Draft
John O’Dea
I
Central to the Knowledge Argument (Jackson 1983) is a thought experiment, in which a future, physically omniscient neuroscientist lives in a black-and-white environment. Although she knows everything there is to know about how the brain works, and in particular about how colour vision works, she has never seen colours herself. When she does, finally, step outside her black-and-white environment, it seems clear that she would learn a fact about the world. What she learns, so it strongly seems, is how the sensation of red has felt to those living in a coloured environment. Since she learns this fact and yet knows all of the physical facts, it must be a non-physical fact that she learns. Some facts about sensation are therefore non-physical facts.
There have been many responses to Jackson’s argument. Some have denied that Mary learns anything, some that she learns something non-factual (such as an ability), some that she learns a fact but not a new fact. And so on. There is near consensus the argument cannot prove what it purports to prove, namely non-physicalism, but no consensus whatever about where it goes wrong.
Michael Tye () has recently proposed an approach that draws on the alleged peculiarity of the concepts we use to talk about – or, to be more precise, think about – the phenomenal character of sensations. It’s proponents (who include Tye, Brian Loar, David Papineau and in some respects Ned Block) describe this peculiarity as the recognitional nature of phenomenal concepts. Here I present some difficulties with this approach.
II
The idea that phenomenal concepts are recognitional gets its intuitive force from the idea that in order to have the concept of a phenomenal state, it is necessary to have had that phenomenal experience; to have been in the state the concept is of. There may be some sort of analogous concept of a red sensation that a person colour-blind from birth could possess, but it would not be the fully blown concept possessed by the non-colour-blind.
Now as much as prior experience might be a necessary condition for the possession of phenomenal concepts, it is certainly not sufficient. One would need, presumably, to remember having had it (in the sense, at least, of remembering what the experience is like, if nothing about the circumstances of its actual occurrence). However, we can also see that falsely remembering – having an apparent memory triggered directly in the brain by an electrode, for instance – may equally count so far as possession of the concept of what the memory is apparently of is concerned. Whether you agree with this will depend on whether you think it is possible to have even an apparent memory of something that isn’t caused more or less directly by that something.
At any rate, sufficiently or not, false memory or not, there does seem to be a link between memories of phenomenal experiences and the possession of phenomenal concepts. To have the concept of a phenomenal experience, one must know what it is like, and to know what an experience is like is to have had one and remember it.
So far I’ve not mentioned recognition in this story. It comes in now for two related reasons. The first is that memory is not conceptual – or, at least, it isn’t believed to be by the major proponents of the recognitional view. Recognition is much less controversially conceptual. We often recognize something qua a member of a category. So you might recognize some bird in the wild as a species of parrot, or a person by his accent as Scottish. This is to recognize some token as falling under a type. Recognition tends to be this sort of thing. The concept of a phenomenal experience is recognitional in the sense that the memory we have of that experience must be such as to allow future experiences to be recognized as falling under the same type. Otherwise it is not really conceptual. The second, related reason for bringing recognition in is that future experiences of the same type must indeed be recognized as falling under that type. My memory of a past experience must not only allow the ability to recognize future occurrences as of that type, this must normally be reasonably automatic on introspection. One does not really have the concept of a sensation of red unless one reliably introspectively recognizes sensations of red as sensations of red. This recognition must, moreover, be direct, in the sense in which my recognition of a person as Scottish by his accent is indirect. Here is a quote from Tye’s paper, “A Theory of Phenomenal Concepts”:
…[W]e do not have introspective knowledge of phenomenal character by inferring that character from something else. We acquire introspective knowledge of what it is like to have such-and-such an experience or feeling via a reliable process that triggers the application of a suitable phenomenal concept or concepts. Phenomenal concepts – the concepts that enable us to form a conception of phenomenal character via introspection – are, in my view, recognitional concepts of a special sort.”
Brian Loar espouses a similar view about phenomenal concepts, though in his 1990 paper he expressed the idea of a recognitional concept in the following way:
Given a normal background of cognitive capacities, certain recognitional or discriminative dispositions suffice for having specific recognitional concepts, which is to say, suffice for the capacity to make judgments that depend specifically on those recognitional dispositions. Simple such judgments have the form: the object… a is one of that kind… It is a basic fact about our cognitive set-up that recognitional dispositions can suffice for mental predicates with specific conceptual roles, and in that way create cognitive content. (p. 87)
So they have in mind that the recognition of phenomenal states is a kind of brute causal mechanism, or the outcome of a brute causal mechanism. This makes the concept refer directly, and without any a priori links with other concepts. It is this lack of a priori links that explains, and indeed, explains away the existence of the explanatory gap.
But how does this idea purport to help answer the Knowledge Argument? In the following way: when Mary the neuroscientist steps out of the black-and-white room and has her first sensation of red, she acquires a concept that she has not been in a position to possess – namely, of course, the concept we all have of red sensations. She is suddenly able to think new thoughts about red sensations. In Tye’s words:
In order for Mary to think a phenomenal thought, she must exercise a phenomenal concept. She does not have phenomenal colour experience concepts in her room. So, when she leaves the room, she starts to have new phenomenal thoughts. Content-wise, it must be granted, in one standard sense of the term ‘content’, these thoughts will not be new, given Mary’s complete physical knowledge in her room. For, on my proposal, phenomenal concepts refer directly and thus the contribution they make to thought content is given by their referents alone – referents that are physical, if physicalism about consciousness is true. But thought-types need not be individuated by their content alone.
Intuitively, phenomenal thought types play a different role in rationalizing explanations than non-phenomenal thought types. If their contents are identical, then a second factor must account for this difference. And intuitively that factor is simply that phenomenal thoughts exercise different concepts – phenomenal concepts… This two-factor view of phenomenal thought types permits the physicalist to maintain that there is a perfectly good sense in which Mary discovers that so-and-so is the case after she is released. For she comes to think new thoughts and thereby instantiate cognitive thought-types (knowing-that thought types) she could not instantiate before, even though, given her exhaustive knowledge of the physical facts, the contents of her thought-types before and after remain unchanged. And if Mary or anyone else knows that p at time t without knowing that p before t, then surely it is correct to say, in ordinary parlance, that the person has made a discovery at t.
There are two main objections to this view that I want to present here. The first calls into questions whether it is right to count our recognition of phenomenal states as concepts, and the second calls into question the utility of this idea, even if we accept it, in answering the knowledge argument.
III
To begin, note that to simply say that Mary, while in the black-and-white room, lacks the experience required for her to possess the right concept, isn’t enough, since the problem isn’t just Mary’s – it is also ours (Tye isn’t saying this, of course, but bear with me). We can’t infer the presence of phenomenal properties from the presence of non-phenomenal ones, although we possess both sets of concepts. If one doesn’t have the concept A, one isn’t going to be able to infer the presence of A’s from anything. It is trivially true that the concept A must be in some form in my conceptual repertoire in order for me to perform an inference that includes it. So it isn’t merely Mary’s lacking phenomenal concepts that disables her from knowing everything about phenomenal states. If this were the problem, then no conclusions would follow about us, since we are not similarly disabled. The point of the Knowledge Argument is to highlight our own inability to make the inferences that we agree Mary can’t make.
Now Tye argues for a real sense in which Mary learns of the existence of a type of object that she didn’t know of previously. What is much more obvious is that she learns of the existence of a token object that didn’t previously exist. Indeed, insofar as we are inclined to say that she learns the existence of the type, it is in the sense that she learns of the existence of the token and, in general, things exactly like it. The token is placed under a category, but for Tye it isn’t a category that abstracts away from the introspectible character of that token. If it abstracted away, then an introspectibly different state might come under the same category, therefore thoughts that employ that category would be blind to this introspectible difference, and therefore couldn’t underwrite the knowledge we actually have, by hypothesis, of that difference.
According to Tye, Mary’s new knowledge arises out of her acquiring new phenomenal concepts, on the basis of which she sees that there are experiences of that type in the world. My objection, then, is this: Mary’s discovery of the type does not really seem to amount to anything more than her discovery of the token. There seems to be no real cognitive distance between the two. It is, after all, not possible to be aware of something and yet be blind, as it were, to its possible exact resemblance to other things. To see something at all surely is to be open to this possibility. But this openness isn’t conceptual – or, if it is, it is only as conceptual as any perceptual experience. Insofar as Tye’s argument relies on this, it relies on the idea that perception is, at its most basic, conceptual. Tye gives no reason to think this, and it is a very tendentious view, to put it lightly. Moreover, it is not even clear that Tye either agrees with it or takes himself to be committed to it.
If what Mary thinks when she has that first red experience is something like “So this is what people have been talking about!”, is it necessary that we think of her as having categorized what she has in mind by “this”? I suggest not. I suggest that it attributes a thought one step too abstract to claim that what she thinks should be described as “So, this sort of thing is what people have been talking about!”
Now Mary may categorize her experience, but no categorization is needed to see something as a duplicate of itself, and it is precisely the existence of duplicates of her new token red sensation that Mary learns the existence of. Mary cannot have an experience and be blind to some other experience being a duplicate, if it is. Mary’s learning that there are duplicates of her token red sensation does not require putting the token under a category, and therefore does not require conceptualisation. Therefore Mary’s new knowledge does not involve the acquisition of a concept. Not, at any rate, of the kind that it is possible to be blind to.To put this point in a different way, Tye’s recognition amounts, in the end, to nothing more than memory, since to remember something just is to be able to recognise its duplicate.
IV
So much for my first objection. Now I’d like to grant, for the sake of the argument, that what Mary learns does crucially involve the acquisition of a new concept. I want to look more closely at the sort of concepts involved and question whether they could really solve the Mary Problem.
Tye claims that what Mary acquired is a phenomenal concept, and that phenomenal concepts refer directly (direct reference is normally to tokens rather than types, but I leave that aside). What makes then refer directly is the close causal connection between the state and the concept. Both Tye and Loar refer to this as the “triggering” of the concept by the state. What makes these concepts phenomenal is that they essentially involve a specific kind of mental image (broadly construed) of the state they are of. Specifically, the mental image involved is a kind of facsimile, or exemplification, of that state type. (Sometimes it will be an actual instance of the type rather than a facsimile, as when I have a thought about an occurent sensation). Tye equates the sense (i.e. as cognitive significance, not reference-fixer) of the phenomenal concepts with this imagery. Here is Tye (2003) on this notion:
On some occasions, when we introspectively recognize a phenomenal type, we make no use of the stored copy or image. On other occasions, when the phenomenal type is absent from our experience but we are thinking about it, we…retrieve a copy of a token of the phenomenal type…an image or quasi-image, which we may then put to cognitive use.
And Loar (1990):
A phenomenal concept perhaps typically involves not merely a recognitional disposition but also an image; and so, as a psychological state in its own right, a phenomenal concept, given its connection with imaging, is far more similar (in some intuitive sense) to a phenomenal state that either is to the psychological state of having a given physical-theoretical concept. (p.90)
There seems to me a worry, which I don’t yet want to call an objection, about the idea that phenomenal concepts are degraded copies of an original. I take it that it is pretty clear that ordinary mental images of pains, itches and so forth are degraded or imperfect exemplifications. Mental images of pain don’t have the same unpleasant quality; thoughts about sunsets aren’t as vivid as the real experience of looking at a sunset. If they did, we would never want to think about pain, and never need to do anything more than think about sunsets. On the other hand, thoughts about mental images – say, thoughts about memories of sunsets – seem to involve those very images themselves, or at least exact copies thereof. But – here is where the worry starts – they don’t seem to be any more truly phenomenal concepts for all that. My mental image of pain, if we call it a concept, is a relatively poor copy of the actual state, pain, that it is of. But how poor could it be while remaining a phenomenal concept of pain? In answering this, note that it is insufficient to simply say that it (that is, the mental image) must itself be a phenomenal state, since the mental image of brain state such-and-such will naturally be a perfectly phenomenal state without being a phenomenal concept of that brain state.
There may be a good solution to this, but note also the difficulties inherent in positing a mental image of a phenomenal property which itself has phenomenal properties that inexactly resemble the phenomenal properties of what it is an image of. It invites the question, inexact in which respect? This is an awkward question, since phenomenal characters are not normally supposed to have respects. Or, relatedly, take some particular colour, which is the kind of featureless property that Tye has in mind – or, at least, has the kind of featurelessness that Tye has in mind. An inexact copy of some particular colour is an exact copy of some other colour. We seem driven to analogously say – as Loar all but said in the above quote – an inexact copy of a phenomenal state is an exact copy of some other phenomenal state. So the problem this account now faces is that there will be no way of telling, introspectively as it were, whether a given mental image is an image of the mental state it exactly resembles, say, an image of an imagining of pain, or whether it is an image of the state it inexactly resembles, as in simply an image of pain. It is for this sort of reason that Tye insists that phenomenal concepts fix their reference causally and directly rather than via these images. Nevertheless, since exemplificatory images are the cognitive significance of phenomenal concepts, more needs to be said about how a single mental image type could transparently play these two different cognitive roles.
These worries point towards my second objection, which concerns the capacity of this idea to answer the Knowledge Argument. This objection re-describes the original thought experiment with one difference, which ought to be the crucial difference if Tye is right, but which I think it is clear makes no real difference to the conclusion we are drawn to.
According to Tye, what Mary lacks while in the black-and-white room are the phenomenal concepts of colour sensations (not simply concepts of colour sensations, since she has those – that is to say, she has concepts of the brain states that are those sensations).
So now we imagine Mary in the black-and-white room with the phenomenal concept of a red sensation. This concept involves a mental image of red which exemplifies, but inexactly, the experience of a sensation of red. Moreover, it is apt to be triggered by a red sensation were she to have one, but it is there, let’s say, by the grace of medical science; the concept has been neuro-surgically implanted. Or call it an innate concept. At any rate, Mary is in full possession of a mental image or the sensation of red, and which she takes to be of the sensation of red, though she has never actually had the sensation the image is of.
Note, before I go further, that we can agree that possession of the concept of the sensation of red requires that one has had it. But this requirement is a causal one and in ordinary circumstances, not a constitutive requirement. In the ordinary course of events, actually having a sensation is causally necessary for having the concept of that sensation. But we can circumvent this ordinary course of events in a thought experiment without danger of incoherence.
So by some deviant causal chain, Mary has the concept but has never actually had the sensation. The question now is, when she steps out of the black-and-white room and has her first actual red experience, does she learn anything? I think it is clear that she would – namely, she would learn the difference between what the mental image led her to expect and the qualities of the fully blown sensation. The image is, after all, an inexact copy, but how inexact there is no introspectible way to tell. If you aren’t quite convinced, change the example from the sensation of red to the sensation of extreme pain, and make Mary’s conception/facsimile of it a rather vague one. So the issue now becomes, does having the mental image of a sensation tell you all there is phenomenally to know about it. The answer must be no. After all, how could it, since the image is a degraded, inexact copy and therefore by hypothesis lacks at least some – presumably phenomenally relevant – features of what it is an image of. So having the experience itself must teach one more about that experience type than a mere mental image of it could. This problem is exacerbated as we imagine mental images that are poorer copies of the experiences they are images of; it is minimized as we imagine them better copies of what they are images of.
Tye argues that what Mary learns amount to her acquisition of a phenomenal concept, but, as I hope I’ve shown, if Mary can have the concept in this sense and yet learn something on first having the sensation the concept is of, then Tye can’t be right. What this says about the Knowledge Argument I don’t yet know, but it shows that the recognitional concept strategy as Tye pursues it must be mistaken.
References
Loar, Brian (1990) “Phenomenal States” Philosophical Perspectives 4; 81-108.
Tye, Michael (), “A Theory of Phenomenal States” Philosophy (also, http://www.utexas.edu/cola/depts/philosophy/faculty/tye/Theory.pdf ).