THE DISUNITY OF COLOR

What is color?  Presumably a characteristic shared by things that are red, blue, green, etc.  The
"etc" marks a crux: to understand what color is, we must know why some qualities are included
in the denotation of "red, blue, green, etc," while others (fifty degrees Celsius, weighs three
grams) are excluded.  It may seem obvious that, even if we are unable to give a verbally
articulated definition of the common characteristic of color qualities, we have an intuitive or
experiential grasp of it.  (Hume thought that this experiential grasp was so strong that even if we
had been presented all our lives with an incomplete array of colors, we would be able to form a
mental image of the missing shades.)  I contest this.  We cannot arrive at an adequate
understanding of color by generalizing from features of human experience. 
My main reason for thinking that our experience is insufficient to give us a proper grasp of
color is that we have access only to one kind of color experience.  Humans see in color.  But so
do other animals.  There are significant differences amongst them.  In the first part of this paper
(sections 1-4), I argue in support of four propositions which together imply that there is neither
an experienced nor a natural unity that determines the membership of the class denoted by "red,
blue, green, etc." in a way that properly accommodates differences amongst the animals that see
in color. 
In the second part of the paper (sections 5-8), I attempt to construct a conception of color
congruent with the facts of evolution and the variability of color vision across the animal
kingdom.  My thesis is that color vision is to be defined not in terms of the experiences,
perceptual discriminations, or concepts that it makes available to organisms   or to put it
another way, not in terms of its output   but by reference to the information it receives as input. 
Nothing else unites color across systems that have evolved independently.  This conception of
color is able to deal naturally with the color perceptions of other kinds of organism because it
does not require a high degree of shared experience or content across broad biological taxa.  And
it allows us to think about color realism in new and helpful ways.

            1.  INTRODUCTION: NARROW ANTHROPOCENTRISM 
Philosophers have standardly defined color in terms of human experience.  This is odd: only a
Berkeleian idealist would define shape or motion this way.  That a class of things evokes similar
reactions from humans is no guarantee of its unity.  According to Scott Atran, pre-scientific
"folk" and other early botanical taxonomies grouped plants together according to their medicinal
properties.  He records a complaint of systematists in the early modern period that "the more
[such human centered taxonomies] enriched medicine, the more they threw botany into
confusion."  Is it clear that anthropocentric conceptions of color definitions in terms of
features of human experience do any better? 
Let's look at a typical anthropocentric definition.  David Lewis has recently proposed the
following as a schema for defining color terms.  

          D1. Red is the surface property which typically causes experience of red in people who have such things
     before their eyes.

          D2.  Experience of red is the inner state of people which is the typical effect of having red things before the
     eyes. 

Lewis claims that the truth of the above pair of statements is preserved when any other color
term, for example 'green', is substituted throughout D1 and D2 in place of 'red'.  So:

          A colour is any first component of a corresponding pair.  A colour experience is any second component.

This is not quite right.  Other terms like 'smooth textured' and 'speckled' would also satisfy D1
and D2 when substituted for 'red'.  But this is no great difficulty.  Lewis avers that colors are
reflectance properties of surfaces.  So to get a general conception of color out of his definitions,
we need to amend D1 accordingly: "Red is the surface spectral reflectance which typically
causes . . etc."  
                                        Consider, then, the following formulae: 

          S1.  x is the surface spectral reflectance which typically causes experience of x in people
     who have such things before their eyes.

          S2  Experience of x is the inner state of people which is the typical effect of having x
     things before the eyes.

Lewis's theory, slightly amended, is that colors are the "first component" of pairs that satisfy
these formulae.
Before we begin to evaluate this theory, let us look briefly at the neurocomputational basis
of human color experience.  Humans possess three types of photoreceptor, or "visual pigment,"
each sensitive to distinct but overlapping wavebands of the visual spectrum.  Call these
wavebands long (L), middle (M), and short (S).  Information about spectral distribution is
extracted from the outputs of these cells by sampling the relative strength of a signal in selected
wavebands.  This process, called "opponent processing," (in effect) computes the following
functions: (L+S)-M, (L+M)-S.  The first function tells us the relative strength of a signal at the
ends of the visual spectrum (L and S) as compared to the middle.  If this function takes a positive
value, that is, if the signal is stronger at one end or another of the spectrum, it will look reddish. 
If it is stronger in the middle of the spectrum, and the value of the function is negative, it looks
greenish.  The second function tells us how much of the signal's strength is in the middle and
long wavebands.  (The effective spread of this waveband does not quite extend to the long
wavelength extreme of the visual spectrum.)  If it is stronger here than in the short waveband, the
signal looks yellowish; if it is stronger in the short waves, it looks bluish.  Color experience is a
composite of these opponent processes.  Consider a short-wave signal.  It is experienced as
reddish because it is at one spectral extreme, and bluish because it is at the short wave end.  Thus
it is experienced as bluish-red, or violet.  Similarly consider a signal that is in the middle towards
the long end.  It will look greenish because it is in the middle, and yellowish because it is at the
long end.  The composite experience is of a greenish yellow.
Return now to Lewis's anthropocentric definition of color.  It immediately bumps up
against the problem of "novel colors."  Imagine an organism, O, very much like the above, but
with its short wave receptor shifted toward the ultraviolet.  It will be capable of seeing ultraviolet
reflectances.  If it used the exactly the same differencing functions as above, it would experience
ultraviolet in much the same way as we see violets and purples (that is, as bluish-red). 
Ultraviolet is not a color according to S1: spectral reflectances in this range do not "typically"
cause "people" to have any sort of experience at all, even when they are "before the eyes." 
However, S2 admits O's experience of ultra-violet; it is the very same "inner state" which, in
people, is "the typical effect of having violet things before the eyes."  Lewis is thus obliged to
say that although O experiences ultra-violet as a color, it is not a color.  This odd result, which
seems incongruent with the general idea of pairing colors with kinds of experience, shows how
his definition turns on contingent features of human color experience.  It does not allow us to
focus on the mechanisms human color vision uses to detect features of the outside world, which
can easily be replicated (even in other humans) in slightly modified form.  Such a definition is
obstructive when we wish to consider color and color vision in a general setting.
The problem becomes acute when we consider color vision in other species.  The pigeon
has eyes with photoreceptors similar to our own, and retinal cells that treat the output of these
receptors by "opponent processing."  However, the pigeon has four visual pigments.  (We are
trichromats, pigeons tetrachromats.)  The pigeon's fourth pigment is sensitive to ultraviolet, and
it is thus capable of seeing reflectances in the ultraviolet range of the spectrum.  Further, it
computes three difference values   it samples three wavebands, not just two.  Consequently,
pigeon colors have three hue-components; the colors they experience are not completely
describable, as ours are, in terms of two such components.  When pigeons look at the feathers of
other pigeons (which tend to have a high short wave reflectance), they not only see reflectances
there that we cannot see, they also experience them in ways we cannot imagine.  We have
already seen that Lewis's anthropocentric conception fails to accommodate ultraviolet "colors." 
In addition it excludes the pigeon's color experiences because they do not correspond to any
inner state of people, and thus fall outside the scope of S2.  This is arbitrary.  A general
conception of color and color experience should allow us to treat the human and the pigeon
systems as instances of a general kind.  Or else it should give us a principled reason for
excluding the pigeon, or for counting the matter as indeterminate. 

          Conclusion 1: Anthropocentric accounts that restrict colors to those detected by humans,
     or color experiences to those that occur in humans, cannot account for color vision as it
     occurs in other species.  

                  II.  FLEXIBLE ANTHROPOCENTRISM
Some philosophers suggest that the way to handle the pigeon's experiences is to include them,
provided they sufficiently resemble the human case.  This is a procedure in broad agreement with
that recommended by C.L. Hardin: 

          We can begin with the human case, in which we know there to be genuine color vision, and extend the
     concept of a color perceiver outward to other species.

To someone new to this area, this must seem rather timid.  Why not eliminate the reference to
human experience altogether?  This is what most color scientists do: they define color vision
functionally.  For example:

          F*.  Color vision is the perceptual discrimination capacity underlying differential
     responses to light differing in wavelength only.

One problem with F*, emphasized by David R. Hilbert, is that it requires very little by way of
an internal representation of color.  Hilbert presents us with the following sort of case: some
invertebrates have wavelength-sensitive behavior that can be modelled on a system consisting of
one receptor sensitive to long wavelengths connected to a motor that makes them move toward
the light source, and another receptor sensitive to short wavelengths that makes it move away. 
Such a visual system responds differently to red and blue, but it is doubtful that it has color
vision.  Since the two outputs are never combined, the organism never achieves a unified
representation of color.  
Hilbert concludes that "the empirical literature is not going to provide us with an
independent characterization of color vision that will be helpful."  This is unduly harsh.  Many
color scientists are in fact aware of the importance of cases like this.  For example, Timothy H.
Goldsmith distinguishes between color vision proper and mere wavelength dependent behaviour. 

          Our own experience tells us that to have color vision is to see colors.  How do we translate that into an
     operational definition that can be applied to other animals?

Goldsmith suggests that if an animal can learn or be conditioned to perform "a simple task like
pushing a bar when it sees the 'correct' stimulus," then it "it must be able to extract from the
stimulus some quality associated with wavelength composition."  In other words, even if an
organism displays innate and unmodifiable behaviour that requires multiple visual pigments etc.,
Goldsmith is not prepared to count it as possessing genuine color vision because it might still not
possess a separate representation of color qualities.  Since conditioned behaviour requires such a
representation, he takes the presence of conditioned responses to color as operationally defining
color vision.  Without worrying too much about the merits of this operational definition, why not
simply demand (pretty much in Goldsmith's words) that:

          F.  Color vision is the ability to extract from light visual representations of qualities
     associated with its wavelength composition?

What is wrong with this?  (In my own view, very little: I shall return to the utility of F in section
6.)
The reason why many philosophers are reluctant to accept F is that it omits reference to the
phenomenology of our visual experience of color.  In this context, they tend to quote or para
phrase P.F. Strawson: "Colors are visibilia, or they are nothing."  (Notice that the quotation
from Goldsmith above contains a paraphrase of this idea.)  This is taken to imply that color
should be defined in terms of its characteristics qua visibilium.  Thus Hilbert, whose guiding
intuition is again a paraphrase of Strawson "Colors that cannot be seen . . are not plausibly
colors at all," insists that, although "color vision plays rather diverse roles in different
organisms," an adequate conception of color must focus on human phenomenology. 
There are several ways in which resemblances to human phenomenology have been
invoked in the recent philosophical literature; and I shall come back to Hilbert's own way in the
next section.  (A principal aim of my sections 2-4 is to undermine such resemblance accounts.) 
But first, I want to consider the views of Evan Thompson, who was among the first to point out
the philosophical relevance of vision in other animals (see note 1 above).  Thompson thinks that
in deciding whether other animals possess color vision, we should stay close to the structure of
color as humans experience it.

          I agree with Hilbert that we can have an independent phenomenological route to colour vision.  What this
     route reveals . . . is that colour vision, whatever else it might be, is the ability to see visual qualities
     belonging to the phenomenal hue categories, red, green, yellow, blue.  

These words suggest a strong phenomenological constraint: if an organism does not experience
the "phenomenal hue categories" it does not experience colors at all.  But in fact Thompson
intends something different.  He is interested in the computational and physiological basis of
human color phenomenology, and he is recommending that color vision should be defined in
terms of resemblances to the mechanisms that produce the "phenomenal hue categories" in
humans.  Since opponent processing is primarily responsible for these categories, his
recommendation is, in effect, that color vision should be defined as the ability conferred on an
organism by multiple visual pigments and opponent processing.  (This corresponds quite closely
to Hardin's intuitions as well.)
To illustrate some of the difficulties this proposal brings, let us first test the strong
phenomenological constraint suggested by Thompson's formulation, by reference to a scenario
described by J.B. Mollon:

          Our colour vision seems to depend on two, relatively independent, subsystems   a phylogenetically recent
     subsystem overlaid on a much more ancient subsystem.

The ancient subsystem, he says, is dichromatic, and it detects a very simple characteristic of the
wavelength distribution of a spectral signal, roughly whether it is stronger in the short or the long
wavelength end of the visual spectrum.

          Subjectively, it is the primordial subsystem that divides our colour sensations into warm, cool, and
     neutral.

The question: Consider a species that possesses the "ancient" subsystem, but does not know
Thompson's "hue categories, red, green, yellow, blue" does it have color vision?  On grounds
of phenomenology, and in isolation, one might well be inclined to say "No": in our color
phenomenology, warm and cool seem to be extrachromatic characterizations that sit on top, as it
were, of the colors.  But when we remember that warmth and coolness is pretty much the same as
the long/short wavelength determination that forms the basis of our blue-yellow categories, we
should hesitate.  Clearly the primordial subsystem has access to one of the two dimensions of our
richer experience of hue.  To deny that it represents the world in color seems as presumptuous as
for a tetrachromat to sniff at us.
Thompson might agree with this conclusion: the ancient sub-system does not know the hue
categories he fastens on warm-cool becomes blue-yellow only when the additional hue
category is in place but it does have opponent processing, the neurocomputational basis of
these categories.  So in order to test the weaker (and much more plausible) hypothesis that
opponent processing is a defining condition of color vision, imagine a primate that during the
course of evolution came to live in caves.  In the dim light available there, its opponent
processing system is of little use, because the color sensitive cone cells put out small values in
poor illumination, and the differences between these outputs are smaller still.  However, the
addition of signals from the three cone cells what corresponds to the achromatic channel in
human daytime vision is still useful.  Let us imagine that the primate retains the achromatic
channel, but loses the differenced chromatic channels with the passage of evolutionary time. 
Now, though the achromatic channel delivers our daytime perception of black and white and
gray, it is most sensitive to light in the wave-band of greatest overlap between the three cones,
i.e. in roughly the middle of the visual spectrum.  This is why yellows and greens seem bright to
us two kinds of cone cell contribute amplitude to our perception of these colors at close to their
peak sensitivity, thereby nearly doubling their brightness while the reds and violets at opposite
ends of our spectrum, which activate only one cone, seem relatively dark.  So, in effect, our
primate distinguishes (what we call) yellow from blue and red (just as our own daytime
achromatic vision does).  But it will do this without experiencing hue; it merely experiences
yellow things as bright.  Does it have color vision?  Does it "remember" an aspect of color much
as the ancient subsystem anticipates it?
Compare the primate's vision with that of a rod monochromat, the latter resulting from a
congenital absence of color discriminating cone cells.  Rod monochromats rely for vision entirely
on the rod cells that most of us use for night vision only.  There is a resemblance between what
such monochromats see and what we experience as black, white and gray: one such individual,
Knut Nordby, says that phenomenologically his vision "has some resemblance to that of an
orthochromatic black-and-white film."  However, there is a difference: what such an individual
(and the rest of us at night) perceives as "bright" is not the same as the brightness we have just
been talking about, that is, that of yellow and the other "bright" colors.  Color information is an
essential part of the brightness a chromatic perceiver sees: yellow looks bright to her because two
cone cells are contributing to her perception of it, and this differentiates it from blue and red. 
Now, the surfaces that the imaginary primate sees as bright correspond to those that the chro
matic perceiver sees as brightly colored.  The rod monochromat's vision, however, is based on
the output of one type of cell only, and so its response is much flatter across its range of
sensitivity; his brightness does not correspond to the brightness of colors like yellow.  Thus the
content of the primate's vision is closer to ours in one crucial respect that is relevant to our
perception of color.  
Now, taking only phenomenological resemblance into account, there may be very little to
distinguish the primate from the rod monochromat.  The range of qualia available to both is
pretty much the same as our own night vision reveals, rod based vision is in black and white
and gray.  Nevertheless, the correspondence that does exist between the primate's vision and that
of a color perceiver is significant with respect to how it is classified.  It is not merely the extent
of the similarities and differences amongst these three systems that makes the difference
regarding which are, in some sense, perceivers of color, but how the comparative phenomena are
explained.  Considering the evolutionary connection it has with other primate color perceivers, it
makes sense to insist that, as in the case of the dichromatic system that experiences warm and
cool colors, this animal's perception of certain surfaces as bright and others as dark is a manifes
tation of one aspect or dimension of color vision. 
The cases we have just discussed show that we cannot define color vision by phenomenol
ogy or by the cognitive mechanisms that produce a particular phenomenology.  Of course, we
already knew this from our consideration of narrow anthropocentrism.  But there is a further
point that we can take from our discussion so far.  When we are dealing with human color vision,
we have a more or less fixed object of investigation.  There is no urgent need to decide what is
essential and what is accidental about this system; we can take it as we find it, subject to such
variation as exists amongst individuals of the same or closely related species.  When we
investigate color vision in general, the point of taking the human case as a paradigm becomes
less clear: why should we assume that the human system is the norm rather than some kind of
oddity or specialized adaptation?  (In fact, primate color vision is odd, having evolved, independ
ently of other such systems, from the vision of early mammals, who, in their early evolutionary
period, were nocturnal dichromats.)  This is why flexible anthropocentrism is not very helpful.  It
does not offer us an understanding of deeper issues.  The complaint is not that it does not offer us
precision perhaps no approach can do that.  The problem is that some kinds of resemblance are
more relevant than others.  Arranging things by similarity to the human case may well be a useful
preliminary to sorting these out, but it is the explanation of similarity and difference that must in
the end form the basis for the classification of visual systems.  

          Conclusion 2: Defining color vision by its overall resemblance to the human case falls
     short of articulating explanatory principles that govern the classification of color-like
     experiences in other species.  

                   3.  ANTHROPOCENTRIC REALISM
It might be thought that the difficulties of flexible anthropocentrism can be traced to its failure to
deal with real color.  As we noted before, motion perception is defined not by our experience of
it, but by the nature of motion.  Perhaps we should take a similar approach to color perception,
and define it by reference to the qualities it detects.  In sections 3, 4, and 5, I shall be considering
successively less demanding versions of this idea, arguing that each is unable to deal satisfacto
rily with the facts of comparative color vision, because it is impossible in principle to set bounds
on the qualities that color vision is able to detect. 
Hilbert responds to the shortcomings of the rigid and exclusionary anthropocentrism
considered in section 1 by formulating an approach he calls "anthropocentric realism."  He starts
with the idea a) that it is possible to identify our own color experiences introspectively as "an
aspect of our visual experience that is clearly different from figure, texture, depth."  He suggests
b) that the human color experiences so identified have the biological function of detecting
surface spectral reflectances. This, he thinks, makes it plausible c) that colors are surface spectral
reflectances.  In short: Hilbert agrees with Thompson in identifying human color experience by
means of its phenomenology, but instead of taking the essence of color vision to reside in the
perceptual mechanisms that produce that phenomenology, he concentrates on its environmental
significance.  He proposes that, regardless of what species it occurs in (and disregarding
phenomenology in species other than humans) a visual system should be identified as color
vision if, as in the human case, it detects spectral reflectances via color-like experience. 
So long as it is applied in a disciplined way, this form of anthropocentrism is harmless, so
Hilbert claims.

          The claim is not that human beings have the best color vision or that human characteristics are more central
     than that of other organisms.  They may possess visual abilities that humans lack and as long as we are
     careful to classify organisms as having color vision only on functional grounds we will only impose
     similarity on organisms that really are similar in relevant ways.

Hilbert's reliance on features peculiar to human color experience is not, at this point, as strong as
Thompson's.  The weight of his approach rests primarily on the environmental function of color
vision.  Hilbert does not discuss the question explicitly, but it is clear that his approach allows
him to extend the scope of a conception like Lewis's to colors not visible to humans.  The
pigeon's detection of the ultraviolet in bird feathers is a color vision state because it is a color-
like experience that detects reflectance.  Presumably, this makes ultraviolet a color.  Further, if
we allow Hilbert the assumption that both the rod monochromat and the imaginary primate
discussed above have "color-like experiences," (a contentious but not indefensible claim), he
might well arrive at the same conclusions as I do with regard to them, since the primate makes
wavelength based distinctions among spectral reflectances, whereas the rod monochromat does
not.  These are good results.  By stressing the reflectance-detection function of color vision,
Hilbert is able to include a wide range of experiences of other species.  And by introducing a
criterion based on function, he goes beyond vague and non-explanatory considerations of overall
resemblance to human phenomenology.
Hilbert's conception is not free of counter-intuitive consequences, however.  Let's look
again at the pigeon's ultraviolet vision.  The atmosphere scatters ultraviolet light more than light
of longer wavelengths; consequently, light coming from a direction perpendicular to the sun
tends to have a high concentration of ultraviolet, while light from the direction of the sun tends to
be unbiassed as to frequency, that is, white.  J.F. Nuboer ("Comparative View") has speculated
that with its ultraviolet vision, the pigeon is able to navigate in an aerial environment which not
only lacks "landmarks" but extends deep into three dimensions (by contrast with our own
environment which does not normally extend far below our feet or above our eyes).  Even in its
trackless space, the bird can determine direction in heliocentric coordinates.  But both the narrow
anthropocentrist of section 1 and Hilbert are committed to saying that it is not thereby detecting
color.  The narrow anthropocentrist is forced to say this for the uninteresting reason that humans
do not see in the ultraviolet.  Hilbert is forced to agree, but his more interesting reason is that he
identifies colors by what human color vision detects, which (according to him) is surface
reflectance.  The pigeon detects a property of direction, and direction is not a surface. 
This is not so good a result.  We discriminate surfaces and locate them by the color-
experiences induced in us by light differentially reflected from them.  This is at least part of our
justification for saying that surfaces look colored.  The bird discriminates directions by the color-
like experiences induced in it by light differentially scattered from them.  This should incline us
to say that to the bird, directions look colored: the direction of the sun looks white, and directions
at right angles to the sun look ultraviolet.  (Directions perpendicular to the sun look the same
color to it as ultraviolet plumage.)  Why should the fact that direction is not attached to a surface
count against its being colored?  We say that the sky is blue, and it is not a surface.  
Hilbert, however, refuses to truck with any of this: 

          The only way an objectivist can meet the challenge posed by the apparent fact that color vision functions to
     detect different properties in different kinds of organisms is deny that this really [is] the case.

He denies that the bird's experiences of direction are experiences in color.  Colors are what
human color experience detects.  We could have called directions colored, but given how we
have fixed the reference of color terms, it turns out that we do not.  Personally, I think that the
conclusion is infelicitous, but part of the point I want to make is that a lot of our intuitions count
for little in the unfamiliar territory of inter-species comparisons.  It would not be wise to exclude
Hilbert's thesis on the flimsy ground that it conflicts with what I think one ought to say.  Still it
is fair to ask: why is it so important to Hilbert that non-reflectances be excluded?  This is what he
says:

          Color objectivism requires that there be some mind-independent property that is color.  If . . . there is no
     single property that all organisms with color vision are capable of visually detecting,  the objectivist . . .
     must either deny that possession of color vision entails that the organism has the ability to visually
     determine the color of a surface or give up his claim that color is an objective property.

And this is a bit obscure; it is far from clear why the objectivist must insist on there being a
single mind-independent property that is color.  Why not allow for a plurality of mind-independ
ent properties?  
Hilbert's intuition is, I believe, a realist corollary of Hume's concerning the connectedness
of color.  We make judgements of comparative similarity among certain terms, but not among
others.  For instance, we are able to attach clear perceptual sense to the judgement "Orange is
more like red than it is like green" but not to "50  Celsius is more like red than it is like green." 
It is this experienced connectedness of the color properties, some think, that makes color itself a
single property one might define color as that set of qualities, x, for which one can attach
clear perceptual sense to "x is more like red than like green."  Further, since such judgements are
comparative, they facilitate the arrangement of  colors in a multi-dimensional map or "quality
space" in such a way that if x is more like y than it like z, then x is placed closer to y than to 
z measure the distances in this map, and you have an inverse measure of color similarity.  The
map defines what has been called the "resemblance ordering" of color.  The realist might seek
to ground this ordering in an objective ordering of the qualities that color vision detects.  He
seeks a field of qualities that a) exists independently of our perception of them, and b) displays
the same resemblance ordering as experienced color. 
This, I take it, is what Hilbert is trying to do.  Spectral reflectances are real, and they can be
arranged in a numerically measured quality space.  Hilbert looks to this quality space to provide
him with a real counterpart of the unity of experienced color   real in the sense that it is
recognizable and codifiable by the scientific investigation of things outside experience.  

          If we think of triples of integrated reflectances as coordinates in a three-dimensional space, then similar
     colors will occupy adjacent regions of that space. . . Every reflectance will have a location in this space.  

          Color space is [thus] specified completely objectively, so the various color relations have an objective
     basis.  Color space is also structured in the same way as our perceptual color judgements so that there can
     be no conflict between true perceptual judgements and the objective facts.  The reference of color terms
     and the relations and properties they instantiate are objective although anthropocentric.      

The properties of directions and the like would not be a part of reflectance space that is, they
are not representable as "triples of integrated reflectances," and they would thus be arrayed in a
different and incommensurable quality space.  This is why Hilbert would like to exclude them
from the field of colors.
The trouble with this proposal is that it leans too heavily on the claim that reflectance is the
one and only thing that human color vision functions to detect.  What we have learned from the
case of the pigeon is that the mechanisms of color vision can be used to detect other things.  So if
human colors are really unified by being reflectances, it is sheer luck.  Let the pigeon try a
"columbicentric" definition in the style of Hilbert's anthropocentric realism let it define p-color
as whatever its own kind of experience properly detects and it would pull in a heterogeneous
collection of properties, not only reflectances but directions as well.  In the case of the pigeon,
then, the introspected unity of color experience fails to guarantee that the external properties that
correspond to its experiences constitute a single resemblance ordering in the strong sense Hilbert
demands.  So, the objectivist would be ill-advised to demand, in general, that the unity of color
experience be replicated in the real world.  If it is so in the human case, it is because of the
fortuitous fact if it is a fact that our color experiences detect one thing only.  If our experience
had pulled in a heterogeneous collection of environmental features, as the pigeon's does, the
resemblance ordering of reflectance would not have provided so neat a way of adjudicating
dodgy colors.
Once this point has been made, it is hard not to notice that in the human case too, color
experience does seem to gather in a heterogeneous collection of spectral emission
properties reflectances, to be sure, but luminances and transmittances as well, for lights and
stained glass windows seem not only to be seen in color, but to be experienced in ways that are
comparable, in exactly the above sense, to the way in which surfaces are experienced.  (The red
of a stained glass window is the same as that in its creator's cartoon.  It might be judged more
similar to the orange of a traffic light than to the green of a leaf.)  Indeed we are more like the
pigeon than Hilbert is ready to acknowledge.  Not all of our colors are reflectances, not all are
properties of surface.  Not every color of which we are aware is even located the blue of the
sky has a direction (like the pigeon's ultraviolet?) but no obvious location.  Is it not wishful
thinking to suppose that all the colors we perceive can be accommodated in reflectance space? 
And if human experience detects all of these different kinds of property, how can we be so sure
that direction is not colored?

          Conclusion 3: Anthropocentric definitions of color in terms of the alleged real unity of
     colors as detected by humans cannot accommodate the diversity of properties detected by
     color vision systems.

                4.  THE VARIABILITY OF RESEMBLANCE
One way of responding to the difficulties of the last section is to abandon the attempt to reduce
color to a single physical variable like reflectance, and to make it instead a quality field on to
which diverse variables get mapped.  Thus we might want to concede that color vision deals with
a variety of things: light reflecting surfaces, luminous, transparent and internally reflecting
objects and films, holograms, diffraction gratings, directions in which light is scattered, and so
on.  We could hold that though these things are diverse in their color-making characteristics,
color is somehow a common measure of all of them.  And there is surely something right about
this: we have just noted that the red of a stained glass window can be compared with that of a
printed paper and that of a light.  But our initial question remains: how is this field of qualities to
be characterized?  What is the domain of colors?  It might seem that the resemblance ordering of
colors gives us an answer at least to this question.  Waiving the question of what real similarity a
red paper and a red glass might bear to one another, why not hold that any quality that is
comparable to red and green etc. in the resemblance ordering of color, for us or for any other
organism, is a color? 
In order to evaluate this suggestion, I turn now to an argument first formulated by Hardin. 
Hardin notes that the resemblance ordering of color as humans experience it is an amalgam of a
number of similarity and contrast relations amongst the colors that do not (pace Hilbert) belong
to any quality field that exists independently of our perception of color.  He concludes that color-
realism is false: none of the things mentioned above has color independently of our perceptions. 
I do not think that this anti-realist conclusion is warranted.  However, the facts do, in my opinion,
warrant a different conclusion that color cannot be defined by a resemblance ordering.  
Here, in summary form, are some of these relations Hardin relies on to make his point.

Opponency As we have seen, opponent processing yields two hue dimensions, named after
the hue experience they cause in humans.  The "red-green" dimension of hue measures the
relative strength of the ends of the visual spectrum compared to the middle; the "blue-yellow"
dimension of hue similarly measures the strength of the short wavelength end relative to the
middle-long.  The structure of experiences occasioned by opponency correspond to nothing that
exists independently of opponent processing perceptual systems.  For example:
(i)  Since violet is a short wave color, it looks bluish, since it is at one end of the spectrum
it looks reddish.  Thus it looks a reddish blue.  Colors at the other end of the spectrum, for
example orange, look reddish too.  Because violet looks reddish, it looks more similar to orange,
with which it shares a component, than to greenish yellow, with which it shares no similarity. 
However, violet is at the opposite end of the visual spectrum from orange.  In actuality, it is
closer to greenish yellows than to orange.  Thus opponent processing distorts the ordering of
colors by wavelength.  Nor does the similarity of violet to orange have a counterpart in the space
of reflectance.  Violet occupies a region of this space that is close to dark blue and further from
orange than green.  
(ii)  Violets are experienced as mixed, since the experience involves a non-zero component
in both hue dimensions.  In this it contrasts with the so-called "unique" colors, in which one
component is null, and which are experienced as "pure."  In the spectrum, however, violet simply
occupies a waveband; that waveband is no more mixed or pure than any other.  Spectral colors
can be mixed, but the results do not parallel the mixture of hue components.  The notion of a
mixed color is ill-defined in reflectance space.
(iii)  Our experiences of black and white are the result of the activation of the achromatic
channel, which carries the result of adding the outputs of the cones.  Black-and-white "bright
ness" constitutes a component of color experience which varies independently of the hue
dimensions.  Consequently, one and the same hue can have variants that are more or less black. 
This dimension of variation has no counterpart in the visual spectrum or in reflectance space:
neither makes sense of the idea that the same hue can have different gray values.  
For these and other reasons, the quality spaces of reflectance and wavelength are simply
incommensurable with that of hue, saturation, and brightness, which is created by opponent
processing.
Categoricity  The colors to which human languages give names are experienced across
cultures as sharply different from one another.  This phenomenon, which is the cause of the
banded look of the rainbow, or of a Munsell color chart, has no counterpart in the real world. 
The real quality spaces which we have been considering as candidate counterparts for color
experience do not display any such discontinuities.
Affect etc.  Colors carry affective associations.  Red has an agitating effect on people, for
instance, while green is soothing.  The warmth and coolness of colors, discussed earlier, also
have affective associations.  Some of these are strongly cross-cultural, and seem to have a genetic
component.  Obviously they have no real counterpart; it makes no sense to say that red would
have been agitating or bold in the absence of perceivers.  Yet it seems as if these qualities are a
part of the way in which the colors are experienced and contribute to their similarity and
difference.  If two colors are experienced as agitating, or as warm, they are experienced as more
similar to each other than is predictable by their hue alone.
Hardin argues on the basis of such failures of correspondence between experienced colors
and real quality spaces that any attempt to associate color with properties that exist independently
of perception would entail the loss of the structure constitutive of color.  The price of identifying
the colors with real properties is that we are obliged to give up well entrenched similarity
judgements concerning the colors because they are based on "subjective" features of color
perception such as those enumerated above, features that originate in us, not "out there."  If we
reckon such well entrenched judgements to be essential to the resemblance ordering of the colors,
and if colors are defined by that resemblance ordering, then no real quality space counts as color. 
Color, Hardin concludes, is subjective.
Now, Hardin's argument does not exactly establish that the colors are subjective. 
Certainly, he makes a strong case for thinking that the resemblance ordering of the experienced
colors has no real counterpart, and hence that there is nothing to unify the colors.  But is it right
to conclude that red and violet and other such shades of color have no real counterpart?  There
seems to be a gap between the two propositions.  One can put the point in this way.  Anti-realists
like Hardin rely on a strong conception of what it is to be red.  If a property is to be counted as
identical with red, it must not only correspond to the experience of red, but stand in the place of
red in the resemblance structure of colors as given by experience.  If mind independent properties
are to be so counted, they must find their place in this structure independently of mind.  And one
might think that this is too demanding.  Why can one not hold that experience of red and
experience of green track properties that exist independently of us, but that the experienced
opposition between them has no real counterpart?  Why can we not say that color experience
misrepresents the true resemblance relations of colors?  (Indeed, in many cases, we are able to
say exactly why color experience does this: for instance, it treats spectral colors at opposite ends
of the visual spectrum as similar because they do not occupy the middle.  This is why it distorts
the natural similarity metric of wavelength.)  
Still, Hardin's argument does show that realists have a problem if they seek to use the
experienced resemblance ordering of colors to define what color is.  For it does show that since
this ordering is not replicable in the mind-independent world, the distinction between color and
other things would then be based not on a "difference in things themselves" but on a "different
attitude we take up" to them.  Even if we restrict color to reflectance, as Hilbert wants us to do, it
would fail the Armstrong test (see note 35 above), and not be a natural type.  But this problem
should vex anti-realists like Hardin and Thompson just as much as it bothers realists.  They have
been no more helpful with the questions of definition "What is color vision?"  "What is color?" 
They insist on the definitional priority of structures that arise out of opponent processing, in
particular, the hue-saturation-brightness resemblance ordering.  But this ordering cannot, for two
reasons, serve to define color.  
The first reason is that opponency is not the only cognitive mechanism that shapes our
experience of color.  Categoricity is an extremely important feature of human color vision
because it is crucial for naming the colors and remembering the colors of objects.  Thus it is the
basis for two of the most important things we do with color.  Yet categoricity clashes with
opponency, since the relatively sharp distinctions we experience as between the named colors are
as foreign to hue-saturation-brightness orderings of the colors as they are to quality fields that
exist independently of perceivers.  Categoricity distorts the hue-brightness-saturation ordering,
and the consequence is that the experienced ordering is different from the latter.  Hardin's
argument against the realists tells equally against his own contention that that opponency is
essential to the colors.
There is a core truth that might be thought to support the Hardin-Thompson notion that
opponent processing is essential to color vision: all actual biological color perceivers do in fact
use opponent processing.  But the second reason for rejecting their thesis is that, as I have
argued, this need not have been so: the imaginary primate of section 2 is a color perceiver though
it does not have opponent processing.  (The force of the example is bolstered by the enhance
ment of its discriminatory abilities proposed in note 21.)  In any case, why should we think that
only biological organisms can generate color concepts?  If a computer were to generate a set of
classifications on the basis of input from a digital "color" camera, it might not come up with a
hue-saturation-brightness structure since the camera does not employ the sampling techniques
and differencing characteristic of opponent processing, but encodes colors in a different way. 
The color classifications generated by the computer might correspond to the wavelength ordering
of the visual spectrum, or to reflectances, or to something else, but they need not be arrayed in a
similarity space of two or more hue dimensions plus black and white.  (We can imagine that the
computer encodes color information in terms just of luminance and a weighted average for
wavelength.)  We might want to deny that the computer had color vision on the grounds that it
doesn't see or experience anything.  But would we want to say that the qualities it detects are not
colors simply because they do not have the structure of colors generated by opponent processing? 
This sounds unreasonable. 

          Conclusion 4: Anthropocentric definitions that define color in terms of the resemblance
     ordering or other property structure found in human experience are defeated by the
     multiplicity of possible color resemblance orderings.  

                     5.  PLURALISTIC REALISM
Lewis offers anthropocentric realists a way out Hardin's difficulty.  As we have seen, he believes
that colors are surface reflectances, and that these exist independently of humans.  Despite this,
he disarmingly concedes Hardin's points about the psychophysical origins of color similarity,
simply allowing that the properties and relations that constitute the resemblance ordering of
color call them collectively R might originate in color experience.  But this is all he concedes. 
Lewis argues that we should not conclude that the field of reflectance properties fails to display
R.  He says that the correspondence between S1 and S2 above "yields relations among colours in
the image of relations among colour experiences."  Hardin assumes (as does Armstrong, among
others) that the properties and relations that structure a real resemblance ordering must arise
from the nature of things in its domain.  Lewis demurs.  Stripped of the set-theory jargon of
"images," his position amounts to this:  if two surface reflectance properties are experienced as
opposed, and reflectance properties are experienced as pure or mixed, that is resemblance
enough.  The resemblance ordering of color can be constructed from experience.  Take each
reflectance triple and plonk it down, according to how it is experienced, in the hue-saturation-
brightness space of red-green, blue-yellow, and black-white.  Then introduce the sharp bound
aries of categoricity, and the commonalities of affect and the like.  There you have the preferred
resemblance ordering of colors.  This suggestion can be extended to the multiplicity of things
that have color.  If a light is experienced as having the same red as a reflecting surface, group
them together with respect to color.  Abandon experience-independent resemblances.  The
similarity between them is simply that they are experienced similarly.  
Lewis's force majeure is tempting.  Why are red and blue similar?  Because the opponent
processing system happens to sample the ends of the spectrum as opposed to the middle.  Isn't
that all that there is to the matter?  Why should this similarity be replicated in that of wavelength
or reflectance?  However, it is difficult to reconcile this move with Lewis's brand of anthro-

pocentrism.  And it is this anthropocentrism that lies behind his own defence of realism.  

          An adequate theory of colour must be . . . commonsensical.  [This] can be compromised to some degree. . .
     But compromise has its limits.  It won't do to say that colours don't exist . . . [I]t is a Moorean fact that the
     folk psychophysics of colour is close to true.

The trouble is that the "folk psychophysics of colour" is not just different from pigeon color
psychophysics, but incompatible with it.  Folk psychophysics does not just tell us that every
reflecting surface is experienced as reddish or greenish or bluish or yellowish; it insists that every
chromatic surface is so.  (This is the sort of claim that can be elicited from "folk" in a psychol
ogy lab.)  If these propositions are close to true, what of pigeon psychophysics?  It proclaims the
existence of hues unknown to humans, and asserts that ultraviolet is a color.  It makes hue
comparisons in a dimension orthogonal to red-green and blue-yellow, and denies that every color
is reddish or greenish or bluish or yellowish.  Pigeon psychophysics must, then, be (close to)
false.  And such psychophysical differences are endemic right across the animal kingdom.  Some
of these differences are small.  The colors different humans experience as unmixed might vary a
bit from one to another, their hue discrimination abilities may be different in different parts of the
spectrum, the brightness and saturation values of their colors may not be the same, the things that
they see as colored may vary.  Even these small differences will result in different individuals
attaching different truth values to propositions like "Light of wavelength 475 nm is unique blue
(that is, blue with no admixture of red or green)."  If different organisms experience colors
differently, whose experiences are we going to use when we construct "relations among colours
in the image of relations among colour experiences?"  
At first glance, this sort of variability does not seem to be a terribly difficult problem for
the realist.  It is natural for her to say: "There is just one thing out there, but lots of ways in
which it is experienced and represented."  She seems thus to be able to account for variability as
deriving from experience not reality, and trace disagreements to the idiosyncrasies of experience. 
This approach, however, faces the difficulty that there is no one thing out there that is the real
counterpart of color experience wherever the latter might occur: the differences between the
pigeon and ourselves cannot be represented simply as different experiences of the same thing.  
The realist does not need to reject the idea that color vision can detect a variety of things. 
Properties are numerous, and it is possible that different color perceivers fasten on different
properties, or even different property structures; apparent variations in color reports need not be
taken as conflicting with one another they can be explained by the fact that people are not
really talking about the same thing.  For example, we have been saying that the red-yellow
dimension of color tracks the relative strength of the ends of the visual spectrum in the input
signal relative to the middle.  But different visual systems might demarcate the ends and middle
differently.  If the middle wave receptor is shifted slightly towards the short wave end, the
middle waveband sampled by the red-green hue component (L+S) -M will move as well. 
This will have the consequence that one and the same light might look reddish to one person,
greenish to another.  There is no contradiction here, because red does not mark exactly the same
classification in both.  In line with this, can we not say that because pigeon color is different from
folk color, the pigeon psychophysics of color is compatible with that of folk?
I am sympathetic to this position, which I will call "pluralistic realism," and what I say in
the rest of the paper is an attempt to make it biologically plausible.  Pluralistic realism stands in
urgent need of theoretical underpinning.  What does it do to the kind of realism that is based on
biological functions, which holds that there is some environmentally important set of properties
on which different color perceivers converge?  What can we say about the variations in the
qualities that are detected by different perceivers of color?  Pluralistic realism also needs to be
clarified with respect to the question of definition that we have been considering: what makes all
of these different properties color?  For it implies that there is no unique property structure no
one resemblance ordering, for example that can be appropriated tout court for the purposes of
defining color.

                  6.  THE SUBORDINATION OF COLOR
Conclusions 1-4 above imply collectively that anthropocentric definitions of color cannot
accommodate the variability of color vision across species.  Pluralistic realism seems to help
with this problem.  But there are obstacles to allowing color any significant variability.  One such
obstacle is this.  Some think that color is important as an indicator of states of affairs in the
environment it is no accident that in the two most important works on induction written in this
century, the central examples are "Emeralds are green" and "Ravens are black."  This suggests
that, despite the difficulties discussed in the last three sections (that is, bracketing questions
about the unity of color), there must be real environmental qualities on which color vision in
different species converges.  Maund captures the thought in this way: 

          Objectivists can legitimately point to the fact that colour vision is a biological endowment not only for
     humans but for many other animal species.  It seems plausible that if we possess colour vision and if we
     have evolved to possess it, then there is some objective feature in the world that colour vision picks up. 
     That is, it is plausible that our colour vision is colour-catching (and not colour generating).

If this is so, catching colors must constitute the essence of color vision.  If there is a fixed
number of colors "in the world," variation is limited: catch them all and you have full color
vision, catch only some of them, and you are color-deficient.  
It is not my aim here to refute or contest the contention that color is an important indicator
of environmental states of affairs.  My purpose is rather to explore ways in which to put this fact
into a context that makes it less of an obstacle to variability.  I mean to question the idea that
there is a fixed number of colors out there to be caught, and thus to make room for pluralistic
realism.  To this end, I shall start by considering the question that is begged by Maund's
objectivist: why do our visual systems bother with color at all?  What sort of advantage does
color vision give an organism? 
Given the degree to which philosophers take it for granted that color is useful, it comes as
something of a surprise to find that, to students of the evolution of color, it is far from obvious
why it should be.  Some think that color perception might be useful for the detection of object
boundaries.  Where adjacent surfaces have the same luminosity, they must be discriminated by a
difference of hue.  But as Mollon observes: 

          [I]n fact, it is rare in the natural world for one surface to lie in front of another in such a way that both have
     the same reflectance, both lie at the same angle to the incident illumination, and the nearer throws no
     shadow on the further.

In other words:  a) the adjacency of equiluminant surfaces is rare, and b) in situations where it
does occur, the environment tends to provide us with other cues relevant to the segmentation of
scenes.  Color is not needed for the construction of object boundaries in most situations.  
Mollon's thinking complements that of Gerald Jacobs, who notes that color vision gives us
"a multi-hued world in which objects appear to merge and contrast by virtue of their differences
in color."  This leads people to assume that color promotes contrast.  Jacobs objects: "It is
simply not appropriate to assume that a difference in chromaticity between two regions automati
cally makes them more discriminable."  "Indeed," he says, "quite the opposite may occur."  

          In testing large numbers of subjects from various mammalian species, some of which have "good" color
     vision, I have frequently observed that given a luminance difference as cue, the animal frequently uses that
     as a basis for discrimination even if a color difference known to be discriminable is also available.

Mollon thinks that color vision comes to its own in a special sort of situation:

          The colour-deficient observer is not often confronted by a homogeneous target of one chromaticity that lies
     on an equiluminant surface of different chromaticity.  Rather, his difficulty arises when the background is
     dappled and brindled, when, that is, luminosity is varying randomly. 

Obviously, this sort of target is not the norm in our environment.  Nor would an organism be
hugely disadvantaged if it were unable to function in this situation.  For his part, Jacobs argues
that the significance of color vision has more to do with what we might call the recognition of
objects than with their discrimination.  When we want to re-identify an object or put it into the
same category as others (i.e. describe it), its color may be important.  Even conceding these
points, however, it is not entirely clear, given how specialized these tasks are, whether they were
the reasons for which color vision evolved in the first place, or merely additional advantages that
followed after it became established.  For all we know, color vision was initially an advantage
not because of the visual discriminations to which it contributes, but simply because it enables us
to see in bright daylight.  
These conclusions challenge the idea that color-vision evolved because there are determi
nate qualities out there, the colors, so universally salient that widely divergent species converge
on them.  Indeed, even the older theorists that Mollon and Jacobs criticize do not make this
assumption: the criticisms are aimed at facile attributions of function to color processing.  It is
more appropriate to think that color vision is advantageous because it enhances visual discrimi
nation abilities already in place.  For example, it helps humans extend their ability to segment a
scene to situations where the background is dappled and brindled, it helps birds find their way
when there is heavy cloud cover obscuring the direction of the sun, and it helps bees find the part
of the flower with pollen.  These different kinds of organism have quite different visual discrimi
nation abilities, and use them for different tasks.  Why then should we expect that they will all be
helped by "catching" some elements out of a fixed store of colors?
In fact, we have every reason to think that this is not what they do.  Some organisms have a
colored oil-drop in front of their retinas, in order to enhance hue discrimination in specialized
lighting situations.  Should we say that they are distorting the true colors out there, like a person
who dons cerulean spectacles?  Our own color-vision systems construct sharp boundaries around
our color terms to aid the re-identification and linguistic description of colored objects.  (See the
paragraph on categoricity in section 4 above.)  Is this a distortion of the smooth resemblance
ordering of the colors?  To both these questions, the right answer is: "In a way, yes."  Both
devices exaggerate.  However, what is exaggerated (or "enhanced," as people in machine vision
say) is a helpful feature that really is there.  (Red is different from orange, though there is no
sharp boundary that separates them.)  This sort of fact should not therefore lead us to an
unqualified subjectivism.  But it should alert us to the dangers of "color-catching" realism.  The
important thing for our immediate purposes is this: each such device helps mold a species-
specific color concept adapted to environmental features and tasks specific to the organism it
serves.  The undistorted catching of qualities out there is not the biological function of color
vision. 
Given the task dependency of visual discriminations, it makes sense to think of color
properties along the same lines as the complex concepts generated by induction and other post
perceptual processes.  Imagine an impecunious but observant tourist touring an unfamiliar
country, and looking to public buildings for facilities such as public washrooms and lounges. 
Such a person might quite quickly become adept at visually identifying buildings that would
possess such facilities and at finding them once inside.  Evidently these abilities depend on some
kind of a classificatory and descriptive schema for buildings, implicit or explicit.  It is highly
unlikely that his schema would coincide, extensionally or intentionally, with those made to serve
different purposes, for example, one constructed by a jewel thief, or by a historian of architec
ture.  It is possible to think of color concepts constructed by different species in evolutionary
times along the same lines.  The practical applications of color concepts vary quite markedly
from species to species, and consequently the content and extension of these concepts does so as
well.  Thus, pluralism with respect to color concepts grows naturally out of a consideration of the
biological function of color vision, provided that we drop the assumption that this function is tied
to catching colors from a bounded store.  Note, also, that even though they are purpose-built and
idiosyncratic, and possibly exaggerated, even though they are actually incompatible, the
architectural schemata of the tourist and jewel thief may each contain some truth.  The same
applies to the color concepts used by different species. Thus we have some reason to turn to the
pluralistic realism broached at the end of the last section. 
Empiricist philosophers have generally assumed that color is a datum of perception which
gets associated with other such data texture, shape, movement, size by means of such post-
perceptual process as "custom and habit," classical conditioning, or induction.  They attempt to
account for the formation of complex empirical concepts as arising out of the association of such
data.  They have tended to overlook the possibility that color concepts might themselves have
been constructed in much the same way, for specialized tasks, not by post-perceptual processes
within an individual, but by adaptation and natural selection in evolutionary time they have
neglected Konrad Lorenz's famous dictum that what is a priori for the individual (for example,
that color resides in surfaces) might be a posteriori for the species.  Moreover, what is simple
and unanalysable for the individual (for example, the experience of yellow) might in fact have
been constructed during the evolution of the species, and what is necessary for individuals in one
species (for example, that there is no such thing as a greenish-red) might be false for the
members of another. 
The Lorenzian shift of perspective makes the color scientist's characterization of color
vision, which we encountered in section 2 above, more attractive than it might have seemed
before.  Since we are thinking of color concepts as emerging from information available to
species, their identity might rest with the information used rather than in what is made from it. 

               F.   Color vision is the ability to extract from light visual representations of qualities
          associated with its wavelength composition.

The objection to F was that it does not encourage a "thick" enough conception of color vision; it
lets in too many concepts.  But we have found plenty of reason to disavow thick conceptions. 
And so I propose, in the spirit of F:

          An organism possesses color vision if it uses wavelength discrimination as an indispens
     able part of its construction of visual representations of environmental features.

          A color quality is one that is generated from the processing of wavelength sensitive data,
     and becomes available only as a result of such processing.

The usual way of characterizing color vision is by looking to central features of its output in
humans, features of the qualia or concepts that it yields.  My proposal is that we look instead to
the information that color vision takes as input.  I have allowed that in the biological domain, the
processing of wavelength sensitive information is always done by multiple visual pigments and
some form of opponent processing.  Consequently, all actual organisms show certain very broad
similarities in their color concepts, that is, they display some version of the opponent hue
categories.  But this does not mean that color vision can be defined in terms of the presence of
these hue categories.  We know from imaginary biological examples and artificial color
processing that opponent processing is not necessary for color.  But the processing of wavelength
sensitive data is necessary for any kind of color concept: that much is agreed on all sides.  
Color vision comes in degrees dichromatic organisms can sense warm-cool, bright-dark,
other organisms being trichromats or tetrachromats can make more distinctions than that.  There
are also lateral differences between color perceivers whose discriminations are equally fine-
grained differences in similarity structures, of things detected, of behavioural significance.  All
these kinds of difference are better accommodated by a set of parameters with respect to which
different systems vary than by a framework that excludes some of them on the grounds that they
are different from the human case.  My definitions mark the ground level of such a parametric
approach, the weakest sense in which we can say that something has color vision.  More
demanding performance standards exist.  The criterion proposed by Thompson, possession of
"the phenomenal hue categories, red green, yellow, blue," is one such; it can only be satisfied by
those biological organisms that are trichromats and better, and use opponent processing, but not
by dichromats, whose hue structure is simpler, and not by color processors that abjure opponent
processing.  Such criteria define classes of systems that conform to parametric standards more
demanding than F.  What they fail to do is provide a general account of color, or color vision.  

            7.  THE VARIABILITY OF COLOR PHENOMENOLOGY
I have suggested that color concepts are to be identified not by phenomenology or content but by
the kind of information on which they are built.  It might seem that this proposal does not do
justice to the kind of question that philosophers of perception really want to answer, and the way
in which they want to answer it.  What has become of the idea that color is essentially a
visibilium, and that it must be defined by resemblances to the human visibilia that take the name? 
The intuition might be generated in this way.  Imagine a richly colored scene, the interior of a
grand cathedral, say,  with stained glass and gilt, polished wood and coloured stone.  Now
imagine a high resolution, natural contrast, black-and-white photograph of the same.  What is the
difference?  Most people would say: color is missing from the photograph.  There is something
chromatic perceivers experience when they view the cathedral in good light, and that is what we
want highlighted in any theory of color.  And it might be thought that there is some such thing
available to all color perceivers regardless of other differences.  A bird may see heliocentric
directions in color, so the intuition goes, but this must mean is that when it makes visual
discriminations of direction, it has experiences similar to those that we have in the cathedral. 
The difference is just that it associates these experiences with directions rather than surfaces. 
And it might be thought that this commonality of experience is a more telling basis for a general
theory of color than information.  
The trouble with this intuitive demand is that it is not actually plausible to think that the
colored scene presents us with a separate sort of datum that is present in all color perceivers, by
definition.  As we saw in the last section, it is misleading to think of color vision as a specialized
module dedicated to catching the colors, and added to the perceptual equipment of color-blind
organisms.  It is more appropriate to think of it as an enhancement of an underlying visual
system that is already present in color-blind competitors.  The point that I want to make here is
that the character of color experience in a given species may derive more from the nature of the
underlying visual system than from the fact that it is a color experience. 
Consider four species: humans and a closely related dichromat species, H', honey-bees and
a closely related dichromat insect species, B'.  Humans and honey-bees have advanced color
vision.  We will suppose that H' and B' do not being dichromats they lack the hue categories
that Thompson considers essential to full perceivers of color.  There are two ways of making
groups out of these species.

Functional groups
          A.  The species with good color vision: humans and honey bees.
          B.  The dichromats:  H' and B'.

Phylogenetic groups
          C.  The primates: humans and H'.
          D.  The insects:  honey-bees and B'.

It might seem obvious that since we are investigating color vision, it is the functional groups that
are important.  Clearly, they are important to the study of color-vision-as-such.  The species in
group A not only satisfy F, they can make complex discriminations of color, which, in the case of
actual biological organisms, means three or more visual pigments and opponent processing.  The
species in group B lack these endowments, or possess them only in an etiolated form.  Neverthe
less, it is a mistake to think that there must be a characteristic kind of datum that is shared by the
organisms in A.  Color phenomenology may arise out of specialized adaptations that do not cross
phylogenetic boundaries.  Because humans and honey-bees have no common ancestor with color
vision, there is no reason to think that the two color vision mechanisms developed as a response
to comparable environmental challenges.  Their color discrimination mechanisms are adapted to
the very different tasks dictated by the disparate visual systems in which they are embedded.  To
illustrate this, humans have single-lens eyes, their color processing extends beyond the retina into
cortical areas of the brain, and they use color concepts in language-like representations of
objects.  On all counts, bees are different.  Further, bees are tetrachromats and use polarization as
an integral part of their visual representations, including color, and use color vision directionally. 
Humans lack these refinements.  
Such differences show how phylogenetic groupings may actually be a lot more predictive
with respect to the phenomenological character of color vision than the functional ones.  As
Sacks tries to demonstrate, color-blind humans are able to form a pretty good conception of what
it is like for other humans to experience color, by projecting from their own visual experience. 
Presumably, they are able to do this because color integrates smoothly with the visual concepts
they share with chromatic humans.  Color-blind humans might not be so good at understanding
the color-deficient visual world of a mole or a cat, which contain visual concepts not easily
projectible from human ones.  By the same token, given the exotic character of the honey bee's
visual concepts, the color deficient visual system of B' might be a better basis for figuring out
what the honey bee's colors are like than our own color vision.  One often encounters ultraviolet
photographs of flowers with a caption like: "This is how a flower looks to a bee."  This is
misleading.  Certainly the photographs display features of the flowers that are evident to bees but
not to us.  But we are looking at those features in an essentially human way, for example with a
single-lens eye.  That gives us a poor basis for projecting how the feature looks to the bee.  An
understanding of the visual representations that are formed by a compound eye, or of the sort of
discrimination task polarized light is used for, is a better basis for figuring out the nature of a
honey-bee's color representations.  
It is true, then, that F is not in itself helpful in getting to a conception of what it is like to
see in color.  The character of color experience as it occurs in humans, or in members of other
species, is shaped by factors that are shared with color-deficient perceivers from closely related
species.  When we have nothing but color to go on, we have no basis for projecting what color
categories might be like in an organism quite different from ourselves.  This does not imply that
F is inadequate; I have argued that it defines color-vision-as-such, and what I have been trying to
show is that color-vision-as-such is not united by a commonality of experience. 

                     8.  COLOR REALISM AGAIN
Recent discussions of color realism have been dominated by the question: "Is there something
out there that correlates with the experienced similarity structure of color?"  This sort of question
does not invite a nuanced response.  Anti-realists think that the structure of experienced color is
so unlike anything out there that a positive answer is absurd: "The four opponent colours and the
perceptual space they circumscribe cannot be characterized, not at all, in terms of a distal
property."  Realists on the other hand are more impressed by the sorts of consideration
rehearsed by Maund: by color constancy and the "color-gathering" that they think it implies. 
Realists assume, often without a lot of argument, that somehow or another a sufficient parallel
ism between experience and reality can be found.  Middle ground is hard to find.  
In the light of the discussion so far, I propose that we should adopt a more complex
conception of the content of color vision.  Not only across species, but within ourselves, color
concepts are constructed from several different processes; these distort reality and interfere with
one another even while they deliver truths about the world.  Our method for discriminating lights
of different colors is opponent processing.  As successful as this method might be, the hue-
saturation-brightness dimensions of color, which emerge from opponent processing, distort the
similarity space of wavelengths.  Thus a process which is capable of delivering true and useful
messages: for example, that a red light is different from a green, or that a ripe tomato has a
surface reflectance more like that of a half ripe tomato than like that of an unripe one, is also
capable of telling us something false or misleading, for example, that violet is more similar to red
than to green.  Again, the categoricity of color perception distorts resemblance orderings to be
found in the world independent of perception by enhancing similarity around the focal colors,
and exaggerating differences across the boundaries that separate categorical colors.  Categoricity
is a mechanism that helps us remember similarities and make inductive inferences.  It also helps
us name the colors.  But it too is a source of false judgements that there is a sharp boundary
between red and adjacent colors, that the tomato is more similar to other red things than to any
brown one.  The same sort of thing goes for the other experienced color attributes.  Warm colors
are felt to be similar to one another, so are the bold colors; these similarities, the salience of
which are attested by interior and fashion design concepts, cannot be mapped onto other
experienced similarities in a smooth way, and give us messages about the world which are, at
best, partial truths. 
This fragmentation of color allows us to clarify the debate between realists and subject-

ivists by making the message of color experiences less laconic.  If we are allowed to suppose that
"color vision" tells us more than one thing, we can allow that it tells us something true as well as
something false.  To this end, we need to treat of color experience as if it were a composite
utterance, like a theory.  We can systematize color vision in terms of multiple propositions
concerning seen objects.  A perceptual state like that expressed by "That tomato is red" seems
simple, but in fact it presents us with a variety of propositions.  With considerable attention to
psychophysics and neurocomputational analysis, we can try to figure out what these are.  (Land's
work on reflectance is a pioneering study of this sort.)  But without being specific as to the exact
content of these propositions, it is important to note that such a perceptual state furnishes us with
concrete abilities to interact successfully with physical objects in the environment.  This is
evidence that color vision furnishes us with some truths, and accounts for our intuition that when
we make a judgement of this sort, then, provided that everything goes well, we learn something
about the world out there. But it is also the case that some of the propositions are false, as we
have seen. 
The view that I am proposing bears certain similarities to a theory of color that is now
attributed to Descartes and Locke.  This theory, which has been dubbed the "Error Thesis,"
maintains, in essence, that perceiving something as colored is tantamount to thinking of it as
bearing a similitude to that experience itself, which is not only false but incoherent.  I have been
telling a somewhat more complicated story.  Descartes and Locke formulated their proposal in
the context of their belief that color is simple.  They allow color perception to tell us no more
than one unanalysable thing that something is red, for example and so if a color perception
turns out to have some falsity in it, they lack the resources with which to support the possibility
that it might also have some truth in it.  I have been urging that color experience is not simple: it
furnishes us with a plenitude of assertions about seen objects.  The question of realism has more
bite in this view of color in this framework that considers partial truth and partial falsity, instead
of correspondence and matching.  
Color is a disunity, then.  There is neither a phenomenology of color vision nor a set of
shared concepts that defines color.  There is a commonality in the informational material from
which color concepts are constructed, and a consequent functional commonality in the mecha
nisms that are needed to gather this information, but no mind-independent property that all color
perceivers track or detect, no one ecological problem they all try to solve.  Color is a heteroge
neous collection of perceptual concepts generated from wavelength sensitive data for a variety of
purposes by cognitive systems with different neurocomputational structures and evolutionary
histories.                              NOTES