Let me begin by indicating where I think Harman and I are in agreement. We both think that "subjective reactions" must come into an account of color, although we have different views about how they do. We both think that perceptual experience has a "presentational or representational character," and that color is represented by our visual experiences as a feature of external objects, not as a feature of our experience. Moreover, we agree that, as Harman puts it, "color is experienced as a simple basic quality, rather than a disposition or complex of causal properties." As Harman emphasized in an earlier paper,1 what we are introspectively aware of in our experience is its presentational or representational content, not any "mental paint" which bestows this content. I shall refer to all of this as Harman's "phenomenological point." Because we agree on this, we also agree that if his characters George and Mary were spectrum inverted relative to each other, supposing that to be possible, this would have to involve their perceiving the same objects as having different properties, this despite the fact that as normal perceivers they would perceive these objects as having the same colors. And I think we agree that in this case the properties would have to be relational ones, defined or constituted by their relations to the experiences of the subject perceiving them.
On all these points I think we are in substantive agreement. There is another point on which our agreement is, I think, merely verbal. We agree that there are no "color sensations." But for me this is just the point that the current meaning of "sensation" makes it an inappropriate term for our color experiences, while for Harman the rejection of color sensations seems to be linked with the rejection of "qualia." Now in agreeing with Harman's phenomenological point, I agree with him in rejecting one conception of qualia -- the "mental paint" conception, according to which we are aware of the representational content of our color experiences by being aware in a quasi-perceptual way of their qualia, in much the way we are aware of what colors are represented in a painting by perceiving the pigments on the canvas. But I think that there is another conception of qualia on which they are required by a satisfactory account of color experience.
Finally, there is one point of difference between us that is not quite, or not yet, a point of disagreement. Harman says that he is "not sure" whether the imagined possibility of spectrum inversion is coherent, whereas I am quite sure that it is coherent. One of the main things I want to do in what follows is to show that if Harman were to climb off the fence on this issue, and climb down in my direction, then given our other points of agreement he could not stop short of accepting the rest of my view, qualia and all. Along the way I will show that the circularities Harman claims to find in my view are simply not there; they seem to be there only because, perhaps out of misguided charity, he states my view in a way that omits its commitment to qualia. That is truly a case of Hamlet without the Prince of Denmark.
So let me start laying out the argument. Since my object is to show what Harman is committed to on the assumption that spectrum inversion is possible, I assume that it is possible. Assuming Harman's phenomenological point, we have
1) If George and Mary are normal perceivers who are spectrum inverted relative to each other, then when George sees a red object it looks to him to have a property, call it Q, which green objects look to Mary to have, and when Mary sees a red object it looks to her to have a property, call it T, which green objects look to George to have.
But Harman thinks, as I do, that
2) If George and Mary are normal perceivers, then (under normal conditions) red objects look red to both and green objects look green to both.
It follows that
3) The Q and T referred to in (1) are distinct from redness and greenness.
It is compatible with (3) that Q and T are in fact not properties that visible objects ever have -- that while George and Mary always perceive red and green objects as having one or the other of these properties, their color experiences are to that extent illusory. There are two versions of this view. One, which I call "literal projectivism," says that Q and T and the like are in fact properties of experiences, qualia in fact, which are projected onto external objects. So on this view we do perceive "mental paint," but perceive it (misperceive it) as belonging to external things. I take this to be the view that Paul Boghossian and David Velleman have advanced in several papers. On their version of the view it is colors that are falsely but unavoidably projected onto objects by our experience, and this conflicts with (1) and (2), which presuppose that objects literally do have colors. But there could be a version of literal projectivism which allows that external objects can be colored but holds that our visual experience projects onto objects properties -- Q and T, and their ilk -- which in fact are instantiated only in our minds. A different version of projectivism is what I have called "figurative projectivism"; according to this, Q and T would be properties that are in fact instantiated neither in external objects nor in our minds, but which certain experiences falsely (but unavoidably) represent external objects as instantiating. This seems to be the view that J.L. Mackie endorses and ascribes to Locke in his book Problems From Locke. But I take it that Harman agrees with me in rejecting all such projective error views. So, still assuming for the sake of discussion the possibility of spectrum inversion, he would agree that
4) Under normal circumstances, when George or Mary perceives an object to be Q or T, that object is Q or T.
This means that when both George and Mary are looking at a red object under normal circumstances, that object is both Q and T. Now one could perhaps tell a story according to which in addition to their colors objects have other intrinsic properties -- or other properties with as good a claim to being intrinsic as colors have -- and that Q and T are among these, and that for some reason George can perceive something to have Q only when the thing is red while Mary can perceive something to be Q only when it is green, and so on. But such a story would have no plausibility at all as applied to actual objects. I take Harman to agree that given the nature of the actual world, the only plausible view about the nature of Q and T is
5) Q and T are relational properties that objects have in virtue of their relations to the visual experiences of subjects viewing them.
It should be emphasized that (5) does not mean that Q and T are perceived as relational properties. Harman says that it is compatible with colors having a complex nature that "the concept of color as it figures in [perceptual representation] is simple and unanalysable in causal terms, because color is experienced as a simple basic property"; and the same could be said about the concepts of Q and T, as they occur in our perceptual representations.2 And certainly one can perceive what is in fact a relational property without perceiving it as relational. When an object feels heavy to one, one does not perceive it as having a property that is relative to people of one's own build and strength, yet it is true, even conceptually true, that the heaviness one feels is such a property.
The question now becomes, what relation is involved in such properties as Q-ness and T-ness, and what it is a relation to? Elsewhere I have used the term "phenomenal property" for properties of the sort Q and T are supposed to be, and I have suggested that a phenomenal property is a property an object has in virtue of actually causing, in a certain sort of way (the way involved in normal cases of vision), an experience having a certain qualitative character, i.e., having a certain quale. The suggestion Harman attributes to me is different. The relevant difference is that instead of specifying the mental term in the relationship in terms of qualitative character, it specifies it in terms of representational content. So, in the case of the phenomenal property of being Q, this would be constituted by a relation to an experience that represents the object as Q. But of course this makes the account circular. As Harman puts it, "In order to understand what the concept Q is, we need to understand what objects are Q to someone, but in order to understand what objects are Q to someone, we need to understand what the concept Q is." But that circularity has nothing to do with me, because the view that is afflicted with it is not mine. In a footnote Harman says that he has "oversimplified," and goes on to say "Shoemaker's actual account supposes that an experience has a certain intrinsic phenomenal feature x that is responsible for its representing something as Q. For something to be Q is for it to be such as to produce experiences with feature x." But the view ascribed to me in the footnote, unlike the view ascribed to me in the text of Harman's paper, is not circular. So "oversimplified" isn't quite the right word.
Let me take stock. We have seen that Harman seems committed to holding that if spectrum inversion is possible then color perception involves the perception of relational properties of objects that are defined, or constituted, by relations to visual experiences. And we have seen that it is unacceptable, because circular, to specify the mental terms of these relationships, i.e., the experiences, in terms of what properties of this sort they represent. It seems obvious that there is no prospect of our being able to specify these experiences in terms of any other representational properties they have. So it seems that it must be the case that having such a property consists in being appropriately related to experiences having a certain nonrepresentational property. I.e.,
6) Q and T are properties objects have in virtue of being related in certain ways to experiences having certain nonrepresentational properties.
It seems to me that Harman is committed to its being the case that (6) is true if spectrum inversion is possible and the story about George and Mary is coherent. But (6) is my view. And properties that can play the role that (6) assigns to the nonrepresentational properties are what I mean by qualia. So I think that Harman is committed to allowing that if spectrum inversion is possible, there are qualia as I understand them.
Let me guard against a misunderstanding. (6) of course will not do as a definition of the terms "Q" and "T". For it does not say what the nonrepresentational properties in question are; it just says that there are such and (by implication) that the one associated with Q is different from the one associated with T. What (6) says, generalized, is that for each different phenomenal property we can perceive an external thing to have, there is a different nonrepresentational property of perceptual experiences, such that having that phenomenal property consists in standing in the appropriate relation to an experience having the corresponding nonrepresentational property. It is a further part of my account that when an experience has a nonrepresentational property of this kind, it represents the experienced object as having the corresponding phenomenal property, and does so in virtue of having that nonrepresentational property. Supposing 'Q' to name such a phenomenal property, there is no suggestion here that the appropriate way to "define" 'Q', in the sense of explaining its meaning, is by specifying the corresponding nonrepresentational property. I might define it, in the sense of fixing its reference, by saying that it is the phenomenal property my experience of red things represent them as having. And while there may conceivably come a time when we can specify the nonrepresentational properties in neurophysiological terms, the only way of picking them out that we are likely to have in the foreseeable future is by means of such descriptions as "the nonrepresentational property that underlies my experience of red things."
Although Harman says that he is uncertain whether the idea of spectrum inversion is coherent, he takes the alleged circularity of my account as raising problems for it. And he mentions other reasons for rejecting qualia. Since the account he finds circular is not mine, and since the notion of qualia he attacks (the mental paint view) is not mine either, I find nothing in his paper that threatens either the claim that spectrum inversion is possible or my account of it in terms of phenomenal properties and qualia. But what is there to be said in favor of my account?
As is implied by what I have said already, any case for the possibility of spectrum inversion will at the same time be a case for my account. At any rate, Harman should think so, for given assumptions I share with him, my account gives the only way of accommodating spectrum inversion. I have believed for some time that reflection on the possibility of intrasubjective changes in color experiences provides strong support for the claim that the notion of intersubjective inversion is at least coherent.3 The conceivability of a partial intrasubjective inversion -- where the appearance of things of most colors remains unchanged, but two narrow bands of colors change place in the subject's color quality space -- should be uncontroversial. Such a change would involve changes in the subject's discriminatory and recognitional capacities that could be detected without relying on the subject's testimony. But it would seem that a series of such partial inversions could add up to a total inversion. And I think that if one allows the coherence of the supposition that there could be a total intrasubjective inversion, one will be hard pressed to find plausible reasons for denying the coherence of the supposition that there could be total intersubjective inversion.
But rather than pursue this line of thought further, I want to argue more directly for the view that a satisfactory account of perceptual experience requires qualia. While, as I have argued, the existence of qualia is implied by the view that spectrum inversion is possible, the converse is not true. I myself am prepared to allow that it may be that the structure of our color quality space is asymmetrical in ways that make it impossible that there could be behaviorally undetectable spectrum inversion for creatures having such a quality space. The asymmetries I have in mind include the fact, or alleged fact, that if we consider the four "unique hues," pure red, pure yellow, pure green, and pure blue, there are fewer discriminable shades between the last two than between the first two. This has the consequence that there is no mapping that maps each unique hue onto a different unique hue, maps any two discriminable shades onto two other discriminable shades, and preserves the similarity ordering of the hues. That our color quality space is asymmetrical in this way -- if it is -- does not rule out the possibility of there being creatures, otherwise very much like ourselves, whose color quality space is not asymmetrical in such ways and for whom undetectable spectrum inversion is possible. But I want to focus on what it is for there to be, or not to be, such asymmetries.
The existence or nonexistence of such asymmetries does not have to do with the structure of the system of colors, thought of as mind independent properties. Nothing in the physical basis of colors in objects or in the light answers to the distinction between unique and nonunique hues, and there are no relations between the physical bases of colors in objects or in the light that fix the similarity relations between colors as we perceive them. We are talking here about something "subjective" -- the color "quality space," to use Quine's term, of creatures having a certain sort of perceptual system. A creature's quality space can be thought of as (among other things) imposing a similarity ordering on possible stimuli; the position of stimuli in that ordering will have to do with such things as whether they are discriminable by that creature, and, if they are, how easy or difficult such discrimination is -- also, as Quine points out, with what sorts of inductions the creature is apt to engage in. We know that in the case of color there are stimuli that are physically very different -- involving combinations of wavelengths that are metamers -- that occupy the same place in the similarity ordering of creatures like us. What I want to suggest is that the notion of a quale is implicit in the very notion of a quality space.
Harman has always been willing to admit that our perceptual experiences have, in addition to the representational properties we are introspectively aware of, nonrepresentational properties that in some sense underlie these. These, he says, are physical (physiological) properties to which we have no introspective access. But he would surely allow that there is no a priori reason to think that whenever two visual experiences are phenomenally exactly the same, their physiological properties are the same; as a good functionalist, he presumably allows that being an experience of a certain phenomenal type is "multiply realizable" at the physical level. And assuming physicalism, having a quality space that imposes a similarity ordering on external stimuli must surely go with having a quality space that imposes a similarity ordering on the physical states of the creature that realize the perceptual experiences produced by the stimuli in the first quality space. This will group these states into equivalence classes, the members of each of which all realize experiences of a single phenomenal character, and it will order these equivalence classes in a way that corresponds to the phenomenal similarity ordering of the experiences realized by their members. Now there is a very strong sense in which we lack introspective access to individual physical states of these kinds. While there is a sense in which our perceptual system is sensitive to the presence of such a state, it is not sensitive in the same way to its absence, since it will be in the same conscious perceptual state whether it is in that state or in some other state belonging to the same equivalence class. But for each of these equivalence classes there is a higher-order state a creature is in just in case it is in some state or other belonging to that class. In the case of these higher-order states, there is sensitivity both to their presence and to their absence. Insofar as one is aware of whether two experiences are phenomenally the same or different, one is sensitive to whether they are realized by members of the same equivalence class or by members of different ones. As a first approximation, one could say that qualia are these higher-order states. Equivalently, they are functional states corresponding to these equivalence classes -- functional states of which the members of the corresponding equivalence class are realizations.
The reason this is just a first approximation is that experiences are phenomenally similar and different along a number of different dimensions. Experiences that are phenomenally different may be phenomenally exactly alike on one dimension while differing on another. So we need a more fine grained account. Not only must there be a similarity ordering on the physical states that realize experiences having complex contents; there must also be a similarity ordering on the physical properties of these states that contribute to their similarity or difference along particular dimensions. It will be equivalence classes of these, as grouped by such an ordering, that correspond to individual qualia.4
What all of this requires is something I have long insisted on -- that there must be functionally characterizable relations of phenomenal similarity and difference amongst experiences. Roughly, these are the functional relations between experiences that bestow the behavioral capacities, the recognitional and discriminative capacities, and the dispositions to belief, that go with having a certain quality space in Quine's sense. These similarity and difference relations are closely related to, but must be distinguished from, what might be called relations of "intentional" similarity and difference between experiences -- i.e., relations that hold between experiences in virtue of similarities and differences in their representational content. The phenomenal similarities and differences significantly constrain what relations of intentional similarity and difference there can be, but they do not by themselves determine them. For the relations of intentional similarity and difference are partly determined by relations to the environment. Supposing that the neural states of a brain in a vat could duplicate mine over an interval, it would have states that stand to one another in the same relations of phenomenal similarity and difference as mine do, but it would not necessarily have states that stand in the same relations of intentional similarity and difference as mine do -- if it were a brain that had always been in a vat, its experiences would not be of color in the sense in which mine are, and so could not be similar or different with respect to what colors they represent. However, given a background of externalist constraints on content, the qualitative similarities and differences can be said to determine the intentional similarities and differences; and, with the same qualification, the qualitative character of an experience can be said to determine its representational content.
I have said that we are "sensitive to" the presence or absence of qualia in our experience. In saying this I am not reneging on my acceptance of Harman's phenomenological point and returning to the "mental paint" conception of qualia. In part this sensitivity is a subpersonal affair; it amounts to the fact that (with the qualification just mentioned) the qualia instantiated in our experiences, and the relations of phenomenal similarity and difference between them, determine their representational contents and the relations of intentional similarity and difference between them. What in the first instance we are perceptually aware of are properties of things in our environments, and similarities and differences with respect to these; and what in the first instance we are introspectively aware of is the representational content of our experiences, i.e., what properties, similarities and differences there appear to be in our environments. That is what I am calling Harman's phenomenological point. But on reflection we can come to realize that the intentional similarities and differences we are aware of are grounded in phenomenal similarities and differences, and that these are grounded in the nonrepresentational features of experiences I am calling qualia. So there is a sense in which we can be aware of the latter; knowing what intentional similarities and differences there are, we can know what phenomenal similarities and differences there must be. But this is a far cry from the quasi-perceptual awareness required by the mental paint conception of qualia.
I should also say that in speaking as I have done of qualia, or experiences of phenomenal types, as "multiply realizable" I am not going back on my denial, argued elsewhere, that individual qualia are functionally definable. That denial was premised on the assumption that "qualia inversion" is possible. It is compatible with this that the notion of being a quale is functionally definable, and that is enough to give sense to speaking of qualia as multiply realizable, e.g., by neurophysiological states or properties. What the denial does imply is that we cannot specify an individual quale in purely functional terms. Any such specification must involve an indexical element, or an element of reference-fixing a la Kripke. For example, a particular quale might be specified as the one that characterizes experiences of red things in creatures neurophysiologically like me.
What does this view say about the nature of color? Given that I think that experiences of the same color can be qualitatively different -- either in different persons, or in the same person at different times -- I obviously cannot accept any view that defines colors in terms of the sorts of experiences they are apt to produce. (It should be noted that one doesn't have to believe in the possibility of behaviorally undetectable spectrum inversion in order to believe in this possibility. It is clear, I think, that there can be creatures who perceive at least some of the same colors as we do but have differently structured color quality spaces -- i.e., differ from us in the similarity and difference relations they perceive between things of the various colors. Clearly, the experiences of at least some colors would have to be qualitatively different in these creatures than they are in us.) What ought to be uncontroversial is that color similarity and color difference are "subjective" in the sense of being perceptual system relative. And perhaps the property of being a color (or a color "realization") is likewise perceptual system relative -- maybe there are reflectance properties of objects that are colors for bees and not for us, and ones that are colors for us and not for creatures of some other species.
As for our concepts of particular colors, I think I must assign them a bit more complexity than Harman does. Consider again George, who perceives redness by perceiving the phenomenal property Q. George, if he is an ordinary person, won't have a name for Q -- the only name he has available for describing his of-Q experiences is "red." But if he is willing to listen to reason, he can come to appreciate that his concept of Q -- looking like that -- and his concept being red are different. For those of us who can see, the reference of a particular color word gets fixed by reference to whatever phenomenal property is involved in perceiving things with that color. But we take the property the word refers to be one that objects have independently of whether they are being viewed, and we take ourselves to be entitled to say that other persons, and other animals, are perceiving something to be red when we are not in a position to have an informed opinion as to whether their experiences are phenomenally like the ones we have when we view red things.
Harman says that objective color "is plausibly identified with a tendency to produce a certain reaction in normal perceivers, where the relevant reaction is identified in part with reference to the mechanisms of color perception." What if normal perceivers were divided equally into the likes of George, who perceives red by perceiving Q, and the likes of Mary, who perceives red by perceiving T? It seems to me that it would then be true to say that red can be "identified" both with the tendency to produce of-Q perceptions in George-like creatures, and with the tendency to produce of-T perceptions in Mary-like creatures. Maybe "identified" isn't the right word; for how can one thing be identified with two different things? But if whatever has the one tendency will necessarily have the other, perhaps "identified" can be allowed to stand. In any case, neither of these "identifications" will be privileged. If in fact everyone is like George, or everyone is like Mary, then we can identify red with a tendency to produce a certain reaction in normal perceivers. But even still this identification will not be privileged; for it may be that red can also be identified with a tendency to produce some quite different reaction in bees. Yet another way of saying what red is will involve picking some object or kind of objects -- e.g., ripe tomatoes -- and saying that something is red just in case it is, for creatures with our sort of quality space, indistinguishable from such objects with respect to color, where indistinguishability with respect to color is in turn explained in terms of indistinguishability simpliciter in certain circumstances, namely those in which other properties are masked. None of these ways of picking out red gives a meaning equivalence; all are "reference-fixings" that depend on synthetic truths about the world. And I think that is all we can expect. Here I don't expect Harman to disagree. As best I can tell, such differences as there are between Harman and me about color, and about the concept of color, go back to the differences about color experience discussed earlier.5
1. "The Intrinsic Quality of Experience," in James Tomberlin (ed.), Philosophical Perspectives, 4, Action Theory and Philosophy of Mind. Ascadera, CA: Ridgeview Publishing Co., 1990. 31-52.
2. Rather than say that our experience represents colors, and properties such as Q and T, as simple, it would be better (if we want to avoid a projective error theory) to say that it does not represent them as complex. Likewise, rather than say that our experience represents such properties as monadic, it is better to say that it does not represent them as relational. Here our experience is best thought of as noncommittal about such matters.
3. See my "The Inverted Spectrum," Journal of Philosophy, 79, 7, 1982, 357-81. Also my "Lovely and Suspect Ideas," Philosophy and Phenomenological Research, LIII, 4, l993, 905-910.
4. See my "The Phenomenal Character of Experience," Lect. III of "Introspection and 'Inner-Sense'", Philosophy and Phenomenological Research, LIV, 2, 1994, 249-314.
5. My thanks to Harold Langsam for comments on an earlier draft of these remarks.