DON DEDRICK University of Toronto

In Color for Philosophers C. L. Hardin argues that chromatic objectivism–a view which identifies colour with some or other property of objects–must be false. The upshot of Hardin's argument is this: there is, in fact, no principled correlation between physical properties and perceived colours. Since that correlation is a minimal condition for objectivism, objectivism is false. Mohan Matthen, who accepts Hardin's conclusion for what can be called "simple objectivism," takes it that an adaptationist theory of biological function applied to colour is able to surmount the problems Hardin describes. It is Matthen's view that I am primarily concerned with in this paper. I will argue that it entails an overly simple view of adaptive value–as, perhaps, do all objectivist views.

In the first section I rehearse Hardin's argument against simple objectivism. In section two, I give a brief account of Matthen's modified biological objectivism. The third section argues that perceptual states need not represent objective properties of the external world to have adaptive value. In section four the value of adaptationist speculation is defended and in the final section, section five, I discuss the basis for a non-objectivist theory of colour perception that is compatible with the adaptationist strategy. My point, throughout this paper, is not to refute a biologically based objectivism but to offer a more sophisticated way of thinking about adaptive value and to formulate, in outline, a non-objectivist account of the function of colour vision.

On the most plausible objectivist view, colour is a property of objects, namely, their surface spectral reflectance (SSR). The reason for preferring this property is straightforward: the ratio of absorbed to reflected light across the spectrum–the "reflectance profile" of an object–is a relatively stable property. While an object's perceived colour may differ greatly in relation to incident illumination, SSR does not. We thus have a physical property–a high-level one–which is undoubtedly relevant to human colour experience. Does this mean that objects are coloured and that their colour is, to speak reductively, identical with their spectral reflectance?

One of the basic principles of colourimetry is colourimetrical identity :

In so far as we recognize this claim to be qualified we can take it to support the reflectance theory, i.e., the claim that colour is to be identified with SSR. The problem comes with the introduction of another notion basic to colourimetry, metameric identity :

Hardin's argument against objectivism places the following burden of proof upon the objectivist: explain how it is that two different samples, with radically different SSRs, can be matched as to perceived colour by an observer. Since there is no identity of surface reflectance and perceived colour to be had, Hardin concludes that

The absence of a principled correlation between spectral reflectance values and perceived colour rules out a simple version of chromatic objectivism. Mohan Matthen has, however, developed an objectivist account of perceptual content which, though it is not particular to colour, takes chromatic perceptual states to be good examples for that account.[4] His view is interesting because it brings together objectivism and evolutionary advantage. Since I believe that this latter notion is an important element in the critique rather than the defence of chromatic objectivism, I want to consider Matthen's view.

I can only sketch Matthen's view here. The essential idea is that of a biological function which is specified in relation to adaptive value. A perceptual system may indicate some objective property of the external environment and, in doing so, confer an adaptive advantage upon an organism. Crucially, the content of some class of perceptual states is specified in relation to this function: "Suppose that the content of a perceptual state is F not when it does in fact indicate F, but when its biological function is to indicate F."[5]

With respect to colour, it is Matthen's view that chromatic perceptual states have, as their biological function, the indication of SSR. Thus, the perceptual content of a chromatic state is identified with the reflectance values for an object as indicated by the colour vision system. But sometimes there will be misperception because this biological function is not realized. There are any number of empirical reasons for this but the essential point, for Matthen, is that the colour vision system need not be, and is unlikely to be, entirely reliable. A perceptual system like colour vision is adapted to a limited range of conditions. Within this range it is mainly, though not wholly, reliable. Outside of it the system is, as a matter of principle, less reliable. Outside and within this range normal misperception occurs. It is misperception because the biological function of the system is not realized, resulting in error relative to that function. It is normal because the system is not malfunctioning but simply limited to information in the distal environment which, due to its nature and the nature of the visual system, makes it impossible to distinguish every difference in SSR. Metamers and other chromatic phenomena can be handled by Matthen's theory of biological function: we should not expect chromatic states to always, successfully, indicate SSR. As Matthen says, "it is better to use an imperfect indicator than to have no access to task relevant properties. To use veridicality as a filter on this range of situations is to show a touching, but quite unbiological devotion to truth."[6]

Though Matthen holds that the class of chromatic states need not be characterized as veridical, veridicality is, in a reconstructed form, essential to his account. It is the biological purpose of the colour vision system to indicate SSR, even though it may fail to do so. In this sense, the colour vision system as a whole is properly characterized in terms of veridicality and chromatic perceptual states are adaptively valuable in so far as they are veridical (or approximate veridical indication of SSR). Matthen's notion of biological function effects a conservative modification to simple objectivism.

Though it is true that this theory gives us a way of dealing with cases that are problematic for simple objectivism, such cases do not, as we shall see, require an objectivist interpretation of biological function. The fact that there are many and systematic instances where chromatic states do not indicate SSR will lead us to examine the claim that the veridical indication or representation of this objective property is in fact the biological function of the colour vision system.

Is the notion of biological function one which fits well or poorly with Matthen's representational realism? Is a conceptual model which relates evolutionary advantage to the veridical representation of "real" properties appropriate? There is, as Jerry Fodor has pointed out, no a priori reason to assume that veridicality and usefulness coincide.

Fodor's conceptual case against the identification of usefulness and veridicality runs as follows. For some organisms, identification of predators is important. If you cannot do it, as a species, you are gone. Suppose the processes mediating predator-detection are very accurate–they always recognize predators and only predators–but are also very slow so that members of a species determine, "yes it's a predator," three seconds before they are gobbled. Other species have processes which are not particularly accurate but are inaccurate in an interesting way: they yield many false-positives (they falsely recognize there is a predator when there is not) and few–one hopes no–false negatives. These processes are also fast–which is one reason why they yield false positives–so the members of this species escape or, at least, are better at avoiding tooth and claw. The moral of this story is that, for some cases, it is in the interests of an organism, as Fodor puts it, "to trade false positives for speed."[7] To put this another way: to trade veridical recognition for ecologically useful recognition.

Fodor's point goes further than Matthen's acceptance of error in perception: a perceptual system may confer adaptive value just because it is non-veridical. It is this potentially radical implication of Fodor's example that I will exploit in my critique of Matthen's objectivism. The study of perception and of cognitive success in general may, in the end, bear no real resemblance to traditional, essentially Cartesian concerns with representation and error. When we begin to talk about perception and cognition in an evolutionary, ecological way we shift our interests away from representing and toward interacting: how does a particular organism fit with the ecological system it participates in? The fact that our colour vision system often misrepresents SSR may not disadvantage us ecologically–adaptively–any more than a recognition system which produces false-predator states does.

The virtue of Matthen's account of chromatic perceptual states is to be found in the explanatory strategy proposed. We begin with the assumption that such states have a biological function and we attempt to construct an account of that function. This is not an easy task, but my dispute with Matthen is not a dispute about basic strategy. The value of the strategy itself is threefold. (1) It will not allow us to rest content with the proximal causes of perceptual states. We are forced to broaden our horizons and consider such states in relation to the environment in which they have arisen and in terms of which they have potential value. (2) It affords the possibility of a deeper understanding of our perceptual apparatus in general. I say this because any attempt to utilize the tools of evolutionary biology in order to explain the perceptual apparatus that we have got will lead us through the thickets of evolutionary history–the tracing of particular aspects of the apparatus; the trichromacy of human colour vision, for example. (3) We will also, and for similar reasons, be interested in comparative colour vision. We are beginning to understand the advantages and limitations of human colour vision via the contrast of that system with others: some more sophisticated, some less, some that appear to be "selected for" quite directly.

One of the pitfalls of adaptationist theorizing is the inference from functions that are realized by some biological system to the claim that such a function is adaptively advantageous: the naturalist's naturalistic fallacy. Why not, then, stick to the description of such functions rather then attribute what may be a fanciful evolutionary value to them? As Daniel Dennett has argued, the separation of an actual function from its value (its biological function) is easier said then done. We need to speculate on value–the "why" of evolutionary theory–in order to come to grips with the descriptive task–the "what" of actual function. Dennett writes:

Take a case from the literature on animal vision. Certain sea birds–some terns–have oil droplets in their eyes. What is the function of these droplets? We find that they screen out a portion of the spectrum. Thus, the actual function of the oil droplets is to act as a filter. Why? Is this just an evolutionary accident? It does not appear to be: the light waves screened out make it possible for the tern to see through the glare reflected from the surface of the water, and this makes it possible for the tern to see its prey–fish.[9]

The problem for those who claim we should stick to actual functions rather than speculate on evolutionary value can be teased out of this case. It is "why" questions which allow us to treat the organism as embedded in its environment. A descriptivist claim to the effect that the actual function of the oil droplet is to screen out a particular portion of the spectrum is true. But as soon as we ask why–why that portion of the spectrum?–we look to the point of the biological facts and reformulate our conception of actual function (i.e., the oil droplet acts as a filter which enables the bird to see through the surface of the water to its prey which is clearly adaptationally advantageous–i.e., birds without the droplets would be at a disadvantage relative to those with them, and so forth). To refrain from asking "why" questions is to restrict oneself to a proximal account of biological function that will import adaptationism covertly in order to relate its claims to the environment (as Dennett suggests, above) or settle for ecologically uninformative (if perfectly correct) actual functions.

One reason why the bird story works well is the fact that we can find a specific biological function for one element of the tern's colour vision system. Is human colour vision like this–are there any elements of it which seem to have specific functions? The answer is not clear. Human colour vision is a good general purpose system. It works well in the day and our achromatic system is relatively acute at night. There is no special purpose hardware, nor any big perceptual "holes." (We lack the capacity for discrimination in the ultraviolet region of the spectrum, but this seems to matter less to us than it does to, for instance, bees.) The inability to identify any particular function will suggest to some that human colour vision is degenerative; a holdover from our evolutionary ancestors who had some specific use for it. Even if this is true, it is not clear that we can avoid the matter of adaptation. Once again, it may lead us to actual functions.

One way to look at Matthen's view is as one attempt in a series of attempts to get clear on what the colour vision system actually does; to determine what sorts of evidence are relevant to that question; to determine how to interpret disparate pieces of evidence. As Dennett says

In attempting to understand human colour vision biologically we begin with a simple objectivism as to chromatic perceptual states and suppose that (a) they represent objective properties of the distal environment, and (b) an adaptive advantage is confered in virtue of such representation. Phenomena such as metamerism will lead us to abandon this simple biological objectivism, and Matthen's modification can be invoked: an account which preserves veridicality as crucial to the biological function of chromatic perceptual states while allowing that such a function need not, always, be successfully realized. Quasi-empirical theories like Edwin Land's[11] are relevant here as are the problems for such views. For instance: Land's retinex computations work well for two-dimensional arrangements of coloured papers and yet they do not work for three-dimensional (i.e., realistic) scenes. The problem is that retinex theory makes certain assumptions (that step-changes in what Land referred to as "lightness values" mark borders between colours, for instance) that are false for scenes viewed in three dimensions, in shadows, etc. This rather significant problem for retinex theories has led to a whole new class of models which attempt to address it and, indeed, to the theoretical claim that a trichromatic system, like the one that humans and many other animals possess could not, in principle, recover reflectance perfectly.[12]

Theoretical claims like this are grist for Matthen's mill but they also suggest that the recovery of SSR may not be the biological function of the colour vision system. Thus, another set of questions open up. How are we to deal with perceptual phenomena like the red-green and yellow-blue colour exclusions? Are metamers really failures relative to a system which seeks to recover SSR, or do they tell us something else about the nature of colour vision? Different views as to why the colour vision system exhibits the properties that it does provide different sorts of interpretive strategies vis-a-vis those properties. In the next section I will consider a non-objectivist strategy that may lead to a more sophisticated understanding of the function of colour vision.

In the preceding sections I have made two main points: {1} perceptual states need not be veridical to be useful; {2} the main value of adaptationist speculation is to be found in its embedding of an organism within that organism's environment. Matthen and I agree on {2} and, in a superficial sense, we agree on {1}. He does, that is, allow that chromatic perceptual states need not be veridical to be useful. But as we have seen, veridicality is an essential feature of his account in that the biological function of the colour vision system is specified in its terms. Matthen, then, interprets {1} in a weak sense: usefulness, when all is said and done, is a function of veridicality. In this section I will sketch a stronger interpretation of {1}; an interpretation that is non-objectivist.

I have pointed out that it is not clear if human colour vision has any particular evolutionary value. This is less damning a criticism of Matthen's view than one might imagine. Even if human colour vision is degenerative it could nonetheless have been selected originally, ancestrally, for its capacity to indicate SSR. This granted, once we have broken the connection between veridicality and usefulness (with respect to adaptation) we have less of a reason to promote an account which involves, essentially, error. The claim that perceptual states are sometimes, perhaps often, in error relative to their supposed biological function suggests a willingness to accept, as Dennett calls it, an instance of "second-rate design." Matthen's view entails that a system which is better at the recovery of SSR than our actual system is an adaptively better system. While we must be careful here–it would be easy as well as naive to argue that the system we possess is optimal–there is an alternative.

In the last section I suggested that that attempts to get clear on the actual function of colour vision will focus our attention on phenomena that Matthen's view ignores. These phenomena are, generally, the consequence of internal neural processing which intervenes, so to speak, between causally relevant external properties and the chromatic states which they are in some way related to. When these processes are considered a different view of the colour vision system's actual (and perhaps biological) function emerges.

On the widely accepted psychophysical theory of colour processing known as opponent colour theory,[13] receptoral outputs from the three cone types in the retina are processed through two colour channels. The so-called "red-green channel" produces either a red response or a green response. Thus, if the red value is 3 and the green value 7, the output is green-4. Because of the opponent nature of this channel there is no composite red-green response (the same goes for the yellow-blue channel). If, on the other hand, both the red-green and yellow-blue channels are active, composite responses result: mixtures like orange and turquoise. One can see, from this account, why metamerism happens. For an indefinite number of response patterns the outputs of the opponent channels will be identical (e.g., suppose that red=6; green=10).[14]

Opponent processes place an indirect constraint upon the recovery of SSR. They are (partly) responsible for the fact that colour samples which may differ radically in spectral reflectance are matched by an observer. One consequence of this constraint is that the number of chromatic discriminations possible for human subjects (and other species with similar vision systems) is diminished relative to a system in which every difference in reflectance translates into a perceptual difference. It is not as if we are perceptually impoverished by this constraint–it is estimated that humans can, in principle, make millions of colour discriminations. More to the point, we need not take it that a system unconstrained in the manner described would have a greater ecological–adaptive–value. Considered in relation to the environment an opponent system realizes an interesting function: it takes a complex set of distal information–SSR among it–and transforms that information into a simplified set of perceived differences. In relation to this process, another interpretation of metamerism suggests itself: rather than being errors relative to SSR recovery, metamers–groups of stimuli which differ in physical composition but not in perceived colour–may be the consequence of a system "designed" to simplify, for perceptual and ecological purposes, a complex stimulus-space.

Am I supposing that the construction of colours, rather than the indication of SSR is the biological function of the colour vision system? We cannot at this point in time decide. It is not enough to simply assert this alternative to objectivism. We need to see how the alternative view fits with extant work in biology on the animal uses of colour discrimination and to consider the possibility of experimental work designed to explore the alternative view. It may turn out that we can make no ultimate sense of colour vision as "selected for." Even so, we will arrive at this conclusion, as Dennett suggests, through a consideration of possible biological functions and their implications.

The emphasis on veridicality that we find in a biological view such as Matthen's has the virtue of simplicity. It tells us that perception aids our survival in the world by latching onto independently existing properties in that world. The point that I am making–and this is the sense in which my view is non-objectivist–is that, while we may need chromatic regularities to survive we may not require "real" ones in the objectivist's sense. It is the discriminability of colours and their relative ecological stability which is important to an organism, not their grounding in some independently existing property. Matthen sees that a strict objectivist reduction of perceptual content to SSR is neither plausible nor required. A further step in the process of understanding color vision biologically may lead us to complex interactive models that neither support nor are supported by any commitment to objectivism. Just what the resulting non-objectivism will amount to (in philosophical as well as biological terms) is, it must be said, unclear.[15] But that is because we have only begun to explore the consequences of {1}, above.

I am not going to argue that SSR recovery is not the function–biological or otherwise–of the colour vision system. I will settle here for the weaker claim that once we have broken the connection between veridicality and usefulness with respect to chromatic states, we must, at the very least, consider more sophisticated accounts of the relationship between these states and the world. I've suggested the direction such an account could proceed in; a direction that takes us away from objectivism, normal misperception, and error while remaining within the bounds of adaptationist theorizing.[16]

1. D. B. Judd and G. Wyszecki, Color in Business, Science, and Industry, 2nd. ed. (New York: Wiley, 1963), p. 103.

2. ibid, p. 103-4.

3. C. L. Hardin, Color for Philosophers: Unweaving the Rainbow (Indianapolis, IN: Hackett, 1988), p. 64-5.

4. Mohan Matthen, "Biological Functions and Perceptual Content," Journal of Philosophy, 85, 1 (January 1988): 5-27. Also: "Intensionality and Perception: A Reply to Rosenberg," Journal of Philosophy, 86, 12 (December 1989): 727-733.

5. Matthen, "Intensionality and Perception," p. 729.

6. Matthen, "Biological Functions and Perceptual Content," p. 13.

7. Jerry Fodor, The Modularity of Mind (Cambridge, MA: MIT Press, 1983), p. 71

8. Daniel Dennett, The Intentional Stance (Cambridge, MA: MIT Press, 1989), p. 278.

9. See W. R. A. Muntz, "Inert Absorbing and Reflecting Pigments," in Handbook of Sensory Physiology, edited by H. J. A. Dartnell (Berlin: Springer Verlag, 1972), p. 529-65. Another account of the biological function of the oil droplets can be found in J. N. Lythgoe, The Ecology of Vision (Oxford: Clarendon Press, 1979), p. 180-83. This account claims that the biological function is to allow the tern to see other terns and, in particular, other terns feeding (through haze; over long distances at sea). For the purpose at hand it does not matter which of these accounts is correct, if either is.

10. Dennett, The Intentional Stance, p. 280.

11. There are numerous papers by Land and his associates. See, for example, Edwin Land, "The Retinex Theory of Color Vision," Scientific American, 237, 6 (June 1977): 108-28.

12. These developments in the computational theory of color vision and their philosophical implications are discussed in Evan Thompson, Colour Vision: A Study in Cognitive Science and the Philosophy of Perception (London and New York: Routledge, in press).

13. A clear account of opponent color theory can be found in Hardin, Color for Philosophers, p. 26-58.

14. I am oversimplifying here. It is often claimed or assumed that the arithmetical values mentioned represent the spiking frequencies of post-retinal opponent cells and, as such, specify a neural code for color. This is a problematic claim, as Hardin himself notes (Color for Philosophers, p. 57). The neurophysiology of color vision is more complicated than the psychophysical model of color perception suggests. For a discussion of this issue see Davida Teller,"The Domain of Visual Science," in Visual Perception, The Neurophysiological Foundations, edited by L. Spillman and J. S. Werner (New York: Academic Press, 1990), p. 13-14.

15. I do not think that, having abandoned objectivism, we must accept Hardin's neural subjectivism. Hardin claims that colours reduce to neural states and that object colours are illusions (Color for Philosophers, p. 111-12). This is just the other side of a traditional ontological coin: either colours are in things, or colours are in the mind/brain. An ecological approach to color vision may do away with this dichotomy. For an argument along these lines see Evan Thompson, Adrian Palacios, and Francisco Varela, "Ways of Colouring," Behavioral and Brain Sciences, 15, 1 (March 1992): 1-74.

16. I would like to thank Mohan Matthen for his comments on earlier versions of this paper. Thanks also to Evan Thompson, Louis Charland, and the unnamed referees.