Sydney Shoemaker – Cornell University [Draft 4/13/02]

[This material is slightly revised and reorganized from two Whitehead lectures given at Harvard in April 2002.  The first NEH presentation (this file) corresponds to the first Whitehead lecture and the beginning of the second.  The second NEH presentation corresponds to the rest of the second lecture, plus “Introspection and Phenomenal Character”.]


            The words “content” and “character” in my title refer to the representational content and phenomenal character of color experiences.  So my topic concerns the nature of our experience of color.   But I will, of course, be talking about colors as well as color experience. 

            Let me set the stage by mentioning some things, some more controversial than others, that I will be taking for granted.  I assume, to begin with, that objects in the world have colors, and have them independently of being perceived to have them, and independently even of there being creatures capable of perceiving them.   I think, and this of course sets me apart from the many color irrealists among philosophers and color scientists, that any reasonable semantics for color terms, and any reasonable account of the reference of color concepts, should  yield the result that colors are properties of external things that are realized in certain of their physical properties, namely those responsible for their reflecting or emitting the light whose impact on our retinas is involved in causing our color perceptions. 

            This brings me to a further assumption that I shall be making, namely the truth of physicalism.  I take physicalism to be the thesis that all properties of things either are or are realized in basic physical properties, where basic physical properties are the properties that underlie the behavior and causal powers of inanimate things.  There are two ways in which the commitment to physicalism will figure in my discussion.  First, I assume that colors are physically realized properties.  Second, I assume that color experiences are physically realized.  These two commitments frame the problem I am discussing – how can colors be properties realized in the microphysical properties of things, and how can color experience be so realized?

            I don’t think there is any generally accepted account of what it is for a property to be “realized in” other properties.  For those who think, as I do, that properties are individuated by their causal features, it should seem plausible to say that property P realizes property Q just in case the forward looking causal features of Q are a subset of the forward looking causal features of P, and the backward looking causal features of P are a subset of the backward looking causal features of Q.[1]  Taking the relation of determinates to determinables to be a special case of the realization relation, we can illustrate this with red and scarlet – the forward looking causal features of red, i.e., the contributions it can make to causing various effects, are a subset of those of scarlet, and the backward looking causal features of scarlet, i.e., its being such that its instantiation can be caused in certain ways, are a subset of the backward looking causal features of red.

            I think the view that colors, insofar as they are properties of material things, are physically realized is one that has been widely accepted even by philosophers who would emphatically reject the view that any mental properties are physically realized.  It is a view Descartes clearly accepted, although in some passages his view about color sounds more like eliminativism than physicalism.  The view that colors are properties of material things, but properties of them that  are caused by, rather than realized in, their microphysical properties, is incompatible with what we know about the role of the microphysical properties in causing our experiences of color.

            That color experiences are realized in the physical, presumably in physical states of our brains, is of course much more controversial.  This is not something I shall argue for.  It figures here as a widely accepted assumption that for those who hold it poses part of the puzzle about color and color experience.  This is a special case of a puzzle that has figured centrally in recent discussions in the philosophy of mind – for example, Thomas Nagel’s discussion of “what it is like” to be in conscious mental states, and Frank Jackson’s example of the black and white room bound physiologist who knows all of the physical facts about color perception but does not know what it is like to see red.  Appropriating Joseph Levine’s term “explanatory gap,” I will refer to this problem as the subjective explanatory gap.[2]   It is, in brief, the problem of how, given what it is like to be in subjective mental states, these states can be realized in physical states of our brains.   Nowhere does this problem arise more vividly than in the case of color experience.  

            But there is also an objective explanatory gap problem, which is in some ways  more fundamental and certainly of much greater antiquity.  This is the problem of how colors, given their perceived nature, can be, or be realized in, physical properties of things, given what we know about these physical properties.  This is a central case of the problem Wilfrid Sellars raised by asking how, if at all, we can reconcile the “manifest image,” embodied in the common sense view of the world and our ordinary experience of it, with the “scientific image.”[3]  And it is a problem that clearly exercised presocratic Greek atomists like Democritus and seventeenth century thinkers such as Galileo, Descartes, and Locke. 

            A time-honored response to this problem, found in all of the thinkers just mentioned, is that of “kicking upstairs into the mind” the phenomenal or qualitative character we confront in our perception of color.  The idea is that what we naturally think of as the perceived nature of the color is really the nature of the experience produced in us by the object perceived as having the color.  Sometimes, as in Democritus and Galileo, and in recent theorists such as Paul Boghossian and David Velleman, this takes an eliminativist form.[4]  Strictly speaking, external objects don’t have colors, and our perception of them as colored is an illusion – perhaps involving the projection onto them of features of our experiences.  Sometimes it takes the form of a dispositional theory of color, according to which an object’s having a certain color is simply a disposition, or in Lockean terms a “power,” to produce experiences of a certain sort in perceivers under certain conditions.  And sometimes it takes the form of a view that allows that colors are objective and nondispositional properties of external things, but holds that the phenomenal character of our experiences of colors is only contingently related to what colors they represent, so that in principle different observers, with different sorts of perceptual systems, could perceive the same colors by means of phenomenally very different experiences, and could perceive different colors by having phenomenally identical experiences. 

            If by “intersubjective spectrum inversion” one means a situation in which two  different observers differ in the way just described in how the phenomenal character of their color experiences relates to the colors of things, and yet both veridically perceive the colors of things, then it is only the last of these views that is committed to the possibility of such spectrum inversion.  The eliminativist view denies that there is any veridical perception of the colors of things, while a dispositionalist view that individuates the colors by the phenomenal character of the experiences they produce cannot allow that the same color is veridically perceived via phenomenally different experiences.  But in a broader sense, all of these views allow the possibility of intersubjective spectrum inversion.  All hold that the relation between the phenomenal character of color experiences and the intrinsic nature of perceived objects is contingent in a way that allows different subjects to differ systematically in the phenomenal character of their experiences of objects of a certain intrinsic nature without these subjects differing with respect to the veridicality of their experiences. 

            As has often been observed, employing the strategy of solving the objective explanatory gap problem by “kicking the phenomenal character upstairs” can seem, when done by physicalists, to be a case of moving the bump in the rug from one place to another.  A dualist like Descartes can construe this as kicking the phenomenal character into the nonphysical realm of the mind; but for the physicalist there is no such nonphysical realm.  The experiences having the phenomenal character will itself  have to be physically realized.  And this of course poses what I have called the subjective explanatory gap problem.  It can seem that we have just traded one explanatory gap problem for another.

            The subjective explanatory gap problem has to do with the phenomenal character of perceptual experiences.   Discussion of this has been transformed in the last decade or two by the widespread acceptance of what has come to be called the “transparency” of perceptual experiences.  The classical expression of this is G.E. Moore’s observation in “The Refutation of Idealism,” published almost a century ago, that the sensation of blue is as if it were diaphanous; if one tries to introspect it one sees right through to the blue.[5]  This has been seen by many as showing that it is a mistake to suppose that the perception of colors involves the possession by perceptual experiences of “qualia” or “sensational properties” that give them their phenomenal character, i.e., determine “what it is like” to have them.  Treating this as a mistake goes with a certain sort of “representationalist” view about the phenomenal character of perceptual experiences, one that says that the phenomenal character of a color experience simply consists in its representing, its being “of,” a certain array of colors in the scene being viewed.  I will call this “standard representationalism,” to distinguish it from a different version of representationalism I will be defending later on.  Standard representationalism has been held by a number of recent writers, including Gilbert Harman, Bill Lycan, Michael Tye, Fred Dretske, Alex Byrne, and David Hilbert.[6]  In his paper “A Simple View of Color” John Campbell gives a vivid expression of an apparent consequence of this view when he writes that “the qualitative character of a colour-experience is inherited from the qualitative character of the color.”[7]  The wording here might suggest a resemblance theory, according to which color experiences resemble the colors of things, but I take it that Campbell’s view is rather that the experiences “inherit” their qualitative character from the colors just in the sense that their having the qualitative character they do is simply a matter of their representing, perceptually, the qualitative character of the colors.

            Standard representationalism has no room for the possibility of spectrum inversion, unless the inversion is taken to involve systematic misperception on the part of one of the invertees.  Given that it identifies the phenomenal character of a color experience with that aspect of its representational content that has to do with its representation of color, it cannot allow that the phenomenal character of such experiences and their color content can vary independently of one another.

            It can easily seem that standard representationalism provides a quick solution to the subjective explanatory gap problem.  The problem of how physical states of the brain, or states realized in these, can have representational content is widely regarded as a tractable one.  There are causal-correlational accounts, teleological accounts, and cognitive role accounts that are compatible with physicalism, and while none of these commands universal assent it is widely thought that something along one or more of these lines must be right.  If so, and if the phenomenal character of perceptual experiences is just a special sort of representational content, there is an explanation in the offing of how it can be physically realized.

            I said that the time-honored response to the objective explanatory gap problem involves kicking the phenomenal or qualitative character of color upstairs into the mind.  One could likewise say that standard representationalism solves the subjective explanatory gap problem by kicking the phenomenal character downstairs, into the external world.  But if moving the bump in the rug into the mind does not get rid of it, neither does moving it in the opposite direction.  For now we have the objective explanatory gap problem back again.  How can color as we perceive it be a micro-physically realized property?  How can it be such a property that our experiences of color represent?   And about this proponents of standard representationalism seem to me to be, for the most part, in a state of denial.

            This sets the stage for the rest of what I will be saying in these lectures.  I think that the Moorean “transparency” claim is, property understood, correct.  And as I have already mentioned, I accept a version of representationalism about phenomenal character.  This is a view according to which the phenomenal character of color experiences consists in an aspect of their representational content, but according to which the properties represented in this part of their content are not the colors but other properties closely related to them.  This view will be presented in my second lecture.  In the remainder of the present lecture I will be explaining why I think that standard representationalism is mistaken.  As I have already indicated, I take standard representationalism to deny the possibility of spectrum inversion, except on an understanding of it on which anyone whose color experience was inverted relative to that of a veridical perceiver of the colors would have to be systematically misperceiving the colors.  So among other things, I will be arguing that an acceptable account of color perception must allow the possibility of spectrum inversion without systematic misperception.  I will postpone till my next lecture a consideration of the question of where this leaves the two explanatory gap problems I have distinguished. 

            There are a number of considerations that seem to show that in some sense the ways things appear to us with respect to color are partly a function of the nature of our perceptual system, and so pose a problem for standard representationalism and for the idea that our experiences simply inherit their qualitative character from the colors they represent.  First of all, there is the enormous variety of the physical bases of what we perceive as a single color.  Many different mixtures of lights of different wavelengths are metamers, meaning that they produce indistinguishable color experiences in human subjects.  Surface color is realized in properties called surface spectral reflectances, or just reflectances, each of which is specified by saying what percentage of the light of each wavelength that surface reflects.  And under normal lighting conditions, vast numbers of different reflectances will be indistinguishable by ordinary perceivers and perceived as the same color.  Moreover, for each reflectance there are a vast number of physically very different microphysical properties that will realize it, i.e., will confer the set of light-reflecting dispositions that define it.  So each determinate shade of color is realized in a number of different reflectances, each of which in turn is realized in a vast number of microphysical properties.  And of course not all colors are surface colors.  The blue we see on some surfaces is realized in yet other ways in transparent glass bottles, in neon lights, in lakes and oceans, and in the sky. 

            Where is the qualitative character that all of these realizers of a particular shade of color share, and is inherited by our experience of that shade?  If we ask why it is that things having these different properties look alike with respect to color, the answer will come in part from the part of physics that deals with the reflection, absorption, and emission of light.  But when we ask why the different metamers look alike, the answer lies in part in the nature of our perceptual apparatus.  It is a system so constructed that for each of a large number of different inputs there is a single conscious response.  And it is so constructed that the various stimuli that impact on it form what Quine called a “quality space” – where the similarity ordering constituting such a quality space is always relative to a certain sort of perceptual system.

            A further point is that there is a structure to the phenomenal character of color experiences that seems to have no counterpart in their physical realizers, and has its explanation in the nature of the mechanisms by which we perceive color.  There is a phenomenal distinction between the “unique hues” – unique red, unique yellow, unique green, and unique blue – and the “binary hues”, such as orange and purple and chartreuse.  Nothing in a particular combination of wavelengths, or in the physical properties that account for the reflection or emission of light with that combination of wavelengths, provides a clue as to whether the reception of light with that combination of wavelengths will produce an experience of a unique hue.  Or rather, none of this will provide such a clue until we combine it with information about the constitution of our opponent processing visual system.  It is the nature of this system that accounts for what combinations of wavelengths, and what reflectances, are perceived as unique hues.[8]

            What I have been saying points to an aspect of the objective explanatory gap problem that has no counterpart in the subjective explanatory gap problem.  An important part of explaining phenomenal character is explaining similarity, including identity, of phenomenal character.  And, as we have seen, a central part of the objective explanatory gap problem is explaining how it is that what are physically very diverse properties can be realizers of the same shade of color, and, more generally, how it is that physical properties can have a similarity ordering, qua color realizers, that  bears no apparent relation to any similarity relations that would seem salient to someone who was taxonomizing them simply as physical properties.  It is hard to imagine a plausible explanation of this that is not “subjectivist” to the extent that it takes the similarity ordering of color realizers as determined by a similarity ordering of the experiences they produce under certain conditions.

In the case of the experiences of color there is not the same problem of accounting for the similarity ordering.   It is may well be that the physical states that are candidates for realizing color experiences would not seem, considered just as physical states, to have a similarity ordering that corresponds to the phenomenal similarity ordering of the experiences they realize.  But the very considerations that would qualify them for being realizers of color experiences would serve as a basis for such an ordering.  For physical states to count as realizers of color experiences they would have to play the functional role of experiences vis a vis mediating connections between sensory stimuli and behavior, including of course the effect on the subject’s perceptual beliefs.  This means, for example, that ones realizing phenomenally different experiences would have to be so related that under certain circumstances their simultaneous or successive occurrence would play a role in producing the behavior that shows that the subject can discriminate between the stimuli that produce them, while ones realizing phenomenally very similar experiences would have to be so related that their occurrence would have a correspondingly different behavioral effect.  Likewise, ones realizing phenomenally similar or identical experiences would have to be so related that when one is followed by the other after an interval there would be, under certain circumstances, the behavioral manifestations of recognition.  And what similarity and difference relations hold among experiences would have to play a role in the fixation of belief, both beliefs about the environment and beliefs about the experiences themselves. 

It seems clear that an account of what makes different physical properties realizers of the same or similar colors has to bring in their aptness to produce phenomenally identical or similar color experiences.  Is there a comparable way in which  an account of what makes different neurophysiological states realizers of phenomenally identical or similar color experiences has to bring in the identity or similarity of colors?  Well, if phenomenal similarity of color experiences is what we might call intentional similarity, similarity in representational content, and if the relevant content has to do with color, then of course it is a requirement on neurophysiological states realizing phenomenally identical or similar color experiences that they realize experiences that represent the same or similar colors.   But that doesn’t mean that we have to settle what physical relations count as sameness or similarity of color before we can settle what counts as sameness or similarity of color experiences.  Suppose that we are investigating the color perception of bees, and undertake the job of discovering two things – what physical properties realize the colors bees see, and what states of bees realize experiences of bee color.  Plainly the first thing we would have to do is find out what kinds of discriminatory and recognitional capacities bees have.  That involves, in the first instance, finding what things look similar and different with respect to color to bees – i.e., what features of things are responsible for the light that elicits certain sorts of discriminatory and recognitional behavior in them.  If we take it as criterial of bee-color similarity that, normally, things that look color-similar to bees are similar in bee-color, this will give us a way of identifying what things in the world stand in the relations of bee-color similarity and bee-color difference, which in turn will give us a way of telling what properties of objects are realizers of bee-colors.  This brings out the sense in which similarity of color experience is more basic than similarity of color.

            But there is a version of standard representationalism that holds promise of being able to acknowledge a kind of dependence of the phenomenal character of  color experience on the perceptual system of the perceiver, while holding that the colors perceived are mind independent and that the phenomenal character of the experiences just consists in their representing these properties.  This is a version of what David Hilbert has called “anthropocentric realism,” and of the view suggested by the title, as well as the contents, of Allan Gibbard’s recent paper “Visible Properties of Human Interest Only.”[9] This view does not hold that colors, color similarities, and the perceived nature of the colors are in any sense the product of our perceptual system.  Rather, it holds that our perceptual system selects certain properties of external things and certain relations between them to be the colors and the relations of color similarity.  What determines what properties and relations are selected is the quality space of the subject, i.e., what color stimuli are indistinguishable by the subject, what relations of relative similarity and difference are perceived between objects, and so forth.   In the words of David Hilbert and Mark Kalderon, who propose such a view in a recent  paper, “a pre-existing relation only counts as similarity in hue, saturation or brightness in virtue of an antecedent classificatory function of the visual system.  Facts about visual experience fix which of the similarities among objective properties count as similarities with respect to color.”[10]  This is presented as a version of standard representationalism that denies the possibility of spectrum inversion.

            Let me say how I understand the idea that our visual system “selects” relations to be relations of color similarity.  Here it helps to remember that there is a sense in which similarity is cheap.  For any ordered set of properties we can define a similarity relation such that the degree of similarity of two properties in the set is determined by how close they are to each other in that ordering.  Perhaps most of these should count only as relations of “quasi-similarity.”  But what determines which of these relations count as “real” or “genuine” similarity relations?  A first step towards an answer is to say that such a relation is a genuine similarity relation if it makes properties similar to the extent that their instantiation bestows similar causal powers.  But what sorts of causal powers are relevant will vary depending on our interests.  In the case of sensible properties of things, the relevant powers include the powers to affect the experiences of perceivers; and in the case of the so-called “secondary qualities” these are close to being the only powers that are relevant.  Powers to affect experiences will be grounded in powers to affect the physical states of perceptual systems.  And given that a perceptual system realizes a repertoire of perceptual experiences standing in certain similarity relations, there is an obvious sense in which its physical nature determines what properties bestow the powers to produce in the possessor of the system experiences belonging to that repertoire, and what relations among these properties bestow similarities with respect to these powers.  In this sense the nature of the perceptual system “selects” what properties are to count as sensible properties, and what relations among them are to count as similarities with respect to these properties.

            Hilbert and Kalderon present the selection view as one according to which color content is Fregean rather than Millian.  It is a natural thought that the Fregean distinction between sense and reference can be used to explain away the disparity between colors as we experience them and the surface properties of objects as we know them from science.  The idea would be that while a color experience’s having a certain phenomenal character  consists in its having a certain representational content, that content is to be individuated not just by what color is represented but by the sense or mode of presentation by which that color  is represented.  This is different from any mode of presentation by which the same property might be represented in the terminology or concepts of physics, and this accounts for the seeming disparity.  Now, the idea that the phenomenal character of color experiences is determined by modes of presentation rather than by properties referred to does not seem to square with the idea that this character is simply “inherited” from the colors represented.  And there is of course a danger that by making the mode of presentation a feature of the perceptual experience this view sneaks in by the back door the qualia which standard representationalism is anxious to reject.  But Hilbert and Kalderon have a view that avoids at least the second of these worries, and might seem to avoid the first.  What partly determines the content of color experiences is their conceptual role, and this is fixed by their position in the psychological color space.  This serves as the mode of presentation.  So it is the role of the experiences in perceiving color similarities and differences that partly determines their contents – and it is this that determines their phenomenal character.  They say that “some particular aspect of color content is responsible for color phenomenology,” and they identify that aspect by saying that on their view “the phenomenal properties of color experiences are identified with the structural constraints governing them (as determined by their position in the color space.”  The same view, that phenomenal character is relational and determined by position in a psychological quality space, can be found in Austen Clark’s recent book A Theory of Sentience.[11]

It falls out from the “selection gambit,” as I will call it, that tetrachromats perceive different colors than any trichromats, and that trichromats with differently structured color spaces will perceive different colors.[12]  The properties selected to be the colors, and the relations selected to be the relations of color similarity, can be different for creatures with different sorts of perceptual systems.   So while it will be true on this view that the phenomenal character of color experiences will be determined in part by the nature of the subject’s perceptual system, this does not mean that it can vary independently of the color represented, as in the inverted spectrum scenario.  If different creatures have phenomenally different color experiences when perceiving the same objects, that means on the selection view that their experiences represent different colors.  And this doesn’t mean that one or the other of them is misperceiving; one creature can be veridically perceiving the object to have a certain color of one sort, say a certain human color, while another is veridically perceiving the same object to have a certain color of a different sort, say a certain bee-color.  The same light can carry information about different colors to different perceptual systems.

It should be noted that as I am interpreting the selection view, it gives an internalist account of phenomenal character – for the structure of the color quality space, determined by a “classificatory function of the visual system,” is an internal matter.  This is one of the attractions of the view; for one thing about which internalism seems intuitively right is the “what it is like” of perceptual experiences.  Most versions of standard representationalism seem committed to an externalist account of phenomenal character, and so, assuming physicalism about color, to the unpalatable view that an experience’s having a certain phenomenal character consists in its representing some physical property of objects, leaving us with the mystery of how any of the candidate properties could be such that its representation confers the phenomenal character we  confront in our experience.  If the selection gambit can combine internalism about phenomenal character with standard representationalism, that will strongly recommend it.  I should mention, however, that despite the passages that support a reading of Hilbert and Kalderon as internalists about phenomenal character, this reading conflicts with their official view that the phenomenal character of color experiences is identical with their representational content.  For this content is not determined by position in the psychological color space alone, but rather by this together with “the relations normal subjects bear to their native environment” (p. 301).  So on another reading the account of phenomenal character is partly externalist after all.  But to adopt the view suggested by this reading is to relinquish the chief benefit promised by the selection view.  So I will stick with the internalist reading suggested by the claim that it is a “particular aspect” of color content, namely position in the psychological color space, that determines color phenomenology.

I should emphasize that I have focused on this view because I think that there is a good deal that is right in it.  The idea that our perceptual systems “select” objective properties and relations to be colors and color similarity relations seems to me right; and it seems to be the view that is needed if we are to reconcile the view that colors are objective properties of things with the view that similarity of color experiences is more basic than similarity of color.  What I question is whether this view will serve as a version of standard representationalism.   It could do so if we could assume that perceptual systems having different color quality spaces would always “select” different properties to be the colors.  But that assumption seems wrong.  It seems possible that the same set of determinate shades of color might have one similarity ordering relative to one sort of perceptual system, and a different similarity ordering relative to another.  Certainly it seems possible that sets of reflectances should have different similarity orderings in the experiences of creatures with different sorts of perceptual systems.  Or it might be that some but not all of the sets of reflectances that are selected as surface colors by one sort of perceptual system are among those selected as surface colors by a different sort of perceptual system, so that in the quality spaces of one of these systems these colors stand in similarity relations to properties that are not selected as colors at all by the other perceptual system.   It is compatible with this that within the experience of a given perceiver or kind of perceiver, each color is picked out by the role it plays in the similarity ordering – so that positions in this ordering serve as “modes of presentation” of the particular colors.  But suppose that subjects S1 and S2 have differently structured color quality spaces, but that one of S1’s colors, call it C1, has as surface color realizers the same set of reflectances as one of S2’s colors, call it C2.  More generally, suppose that C1’s total set of realizers, those for colored lights and transparent or translucent solids as well as surface colors, is identical with C2’s total set of realizers.  Nothing in the selection account rules this out, and it seems perfectly conceivable.  If the possible realizers of C1 are the same as those of C2, it is hard to resist the conclusion that C1 and C2 are the same color.  But if they are the same color, then perceptual systems with differently structured experiences spaces can “select” the same property in the world as one of the colors while selecting different similarity relations between it and other colors.  Assuming that this would not involve systematic misperception on the part of the possessors of one of the perceptual systems, and there is no reason to think it would, this contradicts the view of Hilbert and Kalderon that the colors are individuated by their similarity relations.   And if, as they claim, the phenomenal character of color experiences is determined by what color similarities they represent, it would seem that it gives us a case in which veridical experiences of the same  color, in the same viewing conditions, differ in phenomenal character.  Given this possibility, it certainly does not seem that the phenomenal character of color experiences can be simply inherited from the nature of the colors they represent.

What was just presented was a case in which creature S1’s experiences representing a certain color differ phenomenally from creature S2’s experiences of that same color because S1’s and S2’s visual systems select different similarity relations for that color.  Can we also have a case in which S1’s experiences of  color A are phenomenally like S2’s experiences of a different color B, and S2’s experiences of color A are phenomenally like S1’s experiences of color B?  That is, can we get this sort of spectrum inversion on the assumption that the visual systems “select” properties and similarity relations to be colors and color similarity relations.  I think we can.

To show this I need to introduce a distinction between two sorts of similarity spaces.  The color experience space for a given creature consists in the similarity ordering of the color experience types belonging to that creature’s repertoire of sensory states.   Since any experience of color will at the same time be an experience of other properties as well, what this is really an ordering of is a class of representational features of experiences, namely those involved in the representation of color – I will say something about what these are in my next lecture.  The color property space for the creature will consist in the similarity ordering of the properties selected as colors by that creature’s visual system.[13]  To a first approximation, there will be an isomorphism between these – there will be a way of pairing experience types, or experience features, with properties of objects such that the items in the one set will stand in a network of similarity relations which is the same as that the paired items in the other set stand in.   This pairs experience types or features with properties they represent.[14]  It is now generally accepted that our similarity space for colors has an asymmetrical structure.[15]  Assuming that a creature’s color experience space and its color property space have an asymmetrical structure, and of course the same asymmetrical structure, there will be just one way of pairing the experiences, or experience features, and properties of objects that yields such an isomorphism.  And the property represented by an experience of a certain type under normal conditions, and standardly tracked by it, will occupy the position in the color property space that corresponds to that occupied by that experience type in the color experience space.  The reason this is only a first approximation is that the pairing of color experience types with color properties can vary with viewing conditions.  So it would be more accurate to say that there is an isomorphism of the sort described for each set of viewing conditions.  But this is a complication I will allow myself to ignore.

Now it seems perfectly possible that two creatures, S1 and S2, should have color property spaces that are different but nevertheless have the same structure.  Their visual systems differ in such a way that they select somewhat different properties as the colors, and somewhat different relations as the relations of color similarity and difference.  As we might suppose, this is partly because in some cases what are metamers for one of them are not metamers for the other.  But it might nevertheless be true that the colors perceived by Sl can be paired with those perceived by S2 in such a way that the similarity relations observed by S1 between things with his colors are the same as those observed by S2 between things with the corresponding S2 colors.  This of course requires that the color experience spaces of the two creatures have the same structure.  As we have seen, Hilbert and Kalderon seem to hold, in common with Austen Clark, that because phenomenal character is determined by position in a psychological quality space, in such a case each experience in the repertoire of one of these creatures must be phenomenally just like its counterpart in the repertoire of the other.  I will question that view later on.  But while I do not think that each experience in the repertoire of one of my two creatures would have to be phenomenally just like its counterpart in the repertoire of the other, I agree that this could be the case.  So let’s suppose that it is.  It also seems possible that while the two visual systems select different sets of properties, there is some overlap between the sets they select.  That is, some properties are selected by both of them.  And we can suppose that in some cases a property selected by both visual systems occupies a different position in the color property space of the one than it does in the color property space of the other.  If this is so, then the experiences of S1 that  represent such a property will have to be phenomenally different from the experiences of S2 that represent that same property.   But if S2’s experiences with a certain phenomenal character don’t  represent the color that S1’s experiences with that phenomenal character represent, even though that color is one that both perceive, those experiences of S2 must represent some other color – and why shouldn’t that also be a color that both of them perceive?  If that is possible, it is possible that there should be two such properties, A and B, such that S1’s experiences of A are phenomenally just like S2’s experiences of B, and vice versa.  

            It may be objected that the claim that color similarities are perceptual system relative is at odds with things we know a priori to be necessarily true of the colors, e.g., that orange is more similar to red than it is to green.  For aren’t I saying that there might be perceptual systems such that creatures with those perceptual systems could say, truly, that orange is more similar to green than to red?  Well, no.  They couldn’t say that, truly, and mean by it what those words would mean coming from my mouth.  When I use those words, the similarity relations I am talking about are those selected by my sort of perceptual system – and that makes it necessarily true that orange is more similar to red than to green.  Of course, when the other folk use the same words, the similarity relations they are talking about are those selected by their sort of perceptual system, and that makes it true for them to say “Orange is more similar to green than to red,” where the terms “orange,” “green” and “red” refer to the very same properties they refer to in my statement.

            I said earlier that if the possible realizers of color C1 are the same as the possible realizers of color C2, it is hard to resist the conclusion that C1 and C2 are the same color.  And my subsequent argument has rested on that conclusion.  Those who do not like where that argument has taken me may feel that it is after all not so hard to resist that conclusion.

One way of resisting it is to be a certain sort of dispositionalist about colors.  Let a relational power be one that is grounded not only in the intrinsic properties of the thing that has it but also in the existence of a kind of things that in suitable circumstances would be affected in a certain way by the exercise of the power.  Adapting  Robert Boyle’s famous example, the power of a key to open a certain door will be a relational power if it is one that we can deprive the key of by changing the lock.  If colors were relational powers, two different colors could be alike in what intrinsic properties of colored objects they are grounded in, and yet different because the existence of the powers would involve different sorts of perceivers – one is a power to produce experiences of one sort in one sort of perceiver, and the other is a power to produce a different sort of experience in another sort of perceiver.  They would not be necessarily coextensive, because the existence of perceivers of the one sort would be independent of the existence of perceivers of the other sort.  This view has the cost of implying that before there were perceivers there were no colors.  That seems to me reason enough to reject it.

            Could one maintain instead that colors are intrinsic powers, i.e., powers grounded only in the intrinsic properties of the things that have them, and still maintain that two different colors could be alike in what intrinsic properties ground them, i.e., are their “categorical bases”?  It does seem natural to say that the power to produce experiences of type A in perceivers having one sort of visual system is a different power than the power to produce experiences of type B in creatures with a different sort of visual system, even though these are powers the things have whether or not creatures with those sorts of visual system exist, are grounded in the same intrinsic properties of their possessors, and are necessarily coextensive.

But suppose I am presented with a coin, and told that it has two different intrinsic powers.  It gets you a Coke if inserted into the slot of an Alpha machine, and it gets you a Pepsi if inserted into the slot of a Beta machine.  It might turn out that Alpha and Beta machines are mechanically identical, the only difference between them being that Alpha machines are stocked with Coke and Beta machines are stocked with Pepsi.  So the proximate effect of inserting the coin is the same whether it is inserted in an Alpha machine or in a Beta machine.  Here it seems clear that in no sense are different intrinsic properties of the coin involved in producing the Coke output than in producing the Pepsi output.  If we have different intrinsic powers here, these powers are not different intrinsic properties.  The difference seems conceptual rather than ontological – we have a single property in virtue of having which an object falls under two different dispositional concepts.  And it might be like this with colors.  Let C1 be a color perceived by an observer with one sort of perceptual system, and C2 a color perceived by an observer with a different sort of perceptual system.  It might be that these are grounded in the same intrinsic properties of objects, and that the proximate effects of the presentation of something with C1 on an observer of the one sort are the same as the proximate effects   of the presentation of something with C2 on an observer of the other sort, although the experiences eventually produced are different.   If  colors are intrinsic properties, C1 and C2 are the same color.   We can allow that an object having this property has two different “intrinsic powers,” these being powers to affect the two sorts of observers.  But these won’t be different colors – they will be, perhaps, the same color under two different descriptions. 

The inverted spectrum scenario I have described is not the one that has been most frequently discussed in the literature.  The most frequently discussed case is that in which the inversion is behaviorally undetectable – the two creatures make all of the same color discriminations, and apply color words in exactly the same way.   That possibility requires that the color experience space and the color property space have a symmetrical structure.  As already mentioned, our color experience space and color property space do not have a symmetrical structure.  This comes out in a variety of ways.  One is that there are more discriminable shades between unique red and unique yellow than there are between unique blue and unique green – so there is no mapping which both maps unique hues onto unique hues and preserves the resemblance relations among the hues.  Another is the fact that whereas blackened  blues are perceived as blues, blackened yellows are perceived as a different color, brown; this sort of fact precludes a mapping that maps color categories onto other color categories of the same size.[16]  All of this seems to preclude the possibility that there should be creatures who are spectrum inverted relative to us in a behaviorally undetectable way.  The inversion I have described, which involves visual systems that differ somewhat in what relations they “select” to be the relations of color similarity, would of course be behaviorally detectable.

It is nevertheless a question of some interest whether it is possible for there to be creatures with symmetrical color quality spaces for whom undetectable spectrum inversion would be a possibility.  A number of writers, including Daniel Dennett, Austen Clark, and Hilbert and Kalderon, argue that it is not.   Hilbert and Kalderon point out that given that our own color experience space is asymmetrical, and given that color experiences are individuated by their similarity and difference relations to other experiences in the same experience space, none of our color experiences could be the same as any of the color experiences of the hypothetical creature with a symmetrical color experience space.  They take this to show that we cannot conceive of the experiences of such a creature in the way we would have to be able to do in order to conceive of a creature spectrum inverted relative to it.   But they argue further that it is impossible that there should be such an experience space.   They think that because the phenomenal character of experiences is determined by the similarity relations they stand it, there must be asymmetries in the structure of the space in order for the phenomenal character of different experiences in it to be different.   Here they cite with approval a passage from Daniel Dennett – directed, as it happens, against me – in which he says that “what anchors our naive sense that there are such properties as qualia are the multiple, asymmetrical, interdependent  sets of reactive dispositions by which we acquaint ourselves with the sensible world,”  and that for the hypothetical creatures with a symmetrical experience space “what it was like to have one sort of experience would not differ at all from what it was like to have a different one.”[17]

Let me take up these points in order.  I agree that if there were creatures with symmetrical color experience spaces, none of their experiences could be phenomenally the same as any of ours.  And I agree that we would not be able to imagine what it is like to have their experiences.  Does this show that we could not be in a position to refer to their experiences in a way that would enable us to describe creatures that are spectrum inverted relative to them?  No, it does not.  First of all, no reason has been given why we could not, by testing the discriminatory and recognitional capacities of these creatures, map the structure of their color quality space and find that it indeed has a symmetrical structure.  And we could have good behavioral evidence that certain stimuli always produce in them, under certain conditions, experiences identical in phenomenal character.  This would give us a way of fixing the reference of terms designating their kinds of color experience; e.g., we might fix the reference of the term “red*” with the description “the phenomenal character of the color experiences they have when viewing ripe tomatoes in daylight,” and the term “green*” with the description “the phenomenal character of the color experiences they have when viewing unripe tomatoes in daylight.”  We could then describe the hypothetical case of creatures who also have a symmetrical color experience spaces but who get green* experiences from ripe tomatoes and red* experiences from unripe tomatoes.

I turn to the argument that purports to show that symmetrical color experience spaces are an impossibility.  That argument rests, as I said, on the claim that the phenomenal character of experiences is determined by the similarity relations they enter into.   But that claim is ambiguous.  It might mean simply that the phenomenal character of any experience in a creature’s repertoire is determined by its relations of similarity and difference to the other experiences in its repertoire.  If you fix the phenomenal character of those other experiences, and fix its relations of similarity and difference to them, then you fix its phenomenal character.   So, e.g., if an experience has just the similarity and difference relations to my experiences of red, blue, etc. that my experience of green has to them, then it is phenomenally just like my experience of green.   That claim I accept.  But it won’t serve Hilbert and Kalderon’s purposes, because there is nothing in it to rule out the possibility of a symmetrically structured experience space.  What they require is the stronger claim that given a set of experiences, the holding of the similarity and difference relations amongst them constitutes their having the phenomenal characters they have.  This does rule out symmetrically structured experience spaces.  In such an experience space there would have to be pairs of experiences whose members would be indiscernible by any description of the set of experiences to which they belong which speaks only of the similarity and differences relationships holding among its members – and on the view of Hilbert and Kalderon, and of Austen Clark, the supposedly different members of these pairs would have to be identical.[18]   Clark suggests that to define terms for the qualitative states that figure in the experience space “one must form the Ramsey sentence for the entire structure of qualitative relations, and associate each particular term with a particular Ramsey correlate” (p. 18).   This works only if the similarity and difference relationships give the space an asymmetrical structure; applying it to a symmetrical space identifies qualitative states that are supposed to be distinct.  So the assumption that qualitative states are relational yields the result that a symmetrically structured experience space is impossible.   

I suspect that such plausibility as the relational view has derives partly from the failure to distinguish it from the first reading of the claim that the phenomenal character of experiences is determined by their similarity and difference relations to other experiences.  On the first reading that claim simply relies on the point that phenomenal similarity and difference are internal relations that hold in virtue of the phenomenal character of their terms, and does not imply that phenomenal characters are themselves relational properties.  The view that they are relational properties seems to me phenomenologically implausible.  And it goes against it that, as I claimed earlier, it seems conceivable that we could discover on behavioral grounds that a creature had a symmetrically structured color experience space.[19]

            Rejecting the relational view allows the possibility of symmetrically structured color quality spaces, and with it, I think, the possibility of behaviorally undetectable spectrum inversion involving creatures with such spaces.  It also allows the possibility of creatures with color quality spaces having the same asymmetrical structure as our own who are such that none of their color experiences are phenomenally like any of ours.  This would be a case of what has been called “alien qualia.”  Like the possibility of spectrum inversion, this possibility is incompatible with standard representationalism.  But my rejection of standard representationalism does not rest on the rejection of the relational view and the possibilities it allows.   For the possibility of the kind of spectrum inversion I have argued to be possible in the case of creatures with asymmetrically structured color experience spaces is compatible with the relational view; it is  compatible with the view of Hilbert and Kalderon, and of Austen Clark, that the phenomenal character of color experiences is purely a matter of their occupying a certain place in a similarity ordering.  And that possibility is incompatible with standard representationalism.

            One view about spectrum inversion I have not yet addressed is the view, held by some standard representationalists, that there could be cases of spectrum inversion but that these would involve systematic misperception of the colors on the part of one of the parties.  I will begin my second lecture with a discussion of this; that will prepare the ground for the presentation of my own version of representationalism.


[End of first Whitehead lecture.]


            In my first lecture I made a case against the view I have called standard representationalism – the view that says that the phenomenal character of perceptual  experiences consists in their objective representational content, and in the case of color experiences consists in their representing certain colors.  This view denies the possibility of spectrum inversion, except on an understanding of it on which it would require that one of the invertees be a systematic misperceiver of the colors.  I argued that this view has no satisfactory response to what I called the objective explanatory gap problem, the problem of how we can reconcile the perceived nature of the colors with what we believe about the physical causes in objects of our color perceptions, and, what goes with this, no satisfactory response to the considerations that suggest that the way we perceive the colors, the way things look to us with respect to color, is partly a function of the nature of our perceptual system.  The best hope for standard representationalism as a way of dealing with the latter involves what I called the “selection gambit,” which says that the nature of our perceptual system selects certain objective properties to be the colors, and certain relations of similarity and difference to be the relations of color similarity and difference.  And I believe that this is on the right track.  But I argued that, contrary to standard representationalism, this allows for the possibility of spectrum inversion.  More specifically, it allows for spectrum inversion involving perceptual systems, such as our own, that have asymmetrical color quality spaces.  It allows this because it allows the sets of properties selected to be the colors by different perceptual systems to overlap in certain ways.  I also argued that, contrary to certain views about perceptual content, it is possible for there to be perceptual systems with symmetrical color quality spaces, and that it is possible in principle that there should be behaviorally undetectable spectrum inversion involving perceptual systems of this sort.  But the latter argument is not needed to show that standard representationalism is mistaken; the earlier one will suffice. 

            An idea I did not take up last time is that behaviorally undetectable spectrum inversion is possible but would involve systematic misperception on the part of one of the parties to the inversion.  So if Jack and Jill are spectrum inverted relative to each other, one or the other of them will be perceiving red things as green, blue things as yellow, and so forth.   This is a view compatible with standard representationalism; while the two perceivers standardly get  phenomenally different perceptions from the same objects, this difference in phenomenal character goes with a difference in representational content; that is why one of them must be misperceiving.  On some versions of this view, it may be that both perceivers have correct beliefs about the colors of things.  But if, for example, both report that ripe tomatoes look red, and mean this in the “phenomenal” and not just the “epistemic” sense of “looks,” then one of them must be mistaken about the character and content of her experience. 

            Let our invertees be Jack and Jill.   Michael Tye, in discussing this case, allows that in the “epistemic” and “comparative” senses of “looks,” ripe tomatoes will look red to both Jack and Jill – to both they will look as if they were red, and in both cases the way they look will be like the way standard red things look to the person in question.  But in the “phenomenal” sense, ripe tomatoes will look red to one of them and green to the other.  Let Jack be the one to whom ripe tomatoes look green. It is not to be expected, of course, that Jack will say that ripe tomatoes look green – he certainly won’t if the inversion is behaviorally undetectable.  But if he says that they look red, meaning this in the phenomenal sense, he will be misreporting the contents of his experiences.  According to some proponents of this view,  this needn’t mean that the phenomenal character of his experience is introspectively inaccessible to him – only that he does not fully understand the meanings of his public language color words.[20] One might instead suppose that in his idiolect “red” means green and “green” means red.  Then he could be correct when he says “It looks red” of the tomato.  But then, unless we suppose that there is an equivocation in his use of color words that he doesn’t notice, his attributions of the colors red and green to objects, although phonologically just like those Jill makes to the same objects, would be systematically false.   Given that Jack and Jill learned their color language in the same way, the idea that Jack’s color attributions are systematically false while Jill’s are systematically true cannot be taken seriously.  And to my mind the idea that Jack’s color attributions, and color beliefs, are true, to the extent that Jill’s are, and yet are systematically produced by misperceptions of the colors, is equally implausible.   But let me pursue this a bit further. 

            The obvious question to ask about this view is: supposing that red things look to Jack the way green things look to Jill, and likewise for other pairs of colors, what could make it true that it is Jack, rather than Jill, who is systematically misperceiving things?  There is of course the answer: what makes this true is that red things look red to Jill and green to Jack, that green things look green to Jill and red to Jack, and so on.  But the question is, what makes that true?  This is the same as the question, in virtue of what do  Jack’s and Jill’s color experiences have representational contents that make Jill’s experiences normally veridical and Jack’s experiences normally illusory?   If the contents of their experiences were fixed by what causes experiences of certain sorts under optimal viewing conditions, then the experiences Jack expresses by saying “It looks red” should be alike in content to those Jill expresses by saying “It looks red.”[21]  And that of course is contrary to the idea that Jill’s experiences are veridical and Jack’s are illusory. 

            It is obvious that what is needed here is an externalist view of the determination of the color content of perceptual experiences.  And it must be an externalist view that allows that the experiences of different subjects can differ in color content even though the subjects are identically embedded in identical environments.  Such a view can be found in Fred Dretske and is appealed to by Michael Tye in his account of how a “pure representationalist” can allow the possibility of the sort of spectrum inversion in which one of the parties systematically misperceives.[22]  Dretske holds that a perceptual state represents a certain property if it has the function of indicating that property.  And what bestows this function is natural selection.  In our example, both Jack and Jill have experiences that indicate the presence of something red.  But whereas Jill’s experience has the function of indicating the presence this property, and so represents its presence, Jack’s experience is of a type that has the function of indicating the presence of something green, and so represents, falsely, the presence of that property.   Jack is descended from creatures who evolved in such a way as to bestow on the relevant state the function of indicating the presence of green things; but at some point there was a mutation that resulted in, as it were, some wires getting crossed; the result was that in the ensuing lineage the properties indicated by experiences were different from those they had the function of indicating, and so different from those they represent. 

            Michael Tye thinks that it is metaphysically possible for spectrum inversion, of the sort that involves misperception on the part of one of the invertees, to hold between microphysical duplicates.  We can hardly have Jack and Jill as microphysical duplicates, so let’s make it Jules and Jim.  Suppose that Jim is the one who systematically misperceives.  What we must suppose here, apparently, is something like this.  In the remote past there evolved two sets of creatures, who were alike except that the physical states that in one of them had the function of indicating red had in the other the function of indicating green, and likewise for other pairs of colors.  Despite this difference, it was true in both groups back then that the colors indicated were normally those represented, so on the standard representationalist view there was initially no inversion with respect to phenomenal character.  Jules descended from members of one of these groups, and Jim descended from members of the other.  At some point in the lineage that led to Jim there was a mutation that made subsequent creatures in that lineage physically just like creatures descended from the other group, and just like them with respect to what properties are indicated by their sensory states, but not like them with respect to what properties are represented by these states.  Eventually this resulted in Jules and Jim as microphysical duplicates, with Jules as someone who sees the colors of things as they are and Jim as someone who systematically misperceives them.

            This is, of course, a radically externalist view about phenomenal character.  Not only does the phenomenal character of experiences not supervene on the intrinsic physical states of their subjects; it also does not supervene on the total physical state of the world at a time.   And we can leave out the word “physical” here – it does not supervene on the total state of the world at a time, even if there is more to that state than the way things are physically.  My visual experience’s having the phenomenal character it does  – indeed, its having any phenomenal character at all – is partly an historical fact, one about my evolutionary history. 

This is precisely the sort of view one needs if one holds that cases of spectrum inversion are possible but that in such cases some of the parties to the inversion systematically misperceive the colors.  I do not know how to show that this view is wrong.  It seems to me absurd on the face of it.  But let me add an embellishment to the story just told which I think highlights its absurdity.  Let Jack and Jill be the brother and sister of, respectively, Jules and Jim.  Unlike their siblings, they are of course not microphysical duplicates – but they are as much alike physically as siblings of different sexes normally are, despite being spectrum inverted relative to each other.  And of course on the view under consideration, one of them systematically misperceives the colors.    Suppose Jack and Jill marry, and have a child – call him Tom.  And suppose Tom is viewing a ripe tomato.  The visual state he is in is one that in his ancestors on one side evolved to have the function of indicating red things, and in his ancestors on the other side evolved to have the function of indicating green things.  Which color does it represent, red or green?  Or is there perhaps no fact of the matter about what color it represents, and no fact of the matter about what phenomenal character his experience has? 

            From now on I will assume that it is possible in principle for there to be cases of spectrum inversion in which the invertees are equally good perceivers of the colors.  What I want to show next is that while allowing this possibility is incompatible with standard representationalism, it requires acceptance of a different version of representationalism. 




















[1] See my 2001.


[2] Levine, 1983 and 1993.


[3] Sellars, 1963.


[4] Boghossian and Velleman, 1989 and 1991.


[5] Moore, 1922.


[6] See Harman 1990, Lycan 1996, Tye 2000, Dretske 1995, Byrne 1997, and Hilbert 1997 and 2000.


[7] Campbell, 1993, p. 268


[8] See Hardin, 1988, and Block, 1999.


[9] See Hilbert, 1987 and Gibbard, 1996.


[10] Hilbert and Kalderon, 2000, p. 198.


[11] Clark 2000.


[12] But it should be noted that in his 1992 Hilbert opposes the view that “different organisms use their color vision to detect different properties” (p. 153).


[13] I think this corresponds roughly to Austen Clark’s distinction between the space of “phenomenal properties” and the space of “qualitative” properties.  See his 2000, p. 4.


[14] What in the first instance the experience types will be paired with are with are not the colors themselves but color-like properties I call “appearance properties” (see my second lecture).  In the experience of a particular subject these will be closely associated with the colors, but what appearance property one sees in perceiving a particular color will depend on the background, the intensity of the light, and the adaptational state of the observer.  My second lecture discusses the relation between colors and appearance properties.  In the present discussion I assume, for sake of simplicity, a direct pairing  of experience-types with colors. 


[15] See Harrison, 1973, and Hardin, 1997.


[16] For this and related points, see Hardin 1997.


[17] Dennett 1993, p. 927, quoted by Hilbert and Kalderon 2000, p. 207.


[18]Hilbert and Kalderon seem to think that all of the experiences in a symmetrically structured experience space would have to have the same relational properties of this sort, and so the same phenomenal character.  Here they seem to have forgotten that the similarities among color experiences would be along several dimensions, including one corresponding to saturation; even in a symmetrical experience space some experience would stand in different relations than others – e.g., some would be maximally saturated, some not.  I am indebted to Aaron Zimmerman for pointing this out.


[19]I think that if, as I have claimed, we could discover creatures with symmetrically structured experience spaces, we could discover that two such creatures were spectrum inverted relative to each other.  What in the first instance seems empirically discoverable is a case of intrasubjective inversion involving such an experience space – a case in which someone’s color experience at one time is inverted relative to that same person’s color experience at a different time.  For an account of how we might discover a case of intersubjective inversion, see Shoemaker 1994a.  For doubts, see Stalnaker 1999.

[20] See Hilbert and Kalderon 2000, p.


[21] Here “optimal conditions” includes only conditions whose obtaining supervenes on the intrinsic properties and relations instantiated at the time in question – they do not include, as they do for Michael Tye (2000), such things as the creature’s system performing as it evolved to perform. 


[22] See Dretske, 1995, and Tye, 2000.