From Towards a Science of Consciousness Section 2: Color -- Introduction CogNet Proceedings
Spectral inversion arguments have been used for many purposes. I shall here focus on a version of the argument purporting to show that even an explanatory relationship between qualitative experiences and brain processes-let alone an identity of one with the other-is unintelligible. For instance, Joseph Levine (1983) argues that although mental processes and physical process might in fact be identical, we can never have scientific grounds for supposing them to be so:
Let's call the physical story for seeing red 'R' and the physical story for seeing green 'G'. . . . When we consider the qualitative character of our visual experiences when looking at ripe McIntosh apples, as opposed to looking at ripe cucumbers, the difference is not explained by appeal to G and R. For R doesn't really explain why I have the one kind of qualitative experience-the kind I have when looking at McIntosh apples-and not the other. As evidence for this, note that it seems just as easy to imagine G as to imagine R underlying the qualitative experience that is in fact associated with R. The reverse, of course, also seems quite imaginable.Levine argues that, in the absence of an intelligible connection between seeing red and the 'R' story and seeing green and the 'G' story, we can never be entitled to take seeing red to be identical with having neural processes R. The very possibility that somebody could have had the same physical constitution and display the very same behavior that she does now and yet have seen as red what she now sees as green (and, generally, for the same set of stimuli, experiencing all colors as interchanged with their actual-world complements) is sufficient to show that no physical story can ever capture what it is to experience a color.
I do not think that the prospect for a reduction of color experiencing to neural functioning is so bleak. There is enough factual evidence to suggest that the possibility of an undetectable spectral inversion may be an illusion based upon our ignorance, and that if the facts were to be filled in further, the possibility of an undetectable spectral inversion would come to seem as fanciful as the possibility of a human being having Superman's x-ray vision. In the previous chapter, Stephen Palmer has shown that many mappings of color space onto itself are "symmetry breakers," because they map unitary (unique) colors onto binary colors, and vice-versa. Any "spectral inversion" that would carry a binary color like orange into a unitary color like blue would be readily detectable. But why could there not be a transformation in which elementary colors go into elementary colors? As Palmer pointed out, nothing in the simple opponent-process scheme forbids such an interchange, and this would give us an inversion that preserves symmetry. In the quest to find further symmetry breakers, we must therefore look to other aspects of color phenomenology. Perhaps these will suggest deeper reasons, based in the qualitative features of the colors themselves, for thinking that the colors of human experience are intrinsically not invertable. I believe that these deeper reasons will emerge when we consider some characteristics of color categories. Categories are equivalence classes of items that need not be identical. When we call a particular surface "blue," we do not mean to say that it is identical in color to every other surface that is blue. Things that you take to be blue-your neighbor's car, your boss's dress, the sky, the sea-typically differ from each other in tint, shade, or hue. There are light blues, navy blues, electric blues, powder blues. Yet all of them resemble each other more than any of them resembles something that you see as yellow, or as red. It is important to understand that the resemblance that connects two instances of the same color category is not necessarily a function of the perceptual distance between them. It is not hard to find three color samples A, B, and C, which are such that B is separated from A on the one side and from C on the other by the same number of just-noticeable differences, and yet A and B are seen to belong to the same color category whereas C is seen to belong to a different color category. A spectrum looks banded, even though each of its constituent regions blends smoothly into its neighbors.
Is color categorization exclusively a cultural phenomenon, or does it have a biological component? Let's address this question by considering some data about human infants and other primates. Four-month infants know precious little English, and they cannot describe what they see. Nevertheless, by watching their eye fixations one can tell whether they see two stimuli as similar or different. Infants will lose interest in a stimulus that looks similar to its predecessor, but continue looking at a stimulus that they regard as different from what went before. By exposing infants to sequences of colored lights whose dominant wavelengths are 20 nm apart, and recording their eye movements, Bornstein and his collaborators were able to map out their spectral color categories (Bornstein, Kessen, and Weiskopf 1976). These proved to line up rather well with the spectral categories of adults that are mapped with color-naming procedures (for the latter, see Sternheim and Boynton 1966). In a similar fashion, a macaque was trained to respond differentially to spectral lights that human beings would see as good representatives of their categories, and then presented with randomized sequences of lights that did not match the training lights. These lights were categorized by the macaque in pretty much the same way as adult human English speakers would classify them (Sandell, Gross, and Bornstein 1979).
So there must be innate mechanisms not only for detecting resemblances amongst colors, but for categorizing them as well. We should not of course suppose that color categories are consciously or explicitly born in mind by monkeys or infants, but rather that their brains are so wired as to incline them to respond to certain classificatory demands in a characteristic fashion. This was strikingly demonstrated in a series of chimpanzee categorization experiments by Matsuzawa (1985). In order to see what motivated Matsuzawa, we need to take a brief look at Berlin and Kay's famous work on basic color terms (Berlin and Kay 1969). Basic color terms are distinguished from nonbasic terms by their salience and their generality. Applying criteria based on these characteristics, Berlin and Kay were able to show that, with one possible exception, no language currently has more than eleven basic color terms, that each of the terms has a small set of best, or focal examples, that the focal examples from different languages cluster tightly in perceptual color space, and that, in consequence, basic color terms are readily translatable from one language to another. In English, the basic color terms are, as one might expect, the names for the Hering primaries, "red," "yellow," "green," "blue," "black," and "white," as well as "brown," "gray," "orange," "purple," and "pink." The stimuli in the Berlin and Kay work were a selection of Munsell color chips, a collection of color samples carefully scaled and reproduced to exacting standards. The selection consisted of maximally saturated chips taken from the outer shell of the Munsell color space. Using alternative color order systems, other investigators, notably Boynton and Olson (1987) in the United States and Sivik and Taft (1994) in Sweden, have carefully studied the ranges of these terms with very good overall agreement, exploring the interior of the color solid as well as its outer skin. Among other findings, they showed that some colors, such as blue and green, are seen over wide regions of the space, whereas other colors, such as red, orange, and yellow, are of much more restricted extent. We will look at the implications of this shortly.
In the Matsuzawa experiment that was mentioned above, the chimp, whose name was Ai, was trained on a set of eleven focal samples, learning to press the key that contained a contrived character for the appropriate basic color term. She was then presented with 215 of the Berlin and Kay chips that she had not seen. They were shown to her one at a time and in random order, and she was asked to name them. Following the sessions with the training chips, she did not receive reinforcement for her choices. The experimenter assigned a label to a chip when the chimpanzee gave it that label on at least 75 percent of the trials. The results were compared to those generated in a human color-naming experiment, again using the 75 percent consistency criterion. The outcomes were closely similar. The chimp had generalized from focal chips in essentially the same fashion as the human being.
This is a striking result, but what is its application to our problem? Think of it this way. Ai was presumably not doing what she was doing because of cultural bias, the grammar of color concepts, or any other such fancy hoo-ha. She was guided by what she saw, by what looked like what, by, if you will, the intrinsic qualities of her sensory experience. The array of Munsell chips is scaled so that the samples are a constant number of just-noticeable-hue-differences apart, and a constant number of just-noticeable-lightness-differences apart. At one level of resemblance ordering, everything is smooth and orderly. But at the level of categorization, this is not at all the case, as we have already seen. (I might note parenthetically that other measures of perceptual distance are used in other color-order systems, but the results of categorization are essentially independent of this fact. The principles of scaling in the Swedish Natural Color System yield a solid of entirely regular shape, but the categorized areas are as irregular in shape and strikingly diverse in extent in the Natural Color System as they are in the Munsell solid.) If red occupies a small volume in the solid, and green a large one, what does this betoken but a substantial difference in phenomenal structure between red and green? Moreover, this difference is surely intrinsic to the qualities themselves. What else could serve as the basis for categorization? After all, the whole procedure only involves assessing the qualitative similarities and differences between one color and another.
Here is another categorical asymmetry. Brown is a blackened orange or yellow. An orange fruit has in fact the same chromaticity as a chocolate bar. This assertion is commonly met with disbelief, for brown looks to have a very different quality from yellow and orange. This is why people are surprised when they see a demonstration in which a projected orange spot is first dimmed, looking orange to the very edge of invisibility. The same spot is then blackened by surrounding it with an annulus of bright white light. When the blackening occurs, the orange spot is transformed into a rich brown. It is as if the original quality has been lost, and replaced by another.
This appearance of strong qualitative differences is not a general characteristic of blackened colors, most of which resemble their parent hues. Blackened blues, such as navy blue, continue to look blue, and blackened greens-olive greens-continue to look green. Only oranges and yellows seem to lose the parental connection when blackened. Then what would happen in the hypothetical case of spectral inversion in which hues are carried into their complements? The inverse of orange is turquoise, the inverse of yellow, blue. Therefore the inverse of the browns would be blackened turquoises and navy blues. If you are like most people, you will find brown and yellow to be far more different from each other than light blue is from navy blue. In many languages, as in English, the difference between yellow and brown is marked by the use of two distinct basic color terms, but in no language whatever is the light blue-dark blue difference marked with distinct basic color terms, while the yellow-brown difference is left unmarked. In fact, with the possible exception of Russian, no language even has separate basic terms for light blue and dark blue.
It is thus fair to conclude that something has got lost in the inversion, and that if a human being were to be born with such an inversion, it would not go undetected. More to the point, since the blackness in a blackened yellow is the same as the blackness in a blackened blue, or, for that matter, red, or green, there must be some characteristic of yellow that is not present in blue or in any of the other Hering primaries. This probably has to do with the fact that yellow, unlike any other chromatic primary, is most pronounced only at high lightness levels.
But why is this? The most helpful, indeed, I think, the only helpful explanation of this phenomenon would be in terms of a neural mechanism. Recent neurophysiological evidence indicates that color-sensitive color cells in the cerebral cortex statistically "prefer" their yellows light and their reds dark (Yoshioka et al. 1996). This is of course only the first step in a long journey that will, if we are lucky, bring us to suitably rich mechanisms to account for the properties of yellow. Finding such mechanisms would be the way to understand other phenomenal features of yellow, such as why it is that yellowish greens look as though they ought to be classified as greens, even though we judge the yellow content to be well above 50 per cent. The very fact that the internal relations between yellow and its neighbors do not have the same form as the internal relationships between blue and its neighbors suggests that although yellow may be elementary with respect to phenomenal color mixture, it is not elementary simpliciter, any more than a proposition that is commonly used as an axiom for certain purposes is an axiom simpliciter. The only way to understand why yellow, or any other color, has its particular phenomenal structure is to devise a good functional model, consistent with what we know about the underlying neurophysiology. Such models have already helped us to understand the unitary-binary and opponent structures of phenomenal color space (Hardin 1988). More recently, there has been progress in understanding how the intrinsic similarities and differences revealed in color naming might be grounded in the workings of our visual systems. Guest and Van Laar (1997), using the CIE Uniform Color Space realized on video displays, have investigated the size and distribution of the regions of basic color naming and connected them to features of Hunt's quantitative opponent color model. Hunt's complex model was derived from considerations of color appearance that are quite independent of color naming. Once again, quality is expressed in structure, structure is anchored in functional configuration, and functional configuration is presumably rooted in patterns of neural activity.
As a final example, let us consider the well-known distinction between "warm" and "cool" colors. Although this cross-culturally robust division is commonly believed to arise in consequence of environmental associations, it has always seemed to some people, including me, that it reflects intrinsic phenomenal similarities and differences. Recently, Katra and Wooten asked ten subjects to rate eight color samples as "warm" or "cool" on a ten-point scale, with ten as "very warm." As one might have expected, the mean results gave the lowest rating (3.5) to the unitary blue sample, and the highest rating (6.75) to the orange sample. There was a high level of agreement among subjects. Katra and Wooten compared the group data with summed averaged opponent-response cancellation data, which can be interpreted as giving the level of activation of opponent channels. To quote Katra and Wooten's conclusion:
The remarkable correspondence between the obtained ratings of warmth and coolness and the activation levels in the opponent channels . . . suggests that the attribution of thermal properties to colors may be linked to the low-level physiological processes involved in color perception. Higher ratings of warmth corresponded with levels of activation of the opponent channels in one direction, while cooler ratings corresponded with activation in the opposite direction. This suggests that a link to the activation level of the opponent channels, rather than the psychological quality of hue, drives the association of temperature with color, and that the association is more than simply a cognitive process.They thus trace the connection between the warm-cool of temperature and the warm-cool of color to the corresponding activation levels of their respective neural systems rather than to stereotypical environmental associations such as red with fire and blue with water. This does not by itself warrant the conclusion that the respective intrinsic characters of the warm colors and the cool colors are a function of opponent activation levels, but it is consonant with that stronger claim. Furthermore, if one reflects on just how Katra and Wooten's subjects could gain information about the state of activation of their visual opponent cells, it becomes clear that it could only be by experiencing the colors of which these cells are the neural substrate. In other words, the color qualities themselves are the natural expression of neural activation, and we implicitly read them as such.
Now let us sum up these considerations. Color space has an irregular structure. The structure of color space is arrived at entirely by comparing the colors with each other, so the irregularity of structure is intrinsic to the domain of colors. Experiments with nonhuman primates strongly suggest that this irregular, intrinsic structure is of biological rather than cultural origin. The peculiarities of chromatic structure invite explanation in terms of biological mechanisms, and in some cases it is possible to produce such explanations, at least in outline. The details of the chromatic structural irregularities prohibit putative undetectable interchanges of color experiences: small rotations of the hue circuit carry unitary into binary hues; interchanges of warm and cool colors carry negative opponent-channel activations into positive ones, and vice-versa; interchange of yellows with blues exchanges dark blues and cyans with browns; interchange of reds with greens maps a small categorical region into a large one, and a large region into a small one.
What has here been given is an empirical argument, and as such has predictive consequences. In particular, it should be possible to test it with the pseudonormal observers (if such there be) that Martine Nida-Rümelin describes in the next chapter. A pseudonormal would see green at spectral loci at which normal observers see red, and red at loci where normal observers see green. If the present argument is correct, the region of a standard color space that a pseudonormal observer labels "red" should be large, and the region that such an observer labels "green" should be small. Since pseudonormality could presumably be determined by genetic tests, one would only have to see whether there is a correlation between the genetic marker and differences in the size, shape, and location of color-naming regions.
Some proponents of the possibility of an inverted spectrum (e.g., Shoemaker 1984) have conceded that human color space may not as a matter of fact be invertable. They have, however, urged that this does not show that no creature could possibly have inverted sensory qualities, and some of them (e.g., Levine 1991) have go on to argue that the mere possibility of inverted sensory qualities is sufficient to make any functionalist account of the qualities of experience suspect.
In one respect, they are right. Empirical arguments cannot (nontrivially) yield necessary truths. We philosophers rather tenaciously cling to this truism, perhaps because we sense that the independence of our discipline depends upon it. But we must beware of letting it bear too much weight. That we can in some fashion imagine that water is not H2O, or that heat is a fluid, or that there exists a perfectly rigid body, does not license us to suppose that any of these things is possible in any scientifically interesting sense. At our present state of knowledge, to regard any of them as genuinely possible is to exchange hard-won intelligibility for a murky mess of imagery. Given as background what we now know about fluids and heat, it becomes much harder for us even to imagine that heat is a fluid. Granted, there is still no knock-down argument that there is no possible world in which the heat of a gas is a fluid, but we are not thereby tempted to suspect that the heat of a gas might not after all be identical with molecular kinetic energy. When it comes to scientific identities, logical possibility is trumped by overwhelming implausibility.
The case at hand is similar. Much of the appeal of the inverted spectrum as an antifunctionalist or antimaterialist weapon has lain in its intuitiveness: what looks to me to be THIS color (inwardly ostending red) could have looked to me to be THAT color (inwardly ostending green) without anyone being the wiser. This simple intuition has doubtless been aided and abetted by the wide currency of oversimplified models of color space, such as the color sphere and hue circle, in which the structure of color qualities is presented as smoothly symmetrical. But once we do the phenomenology of THIS and THAT, becoming aware of their intrinsic structure, and elaborating the functional structure that underlies them, the initial plausibility of interchange begins to fade, just as the plausibility of heat's being a fluid begins to fade once one understands how the ideas of the kinetic theory engage the empirical facts about heat. And when this paradigmatic example of qualitative interchange loses its grip on our imaginations, the idea of there being abstractly specified qualitative states being interchangeable in abstractly specified creatures with abstractly specified physical workings ought to lose its grip on our intuitions.
Merely schematic specification of the subject matter plagues both sides of the disputes about functionalism. One the one hand a defender of functionalism (Lewis 1980) gives us Martians with plumbing instead of neurons, and on the other, a critic of functionalism (Block 1980) presents us with the spectacle of the whole nation of China acting as a Turing machine. Amusing though these fantasies may be, they are as desperately lacking in the details of what is to be explained as they are lacking in constraints on the putative explanatory mechanisms. It is as if we were asked to judge the possibility of a physical account of living organisms based only on a thorough knowledge of Lucretius's On the Nature of Things. To judge rightly the adequacy of functionalism or materialism to capture the qualitative character of experience, we must carefully describe both sides of the equation. To do so, we need good ideas, the right distinctions, and lots of careful empirical work. That work must take place on several levels, regimenting the phenomenology, developing functional models that are capable of imaging that phenomenology, and investigating how those models might be realized by the available neural resources (cf. Clark 1993). The patient application of these methods can, in principle, capture any expressible fact about sensory qualities, and bring that fact within the ambit of scientific explanation. Will there be a plurality of plausible functional models adequate to the total phenomenology, or will there be but one? Will the preferred future explanations of sensory qualities take the form only of correlations among the behavioral, phenomenal, and neural domains, or will they involve a proper reduction of phenomenology to neural mechanisms? We are simply too ignorant of the relevant facts to answer these questions now, and we ought not to pretend that clever conceptual analysis can offset this epistemic deficiency. We must go much further in solving the "easy" problems of consciousness before we can clearly understand just what the "hard" problem consists in, or whether there really is a "hard" problem at all. And anyway, aren't the "easy" problems hard enough?
Longer versions of this chapter have appeared under the title Reinventing the Spectrum in the following two books:
Mindscapes: Philosophy, Science, and the Mind. Edited by Martin Carrier and Peter K. Machamer. UVK/University of Pittsburgh Press, 1997.
Readings on Color: The Philosophy of Color. Edited by Alex Byrne and David R. Hilbert. MIT Press, 1997.
1. This presentation is based on Hardin (1997).
Berlin, B., and P. Kay. 1969. Basic Color Terms.Berkeley and Los Angeles: University of California Press.
Block, N. 1980. Troubles with Functionalism, in N. Block (ed.), Readings in Philosophy of Psychology, vol. 1. Cambridge, Mass.: Harvard University Press.
Bornstein, M. H., W. Kessen, and S. Weiskopf. 1976. Color Vision and Hue Categorization in Young Human Infants, in Journal of Experimental Psychology 2:115-19.
Boynton, R. M., and C. X. Olson. 1987. Locating Basic Colors in the
OSA Space, in Color Research and Application 12, no. 2:94-105.
Clark, A. 1993. EM>Sensory Qualities.
Clark, A. 1993. EM>Sensory Qualities.Oxford: The Clarendon Press.
Guest, S., and D. Van Laar. 1997. All Colours Are not Created Equal: The Psychology of Categorical Colour Naming, in L. Sivik (ed.), Colour Report F50: Colour and Psychology. Stockholm: Scandinavian Colour Institute.
Hardin, C. L. 1988. Color for Philosophers: Unweaving the Rainbow. Indianapolis and Cambridge, Mass.: Hackett Publishing Company.
Hardin, C. L. 1997. Reinverting the Spectrum, in A. Byrne and D. R. Hilbert (eds.), Readings on Color, vol. 1. Cambridge, Mass.: MIT Press.
Hardin, C. L., and L. Maffi. (Eds.). 1997. Color Categories in Thought and Language. Cambridge and New York: Cambridge University Press.
Katra, B., and B. H. Wooten. n.d. Perceived Lightness/Darkness and Warmth/Coolness in Chromatic Experience, unpublished ms. 38 pp.
Levine, J. 1983. Materialism and Qualia: The Explanatory Gap. In Pacific Philosophical Quarterly 64:354-361.
Levine, J. 1991. Cool Red: A Reply to Hardin. In Philosophical Psychology 4 (1):27-40.
Lewis, D. (1980) "Mad Pain and Maritan Pain." In N. Block (ed.), Readings in Philosophy of Psychology, vol. 1. Cambridge, Mass.: Harvard University Press.
Matsuzawa, T. 1985. Colour Naming and Classification in a Chimpanzee (Pan Troglodytes). In Journal of Human Evolution 14:283-91.
Quinn, P. C., J. L. Rosano,and B. R. Wooten. 1988. Evidence that Brown is not an Elemental Color. In Perception and Psychophysics 37 (3):198-202.
Sandell, J. H., C. G. Gross, and M. H. Bornstein. 1979. Color Categories in Macaques. In Journal of Comparative and Physiological Psychology 93:626-35.
Shoemaker, S. 1984. The Inverted Spectrum, in Identity, Cause, and Mind: Philosophical Essays. Cambridge: Cambridge University Press.
Sivik, L., and Taft , C. 1994. Color Naming: A Mapping in the NCS of Common Color Terms, in Scandinavian Journal of Psychology 35: 144-64.
Sternheim, C. E., and R. M. Boynton. 1966. Uniqueness of Perceived Hues Investigated with a Continuous Judgmental Technique. In Journal of Experimental Psychology 72:770-86.
Yoshioka, T., B. M. Dow, and R. G. Vautin. 1996. Neural Mechanisms of Color Categorization in Areas V1, V2, and V4 of Macaque Monkey Cortex. In Behavioral Brain Research. 76:51-80