In Defense of Color Realism


In this article, I argue that popular explanatory frameworks in perceptual psychology suggest the truth of color realism. I focus on perceptual judgments and their evidential basis: namely perceptual representation. I first draw a distinction between two sorts of normativities with respect to which we can evaluate representational capacities and systems: biological and psychological normativities. The former is defined in terms of evolutionary fitness, and the latter in terms of representational accuracy. Generally, representational systems achieve psychological and biological success (i.e., facilitate survival and reproduction and accurately represent the world) hand in hand, but in special circumstances, a representational system can be such that it serves the organism best while not generally furnishing the organism with accurate representations. I argue in this essay that an explanation how and why our ancestors developed color vision that cleaves its biological success conditions from its psychological success conditions cannot be given unless we are prepared to say the same about the visual perception of distance and other geometrical features of the world. Moreover, it is difficult to see how the anti-realist could begin to specify psychological success conditions for color representations. Hence, we ought to accept the biological utility of color vision as evidence that it is typically representationally successful, and regard our perceptual judgments about color as generally true.

In what follows, I will give a positive argument for color realism informed by developments and theoretical commitments in contemporary perceptual psychology. I argue that the best explanation that we take our judgments about color to be veridical is that they are veridical. I begin by distinguishing two sorts of norms: psychological and biological. I then propose a particular way of thinking about the way psychological and biological norms govern perceptual judgments. In the following section, we look at how visual functions are carried out in a case where psychological and biological norms uncontroversially coincide in humans, the visual representation of size and distance. I then survey some of contemporary frameworks to understand the nature of color perception. In the penultimate section, I argue that we have no better basis for claiming that color vision, understood in some of the ways contemporary psychologists have proposed, is systematically misleading in a way that perception of geometrical features such as size and distance is not. Specifically, we can only reject the coincidence of psychological and biological norms for perceptual judgments about color if we do so in the case of size and distance as well. As this is a consequence few would accept, we ought to be realists about color.

Perception and Veridicality

My overall aim in this article, as stated, is to defend the claim that colors exist. The overall strategy for doing so will be to vindicate perceptual judgments, and in particular first-order judgments, about color as true, i.e., “This [demonstrating] is blue,” “This [demonstrating] is red,” etc. We will begin by discussing perceptual judgments and the norms that govern them. Psychological norms govern not only judgments but all forms of mental representation, while biological norms govern living organisms and their parts. Each has application to the notion of a perceptual judgment. We will first provide an explanation of each of these sorts of norm.

Psychological and Biological Normativities

Psychological normativity is defined in terms of veridicality; a representation is “successful” insofar as it is accurate. Biological normativity is defined in terms of survival and reproductive success: obtaining food, avoiding predators, and propagating the species. Examples of non-representational biological norms are obvious and plentiful: the heart functions well when it circulates blood; muscles when they facilitate bodily motion; and so on. It is quite clear that the two regularly coincide — if I falsely represent a stable and sturdy surface in front of me where there is in fact a steep dropoff, I may easily come to injury or death. In such a case, the failure to produce a veridical representation impedes the capacity to survive. In other words, failure to meet a psychological norm entails the failure to meet a biological norm.

Shepard (2001: 581) proposes, along such lines of reasoning, that organisms that have the capacity for representation, i.e., those who are subject to psychological norms about veridicality, have representational capacities shaped by the environment along biologically normative lines. He writes,

The ways in which genes shape an individual’s perceptual and cognitive capabilities influence the propagation of those genes in the species’ ecological niche just as much as the ways in which those genes shape the individual’s physical size, shape, and coloration. A predatory bird has come to have not only sharp talons but also sharp eyes, and a small rodent has come to have not only quick feet but also quick recollection of the location of its burrow.

A bird’s capacity to represent visually and a rodent’s ability to recall an important location are psychological capacities that promote biological success. When the bird sees its prey accurately (psychological success), it is able to acquire food (biological success).

Psychological and biological normativities do not align perfectly, though. Burge (2010: 302) presents us with a possible sort of example in which biological success is not only independent of, but facilitated by, representational failure, in which an avoidance mechanism functioned to increase strength and agility—in avoiding the predator—even in cases in which the animal engaged in avoidance behavior, because of a misrepresentation as of a predator, when no predator was present. Suppose that in each case, whether or not the predator is present, the avoidance mechanism contributes to the animal’s fitness for avoiding predators. Then, although the ultimate raison d’etre for the mechanism might be absent in a given case, there would be no biological sense in which the mechanism failed to fulfill a biological function when it effect avoidance behavior in cases where the distal condition was not present. ...Failure of accuracy need not be failure to realize any biological function.

We can see then that psychological norms are not just a special case of biological norms. A representational faculty can be faulty with respect to psychological norms, yet functionally adequate with respect to biological norms.

Whether we take a generally biologically successful representational capacity to be generally psychologically successful or not will depend on the question whether the presence of the capacity in the species that have it is best explained by its being psychologically adequate. Other things being equal, this is the case. Things are not equal in the case of the capacity that Burge describes. Avoidance behaviorFootnote 1 is best explained by the representation of the thing to be avoided. It is hard to see what veridical representation could be in place to cause similar behavior in a way that would fit neatly into our present understanding of the biological sciences.

This clarifies the distinction between between psychological and biological norms. The former concerns how we represent the world, and the latter our ability to survive and reproduce, hence our abilities to successfully navigate the world, obtain nutrition, and mate. As a defeasible rule, representational capacities help us fulfill biological imperatives, e.g., by representing the environment to be navigated, classify objects (e.g., as food or not-food), and recognize conspecifics. We can turn now to our bases for assessing perceptual judgments with respect to these norms.

Perceptual Judgments

What I call perceptual judgments are demonstrative judgments about how things appear visually to a subject. Some examples of perceptual judgments are (where, in each case, the demonstrated object is present to the subject):

  1. 1.

    That’s an elm tree.

  2. 2.

    That’s a pickup truck.

  3. 3.

    That object is nearby.

  4. 4.

    That object is red.

Each of these judgments picks out a perceived object by way of perception and categorizes it. What distinguishes such judgments is the use of perception to fix the reference of the demonstrative in order to pick out the object to be categorized. This is what is distinctive about perceptual judgments—the use of perception to fix the reference of a demonstrative in order to cateogorize an object.

The third item on the above list is distinctive with respect to the previous two. The categories which those judgments place objects into—elm tree, pickup truck— are categories that no human being (or other actually existing organism) possesses without a great deal of learning. A geometrical concept such as nearby, on the other hand, seems to be a sort of concept that a subject will possess in virtue of their ability to make perceptual judgments at all. I will argue below that the same is true of perceptual judgments about color.

My reasons for thinking that perceptual judgments about geometrical features of one’s environment are foundational to the capacity for perceptual judgment have to do with the evolutionary history of perception and perceptual judgment in our ancestors. I will use the example of depth perception to begin to make my case. Heesy (2009: 28) states that “interposition, perspective, motional parallax, vergence eye movement cues, accommodation, and optic flow are almost certainly used by arboreal taxa with divergent orbits to determine depth during locomotion,” where divergent orbits refer to those on organisms such the squirrel or tarsier, whose eyes point away from each other. The more complex mechanisms of depth perception, such as use of stereoscopic information, are believed to have been an adaptation in mammals that provided visual information permitting arboreal predatory mammals to “target, track, and seize prey” (Ibid.: 24). The upshot of this is that our capacity for depth perception phylogenetically predates the capacity for judgment.

Information provided by depth perception in creatures that lack the capacity for judgment is used to guide action. Specifically, the animal is able to navigate its environment appropriately where depth and distance are ecologically relevant. A predator, for instance, is better able to hunt because it moves its body in a way to capture prey where it is, rather than fail to do so where the prey is not. An animal needs to be able to locate where it needs to be, or not be, in order to carry out its biological imperatives, and depth perception, like other forms of perception of geometrical features, facilitates this.

Now, consider what we get when we add to this array of representational capacities something further, the capacity for propositional thought. This capacity will not come about in a vacuum, but as an adaptation that builds upon those features an organism has inherited from its evolutionary ancestors, just as the mechanisms of depth perception imbued upon mammals by convergent orbits built upon, rather than replaced, those mechanisms shared by all arboreal taxa. The capacity for propositional thought will be integrated with the representational capacities an organism has inherited.

Consider the sorts of judgment that a creature of a species who, having inherited the capacity for depth perception from distant evolutionary ancestors, also acquired the capacity for propositional thought from recent evolutionary ancestors. In virtue of the latter capacity’s being integrated with the former, it is able to think certain thoughts. Those thoughts are going to be all and only the thoughts whose representational content is commensurable with that of the visual representations licensed by the capacity for depth perception, i.e., objects, surfaces, and distances: “This object [say, a branch] is closer to me than that object [another branch].”

What goes for perceptual judgments about distance goes for perceptual judgments about geometrical features of one’s environment in general—when we make them rationally, we make them on the basis of the content of visual representation alone. In essence, in making such a judgment one is simply consciously endorsing what is already purported to be the case by one’s visual representation.

The route has been circuitous, but we have reached the lynchpin of this chain of reasoning: the sort of perceptual judgment that I have in mind, of which those concerning geometrical features are paradigm cases, are such that they consist in no more than affirming that which is represented to one in perception.Footnote 2 It follows that if the visual representation p concerning the geometry of the environment is veridical, a judgment that p based upon the representation will be so as well.

We return now to the distinction between biological and psychological normativities outline above. Suppose organism O has a representational system R whose biological utility coincides with its psychological success conditions, i.e., the veridicality of the representations it furnishes O with contribute to its ability to survive. R’s survival value consists in its furnishing O with accurate representations. Then the veridicality of any representation r furnished by R will, in normal conditions, be demonstrated by the biological success of O in acting on r. For instance, suppose that the depth perception of the arboreal predatory mammals mentioned above is a system like R. In acting on the representation of prey at such-and-such a distance, such predators prove the representation of the prey’s distance accurate by striking and catching its prey.

I believe that the evolution of depth perception in human beings and their ancestors is best explained by its ability to furnish us with accurate representations of the environment. I believe the same about our visual representation of environmental geometry in general. I belabor this point because I wish to draw a compelling analogy between the perceptual judgments #3 and #4 in the list that began this discussion of perceptual judgment, and more generally between perceptual judgments about geometrical features and colors. In order to do so, I will first compare and contrast one relatively simple mechanism of depth perception with some models of color perception. This will provide a framework within which to make the case for color realism.

Perceiving Environmental Geometry

In this section, we will first motivate and lay out the general problem facing visual system in the various tasks it carries out. The question is, in essence, how the eye and brain are able to recover accurate information about the immediate environment from an optical signal that contains infinitely less information than vision makes available. We will then look at some of the mechanisms that perceptual psychologists propose in the case of some geometrical features of the environment: size and distance.

Optical Information and Perceptual Representation

The Necessity of Information Processing

Perception consists in an outer and an inner phase. The outer phase consists in the illumination of the environment, the refraction of light from surfaces, and the convergence of light in an optic array upon the retina. However, it is the inner phase that we are interested in. We can think of the inner phase of visual perception as a process by which the visual system of a perceiver extract information about the environment from the optic array with which it is presented. It is a causal chain beginning with the stimulation of the retinal field, and terminating with the perceiver being in a cognitive state that represents the environment as being a certain way, which licenses judgments about the world on the basis of said visual representation. Since we are concerned to examine whether there is some disanalogy between the process by which we come to have visual representations of color and of geometrical features of the environment, such that only the latter are generally veridical, we will bracket the question with how much accuracy human vision represents the environment, and focus on the functions that characterize such a capacity.

The way we will characterize the inner phase of visual perception can be motivated by contrasting it with that of the psychologist James Gibson. As mentioned above, he eschewed the information processing model of visual perception. “Being intellectually lazy,” he writes, “we try to understand perception in the same way we understand communication, in terms of the familiar.” But, “we cannot explain perception in terms of communication....” (Gibson 1979: 63) But this will not do. Without delving into the debate that continues today between Gibsonian and mainstream perceptual psychologists, I briefly mention Gibson’s alternative approach. I do so not to give a definitive refutation of any particular claim of his, or to give a sustained critique of the Gibsonian paradigm, but to set the stage for a precise expression of the problem that an representational, information processing approach to visual perception takes to be central to the study of vision.

Gibson takes perception to be “an activity of the whole [animal] system,” or at least the “retino-neuro-muscular” system, hence that perception does not consist in the receipt and processing of information from light. Light certainly does convey information obtained through perception, but that information is only ever obtained by the perceiving animal per se, not the retina and visual system of the perceiver. Features of the environment are not, for Gibson, discovered by the eye, optic nerve, and brain; they are merely “attended to” by the perceiver.

Gibson (2002: 85) would thus deny that there is a problem with depth perception of the sort we described in the previous section. Regarding the “false problem” of depth perception, he writes:

Perception of the environment differs from a perception of space. An environment implies points of observation in the medium, whereas a space does not. The points of geometrical space are abstract fictions, whereas the points of observation in an environment are the positions where an observer might be stationed. Perception of the environment is thus accompanied by an awareness of the perceiver’s existence in the environment... whereas a perception of space in its purest form need not be accompanied by any awareness of the thinker’s existence in that space.

So Gibson believes that we misconstrue the task of vision when we describe its accomplishment, its informational deliverances, in abstract spatial terms. Rather, through vision, we learn our position in an environment consisting of a particular layout. By possessing a visual presentation of the layout, we are able to judge distance, for if we weren’t, we wouldn’t possess the information that allows us to navigate the environment, which for Gibson, is tantamount to saying we have the visual capacities we in fact have.

If this explanation for the capacity to judge distance sounds a bit hollow, that is because it is. To Gibson’s credit, he has the beginning of an explanation for how we are able to perceive depth—he believes that among the pieces of information carried by ambient light is the perceiver’s location relative to the rest of their environment as well the information necessary in order that they navigate that environment, and that the animal is, by nature, visually sensitive to such information. But the question remains, how is the animal sensitive to such information?

Marr (1982: 33), in responding to Gibson’s account of our ability to perceive objects has having a stable and consistent size, wrote “Yes, to be sure, but how?” The rhetorical point of the exclamation is that recovering the necessary information, and only the necessary information, is a task that requires some processing of the information. Recovering the information is a problem that requires a solution because the information we require is underspecified by light itself in the way mentioned in the previous section; the same ambient array could be produced by infinitely many environments, and we only arrive at a single, determinate environment by discounting all but one of those possible environments. Hence, we do not posit an information processing or “communication” model of vision because of intellectual laziness, as Gibson suggested; we do so because there is no other way to account for the reduction of possible environments causing the ambient array from infinitely, perhaps uncountably, many, to the single environment we generally correctly perceive.

Information and Representation

In order to precisify this account of where Gibson’s theory of vision falls short, I will introduce, briefly, a distinction between information and representation. We will discuss information primarily in respect to its epistemic significance. It is a necessary and sufficient condition for an event E to carry information that p relative to a set of true propositions K just in case given knowledge of K, I am in a position to deduce p on the basis of my awareness of E.Footnote 3 If my doorbell rings, then given my knowledge how doorbells work, and supposing it couldn’t have rang due to a glitch, I am in a position to come to know that the doorbell button has been pressed. So, when I hear the doorbell ring, it carries the information for me that the doorbell button has been pressed. If I know that the doorbell button could only have been pressed by a person, then I am in a position to know that someone is at the door—and then, the doorbell ring carries for me the information that someone is at the door.

Representation, in the sense we have been using the term, is a matter of an item’s possessing conditions of satisfaction. Canonical examples of conditions of satisfaction include a declarative sentence’s being true, or a map corresponding correctly to the actual lay of the land it purports to represent. Any item of representation will generally carry more information about the world to a subject than it will have representational content. As Dretske (1981: 43) points out, “If I already know that someone lives in Wisconsin, but you do not, hearing him say he lives in Madison tells me where he lives but not you. A glance at the calendar tells you the date only if you already know the day of the week.” And, as the doorbell example shows, nonrepresentational phenomena can carry information just as well as representational phenomena.

Information cannot have conditions of satisfaction. Dretske (1981: 44) writes that “[w]hat information a signal carries is what it is capable of “telling” us, telling us truly, about another state of affairs.” He continues,

When I say, “I have a toothache,” what I say means that I have a toothache whether what I say is true or false. But when false, it fails to carry the information that I have a toothache because it is incapable of yielding the knowledge that I have a toothache. No one can learn that I have a toothache when I do not have one.

Hence, “false information and mis-information are not kinds of information - any more than decoy ducks and rubber ducks are kinds of ducks. And to speak of certain information as being reliable is to speak redundantly.” Whatever the properties of our ordinary concept of information, in the sense intended here, the notion of carrying information is factive.

Optical Information and Perceptual Representation

We can return now to the problem of information transmitted by ambient light. I begin by stating two principles. First, if one event causes another, then as a signal the caused event carries all the information that the causing event does, for a suitably informed subject. To be specific, if an event E carries information for S that p, E causes E\({~}^{\prime }\), and S knows that E caused E\({~}^{\prime }\), then E\({~}^{\prime }\) carries the information for S that p. I will call this the principle of causal information sufficiency (“causal sufficiency” for short).

The second principle is what Burge (2005: 22) calls the “Proximality Principle.” He states it as such:

Holding constant the antecedent psychological set of the perceiver, a given type of proximal stimulation (over the whole body), together with associated internal afferent and efferent input into the perceptual system, will produce a given type of perceptual state, assuming that there is no malfunctioning in the system and no interference with the system.

To simplify, we can say that the Proximality Principle tells us that for any retinal image (that is, any set of values representing degree of stimulation of the photoreceptors in the retina), there is one and only one representational visual state that will be caused by a perceiver’s having that state.

Whatever information about a perceiver’s environment is carried by the retinal image, the same information is carried by the ambient optic array of the perceiver an instant earlier in the causal chain leading up to the production of that retinal image, modulo signal noise. By causal sufficiency, we can say that the retinal image carries all of the information (modulo some interference due to passage through the eye) that the ambient optic array does. But we have already said that a particular ambient optic array is consistent with infinitely many possible environments existing outside the perceiver that it carries an infinitely disjunctive set of information about that environment. So, similarly, does the retinal image.

It follows from the principle of causal sufficiency that for any neural event in a perceiver in a causal chain beginning with the obtaining of a particular image on their retina, that event carries information consistent with infinitely many possible environments existing without the perceiver. In conjunction with the Proximality Principle, it follows that whatever neural events or states form the supervenience base for the obtaining of a particular representational visual state carry carry information consistent with the environment without the perceiver being infinitely many ways. This poses us with a question: how do we represent the world in a determinate way, much less a generally veridical way, given a seemingly incorrigibly, interminably ambiguous proximal stimulus?

To answer this question, we need to pursue explanations about and beyond the terms of chains of neural events beginning with retinal image. I have argued here that such terms will fail to narrow down the range of possible environments without the perceiver. As far as the sorts of explanation that will suffice to answer the underdetermination problem, I will proceed by presenting specific examples. First, we will return to our initial consideration about underdetermination, the way the visual system comes to represent distance. Then, we will consider the underdetermination problem with respect to color. Finally, we will consider the general features of such explanations and whether it would be plausible to introduce some disanalogy by which color perception is specifically untrustworthy.

Perceiving Distance

There is an underdetermination problem in the perception of distance. Palmer (1999: 201) states the problem as such:

...the optical processes of surface reflection and image formation project light from a 3-D world onto a 2-D surface at the back of the eye... such projections can be inverted in an infinite number of ways. Depth perception is thus the paradigm example of logical ambiguity in perception.

In essence the problem is this: any given retinal image is consistent with a perceived object being x distance away given the angle formed by the ray between the light source (e.g., the sun) and object and the ray between the object and perceiver is n degrees. However, the same retinal image is consistent with the same object being 2x distance away if the angle is (n)(.5) degrees, 3x if the angle is (n)(.25) degrees, and so forth. The question is, given that we can veridically perceive objects at distance n, how does the visual system discount the false possibilities that the object is at 2n, 3n, etc.?

There is, in fact, no single answer to this question. Perception of distance makes use of a information pertaining the the state of the eyes themselves (“ocular information”) as well as information garnered from the retinal images (“optical information”) (Palmer 1999: 203). Burge (2010: 347) notes that “Standard theory currently attributes to the human visual system as many as twelve basic capacities for determining distance and depth....” We will focus on the use of one form of ocular information, convergence. Convergence is the “extent to which two are turned inward (toward each other)... when both of them are aimed directly at the point so that light coming from it falls on the centers of both foveae simultaneously.” (Palmer 1999: 205)

The “crucial fact about convergence” that allows the visual system to exploit it for the purpose of representing distance is that “the angle formed by the two lines of sight varies systematically with the distance between the observer and the fixated point.” (Palmer 1999: 205) The visual system can exploit this fact to represent distance up to a few meters. Burge (2010: 349) elaborates:

The constant distance between the two eyes is coded into transformational processes. The fixation point of the two eyes is determined as follows. There is only one central point in the fovea of each eye. Each eye can be pointed in one direction at a time. The two lines marking those directions intersect at one point. The visual system has information that statistically correlates strongly with the angle of each eye and the stimulation registered at the center of each fovea. So (given the length of a line and two angles of a triangle) the visual system has sufficient information to determine the location, hence distance, of the fixation point - the position of the represented object. A perceptual representation as of a representatum at a given location, hence distance, relative to the viewer is formed from this information in accord with the relevant geometrical principles.

We see here—a point we will return to shortly—that the visual system encodes assumptions in order to solve an underdetermination problem. It exploits a regularity that holds between distance and variables the system has access to convergence.

Color Constancy and Representation

The visual system is able to represent features of the environment as remaining constant despite great changes in the retinal image. It is also able to represent, e.g., surfaces, as being in some way homogenous despite being differently illuminated at various parts. As an example of the former, consider the case in which I bring a red tomato from the grocery store to an outdoor parking lot. The composition and intensity of light changes dramatically when I walk out the door, and consequently my retinal images change dramatically. Yet I perceive little to no difference in the tomato itself. As an example of the latter, consider a white wall in a well lit room with a shadow cast upon some portion of it. The wall will be seen as homogenous, with portions differing only in the intensity of their illumination. The phenomenon of perceived sameness in objects despite synchronic or diachronic differences in illumination in emph{color constancy}. Brainard (2009: 1) concurs with this definition; he regards color constancy as our capacity to perceive colors as “stable against transient features of the environment in which the objects are viewed.” And Foster (2003: 1) defines the phenomenon similarly, as “the constant appearance of object or surface color despite changes in the colour of the illumination, and... scene composition and configuration.” Beyond this initial definition, however, we will see that there are significant diversity of opinion as to the nature of color constancy and how it is achieved by the visual system. Color constancy is critical for our discussion because it tells us what is the content of our perceptual judgments about color. As we will see, color constancy either determines the content of color perception as such, or it determines how we arrive at perceptual judgments about color on the basis of conscious experience. Either way, it is key for understanding the veridicality conditions on perceptual judgments about color, i.e., what it would take in order that realism be true.

Projective and Phenomenal Color Constancy

Projective Constancy

To regard color constancy as projective is to maintain that the purpose of color constancy is to recover information about stable surface properties in a perceiver’s environment despite variance in the qualitative appearance of objects in that environment. The putative distinction between stable color (putatively represented by way of projective constancy) and such qualitative appearance was proposed in contemporary empirical literature by Arend and Reeves (1986):

...hues and saturations might change when the illuminant changes but be perceived to result from constant surface colors and varying illumination. The paper that looks unique yellow under direct sunlight might look greenish yellow under the tree and yet might be clearly identifiable as a yellow paper. That is, perfect constancy could still obtain if the viewer, by a perceptual computation, were able to see the paper as an object of the same surface color under illumination perceived to be greener than the direct sunlight. The difference between this type of mechanism and the sensory mechanisms is not in the computation required but in the perceptual representation of the information computed.

It is quite straightforward to see how a project in the ontology of color could do well to embrace this picture of color constancy: in a veridical representation of color, some stable property of a surface is successfully represented. We need only ask what this property is.

Brainard (2009: 395) regards the primary source of confounding in the signal that reaches the retina as the fact that “the cone responses do not completely determine the reflectance properties of the object.” He thus builds on our initial characterization of color constancy: “To provide a representation of object reflectance that is stable across changes of illuminant, the visual system must process the color signal to separate the physical effects of illuminant and object surface.” If we regard the task of color vision in this way, it is a natural conclusion that it represents reflectance properties of surfaces, hence that colors are reflectance properties.

The common thread among accounts of projective constancy is the attempt to explain the visual system’s solution to the problem of color is that some information about what the environment is like is assumed by the visual system. Such information might pertain to the ratios of light reflected between a surface and one or more additional surfaces in the visible environment (the surfaces adjacent to the one under consideration, to be specific), as in Land’s Retinex theory. The idea here is to estimate the contribution of illumination to the retinal image, and discount that information in representing the colors of surfaces. If the visual system can figure out at each wavelength how much ambient illumination has contributed to the intensity of the signal, it is able to solve for the contribution of a surface in the environment.

An alternative approach is to look for some particular value by which the visual system can calibrate its estimation of the illuminant. Evans (1951:3) proposed that the contribution of illumination to the retinal image can be approximated by taking the spatial average of the light reflected from a scene to be neutral across the spectrum. Another suggestion is that the visual system treats the surface in the environment that reflects light most intensely as spectrally neutral.

Brainard’s own proposal is that the computation of surface reflectances can be modeled using the apparatus of Bayesian decision theory. Rescorla gives a summation of the color-perceptual process according to the Brainard’s approach. The visual system, through a Bayesian inference, based upon overall retinal stimulation, that deploys a prior probability over possible illuminants and possible surface reflectances. To a first approximation, the illuminant prior assigns higher probability to illuminants that resemble natural daylight, while the surface prior assigns higher probability to surface reflectances that occur more commonly in the natural environment. This framework can explain both the success and the failure of human color constancy under various conditions.

The primary remaining question, at this point, is just what probabilities are assigned, and why. Brainard proposes that the visual system assumes priors that capture “statistical regularities in the natural environment” that are “deeply embedded in visual processing.” Just what these regularities might be is the empirical question that the Bayesian approach has made into an empirical research project.

Phenomenal Constancy

Arend and Reeves (1986: 1749) note that subjects’ judgments about sameness of color are sensitive to the instructions they receive in experiments. Specifically, their judgments diverge when they are asked to match surfaces according to hue and saturation, rather than whether they are “cut from the same piece of paper.” Thus they describe another, non-projective way of regarding color constancy:

Constancy of perceived surface color might occur because hues and saturations are invariant under illuminant change (spatial or temporal), with no perceptual representation of the illuminant difference in terms of local color qualities. In this case the invariance is produced by a sensory mechanism (e.g., adaptation or contrast) that adjusts the response to cancel the proximal stimulus difference produced by the illuminant change. For example, a paper that looks unique yellow under direct sunlight might continue to look unique yellow in the greenish reflected illuminant under a tree.

On this conception, color constancy is achieved when the qualitative appearance of the perceived object is preserved despite changes in illumination. This is phenomenal color constancy.

Phenomenal color constancy seems to be quite weak in normal human subjects, compared to projective color constancy (Arend and Reeves 1986: 1747).Footnote 4 Wright (2013: 438) entertains the hypothesis that it plays no role at all in attributions of color to objects:

That subjects can separate their judgments about phenomenal color and surface properties, with (at times) the former leading to poor constancy and the latter to good constancy, challenges the idea that color constancy is rooted in stable phenomenal color. On the basis of such instruction effects, projective understandings of color constancy come out as the main alternative to an appearance-based conception. The non-phenomenal attribution of color to objects may be so effortless and ubiquitous that it swamps awareness of variations in phenomenal color, thus explaining the intuitive sense that things look to have the same color when viewing conditions shift.

Nevertheless, this understanding of color constancy has been defended both by philosophers and psychologists. Understanding color constancy phenomenally should not be written off entirely not only for this reason, but also because there are serious doubts about the empirical basis for projective color constancy.

Hilbert (2005: 151) touts theoretical benefits to regarding color constancy as phenomenal along more than the usual three dimensions (brightness, hue, and saturation) in order to accommodate failures of constancy due to qualitative variation. He posits non-chromatic dimensions of qualitative appearance that represent changes in illumination (Ibid.: 150):

When we look at the printed page under indoor illumination we see not only that some parts of it are white and others are black but that the whole of it is dimly lit. When we look at the same page outdoors we see the same distribution of colors but also see that it is brightly lit. When I look at the keyboard of my computer I see both that the keys are a particular shade of white all over and that the illumination on the sides is dimmer than the illumination on the tops. The intuitive description of the pattern of stability and change in our perception of color is, according to this proposal, the right description.

The idea is, in essence, that visual representation of color and illumination are carried out together. Any representational system that is in a position to separate information that reaches the retina due to illumination and reflectance is in a position to represent both. When we take into account that qualitative changes are often caused by changes in illumination, we might take the change to be one in represented illumination (or, to be more precise, a change in the effect of illumination).Footnote 5

Wright (2013: 450) brings to bear results from Reeves, Amano, and Foster (2008: 225) against Hilbert’s proposal. Even if we introduce further dimensions into the representation of color in order to account for qualitative variation without difference in color, there are conditions under which attending to hue and saturation when making color matches fails to track surface properties. Rather, it follows that illumination and surface properties are confounded in the subjects’ appearance-based matches, casting doubt on the idea that phenomenal constancy is properly considered color constancy.

While Hilbert takes color constancy tout court to be phenomenal, Reeves, Amano, and Foster (2008: 226) propose that projective and phenomenal constancy both exist and are distinct.Footnote 6 They present a useful comparison to the perception of heat:

As one moves toward a fire, the sensation of warmth on the skin increases, but the “heat” that one attributes to the fire does not; the projected quality (heat in the fire) is clearly distinguishable form the sensation. Interestingly, constancy is also illustrated here, since distance is discounted when the inferred heat is projected back into the fire and experienced as a property of the fire.

Ignoring the odd metaphor of “projecting the heat back into the fire” (we might take this to mean that the visual system represents the heat as a stable property of the fire), we see how there may be dual representation in the case of heat: projective heat constancy permits one to represent a stable thermal property of the fire, while phenomenal heat constancy facilitates a representation that takes into account both the thermal property and one’s proximity to it. Similarly, projective and phenomenal color constancy may work in tandem.

There are three points to make in distinguishing these two forms of color constancy, projective and phenomenal. First, to stake out the sorts of mechanism possibly underwriting and explaining our capacity to make perceptual judgments about color. The case that those judgments are by and large veridical will depend upon the nature of that mechanism, thus each ought to be addressed. Second, to illustrate that it may be there is no one mechanism that underwrites and explains our capacity to make perceptual judgments about color. This is a possibility that we ought to address in arguing that our judgments about color are, by and large, veridical.

The third point to make is that there is empirical evidence against either of these forms of color constancy, qua perceptual mechanism. Should it turn out that the visual system does not make use of projective or phenomenal color constancy, in the absence of an alternative explanation how color constancy is achieved in vision, we are forced to conclude that constancy in perceptual judgments about color is a purely cognitive phenomenon—that it is a matter of explicit reasoning by the subject, rather than computation by the visual system. We will return to this possibility below.

Relational Color Constancy

I turn now to a phenomenon that calls into question the empirical data in favor of either form of color constancy. Foster (2003) asks the question “Does color constancy exist?” after noting that experiments plausibly fail to measure color constancy as such, registering instead relational color constancy, the “constancy of the perceived relations between the colors of surfaces under illuminant changes, rather than of the perceived colors themselves” (Foster 2003: 680). It merely “enables judgments about whether chromaticity changes in a scene are due to an alteration of illuminant or material properties” (Wright 2013: 445) rather than visual representations and judgments about surface properties as such.

Foster (2003: 439) sets out as a requirement for color constancy to be achieved that its mechanism must be sensitive to changes in surface properties.Footnote 7 One of the primary experimental methods in color constancy research, and the most fruitful extant method, is color matching, in which subjects are asked to manipulate controls that adjust the color of one surface to match that of another, differently illuminated surface.

The problem, Foster (2003: 441) argues, is that “Because information about individual surface reflectance is not needed to succeed at asymmetric colour matching, it cannot measure colour constancy....” Consider the geometric analogy he offers: we can determine whether or not the angle formed by any pair of lines is the same as that of another pair without having any knowledge of the orientation of any member of either pair. In that case, we need only compare the angles to each other, without any sort of calibration and measurement of either angle. Similarly, the color matching exercise can be carried out by comparison of the two surfaces alone, and not by measurement of each surface for comparison of its illumination-independent properties as such to those of the other surface.

The upshot of this is that we must regard the two ways of regarding color constancy just mentioned as tentative. Improved experimental methods and further research will perhaps vindicate an attractive picture of the operation of the color-visual subsystem, but in arguing for color realism, we must also cover the eventuality that our present frameworks for understanding color constancy, and the bases they provide for an account of the content and veridicality conditions of judgments about color, may be misguided and unilluminating. This eventuality is compounded by the question whether color constancy can be regarded as purely perceptual, rather than partially or wholly cognitive, a problem that we now turn to.

Projective Constancy and Higher Cognitive Functions

Recent findings from Radonjić and Brainard (2016: 862) suggest the possibility, though inconclusively, that judgments associated with projective constancy are in fact the product of explicit reasoning. He found that when asked to make judgments associated with projective constancy, subjects often indicated that they employed explicit reasoning. Here is a sample response from a subject who was asked to match apparent reflectance:

First I decided whether the illumination was the same on the two sides of the cube. If it was the same, I looked between the target and the test button several times to decide which matched. If the illumination was different, I tried to identify how: I identified types of illumination by lighter or darker, more yellow or more blue. Then I tried to keep the image of the target button in my mind’s eye while imagining how it would look under that type of illumination to find the right test button.

It is clear that the subject is describing an explicit reasoning process.

This does not show that there is no projective perceptual color constancy, but it does show that empirical support for the existence of projective constancy is more limited than previously thought. We should be wary, in any case, of adopting the stance that perceptual judgments about color are, categorically, inferences from beliefs. For one, it is widely believed that color perception is a faculty had by organisms certainly not capable of inference, such as honeybees (Neumeyer 1980). In fact, the trichromatic color vision honeybees of honeybees is quite similar to that of humans. It would be odd if honeybees, with perceptual apparatus similar to our own, were imbued with visual representations of color while we are not. It would be a difficult case to make, as well, that we are cognizing about the same sort of property that honeybees are representing in perception. A revision of our representation of color, as well as that of other species, would be in order if a purely cognitive picture of human representation of color turned out to be true.

Another reason to be skeptical of the idea that we always infer perceptual judgments about color comes from the ostensive fact that there are perceptual illusions in which color is inaccurately represented. The Benham disk illusion, for instance, seems to be one in which a black surface is periodically falsely represented as being red. Even if one were prima facie inclined to judge that the surface was periodically blinking red, upon learning that the appearance of the surface is misleading, its seeming to blink red would persist. If false representation of color were always a matter of inference, then knowledge that the appearance of the surface is misleading would preclude all misrepresentation of it as red, as such knowledge would, other things being equal, render inferences that the surface was red unwarranted and irrational.

Regardless, future research will have to control for such reasoning processes in order to determine how much of projective color constancy is purely perceptual. While the general idea of projective color constancy, understood as a perceptual constancy, remains intuitive and appealing, a defense of color realism that depends on the rational basis of perceptual judgments about color must be prepared for the eventuality that judgments of the sort associated with projective constancy may be the product of explicit reasoning.

Arguing Against Color Realism


Two sorts of preliminary are in order before we motivate color realism. First, I wish to mention a sort of argument that could be given in favor of realism in order to distinguish my argument from that one. Then, we will set the stage for two arguments. One, that it is doubtful a case could be made against realism if color constancy is accomplished by a subpersonal, perceptual mechanism; and another, that if color constancy is a matter of explicit reasoning, then we ought to think that conclusions based on that reasoning are sometimes true.

The Externalist Argument

Consider the following argument, a variation on Putnam (1981), that goes something along these lines:

Let a “Brain in a Vat” be a brain kept alive while suspended in a vat of nutrient-filled liquid, hooked up by wires to a supercomputer programmed by incredibly knowledgeable and able psychologists so that it experiences a life just like an average 30-something white male doctoral candidate in 21st century USA. Either Corey is a brain in a vat, or Corey is not a brain in a vat.Footnote 8 Suppose it seems exactly as if there is a brown dog in his lap. If Corey is not a brain in a vat, then his judgment “There is a brown dog in my lap”, ceteris paribus, is true. If Corey is a brain in a vat, then his concepts “brown”, “dog”, and “lap” refer to whatever is causally responsible, in the right way, for his employment of those concepts: lines of code in a program, patterns of activation in the computer circuitry, or the like. Hence there are some conditions under which “brown”, “dog”, and “lap” can be employed correctly, and in some cases Corey’s judgment that there is a brown dog in his lap is true.

Putnam’s original argument was intended to show that I am able to have knowledge that I am not a brain in a vat, since the “brains” and “vats” of my conceptual repertoire are in the supercomputer, while my brain is in the vat. But what goes well for those concepts goes just as well for my concept “brown,” among the others mentioned in my variation on Putnam. The possibility that there are true judgments about color is just what we need in order to establish realism.

The primary reason that I believe an argument distinct from this one in favor of realism is the premise that my concepts “brown,” etc., refers to whatever is causally responsible for my acquisition of those concepts. Committed anti-realists could simply reject this claim. They ought to, since it is inconsistent with anti-realism: one’s concept of color cannot represent what is causally responsible for possession of the concept if what it represents does not exist. Pautz (2006, 2014, 2014) explicitly denies that visual representation of color has its content determined by a perceiver’s evolutionary ancestors’ interaction with the environment.Footnote 9 According to this denial, whether I am in a brain in a vat or not, my concept “brown” may refer to the same sort of thing. Whether it does so refer depends upon one’s further philosophical commitments.

In what follows, I assume that our visual representations can but need not acquire their content through some process of adaptation. I intend to show that if we take the Shepherd-style picture of the acquisition of visual content to be one of internalizing environmental regularities in order to represent features of the distal environment as generally correct, and for all we know mistaken in the case of color, then after consideration, we ought to regard it as not mistaken in the case of color. The case rests on the capacity for visual representation to furnish us with a basis upon which to make true judgments about the environment. We turn to this capacity now as our second preliminary.

Perceptual Judgment and Error

I return now to the perceptual judgments I presented at the beginning of Section 1.2, specifically #3 and #4:

  1. 3.

    That object is nearby.

  2. 4.

    That object is red.

Typically when one makes a judgment like #3, the judgment is veridical because of mechanisms like convergence. These mechanisms are reliable given that certain conditions obtain. The main condition on the veridicality-conduciveness of convergence, for instance, is that the object is within a couple of meters from the perceiver. In the case in which this mechanism fails, the human visual system has, as noted, many others to make use of. On the other hand, in the conditions under which no mechanism for depth perception is effective, we are subject to inaccurate perception of depth or distance. In this case, we are subject to perceptual illusion.

Illusions occur when the visual system is not able to discount possible environments consistent with the retinal image a priori. Typically, the visual system is able to exploit regularities in the environment that correlate with features the visual system is attempting to accurately represent. In the case of distance, the visual system exploits geometrical regularities that can be used to carry out a calculation that generally provides correct results. When such regularities do not obtain, distance is misrepresented. Since such situations are the exception, rather than the rule, we generally represent distance accurately in visual perception. In the cases in which regularities do not obtain, we experience illusion.

If the troubling data in Section 3.3 do not discount the idea that color vision can be characterized in a similar way, then we can say the following about color vision. First, take perception of color accomplished by projective or phenomenal constancy. Rescorla’s examples, that illuminant most likely resembles natural daylight, and that surface reflectances are more likely to be those commonly found in natural environments, are paradigm cases of the sort of regularities that the color-visual system might exploit.Footnote 10 There are other proposals, as we saw, but we need not pick between them. The point is only to set out the framework from which to argue below for realism. In any case, whatever the conditions under which constancy is achieved, we could say that veridical perception of color is as well. Where constancy mechanisms fail, we are subject to illusory, inaccurate perception.

As in the case of distance, errors in color perception can arise when conditions are other than the visual system assumes. In such cases, we suffer visual illusions—we represent objects in the environment as being some way other than they are, more or less distant than they are, or having a color other than they do. Such cases are ones in which vision is systematically misleading. Persistent visual illusion, though, is not sufficient to establish anti-realism about color. Misperception of some feature of an object by way of illusion is a case where the visual system wrongly attributes one feature to an object rather than another. If perception of distance is illusory, this is because the visual system attributes one distance to it rather than another; if perception of color is illusory, this is because the visual system attributes one color to it rather than another. Hence, the existence of color illusions requires that color exist in the first place.

In order to establish anti-realism about color, one would have to argue that there is no property such as the one color perception represents. Color perception must not be wrong simply because conditions are not optimal or the surfaces one is viewing are, in their optically salient features, quite different from those normally encountered. Rather, there must not be a property like the one the visual system is attempting to hone in on. Rather than illusory, we must show that color perception is always hallucinatory.

Suppose, though, that research spurred by Brainard’s findings indicates that color constancy is a cognitive phenomenon, a characterization of explicit reasoning. Then the task is quite different. The anti-realist, rather than having to propose that color perception is categorically hallucinatory, must argue that the reasoning that terminates in perceptual judgments about color is categorically faulty. That is, either the reasoning must proceed from false premises, or the inferential steps must inevitably lead one to falsehood. This is a distinct line of argument that will be dealt with separately.

True Judgments Without Realism?

Could the anti-realist protest some part of our preliminary argument? Perhaps. The anti-realist may say something along the following lines: “I agree that our judgments about color are generally true, but I question whether this fact alone deserves to be called “realism.” When I accurately judge an object to be red, what makes my judgment true is not any fact about the object per se, but certain facts about the state of my brain as I observe it. Or, as the case may be, the state my brain would be in were I to observe it. And this fact, so contrary to everyday thinking about color, hardly deserves to be called realism.” Such a line of reasoning has not been explicitly advanced in the literature, but it is a simple twist on some existing anti-realist views.Footnote 11

After making his neurobiological case for anti-realism, Hardin (1988: 111) concludes that “Colored objects are illusions, but not unfounded illusions. We are normally in chromatic perceptual states, and these are neural states.” He does not state explicitly that judgments of the sort we are concerned with are false, but it seems to be a relatively straightforward consequence of the claim “No object is colored” that these judgments are false. Maund (2006: 248), sharing Hardin’s view that objects are not colored, does draw this conclusion. He states that our judgments about colors in the environment surrounding us are false, but that it is “as if they are true.” That is, he holds the view we wish to rule out: that our judgments about color are false, but that it is advantageous for us in navigating the world that they seem to be true.

An interesting case is Averill (2005: 225), who proposes that colors are properties of opponent cell firings in the visual system. However, he denies that what is represented linguistically through color terms is identical to what is represented by color vision (Ibid.: 234). If, quite reasonably, we hold that what goes for representation in language goes for representation in thought as well, it is consistent to say that there are true “color” judgments in spite of the fact that colors reside within the perceiver. Yet if Averill is correct, these true judgments are not even about color. It should be clear from the relationship between perception and judgment that has been proposed in this paper that we cannot follow Averill in this line of reasoning regardless. More generally, it is difficult to make sense of the sorts of color matching experiments that allow us to theorize about color constancy if the judgments that form experimental data concern something other than the colors subjects are perceiving. So we will set Averill’s proposal aside.

Excepting Averill, can the view that colors are “in the head,” that they are features of our neural states or our minds, be reconciled with the idea that judgments like #4 are true? Proposals for the former thesis typically do not try to preserve the truth of the latter. On the face of it, an utterance of “This [demonstrating an object] is red” predicates of the object, an entity dwelling in the distal environment of the speaker, the property of redness. One could try to argue that the structure of such a sentence is not what it seems, but it is difficult to see how one could do so without making an ad hoc distinction between attributions of color to demonstrated objects and attributions of other properties to demonstrated objects. Because such demonstrative perceptual judgments seem to be true just in case the perceptions they are based on are accurate, we treat the possibility of true judgments about color as sufficient to establish realism about color.

Ubiquitous Perceptual Error?

With the distinction between biological and psychological normativities in tow and the qualification that the anti-realist must show ubiquitous perceptual hallucination, rather than illusion, we are able to spell out two problems that stand in the way of establishing anti-realism. The distinction between the two opens up the possibility that our ancestors evolved some representational capacities that systematically misrepresent the world around us because such misrepresentation is evolutionarily advantageous—such a representational system would exist because the representational states it furnishes us with aid us in carrying out our biological imperatives for some reason other than their veridicality. Establishing that this is the case with respect to color, I argue, has the problem that we cannot spell out what it is that color vision fails to do when it fails to furnish us with veridical perceptual representations. Second, by parity of reasoning, if we maintain that color vision is systematically misleading, we are committed to saying that perceptual judgments as such generally lack warrant. Or so I will argue.

Beginning with the first problem: suppose that, for our primate ancestors in the forests of the Old World, when trichromatic color vision first became present in them, it provided them with false visual representations that were useful for them to have. What would have taken for those representations to be true? The anti-realist cannot formulate an answer, because there is no principled way to determine what a veridical color representation would be. The realist, of course, can appeal to a number of bases: the judgments given by normal subjects in optimal conditions, either in everyday life, or in psychophysical experiments; the judgments that help us successfully navigate the environment; and the like. The realist could say as well that color representations that helped our newly-trichromat primate ancestors successfully navigate the world were veridical and those that were not were illusory. The anti-realist, on the other hand, has no way to specify what the world would have had to be like in order that perception of color is veridical. The reason is simple: to specify real-world features that would have to be in place for a representation of color to be accurate would be to specify actually existing colors.

The line of reasoning just presented requires the anti-realist to specify what is being represented as present in the environment that is, in fact absent. Perhaps the anti-realist would protest this requirement and say that rather, it is sufficient to maintain that the accuracy of representations of color only requires that properties of a certain sort must exist that do not exist. That is, we could describe the sort of property that colors would be were they to exist, to give a definite description of some sort that applies to a contingently nonexistent property. To take this route is somewhat similar to specifying what it would take for “Santa Clause exists” to be true in the manner of Russell, by stating that there would have to be an old man who lives at the North Pole, wears spectacles, has elves, etc. But this will not do. If we specify the sort of properties color vision represents by description, say of their structural features (perhaps along with their intrinsic, qualitative natures), we haven’t yet specified the content of any visual representation of color. We have only specified a background condition according to which color vision would veridical.Footnote 12

We move now to the second problem for the anti-realist. Above we saw two examples of perceptual judgments, predicating being nearby and being red of visible objects, and referred to them as #3 and #4 respectively. The question is: on what basis can we say that a judgment like #3 is both warranted by the perceptual evidence, and sometimes accurate, without thereby being committed to saying the same in the case of #4? If there is some relevant difference between the two, it must be found in one of three domains: the nature of the properties themselves, the perceptual system that furnishes us with the representations upon which we base our judgments, or the conditions of the evolutionary development of the perceptual system.

The idea that there is some important difference between colors and shapes loomed large in discussions about perception and reality in the early Modern era, but we would be hard pressed today to point to a difference between the two that would entail the reality of shape and unreality of color. Locke’s resemblance thesis, for instance, is outmoded now that we do not seriously suggest that there is, as a matter of principle, literal resemblance between veridical representation and representatum. And while it is sometimes suggested that causal overdetermination rules out the existence of colors (e.g., Hardin 1988), it likely rules out the existence of many things that most of us would like to endorse the existence of, such as ordinary medium-sized objects. After all, it seems as though microphysical goings-on are sufficient on their own to cause any macroscopic event. An anti-realist appeal to overdetermination thus brings us into controversial territory. A full survey of proposed disanalogies, however, is beyond the scope of this article.Footnote 13

Is there some important disanalogy between the perceptual subsystems responsible for furnishing us with representations of color and spatial relations? I have attempted to motivate a negative answer to this question so far, and while the details are forthcoming, the research project of modeling color constancy along the lines that we are able to model other forms of perceptual constancy is ongoing and shows no signs of being a misguided one. This being the case, we have no reason to propose a significant disanalogy between judgments #3 and #4 along these lines. So I turn to the third area where we might locate a disanalogy between the two.

It may be that the perception of distance and of color differ in some respect in their evolutionary histories—that the conditions under which one became evolutionarily advantageous aligned with the conditions under which that perceptual faculty furnished our ancestors with accurate representations, while not doing so for the other. If we say the former in the case of the perception of distance and the latter in the case of perception of color, then we can give a reason to think that judgments like #3 are generally accurate, and judgments like #4 are not. Color vision would then be like the overactive capacity to detect predators discussed in Section 1.1. Yet this brings us back to the fact that we noted just above: we cannot spell out what representational success or failure would mean in the case of color vision without betraying anti-realist commitments in the first place. Hence, the story of how color vision could become biologically advantageous without being representationally successful cannot be spelled out. Note that in the case of detection of predators, actually existing predators play some role in explaining the how and why of the system of detection. The misrepresentations are illusory, not hallucinatory. But for the anti-realist, colors cannot play an analogous role for color vision.

Now, given that we have covered all of the bases as far as possible disanalogies between judgments like #3 and #4, we have the following result: either both of the visual subsystems that furnish us with representations upon which we base each are generally accurate, or neither is. This is because there is no domain in which we can propose a relevant disanalogy between the two. Extrapolating, we could propose that we cannot call color vision into question without thereby calling into question the reliability of the visual perception of geometrical features of the environment, and the existence of those features. To do so would be to draw an ad hoc distinction between color and color vision, on one hand, and those features and our perception of them on the other. That is a commitment that very, very few people would prefer to make. Hence, I recommend that we deny anti-realism and accept realism. I turn now to the possibility that color constancy is a cognitive, person-level phenomenon.

Categorically Faulty Inference?

When color matches have been made by subjects on the basis of explicit reasoning, as in Brainard’s example discussed in 3.3, the basis of their reasoning involves how surfaces appear to them, color-wise.Footnote 14 The arguments for simultaneous matching, fully stated, are along the lines of the following schema:

Premise 1 :

Surface S currently appears to have color C to me.

Premise 2 :

Surface S currently appears to have color C to me.

Conclusion :

Surfaces S and S both currently appear to have color C to me (i.e., they match).

In the case of diachronic matches, the arguments are along the lines of a slightly different schema:

Premise 1 :

Surface S currently appears to have color C to me.

Premise 2 :

Surface S appeared to have color C to me at t1.

Conclusion :

Surfaces S currently appears and S appeared at t1 to have color C to me (i.e., they match).

Vindicating these forms of reasoning is enough to get us the veridicality of the conclusions, but isn’t enough to establish that the sort of first-order judgments we would like to say are veridical are veridical, i.e., “That object is red.”

The capacity to match colors despite variations in illumination on a perceptual basis was enough to establish that such judgments are veridical because in that case the judgments merely affirm what content in given in perception. If the basis of such a capacity is explicit reasoning, we need something more. What this reasoning would actually consist in is an empirical question, but I take it to be uncontroversial that the reasoning would proceed in part from the color appearance of a surface. In conjunction with the aforementioned process of recognizing sameness or difference in color between surfaces, we can account for our practices of color categorization in general.

The reasoning from surface appearance to surface color can be made explicit as follows:

Premise 1 :

Surface S currently appears to have color C to me.

Premise 2 :

In circumstances such as mine right now, surfaces that appear to have color C to me have color C.

Conclusion :

Surfaces S has color C.

The circumstances relevant in Premise 2 are, of course, the viewing conditions and the state of the perceptual system. Viewing conditions must be conducive to the accurate visual perception of one’s environment; the surface must not be too far away or too close, the illumination should be normal, and so on. One’s perceptual system must be functioning normally; one should not be on hallucinogenic drugs, or subject to awake brain surgery in the visual center.

It is certainly uncontroversial that we can have knowledge of propositions of the form “Surface S appears to have color C to me.” This accounts for the veridicality of Premises 1 and 2 of the first sort of argument above, and the first premise of the subsequent two sorts of argument. The second premise in an argument regarding diachronic surface matching requires only that working memory is truth-preserving in the circumstances of experiment. Visual working memory is quite reliable in the case of color, so long as only a small number of colors need to be remembered.Footnote 15 If we can have knowledge of the present appearance of surfaces, and if in ordinary cases memory preserves knowledge, then we can have knowledge of the sort in the schema for the second premise.

What has been said vindicates the reasoning in the cases of color matching experiments. In order to move from apparent to actual color matches, we need to establish the truth of the second premise-schema of the third argument. I take it that when we are in normal viewing conditions, we are usually in a position to know that this is the case. When one’s visual system is working normally, I take it that we are usually in a position to know that this is the case. That is, ordinarily, when things are “going wrong,” we are in a position to know so. There are possible contrived scenarios in which this is not the case, but these cases resemble those of Cartesian error, and present us with forms of skepticism far more extensive that anti-realism about color. Such possibilities should be argued against in the context of the debate about Cartesian skepticism rather the existence of color.

In saying that we are generally able to distinguish normal from abnormal viewing circumstances, we presuppose that there are circumstances we can be in that are conducive to true perceptual judgments about color. The anti-realist will not accept such a presupposition. The premise-schema itself presupposes that there are such circumstances. Hence, in order that it be true, such circumstances must exist. This is more difficult to demonstrate than that we are able to know we are in normal circumstances given that they exist. However, I think that it is incumbent upon us to accept the premise.

Regardless of whether perceptual judgments about color ever are true, they are evaluated. The subjects in constancy experiments, for instance, can perform better or worse on experiments. The color of the sky is treated as not matching the color of the ground, nor the color of a ripe tomato matching the color of an avocado. In general, we treat some conditions (such as the ones mentioned above) as misleading for the making of color judgments, and some as not misleading. The evaluative criteria that we bring to bear in such cases either are or are not those of accuracy conditions. We treat the criteria as equivalent to accuracy conditions—i.e., there is no indication from within our linguistic practices that the criterion for appropriateness of a perceptual judgment about color is anything other than truth or falsity.

Is there any reason that could be brought to bear from outside of our linguistic practices that the criteria brought to bear on these judgments pertain to anything other than their truth or falsity? In the absence of a proposal, it is difficult to say. There is, of course, one reason that could be brought to bear—in order for such judgments to be true, their subject matter, color, must exist. Anti-realism itself could be a reason to reject that we evaluate color judgments by their accuracy. Yet this is precisely the view that anti-realists assert and realists deny. I conclude, then, that we are warranted in admitting that there are optimal conditions for making perceptual judgments about color, i.e., conditions conducive to true judgments. The third argument-scheme above has sound instances.

If we take it that our inferences along the lines given in the above arguments are generally sound, then our judgments about color are generally true. If our inferences from the appearance of color in some surface to its actual color, as in the third argument-schema, sometimes yield knowledge, then we have basis for some knowledge of the colors of surfaces around us. The capacity to match colors with one another across differences in illumination, as in the first two schemas, extends this knowledge beyond whatever optimal conditions are sufficient to bridge the link between surface color and the appearance thereof. Hence true perceptual judgments about color are ubiquitous, a fortiori color realism is true.


I have attempted to argue in defense of color realism by appeal to the different sorts of norm according to which we may evaluate perceptual representations, and the conditions under which those norms diverge. The claim that other things being equal, biological and psychological norms coincide is one that is neutral between externalist and internalist views about perceptual content, giving the argument force against a broad contingent of anti-realist positions. I have supplemented this argument with one that anticipates the possibility that the representation of color occurs at the cognitive, rather than perceptual, level. If these arguments are persuasive, realism about color is the rationally obligatory position.Footnote 16


  1. 1.

    I assume here that avoidance behavior is an exercise of agency, and that an exercise of agency must be guided by representations in the agent’s mind.

  2. 2.

    This is perhaps a bit oversimplified, as there is some parasitic cases: doubting that what is represented in perception, for instance, or judging contrary to perception, but these cases will not concern us in anything that follows.

  3. 3.

    This is not a metaphysical theory of information. How exactly to understand the nature of information per se is a matter of dispute. The dominant, perhaps classical, understanding of information is as a probabilistic notion, as in Shannon (1948) and Dretske (1981). An alternative proposal due to Cohen (2003) analyses information in terms of counterfactual relations between events.

  4. 4.

    This result has been reproduced many times. See Wright (2013).

  5. 5.

    For misgivings about motivating phenomenal constancy by appeal to the dimensionality of color, see Wright (2013: 448)

  6. 6.

    Wright (2013) and Hilbert (2011) each propose alternative ways that projective and phenomenal constancies may work together.

  7. 7.

    Foster specifies that the property be surface reflectance, but this isn’t a necessary specification — changes the represented sort of property may be merely correlated with changes in reflectance, so long as they are illumination-independent. It is important to note this in order to stay neutral between the various theories of color ontology.

  8. 8.

    I will ignore other skeptical possibilities, since the point about the brain-in-vat scenario is meant to generalize.

  9. 9.

    Or, in the case of the brain in a vat, some connection between the perceiver and supercomputer.

  10. 10.

    One task for the empirical project of studying color perception is to unpack just what “natural daylight” and “commonly occurring surface reflectances” could mean to the visual system. That is, by what parameters would the visual system recognize such features of the environment? I bracket those questions here.

  11. 11.

    Recently, Stov̌er and Bregant (2017: 91) have advanced a form of realism in which colors exist in the brain, while remaining silent on the status of ordinary demonstrative judgments about color. They note that “the latest neurophysiological research suggests that the activity of certain neurons in the brain is enough to realize the colour experience” and conclude that because this is all that is required in order that one experience colors, it must be that colors are located in the brain. According to what we have advanced above, this line of reasoning seems specious. It mistakes the visual representation of color for the instantiation or existence of color—consider that our experience of color can be inaccurate. Were the experience of any given color sufficient for it to be found in the world, there would be no color illusions nor hallucinations. In fact, we ought to count artificially created circumstances like those cited by the authors as paradigmatic of those that give rise to illusion.

  12. 12.

    I have in mind here something along the lines of Burge’s (1991) criticism of Searle’s theory of perceptual content according to which the causal chain between perceived objects and the perceiver is part of the content of a perceptual state. Burge claims that this is simply something that must be in place in order that the content be veridical, but not part of the content itself. It is in this sense that I use the phrase “background condition.”

  13. 13.

    Of note are anti-realist arguments pertaining to the structure of color space put forward in Boghossian and Velleman (1991: 85-86) and Hardin (1988: 66), and addressed in Cohen (2003), Byrne and Hilbert (2003), and Tye (2003).

  14. 14.

    Here for a surface to “appear” some color is for it to be represented as having that color. If we understand color vision as relying on projective color constancy, this does not necessarily entail that the qualitative appearance of the surface has a particular “phenomenal feel.”

  15. 15.

    “Performance was nearly perfect for arrays of 1–3 items and then declined systematically as the set size increased from 4 to 12 items.” (Luck and Vogel 1997: 279)

  16. 16.

    Thanks to Jonathan Cohen, Kevin Falvey, Michael Rescorla, Wayne Wright, two anonymous referees, and an audience at UC Santa Barbara for many helpful comments and suggestions on earlier drafts of this material.


  1. Arend, L., & Reeves, A. (1986). Simultaneous color constancy. Journal of the Optical Society of America A, 3, 1743–1751.

    Article  Google Scholar 

  2. Averill, E.W. (2005). Toward a projectivist account of color. The Journal of Philosophy, 102(5), 217–234.

    Article  Google Scholar 

  3. Boghossian, P., & Velleman, J. (1991). David Physicalist theories of color. Philosophical Review, 100, 67–106.

    Article  Google Scholar 

  4. Brainard, D.H. (2009). Color constancy. In Goldstein, B. (Ed.) In the sage encyclopedia of perception (pp. 253–257). Los Angeles: SAGE Publications.

  5. Burge, T. (1991). Vision and intentional content. In LePore, E., & Van Gulick, R. (Eds.) John Searle and his critics. Wiley: Blackwell.

  6. Burge, T. (2005). Disjunctivism and perceptual psychology. Philosophical Topics, 33(1), 1–78.

    Article  Google Scholar 

  7. Burge, T. (2010). Origins of objectivity. London: Oxford University Press.

    Google Scholar 

  8. Byrne, A., & Hilbert, D.R. (2003). Color realism and color science. Behavioral and Brain Sciences, 26(1), 3–21.

    Article  Google Scholar 

  9. Cohen, J. (2003). An objective counterfactual theory of information. Australasian Journal of Philosophy, 84(3), 333–352.

    Article  Google Scholar 

  10. Dretske, F. (1981). Knowledge and the flow of information. Cambridge: MIT Press.

    Google Scholar 

  11. Evans, R.M. (1951). U.S. Patent No. 2,571,697. Washington: U.S. Patent and Trademark Office.

    Google Scholar 

  12. Foster, D.H. (2003). Does colour constancy exist? Trends in Cognitive Sciences, 7(10), 439–443.

    Article  Google Scholar 

  13. Gibson, J. (1979). The ecological approach to visual perception. Boston: Houghton Mifflin.

    Google Scholar 

  14. Gibson, J. (2002). A theory of direct visual perception. In Noe, A., & Thompson, E. (Eds.) Vision and mind: selected readings in the philosophy of perception (pp. 77–89): MIT Press.

  15. Hardin, C.L. (1988). Color for philosophers: unweaving the rainbow. Hackett Publishing.

  16. Heesy, C.P. (2009). Seeing in stereo: the ecology and evolution of primate binocular vision and stereopsis. Evolutionary Anthropology, 18, 21–35.

    Article  Google Scholar 

  17. Hilbert, D. (2005). Color constancy and the complexity of color. Philosophical Topics, 33, 141–158.

    Article  Google Scholar 

  18. Hilbert, D. (2011). Constancy, content, and inference. In Hatfield, G., & Allred, S. (Eds.) Visual experience (pp. 339–361). London: Oxford University Press.

  19. Marr, D. (1982). Vision: a computational investigation into the human representation and processing of visual information. Henry Holt and Co. Inc.

  20. Maund, B. (2006). The illusory theory of colours: an anti-realist theory. Dialectica, 60(3), 245–268.

    Article  Google Scholar 

  21. Luck, S.J., & Vogel, E.K. (1997). The capacity of visual working memory for features and conjunctions. Nature, 390, 279–281.

    Article  Google Scholar 

  22. Neumeyer, C. (1980). Simultaneous color contrast in the honeybee. Journal of Comparative Physiology, 139, 165.

    Article  Google Scholar 

  23. Palmer, S. (1999). Vision science: photons to phenomenology. Cambridge: MIT Press.

    Google Scholar 

  24. Radonjić, A., & Brainard, D.H. (2016). The nature of instructional effects in color constancy. Journal of Experimental Psychology: Human Perception and Performance, 42(6), 847–865.

    Google Scholar 

  25. Pautz, A. (2006). Sensory awareness is not a wide physical relation: an empirical argument against externalist intentionalism. Noû,s, 40(2), 205–240.

    Article  Google Scholar 

  26. Pautz, A. (2014). The real trouble for phenomenal externalists. In Brown, R. (Ed.) Consciousness inside and out: phenomenology, neuroscience and the nature of experience. New York: Springer.

  27. Putnam, H. (1981). Reason, truth, and history, cambridge. Cambridge: Cambridge University Press. Chapter 1, 1–21.

    Google Scholar 

  28. Reeves, A.J., Amano, K., Foster, D.H. (2008). Color constancy: phenomenal or projective? Perception & Psychophysics, 70, 219–228.

    Article  Google Scholar 

  29. Shannon, C.E. (1948). A mathematical theory of communication. Bell System Technical Journal, 27(3), 379–423.

    Article  Google Scholar 

  30. Shepard, R.N. (2001). Perceptual-cognitive universals as reflections of the world. Behavioral and Brain Sciences, 24, 581–601.

    Article  Google Scholar 

  31. Stožer, A., & Bregant, J. (2017). Physicalist and dispositionalist views on colour: a physiological objection. Acta Analytica, 32(1), 73–93.

    Article  Google Scholar 

  32. Tye, M. (2003). Consciousness: color, and content. Cambridge: MIT Press.

    Google Scholar 

  33. Wright, W. (2013). Color constancy reconsidered. Acta Analytica, 28(4), 435–455.

    Article  Google Scholar 

Download references

Author information



Corresponding author

Correspondence to Corey McGrath.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

McGrath, C. In Defense of Color Realism. Acta Anal 35, 101–127 (2020).

Download citation


  • Color
  • Perception
  • Realism
  • Perceptual constancy
  • Normativity