating conscious experiences may contain more indeterminacies than we imagined but we still get to a stable state at the end: a clearly misremembered lady, a pain of a particular intensity, and so on. Maybe getting there is less than half the fun but we still get there. If so, there is no problem here for the existence of Seemings.

2. Real and apparent time-order.

An example: the phi phenomenon. The phi phenomenon is very simple. A red dot is flashed on a screen. A short time later and an appropriate number of degrees of arc away, a green dot is flashed on the screen. Everyone sees the red dot moving and changing into the green dot - and everyone sees the one changing into the other before the green dot appears. This, of course, is impossible - the brain must be reconstructing the apparent time order after the actual sequence of events in their actual time order is completed.(9)

Does the phi phenomenon pose a problem for Seemings? No, only for certain theories of Seemings. There is a perfectly good state of how things seem to be unfolding - i.e., a perfectly good Seeming. We can even date it - the latter half of the state giving us how things seem (the part from the point at which the dot begins to appear to change color) cannot begin till after the green dot has actually been perceived.

3. Misrepresentation

The classic example of misrepresentation is the frog flicking out its tongue at an object that is in fact a black BB thrown across its visual field. Does it misperceive the BB as a fly, or correctly perceive it as a black object?

The putative problem for the existence of Seemings is that if there is no way to answer this question, it is natural to conclude that there is no fact of the matter about what the frog sees. I think that this conclusion would be a mistake, that there is no problem for the existence of Seemings here. However, the story is trickier than it was in the cases above.

First we need to distinguish between the state of something appearing and what that thing appears to be like. In the case of the frog and most or all other cases like it, there is a quite determinate state of something appearing. It simply cannot be resolved what the thing appears to be like - not by us and, we can perfectly well allow, not by the organism either. What makes this possible? To be aware not just of something appearing but of what the thing appears to be like, we need not one thing, information about something, but two - information about something plus application of apparatus for determining what the thing appears to be like. Whatever the truth of Dennett's claim that all Seemings are judgings, perception clearly requires judgings. We must both be presented with something and characterize it in some way (one of Kant's most basic claims: 1781, A51=B74).(10) Where characterization is needed, one can have a determinate presentation while our view of what precisely is being presented is undetermined (a point that will become significant later). If so, misrepresentation is no threat to the existence of Seemings.

It may appear that I am begging the question against Dennett's theory that Seemings simply are judgings. I don't think I am. On Dennett's theory, the distinction I am after would come out as something like this: the judgment that something is appearing is different from judgments about what the thing appears to be like. The first judgment yields a perfectly determinate Seeming even if determinate judgments of the second kind cannot be made.

4. Boundless descriptions.

Dennett holds that an intentional object can be embedded in propositional attitudes in endless ways, carved up into ever finer representational states, and so on. This phenomenon gives rise to two different problems. First, it creates a risk of an "explosion of distinct 'representational states'" (p. 316). Second, since one of the places where it shows up is in memory, it is a new venue for S/O impasses. We have dealt with S/O impasses; what about the representational explosion? Well, let's suppose that there is some indeterminateness in the range and extent of the attitudes we can take to some represented content, that here things seem to be more determinate in folk psychology than they are. Would that show that the content itself could not be a state or event in which something seems like something? No.

5. Inverted spectra.

Supposed example: where you see red, I could see green, even though all my speech and behavioral dispositions and therefore all my behaviour, including all my statements about color, are the same as yours. I entirely agree with Dennett that no good sense can be made of the idea of an inverted spectrum or of inverted qualia, absent qualia, or any similar produce of credulous faith in the power of the thought-experiments to reveal genuine possibilities. Dennett suggests that inverted qualia, etc., require real Seemings; no real Seemings, no possibility of inverted qualia, etc. (p. 393) and again I am happy to follow him. But how about the other way around? Would the incoherence of inverted qualia, etc., rule out real Seemings? I am not sure whether Dennett wants to go from the incoherence of the idea of inverted qualia, etc., to the impossibility of real Seemings (see Ch. 12, especially Sections 4 and 5). For my part, I do not see how such an inference could go through. How would the impossibility of inverted qualia, etc., rule Seemings out? That colors as they appear to us cannot be inverted, etc., would seem to be entirely neutral with respect to whether those appearances are real events and if so what kind of event.(11)

The five challenges to traditional ideas about the determinateness of various aspects of conscious experience that we have just examined certainly open up new ways of thinking about conscious experience. They just happen not to give reasons to reject the traditional F-level notion of a Seeming. Dennett has another argument against Seemings. What is really going on when people seem to having Seemings, he says, is - judgings! "There is no such phenomenon as really seeming - over and above the phenomenon of judging ... ." (p. 364). Let's turn to that argument.

A theory

Here are some claims that illustrate the central idea behind the theory. When we seem to be surveying an already-existing "field", what we are in fact doing is judging that the field is filled in (1996). Often when we seem to be reporting something "presented" to us, what we are actually doing is making a judgment about the thing. Indeed, often it is hard to apply the perception/inference distinction to experience at all. This rejection of presentations in favor of judgments goes with another idea. Often when we think our "mind" is showing us something, it is in fact telling us about it - the thing is not appearing in an image or anything image-like, we are describing it in some code.

The famous Warhol room of Marilyn Monroe posters is an example. All the Marilyn posters in the room seem to be alike. We think that we are perceiving this. However, a poster off to the left actually has Mickey Mouse ears and a Hitler mustache. We do not notice this. The reason is that we can perceive items clearly only through 2o of arc or so. We are actually judging that all the posters are alike. We do not find out what is in the rest of our visual field, we fill it in. Filling in is a process of judgment, not anything image-like.

What can be said for this theory and the inference that Dennett draws from it? The theory is empirical and stands or falls on the evidence, which I cannot summarize here. It is probably at least partly right. On the other hand, even in the Marilyn Monroe room there is still a lot of real seeing going on - thanks to rapid eye movement, much, much more than our 60 of focal vision would give us.(12) In at least one case, moreover, the theory seems plain wrong, the case of pain (of which more later). And it does give rise to a question about Dennett's real target.

By 'Seemings', I have so far meant simply states or events in which something seems like something. All sorts of mental states are Seemings: judgments, thoughts, and desires as much as perceptions and sensations. Even abstract concepts and ideas appear to us in a certain way.(13) This broad view is Kant's view and it is also Dennett's official view: seeming is simply "the way things look to us"' (p. 373; see also pp. 58, 319, and others). But his examples often take a narrower view, one more like the notion of Seemings in the Empiricist tradition. Here Seemings are thought of in terms of images, i.e., states fully formed prior to cognitive processing. Think for example of the importance he attaches to getting representation in images out of his CADBlind system. Think too of Dennett's general iconophobia, as he called it, throughout the 70's (1978a, 1978e, for example). The answer, I think, is that Dennett is after both: both the idea that mental images are central to Seemings and something about Seemings as such. The distinction, however, is important for the scope of the theory.

Now suppose not just that Seemings are not images (as I and many others would grant ) but that Dennett's whole theory is right: Seemings are simply judgings. Would that entail that Seemings in the F-level sense do not exist? Certainly not. Judgings are themselves Seemings! When I judge, both what I am judging and the act of making the judgment appear to me, are like something for me. More than once Dennett says that "there is nothing more to phenomenology" than judgment (p. 366; see also 364). This is a peculiar thing to say. Judgings themselves have phenomenology - full, rich, robust folk psychological phenomenology. There is no argument against the existence of Seemings here!

So what is going on? It seems to me that the theory of phenomenology as judgings is aimed at a certain theory of what Seemings are like: passive presentations of already stable and well formed material about how something seems. When Dennett says that there is no more to phenomenology than judgment, what he means is that all states having phenomenology are judgments. That is to say, he is offering a P-level theory of what Seemings consist in. I don't know if this has been noticed before but Dennett's pronouncements against Seemings in this part of the book all have a theory of Seemings as their target, the theory that Dennett calls Cartesian materialism, not the idea that Seemings are real states of some kind. Recall the claim that the category of the objectively subjective is bizarre. This claim is aimed at the notion that there are states that carry a certain phenomenology around with them "intrinsically". (This has to be false. How something seems in some representation is a relational matter affected by, e.g., how the state or event is taken by the cognitive system in which it occurs.) Equally, when Dennett urges that there are not two stages, a presentation stage and a stage of judging the presentation, but only one, he is also after a theory of Seemings, not the idea that there is such a thing.

To sum up: in the theory of Seemings as judgings, Dennett is offering a theory of what the real P-level processes realizing Seemings are like. This theory might support other aspects of the metaphysics of the intentional stance, the idea that individual Seeming states cannot be identified and reidentified independently of other such states, for example, but it provides no independent support for the idea that there are no real Seemings.

We can imagine how this new theory of Seemings as judgings might begin to work itself out in connection with perception. We could investigate how information about an object is coded in the visual cortex and, from the structure of this coding, make precise inferences about how the perceiver will judge the object to appear. 'From the fact that this edge is coded as covered, this corner is coded as having a 600 angle, the organism will seem to be seeing the object from underneath and in front.' Or whatever. This analysis could be very precise. What we would have been doing is exploring the specific states and events in the brain that a certain Seeming is based on.(14) Dennett would accept all this without any qualms. When he urges that "there is nothing more to phenomenology" than judgings, among the things that judgings consist in are "various events of content-fixation" (pp. 366 and 365), that is to say, events exactly like the ones we just sketched for perception.

By now unease will be growing. "While you have been pulling this realist story out of the hat," I can imagine someone objecting, " haven't you rather ignored the 'theorist's fictions' of heterophenomenology? Surely in the latter at least Dennett is flatly denying that Seemings exist. In your introductory section, you yourself said that the doctrine of heterophenomenology is where the hard objections to the reality of Seemings are found." Many people have read Dennett this way (including me) but it is at least a partial misreading (as Dennett has prodded me into seeing). When Dennett talks about a theorist's fiction, he is talking about the content of a Seeming - the way something seems to someone. This could be a fiction, even if the person is seeming a certain way to him- or herself. But this is not to deny that there are vehicles of Seemings. There are indeed vehicles of Seemings: judgings.

He does not say straight out that he accepts that there are vehicles for Seemings in CE but he is a realist about processes of content-fixation,

Conscious experiences are real events occurring in the real time and space of the brain [1994a, p.135],

and the only candidate for content-fixers in his theory is judgments. (Dennett develops the theory of the whole middle part of CE in an entirely realist vocabulary.) He clearly accepted the reality of vehicles in (1978a), the paper for which he invented the story of Feenoman and the Feenomanologists so central to CE, and he has expressed the same view again recently (private communication). Folk psychology may well over-simplify Seemings, as Dennett says (p. 320), but it is not wrong when it takes them at least to exist.

So what is Dennett after when he says that the contents of Seemings are abstracta, that the heterophenomenological world is a theorist's fiction (pp. 95, 98)? One thing is the truth of the contents of a Seeming (as opposed to the existence of a vehicle of Seemings). In the same way that the world need not be as it appears to us, we need not be as we appear to ourselves, not even when we appear to have basic things like Seemings and beliefs and desires (p. 319). Another is that in heterophenomenology, we devise sense-making interpretations; we do not "read off" preexisting patterns. Dennett does not say a lot about how this aspect of heterophenomenology relates to the reality of Seemings (p.319 illustrates the gap) but here is how the story would go if told under the constraints of the intentional stance.

When a bit of content is fixed in brain by a micro-judgment, that gives us some organized information to use. However, it is a long way from this information to a fully developed Seeming.(15) History, environment, knowledge of the subject's psychology, how the subject is using the content, causal connections running from the content-bearing state to the world, etc., etc., all enter into both the subjects's first-person and our third-person judgments about how things seem based on this information. How these factors will seem when bound together is not "built into" them; it is how we judge them to be.(16) If so, states or events of things seeming like something are very unlikely to correspond to an aspect of judging events or events of any other kind in that person's brain, in exactly the same way that there is unlikely to be any "similarly structured" events underlying the shared believe of Tom and Sherlock and Jacques and Boris. Even a token-wise mapping is unlikely.

None of this is to deny the existence of real vehicles for Seemings; the judgment about how things seem is the vehicle. It is to argue two things. First and obviously, the business of settling how something seems is vastly more complex than dreamt of in Descartes' philosophy. (Kant came closer to getting it right, or so I have argued [Brook, 1994]). Second and less obviously, the "how things seem" aspect of the judgment has a special reality-status. (This is often called the 'content' of the judgment but this very term begs the question in favor of a stronger brand of realism than Dennett would accept. Though he uses the term himself, he tends to restrict his use of to the information "locked in" by micro judgment. There it is OK.)

We can now fit the realism about context-fixation and the interpretationism about hetero- (and auto-) phenomenology together. Put in the jargon of "syntax" in the inflated sense mentioned earlier, what Dennett calls the 'conscious experiences' in the passage quoted above would have to be the syntax, the physical structure on which a Seeming is based. It might be the acts of content-fixing judgments, the content thus fixed, or both. The "how things seem" of a Seeming would then be free to float off into the world of 'Real Patterns' or something even less brain-state real than that. Surely, it will be objected, this has to be wrong. Surely when a subject judges that so-and-so is such-and-such, the "so-and-so is such-and-such" part has to be something real, some proposition-like structure just as real as the judgment that arrived at it. But that is not so. Here's the question to ask: How do we - how does anybody - fix this aspect of judging, generate something determinate to ascribe to the judgment?

It can happen in two ways: in autophenomenology and in heterophenomenology. In autophenomenology, it has already been done: when the subject arrives at a judgment about how something is (that is, how it seems to her), she has also arrived at a judgment about what that "seeming to her" is like, namely, how that thing seems to her.(17) This is an interpretation of how things seem to her. This interpretation will be constrained, to be sure, in various ways (by information available - not to her but - in the system that is her), history, etc., etc., but it is not clear what it would mean to say that there is something more than the interpretation, something outside the interpretation for it to correspond or fail to correspond to.

Well, what about the interpretation itself? Do you mean the act of interpreting or the interpretation achieved? There is no problem with the reality of the act of interpreting - and the interpretation achieved is merely how something seems to her. In autophenomenology, we never shake free of interpretation to reach something that is being interpreted. The interpretation and the interpretation of the interpretation are all merely the interpretation of how something seems to us. The "content" of a judgment is merely how something is being interpreted as being, there is no "what the interpretation actually is" (over and above the vehicle of the interpretation; the vehicle is more than how something is interpreted).

The heterophenomenological story is similar and much more straightforward. Here someone else is interpreting how things seem to the subject. This is an interpretation of the same thing, namely, how something seems to her, but from the standpoint of another. And again there is nothing that the interpretation itself is. If I on the outside want to know what my interpretation itself is, there is nothing I can do but go auto - and the story we just told for the subject will now unfold for me. All I have to work with is my interpretation of how something seems - in this case, how something seems to me about how something seems to her. And if I want to know what this interpretation is really like, all I can do is "interpret my interpretation", i.e., pay attention once more to how things seem to me about her. Again, all I have is the interpretation itself, i.e., how something seems to me (in this case, about how something seems to her). Again we do not get beyond interpretations to something being interpreted. In fact, interpretations of how things seem have less reality than intentional stance patterns on Dennett's more recent view of them. The latter can at least be discerned in behaviour. How something seems need not even have any expression in behaviour. (When he treats heterophenomenology as about behaviour, Dennett might seem to be denying this. I don't think he is. He certainly does not have to. By definition, heterophenomenology is limited to behaviour - what else could it use? - and therefore possible of zombies (p. 95) but the same is not true of autophenomenology.)

Earlier we left two threads dangling: the air of self-refutation of 'that we seem to have Seemings does not entail that we do" and the question of what to do with Seemings lacks brain state realism. It turns out that the same point ties up both threads: the "how things seem" aspect of a judgment is merely an interpretation of something and the only awareness we have of it is either exactly the same interpretation. This could be called 'self-interpretation' but it would be misleading to do so; it would suggest that there is something there both to do the interpreting and be interpreted. All that is there is act. It interprets the world - which, as Dretske puts it (1995, p. 56), is also an interpretation of what is being interpreted. To sum up, there are only interpretations and constraints on interpretations. These interpretations is not also things to be interpreted - which removes the air of self-refutation and explains what to do with Seemings lacks brain state realism.

Can auto- and the hetero-interpretations come apart? If so and if there could also be reason to favor the hetero-interpretation (we usually favor third-person interpretations over first-person ones), this would threaten first-person authority. Dennett says that we could do heterophenomenology on zombies (p. 95), so it would seem that he would hold that heterophenomenology can override autophenomenology and there is no special first-person authority. In fact, this is the very opposite of his position.

... when we put you in the phenomenologist's clutches, you get the last word. ... we are giving you total, dictatorial authority over the account of how it seems to you, about what it is like to be you. [p. 96, his emphases]

When there is a first-person point of view (autophenomenology), it trumps the third person one. First-person authority, as Dennett interprets it (pp. 77, 81), is the idea that I am the best authority about how things seem to me. And there does seem to be something to this idea. I can be wrong about a lot of things but how could I be wrong about how things seem to me? This would require a distinction between how things seem to seem and how they really seem?! So the question is: Can we keep heterophenomenology and first-person authority, too?

I think we can but we need to understand first-person authority a bit better. Why is it that while we can distinguish between how something is presented to someone, how it ought to seem to them, and how it does seem to them, we cannot distinguish between how it seems to them and how it seems to seem to them? In the literature, it is now common to attribute the impossibility of the latter to the conditions of the agent's words having the meaning they do (Davidson, 1986) or the conditions of holding one another morally responsible (Bilgrami, 1998) or some other social factor. I don't think first-person authority has anything to do with social factors. We are authoritative with respect to how things seems to us because, roughly, how things seem to us is simply and always the first stage in conscious belief acquisition. How things seem to us is - how things seem to us. There is no room, not even a concept of room, for anything to intervene between how things seem to us and how they seem. Anything intervening would by that intervention immediately replace the old state and itself become how things seem to us. For this reason, first-person authority is inviolable.

Of course, we can learn more about how something seems to us, for example by acquiring better concepts for thinking about things of that kind. But this would be for something new to intervene, for how the things seems to change. Notice that as I've characterized it, this first stage is completely content-neutral. It is also neutral with respect to whether a given Seeming has a clear start point or is smeared out temporally as underlying bits of information fight to shape it.

Blindsight and other pathologies of consciousness are no counter-example. A blindsight patient seems not to being seeing anything in the scotoma and is authoritative about this. What makes their case special is that what they seem to themselves to be seeing and what they have some measure of informational access (Flanagan) or access consciousness (Block) to diverge more widely than they do in the normal case.

This may protect first-person authority but at the cost of raising another question: If the subject is authoritative about how things seem to her, what is left for heterophenomenology to do? Surely all a third party has to do and even can do is ask the subject. Well, perhaps; but we also have to understand the answer. Even if the subject is authoritative for how things seem to her, she is not authoritative about her descriptions of these Seemings. When she describes how things seem to her, she is using public language and every element of the description requires interpretation. This makes room for heterophenomenology.

Some commentators have wondered if the realism of CE and Dennett's more recent writings on consciousness can be reconciled with the interpretationist metaphysics of mind of the intentional and heterophenomenological stances (Sedivy, 1995, for example). To be sure, by using words like 'experience' to describe the processes in the brain, Dennett himself muddies the waters, but they can be. We merely have to take both the realism and the interpretationism seriously.

So far we have examined how much and what kind of hostility to Seemings is built into Dennett's picture of consciousness and how the two sides of that picture, the realist side and the interpretationism side, hang together. Does CE offer arguments for the interpretationism about heterophenomenological content not found in earlier works? The claim that Seemings are judgments is new, the interpretationism of heterophenomenology is not. Dennett's way of arguing for the seemings-as-judgments view is largely to urge is merits as a problem-solver as against any of the alternatives.(18) The theory faces a problem.

Judgments and pains

The problem is this. In at least one case, the theory of Seemings as judgings appears to be wrong. Dennett does not shrink from hard cases and the hardest case for his kind of theory is pain. Here is a passage on it's close companion, suffering:

Suffering is not a matter of being visited by some ineffable but intrinsically awful state, but of having one's life hopes, life plans, life projects blighted by circumstances imposed on one's desires, thwarting one's intentions ... . The idea of suffering being somehow explicable as the presence of some intrinsic property is as hopeless as the idea of amusement being somehow explicable as the presence of intrinsic hilarity. [p. 449]

I don't know about the awfulness of suffering being ineffable or intrinsic- indeed, I am not even sure what these terms mean - but the idea that suffering is merely a matter of judgment does not seem right. However, let us grant it. Dennett distinguishes suffering and pain rather sharply (1995, p. 352). Let us grant this, too. Now, what about pain? What does Dennett have to say about it?

No theory of pain as thwarted projects is going to do. If I hit my thumb with a hammer, what I feel will not be a gloomy sense of a reversal in my life fortunes! Indeed, a short, sharp pain like that will probably have no implications for my life fortunes (beyond the five minutes it takes the pain to fade). Yet it still hurts. Dennett himself offers an exactly parallel example (1995, p. 352). So what does he want to say about it?

The only positive account of pain that he has given puts pain squarely on the interpretation side of the interpretation/physical realization split.

Are pains real? They are as real as haircuts and dollars and opportunities and persons, and centers of gravity ... [p. 460].

A motley collection but (almost) all matters of interpretation - remember how Dennett views the reality of persons (I am not sure what haircuts are doing here). He seems to push the pain-as-judgment line in (1995), too, even thought he explicitly talks about short, sharp pains there.

The case of morphine suggests that no such account will work. As Dennett himself has articulated very nicely (1978b), morphine acts against pain in a peculiar way. Subjects claim that it does not remove the sensation of pain - but the pain no longer hurts. (Subjects say things like, "I still have the pain but it no longer hurts!") That is to say, it seems to the subject that the pain is still there - but a feeling-state characteristic of pain is not, namely, the hurting, the awfulness. This example would seem to cut against any reduction of pain to judgment decisively. Yet like the short, sharp pain, again it is Dennett's own example.

Dennett's approach to this example (private correspondence) is to suggest that how things seem to a subject is not definitive (except for how things seem to the subject). Moreover, there is a difference of judgment before and after: as we just saw, the subject judges her situation after the morphine to be different from her situation before. All true - and important. There is more judgment involved in even sharp, emotionally uncomplicated pain than many have thought. But! First, if even a difference as large as the difference between what I judge a feeling to be like and whether it hurts does not count, if even excruciating sensations come out as judgments, what is the force of the claim that all phenomenology is judgment? There is a risk in an 'all' claim like this. Is Dennett reducing phenomenology to judgment - or merely bending the notion of judgment out of shape, expanding it so that it just becomes another name for phenomenology? What is being denied? What would a conscious state that was not a judgment be like? (This move should appeal to Dennett.)

I don't think that that this worry is real. The claim that all phenomenology is judgment includes, for example, a claim that there is always description or other encoding involved. If so, Dennett is not simply redefining 'judgment'. But there is a weaker worry. What is it about some judgments, descriptions, in virtue of which they hurt, others don't? Absent an answer to this question, we have not made much progress with pain (or, mutatis mutandis, any other sensation).

In general, Dennett offers a pretty bloodless account of sensations (and also, I think, affects). Or worse. At one point in his long campaign to wean us away from the traditional theory of Seemings, Dennett suggests that the traditional notion needs something like pseudo-pigment, an illusory substance that he calls figment (p. 346). I am inclined to turn this move around. Dennett attempts to give us an account of pains but what he ends up with seem to be - fains.(19) Fains may make us act like they're pains and even thwart our intentions like they're pains (to the extent that pains do thwart our intentions) but they're not awful to have, so they're not pains. (This is one place where looking like us and "quacking" like us is not enough for being like us.)

Dennett is perfectly well aware of what pain is like, of course. Early in CE (pp. 25, 60-4) and in more recent writings (1998, pp. 174, 280), he speaks very eloquently about it. Part of the problem may be a nasty lurking dilemma. As Dennett says, any account of what the awfulness of pain consists in must, in some way, break it down into elements that are not themselves awful to have. If your analysis merely breaks it into elements that are themselves awful, you haven't analyzed the awfulness (p. 64). (Compare Fodor on propositional content: "if it is something, then it is something else" [1985, p. 9]). This creates a prima facie dilemma: if you keep the awfulness, it will be at the cost of giving no analysis of it; but if you analyze the awfulness, it will be at the cost of losing "the thing itself, ... the pain in all its awfulness", as Dennett has his alter ego put it (p. 64). Over the years, Dennett has gone both ways. In (1978b), the awfulness is essentially unanalyzed and Dennett, in a fine application of what I earlier called the "outside in" strategy, focuses on how the awfulness and whatever it is that we are aware of as pain could be connected to the brain. In CE, he tries to say something about what pain, including its awfulness, might consist in - and loses the awfulness. Nasty.

Can we rescue Dennett on pain? (Not that he thinks that he needs any rescuing.) Notice that he treats pain as an intentional phenomenon. If it is, it has to be handled like belief and perception and seeming and that is how Dennett handles it. Instead of treating pain as intentional and forcing it into the metaphysics of the intentional stance, why not view it as like processes of content-fixation, i.e., as real events occurring in the brain. On this, I think more plausible view, certain brain processes would simply be sensations that are awful to have - sensations that obliterate concentration, cause nausea, disappear with analgesics, etc., etc. Most pains are not intentional, i.e., about anything, and treating them as real brain events seems preferable to treating them as having the reality of blighted hopes, dollars, centers of gravity, etc.(20)

There is a lot to commend this view. Moreover, we can accept it and still keep everything Dennett would (should?) want to say about heterophenomenology and pain. Exact what kind of pain and what implications it has for our lives could still be a matter of judgment, as lacking in brainstate reality as the result of any other heterophenomenological judgment. (A pain can be awful, indeed truly horrendous, without us having any precise view of what kind of pain it is.) In fact, we would end up exactly where morphine patients say they are: "I still have the pain" (heterophenomenological judgment) "but it no longer hurts" (the awful-feeling brain state has been altered). And we retain both blighted hopes and short, sharp pains.


We turn now to Dennett's theory of the Subject and the multiple drafts model. Is Dennett an eliminativist about Subjects? Recall our F-level list of features of the Subject.

List 1: F-level concept of the Subject. A Subject

� operates by forming things like beliefs and desires about objects, events, etc.

� marshals various cognitive resources (beliefs, desires, values, concepts, etc.) to generate an intention to do something and then marshals diverse bodily resources to do it

� is aware of a number of things at once in some way that allows us to relate things to other things

� consciously focuses attention on tasks

Dennett is clearly hostile to the first item on this list as anything more than an explanatory abstracta of the intentional stance and states that we seem to ourselves to have. However, this is not new to CE, being fully present in the metaphysics of the intentional stance. Dennett carries it over to CE (p. 319 for example) but does not offer new arguments for it. So what about the rest of the list?

A notion of mental unity is at work in all three items. Now, if Dennett's multiple drafts model is hostile to anything in the traditional F-level notion of the subject, it would seem to be the idea of mental unity. So mental unity is a good test case. Six kinds of mental unity can be distinguished:

Cognitive unity

Cognitive unity consists in our ability to bring an extremely wide range of cognitive resources to bear: what we want; what we believe; our attitudes to self, situation, and context; input from each sense; information of all kinds; language; memory of various kinds; bodily sensations; our skills; and so on. And we bring all these elements to bear in a highly integrated way. In this regard, Dennett himself mentions binding of textures, shapes, etc. into representations of constant objects (pp. 47, 258).

Unity of consciousness

In addition to cognitive unity, consciousness displays unities of various kinds and in various places. Let's give them separate names, unity of simple consciousness, unity of focus, unified consciousness of one's psychological states, and unified consciousness of self.

i. Unity of simple consciousness Unity of simple consciousness starts from the intuitive idea that we are aware of a great many things at once. Here is a better definition:

The unity of consciousness =df. a representing in which a number of items as represented are combined in such a way that to be aware of any of these items is also to be aware of other represented items as connected to it and of the whole as a single complex represented "world".

ii. Unified consciousness of one's conscious states All the other unities of consciousness can be thought of as special instances of unified simple consciousness. The second is unified consciousness of our own conscious states. As well as having unified consciousness of situations and events not themselves conscious, we are able to be aware of seeing them, having feelings, thinking things, remembering things, etc.

iii. Unity of focus The third instance of unified consciousness worth distinguishing is unity of focus. In unity of focus, some items from unified simple consciousness are selected in a specially focused kind of consciousness and we are able to focus a number of considerations at the same time on these items - desires, beliefs, situations in the world, probabilities, etc. Focal attention comes up once in CE (p. 397).

iv. Unified consciousness of self The fourth and final form of unified consciousness that I will distinguish is unified consciousness of self. In unified consciousness of self, each of us is aware of him- or herself as him- or herself and as apparently the single subject and agent of the representations giving us unified simple consciousness of the "world", the single "common subject" of one's mental states, as Kant put it (Kant, 1781, A350).

Consciousness of one's psychological states and oneself is different from consciousness of the world. All animals are conscious of the world or almost all but it is at least debatable whether any animal other than Homo sapiens is conscious of self.(21) Unified consciousness of one's own psychological states and of oneself as the single "common subject" of one's mental states figures centrally in Kant (1781/7, A350; see Brook, 1994, Ch. 3). Dennett goes both ways on the unity of consciousness in general. On the one hand, he treats it as merely how we seem to ourselves (pp. 74, 108; 1994a, p.134). On the other hand, he discusses pathologies of unified consciousness a number of times (pp. 326, 358, 419-25) and pathologies are breakdowns, hence require something that has broken down. Consciousness of self in particular is a theme throughout (e.g., pp. 67, 77, Ch. 10).

Unity of Behaviour

Finally, our behaviour is highly unified. In doing what we do, we coordinate our limbs, eyes, bodily attitude, etc., in ways the precision and complexity of which would be difficult to exaggerate. Think, for example, of a concert pianist performing a complicated concerto. Or Dennett sitting in his rocking chair, rocking, reading, listening, and looking all at the same time.

Focused, deliberate, integrated reflection and action requires all the items on list (1) except (if Dennett is right) the first one. Dennett would accept that the rest of the list refers to something real. What about the mental unities? As will be clear from the views I have just summarized, he is at least not deeply hostile to them.

He is quite hostile to a lot that has been said about them, of course. Indeed, one of the main objectives of CE is to show that the P-level system realizing any such unities (and everything else to do with consciousness) could be and most probably is very different from anything conceived of in the tradition.(22) According to the multiple drafts model, all mental unities and all F-level features of conscious Subjects could be realized in a system having the following P-level properties:

The whole of this model has come to be known as the multiple drafts model, even though a specific provision concerning multiple competing drafts is just one part of it. (The model has important implications for the nature of individual conscious experiences, too, so my tying it so closely to Dennett's views on the Subject is a bit artificial.) So far as I know, it is at least as likely to be true as any other model of the conscious Subject and more likely than most. At any rate, I won't question it here. Among the most interesting claims Dennett makes based on it are that:

On the multiple drafts model, the P-level procedures and structures that realize consciousness are very different from what the mental unities might lead us to expect. Dennett also urges that unified consciousness has been overrated. Indeed, he wants to downsize the role of mental unities in consciousness quite radically (see, e.g., his comments that conscious practical reasoning is relatively rare and that it is dangerous to take such rare occurrences to be a guide to the whole [p. 252]). But none of this is to deny that mental unities or the Subject exists. In his view, we are focused, unified, able to coordinate rich cognitive resources, and able to report with all the subtlety of the intentional stance on the results of many of these goings-on in us. It would seem, then, that Dennett is not an eliminativist about anything in the traditional list of F-level features of the Subject except the first item, which the metaphysics of the intentional stance had already attacked. Nor is he an eliminativist about the mental unities closely associated with that list.(23) He just wants a better P-level theory of all these things. The one he comes up with is radically different from anything in the tradition but it is still a theory to explain these features, not an argument that nothing has such features.

More generally, if we accept that the traditional Subject exists, we must not prejudge what it might be like (a point already made by Kant and, more recently, by van Gulick, 1988). Who knows what kind of structures and procedures could result in and/or go with being a Subject? What we observe in ourselves and others are cognitive states and activities tied together in various ways. It is an open question what the procedural and neural 'substrates' of these representations and activities might be like, to use Kant's word (1781/7, A350). In particular, the substrate may well be processes of content-fixation generating multiple drafts of narrative fragments, just as Dennett says. Accepting that something exists answering to list (1) and the sketched unities does not commit us to any P-level conception of the Subject, traditional or contemporary. In particular, the traditional F-level conception of the subject does not commit us to such postulated P-level entities as homunculii, Cartesian theaters, or the other weird and wonderful things that Dennett looks upon with such deep suspicion.

Dennett has made important suggestions about what can and cannot appear in an acceptable P-level account in other works, too. In particular, he has been urging for decades that there can be no undischarged homunculii, no exempt agents, in such an account (1978c, p. 102;1978d, p. 124). He has also made a suggestion for how we can discharge the homunculus without abandoning the Subject of list (1). Following Fodor (1975, 74fn.), he offers the idea of representations that can represent to themselves as a possible way out (1978c, pp. 101-2). Again, however, this is a recommendation for P-level theory, not an attack on the existence of the F-level Subject. Dennett's notion of the virtual captain (p. 228) is a theory of what a unified Subject consists in, not an argument that unified Subjects do not exist.(24)

To conclude: Dennett's multiple drafts model does not pose a threat to the F-level Subject of folk psychology, though it is a massive rejection of most previous P-level theories of it.

Two last issues

I will close with a brief look at one interesting implication of Dennett's model and one omission. The implication arises out of this question: What are we doing when we attribute the items on list (1) and the attendant mental unities to someone? We are not just describing how people seem to themselves, so we are not just doing heterophenomenology. Indeed, if the person in question is not "psychologically minded", he might very well not seem to himself to have these features at all. Anyway, the consciousness in question is not merely a matter of how subjects seem to themselves. Not just consciousness but unified, focused consciousness makes a difference to performance so it is more than merely how we seem to ourselves (performance is generally worse when conscious attention is absent, distracted, split between two tasks, etc.). Think of the work of theorists such as Baars (1988 and many other works). If so, what are we doing? At minimum, we are attributing properties that explain what the person is and does, not merely how he seems to himself. This sounds more like the intentional stance. But we are not doing that, either.

Intentional stance attributions are governed by considerations of the appropriateness of beliefs and desires and the practical rationality of actions. But the unity of consciousness, unified focus, etc., are not reasons for action. Thus, when we ascribe them, we are doing something more than and different from adopting the intentional stance, too. When they ascribe consciousness to some subjects, lack of consciousness to others, they are not postulating states and processes that rationalize behaviour. In short, attributions of consciousness use neither heterophenomenology nor the intentional stance.

What is going on, then? Something more like inference to the best explanation. In Dennett's jargon, attributions of consciousness are done from something more like the design stance or even the physical stance than the intentional stance.(25) Inference to the best explanation, however, is held by many theorists to be entirely compatible with many kinds of realism about the inferred states and processes - which returns us to a point noted a couple of times already: the traditional realism of the vocabulary of much of CE. Of course, even if inferences to the best explanation are compatible with traditional realism, it is does not follow that they require it. But Dennett has always been a traditional realist about the physical stance and, in a way and at times, about the design stance, too (see, e.g., 1987c, p. 39). If so, Dennett should be a traditional realist about consciousness. And, as we saw, he is, at any rate about conscious experiences and sensory qualities (though not, of course, about how things seem in them). From where I sit, that is all to the good. The more realism, the better. All that is left for the intentional and heterophenomenological stances is intentional content and judgments about how things seem.

That's the implication. Here's the omission. If the Subject is unified, it is unified both at a time and over time. The most significant element in unifying the Subject over time is autobiographical memory, specifically, memory of having experiences and doing actions (Kant, 1781/7; Parfit, 1984). Whatever the implications of such memory for "personal identity" (being one person over time), it is central to personal unity over time. Indeed, it is so central that, as Parfit (1970, p. 15) has put it, if I remember doing something done by an earlier person or feeling something felt by an earlier person, I will automatically assume that that person was me. Dennett discusses memory a number of times in CE (as we saw, he even calls it criterial for consciousness). Indeed, semantic memory, both short- and long-term, has been central to his model of consciousness since at least 1978e. Yet there is hardly a mention of autobiographical memory in CE.

Summary box score: 1. Dennett has no quarrel with the existence of vehicles of Seemings; he does not even question the idea. His target is a certain theory of Seemings. 2. Likewise with Subjects. The multiple drafts model is aimed at the traditional theory of the Subject, not at the idea that Subjects exist at all. 3. Dennett's theory of Seemings as judgments and the multiple drafts model are generative and suggestive. 4. The account faces some problems. Nothing I have said undermines Dennett's account but we may have domesticated it a bit. It has been thought to have very radical implications. I am not sure that it does.(26)


Akins, K., ed. 1996. Perception. Vancouver Studies in Cognitive Science, 5. Oxford University Press, pp. 158-172.

Baars, B. 1988. A Cognitive Theory of Consciousness Cambridge: Cambridge University Press

Bilgrami, A. 1998. Self-knowledge and resentment. In: C. Wright, B. Smith, C. MacDonald, eds. 1988.

Brook, A. 1994. Kant and the Mind New York: Cambridge University Press

Davidson, D. 1986. Knowing one's own mind. Proceedings and Addresses of the American Philosophical Association 60, pp. 441-58.

Dawson, M. 1998. Understanding Cognitive Science. Blackwell Publishers

Dennett, D. 1978a. Two approaches to mental images. Reprinted in: Dennett, 1978f

Dennett, D. 1978b. Why you can't make a computer that can feel pain. Reprinted in: Dennett, 1978f

Dennett, D. 1978c. A cure for the common code. Reprinted in: Dennett, 1978f

Dennett, D. 1978d. Artificial intelligence as philosophy and psychology. Reprinted in: Dennett, 1978f

Dennett, D. 1978e. A cognitive theory of consciousness. Reprinted in: Dennett, 1978f

Dennett, D. 1978f. Brainstorms Montgomery, VT: Bradford Books

Dennett, D. 1987a. Evolution, error and intentionality. Reprinted in: Dennett, 1987d

Dennett, D. 1987b. Three kinds of intentional psychology. Reprinted in: Dennett, 1987d

Dennett, D. 1987c. Real patterns, deeper facts, and empty questions. In: Dennett, 1987d

Dennett, D. 1987d. The Intentional Stance Cambridge, MA: MIT Press

Dennett, D. 1987e. Consciousness. In: The Oxford Companion to the Mind, ed. R. Gregory. Oxford: Oxford University Press

Dennett, D. 1988. Quining qualia. In: Marcel and Bisiach, 1988, pp. 42-77

Dennett, D. 1991a. Consciousness Explained. Boston: Little Brown and Company

Dennett, D. 1991b. Real patterns. Reprinted in: Dennett, 1998

Dennett, D. 1994a. Real consciousness. Reprinted in: Dennett, 1998.

Dennett, D. 1994b. Instead of qualia. Reprinted in: Dennett, 1998.

Dennett, D. 1994c. Get real. Philosophical Topics 22, pp. 505-68.

Dennett, D. 1995. Animal consciousness: what matters and why. Reprinted in: Dennett, 1998.

Dennett, D. 1996. Seeing is believing ... . In Akins, 1996.

Dennett, D. 1998. Brainchildren Cambridge, MA: MIT Press

Dretske, F. 1995. Naturalizing the Mind. Cambridge, MA: MIT Press

Flanagan, O. 1992. Consciousness Reconsidered Cambridge, MA: MIT Press

Fodor, J. 1975. The Language of Thought New York: Thomas Y. Crowell

Fodor, J. 1985. Fodor's guide to mental representation. Reprinted in: Fodor, 1990.

Fodor, J. 1990. A Theory of Content and Other Essays Cambridge, MA: MIT Press

Harman, G. 1990. The intrinsic quality of experience. In J. Tomberlin, ed. Philosophical Perspectives 4. Atascadero: Ridgeview. pp. 31-52.

Harman, G. 1996. Explaining objective color in terms of subjective reactions. In: E. Villaneuva, ed. Philosophical Issues. Atascadero: Ridgeview, pp. 1-18.

Kant, I. 1781/7. Critique of Pure Reason (1781 and 1787). Trans. Norman Kemp Smith as: Immanuel Kant's Critique of Pure Reason. London: Macmillan, 1963.

Marcel, A. and Bisiach, E. 1988. Consciousness in Contemporary Science. New York: Oxford University Press.

Marr, D. 1982. Vision: A computational investigation into the human representation and processing of visual information New York: W. H. Freeman

Palmer, S. 1999. Color, consciousness, and the isomorphism constraint. y Behavioral and Brain Sciences 22 (6): forthcoming.

Parfit, Derek. 1970. Personal identity. Philosophical Review 80, pp. 3-27.

Parfit, Derek. 1984. Reasons and Persons. Oxford: Oxford University Press

Sedivy, S. 1995. Critical notice of D. Dennett, Consciousness Explained. Canadian Journal of Philosophy 25 (3), pp. 455-83.

Sellars, W. 1963. Philosophy and the scientific image of man. In: Science, Perception and Reality London: Routledge and Kegan Paul

van Gulick, R. 1988. Consciousness, intrinsic intentionality, and self-understanding machines. In: A. J. Marcel and E. Bisiach, eds. Consciousness in Contemporary Science Oxford: Oxford University Press

Wilkes, K. V. 1988. ____, yishi, duh, um, and consciousness. In: A. J. Marcel and E. Bisiach, eds. 1988.

Wright, C., Smith, B., MacDonald, C., eds. 1988. Knowing Our Own Minds. Oxford: Oxford University Press, pp. 207-41

1. Page-references without a date of publication will be to this work.

2. See Wilkes, 1988, for my reasons for limiting the claim to the last few hundred years.

3. Another version of the triple, one associated with Newell, is the trio of knowledge, physical symbol procedure, and system implementation. Dennett claims that his triple of intentional, design and physical stances is closely related (p. 276) but his notion of the intentional is different from the original notions of task or knowledge level in some respects (note 25 gives my reasons for saying this).

4. Sellars' (1963) manifest image is a rough parallel. A host of issues about lexical semantics, defining characteristics, essences, etc., arise here. I am going to duck them all.

5. What if there is no clear breakpoint between list (1) and list (2)? Perhaps folk psychology is itself infected with theory. That would be no surprise. My question would then become: Starting with the most general, least theoretic features, how far down does Dennett want to go before he starts to urge outright elimination? Short answer: maybe little distance for Seemings, quite a long way for Subjects.

6. Dennett does not advance this claim in CE but he makes it in1987a.

7. The recently advanced view of some pain specialists that pain can arise in the brain entirely endogenously gives this question some urgency.

8. Flanagan (1992, pp. 81-5) has an excellent discussion of the "outside in" strategy so I will not pursue it further here. In a response to Flanagan in his wrap up paper for the volume on his work in Philosophical Topics, Dennett give this as his reasons for holding that there is no resolution of S/O and the other impasses: "tremendous progress is made by assuming that there isn't". (1994c, p. 532). Thus, his reason is pragmatic not principled. He is says that simplification is what generates the progress. But is the simplification based on falsehood? That would not be progress. As I said, I won't pursue the issue here.

9. As Dennett notes (p. 139), Kant was already aware that the time order in which events are presented to us can be different from the time order in which we experience them (Kant, 1781/7, fn. to A37=B54).

10. As Kant saw, the same is true of our awareness of our own self. We are aware of ourselves only as we appear to ourselves, not as we are; and appearance of self requires characterization just as much as appearance of anything else does. Dennett of course agrees (p. 67).

11. As Palmer (1998) and others have pointed out, the possibilities for inverting qualia of colour are in fact quite limited.

12. I owe this point to Zoltan Jakab.

13. Flanagan makes this observation in (1992), pp. 67-8. My Kantian/Empiricist distinction is similar to his distinction between the wide and the narrow senses of 'qualia'.

14. What we are doing here is closely related to the old project of figuring out the "existential status" of intentional objects - and may be a first step towards a solution (Dennett,1978a, p. 181). More recently, Dennett has said that attributions of information or how things would seem to the perceiver here are done from the intentional stance and so presumably have all the holism and indeterminacy that goes with this stance. Given the realist presentation of the whole idea of content-fixation in CE, this comes as a surprise. The text of CE suggests exactly the opposite. But, he says, it is his view (1994c, p. 528).

15. Even though, according to Dennett 1994c, p. 528, the determination of what content has been fixed even by a microjudgment itself requires the intentional stance. As I said in note 14, this claim is a surprise. We would expect the intentional stance to kick in only about when autophenomenology kicks in: when we start to judge how things are (i.e., how they seem to us).

16. I mean to be using a concept of binding broader than the one found in contemporary vision theory here, one more like Kant's notion of synthesis.

17. Harman 1990, 1996 and Dretske 1995 both emphasize the idea that when we are aware of a Seeming, as we would put it, what we are aware of are the properties of what is represented in it. So long as the idea is not overgeneralized to our awareness of the vehicles of Seemings, too, it seems right.

18. There is an argument that I have not considered: Dennett's repeated invitation to ask whether there could be anything in the brain that has the properties of how Santa Claus or some other intentional object seems to us (p. 96; see pp. 85, 134). Two problems. (1) Distinguish the referent of a phenomenological item from the vehicle of that item and this argument disappears. What we want to investigate are the brain events (if any) that a phenomenological item consists in, not the events that it represents (brain events or otherwise; they are rarely brain events, of course). Who thinks that the way we seem to ourselves is apt to be anything like what we actually are? (2) It isn't a new argument. Dennett has been using it since (1978a).

19. I owe this lovely neologism to Chris Viger.

20. In a roundabout way, I owe this view to Don Ross - roundabout because it came to me when he was once trying to put pain firmly on the other side of the intentional/context-fixing divide.

21. With respect to distinguishing consciousness of the world from consciousness of self, Dennett has gone both ways in different writings. In (1987e), he carefully distinguishes them. In CE, he never does so explicitly though he toys with the distinction in a few places (e.g., p. 45).

22. Here I am treating the unities as F-level phenomena. They are so integral to, indeed so close to being simply a redescription of, the items on list (1) that it is hard to see how we could have one without the other. Plus, treating them as P-level would complicate the telling of the story I want to tell without changing anything but the telling.

23. One of the interesting implications of what I have just been saying is that unified conscious Subjects and any belief/desire structuring of cognitive processing turn out to be quite independent of one another. Robust conscious Subjects could exist even if belief/desire psychology turns out to be false.

24. I have been known to suggest that Dennett's notion of the virtual captain is a perfectly good P-level sketch of what Kant's transcendental unity of apperception might consist in (Brook, 1994, p. 229)! I meant to be provocative but I also endorse the idea. I also argue in Ch. 9 of that book that a Kantian spin on the idea of a self-representing representation can take us a long way.

25. As we noted earlier, Dennett (p. 276) claims that the intentional stance is quite closely related to the task- or knowledge-level of Newell and company. Since attributions of consciousness are often F-level - i.e., task-level - attributions, we can now see why I said earlier that the intentional stance seems quite different from the task or knowledge level of Newell and others.

26. I would like to thank Pamela Goold, Don Ross, Robert Stainton, and especially Zoltan Jakab for probing and helpful questions and suggestions. I owe a special debt to Dan Dennett for correcting a mistake that I was caught in and doing so in a most generative way.