|Why Do Colours Look The Way They Do?
*This is an unpublished draft. Please do not quote without permission.*
Some links between colour phenomenology and its physiological basis are examined in detail, in particular concerning a kind of hue-inversion where blue is exchanged with green and yellow with pink. A project to develop an appropriate phenomenal language is outlined. The impact of such links on the ‘explanatory gap’ and the ‘knowledge argument’ is considered in the context of the debate between Hardin and Levine. I agree with Hardin that the ‘gap’ is highly exaggerated, but conclude instead that it is for just this reason that we do not need to be physicalists. Instead, a kind of dualism is shown to be scientifically respectable.
It is widely held that there is a fundamental ‘explanatory gap’ between physical and mental phenomena, and that this seriously compromises physicalist theories of the mind.1 Given that the only alternative to physicalism seems to be some kind of dualism, which is generally held to be explanatorily useless, this means that, unless the gap is somehow traversed, mentality is ultimately mysterious.
The gap is typically illustrated in terms of colour qualia. Thus, whilst physics can explain quite adequately how the electronic structure of the surface of a coloured object causes light of only certain wavelengths to be reflected, and whilst there seem to be no fundamental obstacles to providing a complete neurophysiological account of how the eye and the brain process incoming light, there is (it is alleged) something vital that cannot be so explained, namely why colours actually look the way they do. The immediate phenomenological character of green, for example, is something whose explanation lies beyond ordinary physical science.
This gap is supposed to be highly general, but it is significant that most examples given tend to concern colours—more specifically, their hues. For example, nobody finds it intuitively odd that an increase in luminance or light intensity should correlate with an increase in brightness or perceived luminosity, and I have never heard it seriously suggested that you might see things going dimmer where I see them going brighter.2 Indeed, it is not even that easy to explain why luminance (a photometric quantity) and brightness (roughly, its psychosensorial correlate) are different magnitudes.3 Thought-experiments about inverted colour-hues, by contrast, go back at least as far as the Cyrenaics, and are now very widely discussed: most notably the thought that you might see green where I see red (and vice versa). Presumably, then, there must be something especially peculiar about hue as opposed to the other dimensions of colour, such as brightness or saturation, and it is worth exploring why this should be so.
Frank Jackson’s famous ‘knowledge argument’ is often used to demonstrate this explanatory gap.4 The argument is that if a subject, Mary, has never been allowed to see colours other than black, white and grey, then she will not be able to work out what green looks like until she actually sees it, even though she may have as much physical and neurophysiological knowledge as is relevant. No matter how well versed she is in physical science, there must always be an ‘Aha!’ moment when she first sees something green, since she will then become acquainted with something whose nature she cannot predict in advance. The ‘Aha!’ moment seems to bring in new knowledge, and since all physical knowledge is already accounted for, it apparently follows that there must be facts about colour which cannot be accommodated by the physicalist. Similar thought-experiments work tolerably well with other sensory modalities, but it is significant, once again, that within the sphere of the visual, it is only hues that yield psychologically convincing thought-experiments.
The knowledge argument has been much discussed. However, our version of it is different in two ways. Firstly, we are not overly concerned with how much Mary can know before ever seeing colours; rather, we shall be concerned with whether she can tell which colour is which after seeing them all for the first time. Secondly, our Mary is not a super-scientist who automatically knows what is going on in her brain when she sees something. If she can tell one colour from another, it really is in the same way that we can: by simply looking. It might be insisted that she cannot possibly do this. However, colour vision science has demonstrated that there are some very intimate connections between hue-phenomenology and physiology, connections which tend to undermine many of the Cartesian intuitions at work here, as has been shown in considerable detail by C.L. Hardin among others. This paper will explore further the extent to which such empirical research could conceivably go in addressing this aspect of the mind-body problem, in particular concerning hue-inversions. We agree with Hardin that difficulties in connecting the physical and the phenomenal have been greatly exaggerated. However, we shall draw very different metaphysical conclusions.5
Some Results from Colour Vision Science
The perceived hue of an object is primarily associated with the dominant wavelength of the light it reflects. The human eye is sensitive to electromagnetic radiation of wavelengths between roughly 400nm and 700nm, the former, short-wave end corresponding to violet, the latter, long-wave end, to red, and with the familiar rainbow spectrum spreading through intermediate values (most purples are non-spectral, however, and can only be produced by mixture). The retina contains colour-receptors, or ‘cones’, of three types, usually called ‘L(ong)’, ‘M(edium)’ and ‘S(hort)’ whose sensitivities have different peaks within the visible range. This corresponds to the familiar fact that only three basic hues (typically, red, green and blue) are needed to generate all the others (‘trichromacy’). However, although the M-receptors do indeed have peak sensitivity at around 530nm (green), L-receptors peak at around 560nm (greenish yellow), and not at red, as might be expected; so their sensitivity curves are actually very close together. This, together with the fact that cones of all types are sensitive in varying degrees to light of a very wide range of different wavelengths, and also the fact that the neural signal generated is always the same regardless of the wavelength of the light which actually triggers it (the ‘principle of univariance’), ensures that our ability to make precise hue-discriminations across the whole visible spectrum must arise from additional, postreceptoral mechanisms, contrary to the views of early theorists such as Young and Helmholtz.
What all colour vision scientists now agree on is that it is actually the small but crucial difference between the levels of L and M stimulation at a given point in the visual field, rather than their absolute values, that the brain is geared to notice. If L – M is positive, that is if the L cones are stimulated more than the M cones, then the hue at that point looks reddish; if it is negative, then it looks greenish; and if it is zero, then the hue has neither a reddish nor a greenish phenomenal component. The S cones, which appear to have evolved considerably before the time when the photopigments in the L and M cones diverged, operate similarly. Here, the relevant difference is between their stimulation-level and that of all the others: so that if L + M – S is positive then the hue at that point looks yellowish; if negative, then it looks bluish; and if zero, then the hue has neither a yellowish nor a bluish phenomenal component.
We have an example here of what in physiology is called ‘opponent processing’. Where L – M is positive, then chemical activity along the red–green (or ‘r–g’) neural pathways between the eye and the brain is excited; when negative, it is inhibited. A similar result holds with L + M – S and the yellow–blue (or ‘y–b’) neural pathways.6 (There is also a third system, L + M, which corresponds to white versus black, or overall brightness at a given point.) Excitation–inhibition dualities are very common in physiology; other examples of opponent processes include ‘push–pull’ couplets, such as muscular contractions. Thus I can move my forearm up or down, left or right, or can combine the two movements to varying degrees. However, I cannot move my arm both up and down, nor both left and right simultaneously. In so far as a given neural channel cannot be both excited and inhibited simultaneously, there is therefore an analogy between left–right and green–red, and between up–down and yellow–blue. This can be illustrated by means of the following colour circle:7
The fully saturated spectrum is spread out continuously around the circumference. Desaturation can be represented by extending the circle inwards to a disc: thus, a given hue will be desaturated as it moves in along a radius towards pure grey, which is located at the centre. Brightness can be represented along a dimension perpendicular to the disc, which is now extended to a solid cylinder with an indeterminate upper boundary (there is no specific maximum brightness). (The purpose of the dotted diagonal axis will be explained in the next section.)
The opponent-processing model of colour perception is now universally accepted, thanks largely to precise quantitative experiments performed by Dorothea Jameson and Leo M. Hurvich in the 1950s.8 However, the idea was originally mooted by the nineteenth century physiologist Ewart Hering. What is philosophically significant is that his reasoning was almost entirely phenomenological. What the original Young-Helmholtz model failed to explain (among other things) were some basic facts about hue-appearances. The first point is that there are four hues, not three, namely red, blue, green and yellow, which are unique in the sense of looking unmixed. Unique green, for instance, looks just green and nothing else: it does not look either bluish or yellowish, for example, let alone both.9 By contrast, orange can only look simultaneously reddish and yellowish, even if it is exactly midway between red and yellow; moreover, this phenomenal combination is all there is to orange. Thus it, together with purple (bluish red), turquoise (greenish blue) and chartreuse (yellowish green) are binary hues. This unique–binary distinction needs to be accounted for. Merely having three types of receptors in the eye will clearly not do this, but the opponent processing model will, for it implies that there are exactly four points on the visible spectrum where only one of the differences, L – M and L + M – S, is non-zero (i.e. where only one of the channels is operational), and they occur in the right places.10
The second, and related point is that while some combinations of unique hues are possible, others are not, namely reddish green and yellowish blue. This can be difficult to see because it is easy to confuse purely phenomenal composition with ordinary colour mixing, either additive or subtractive. Thus, red and green light can be mixed (additively, as with a colour television set) to produce yellow light, but yellow does not look both reddish and greenish. Likewise, blue and yellow paints can be mixed (subtractively) to produce green; but, pace Brentano, green does not look like yellowish blue. Indeed, everyone is surprised when they first see the results of such mixing: they are intuitively odd in the way in which the mixing of red and yellow to produce orange, for example, is not. This needs to be explained, and the opponent-processing model does this. It likewise explains very neatly why nobody can see hues that actually look reddish green or yellowish blue (except in extraordinary circumstances).11
These links between hue-phenomenology and physiology are well established, and they surely cast some doubt on the knowledge argument. For example, if Mary is presented with a turquoise banana, then she should instantly be able to see that she is being tricked. This is not, pace Dennett (1991), because she has ideal physical knowledge which enables her to tell that the banana is reflecting light of the wrong wavelengths (or that her brain is undergoing the wrong processes). On the contrary, she can just see that the hue of the banana is binary whereas it ought to be unique. Pure phenomenology will tell her that the banana’s hue is not what real yellow looks like, even if this is the first time she has seen colours. It may be protested that the unique/binary distinction itself cannot be grasped without phenomenal background knowledge unavailable to Mary, but this seems unlikely. The distinction is of too general a kind, and—crucially—has a direct link to physiology. Admittedly, the opponent process theory will probably not tell her everything she needs to know, but the idea that we have some kind of absolute gap has surely been undermined. If Mary’s first chromatic experience consists of viewing simultaneously some squares of many different colours (say, red, orange, yellow, chartreuse, green, turquoise, blue and purple, but arranged in no particular order), then it is just not true that her general physiological background knowledge will give her no information whatsoever about which hue is which, even if she relies only on immediate phenomenology and not on physical analysis of the reflected light (or knowledge of her own specific physiological processes). The unique and the binary hues can be directly distinguished.
This might seem not to amount to all that much. However, there are further, though more controversial, connections between hue-phenomenology and physiology. Reds and yellows are generally perceived as ‘warm’ colours, whereas greens and blues are generally perceived as ‘cool’. Sometimes the contrast is expressed as ‘advancing’ versus ‘receding’, or ‘positive’ versus ‘negative’. It might seem that this is merely cultural, or does little more than reflect the fact that archetypal warm things, such as fire, are typically red, orange or yellow, whereas archetypal cool things, such as lakes, are typically blue, turquoise or green. However, there is evidence that the excitatory and inhibitory features of the r–g and y–b channels have a wider physiological impact, as revealed in increases and decreases in ordinary vital signs (e.g. blood pressure, oxygen uptake, heart and breathing rates), alpha wave activity, galvanic skin response, the effect on hyperbilirubinemia (jaundice), the frequency of eye blinks and epileptic seizures, and so forth; so there may be well be direct connections, both physiological and phenomenological, between seeing red/yellow and feeling warmer, and between seeing green/blue and feeling cooler.12 If this is so, and if Mary has been allowed to feel ordinary warmth and coolness, then she could perhaps directly distinguish yellow not only from the binary colours, but also from the cool unique hues, blue and green. This leaves only the distinction between yellow and red unaccounted for. However, it could be argued that even this can be handled by the fact that yellow is essentially a much lighter colour than red with less intrinsic saturation (or ‘chroma’), a phenomenon that can be explained by the fact that the L + M light–dark channel peaks at the yellow part of the spectrum. If so, then it now looks as though physiology alone is sufficient to tell Mary what each and every colour will look like—or, at least, how to identify which is which should they all be simultaneously presented to her. Of course, there will always be some sort of ‘Aha!’ moment when one undergoes some essentially novel experience, but this need not involve any significant new knowledge. When confronted with Hume’s missing shade of blue for the first time (to take an extreme example), one need not learn anything that one could not work out quite adequately in advance. And even when we do have a more interesting ‘Aha!’ moment, there remains an important difference, if only of degree, between ‘Gosh, this is unimaginably strange!’ and ‘Ah yes, this is more or less what I thought it would be like’. Colour vision science can push us a long way from the first sort of reaction to the second, and this undermines the thought that there must be something irreducibly odd and inexplicable about the link between the physical and the phenomenal, even if we concede that there will usually be something new that is learnt when one undergoes a radically unfamiliar kind of experience.
Supersaturated Yellow and ‘Diagonal’ Inversion
Nevertheless, the claim that colour vision science has already managed to characterize all colour hues in purely physiological terms is certainly overstated, and one problem concerns the link between yellowness and chroma. Yellow does indeed have a much lower chromatic content than, say, red or blue, and we can explain this easily enough, as mentioned; but this is not enough to rule out the possibility of alternative hue–physiology correlations should yellow no longer be the hue at which the L + M channel peaks. For imagine a person with a more subtle hue-inversion than the standard red–green variety, namely one where hues are reflected in the diagonal axis from turquoise to orange in the Hering colour circle depicted above. In this case, green maps into blue and vice versa, and red maps into yellow and vice versa. Likewise, purple will be exchanged with its complement, chartreuse, whereas turquoise and orange will each stay fixed. This ‘diagonal’ inversion is more interesting than the standard red–green variety, for not only do unique hues remain unique and binary hues remain binary; also warm hues remain warm and cool hues remain cool (and the neutral binary hues, purple and chartreuse, remain neutral). This ensures that diagonal-inverts (or ‘d-inverts’) are less easily detectable than red–green ones. The chroma problem remains, of course, but it could perhaps be solved by supposing that d-inverts can perceive ‘supersaturated yellow’, a colour that relates to yellow as red relates to pink.13 We cannot perceive, and can only barely imagine, such a colour. Conversely, d-inverts cannot see or readily imagine a supersaturated pink (what we call ‘red’), since the pink/red hue is where their L + M channel peaks. The point, however, is that supersaturated yellow does not seem to be phenomenally impossible in any very profound sense, and we might even one day be able to experience it ourselves should we manipulate our brains appropriately.14 It is certainly hard to see how Mary (or, indeed, anyone else) could know simply by looking at pink and yellow that supersaturated pink is ordinarily perceivable whereas supersaturated yellow is not.
It may still be feared that countenancing ‘alien’ hues of any kind will lead us into dangerous territory. Joseph Levine (1991), for example, has criticized Hardin’s arguments for spectral asymmetry by suggesting that reds could be perceived as cool and greens as warm. If we lose the inter-modal links, then our ability to explain why colours look as they do will clearly be damaged irrevocably. It could be protested, with Hardin, that ‘cool red’ is a contradiction in terms, and that the feature which we naturally call ‘warmth’ is visibly an essential constituent of redness. Yet it is not all of redness (otherwise yellow could not also be warm), so the possibility remains that the residue, whatever it is, could be combined with coolness to produce a new hue. We cannot imagine this, of course, but if we allow the unimaginable supersaturated yellow to be possible then we cannot put much weight on that fact alone. And if low chromatic content is not essential to yellow, why should warmth be?
Given some actual experiences, such as those of human tetrachromats (not to mention pigeons, who may even be pentachromats), it would be rash to reject the possibility of alien hues just because they are unimaginable and barely describable. We might well allow there to be something a bit like ‘cool red’, though it is debatable whether it deserves the name, given that its resemblance to real red would be, at best, extremely slender. But however we describe it, a problem is clearly emerging, for if Mary’s first chromatic experience is of an array of coloured boxes, as suggested above, and yet we include these alien hues as well, her ability to tell which hue is which will surely be seriously compromised—and with it, our thesis.
However, it must be remembered that we should not be expected to explain everything all in one go. It may be that if we restrict ourselves to ordinary hues, then we shall not be able to characterize everything that is possible, and consequently will not be able to explain every aspect of how colours look. It does not mean that we have not made a useful start. Moreover, our explanations are required to concern only what is empirically possible, and it is, in fact, rather unlikely that there could be creatures who could perceive both ordinary and alien hues. In any case, supersaturated yellow is only modestly alien. It is, after all, very much a part of our ordinary concept of a hue that two colours may have the same hue and yet differ in saturation. All we are doing is stretching things a bit.
If we are allowed to do this, then we might also be able to answer another objection to the claim that d-inversion is undetectable, namely that there is a far greater perceptible difference between red and blue than there is between yellow and green. This is reflected in the fact that we have many words for bluish reds and with highly specific meanings (we can say, with assurance, that violet is bluer than magenta, for example). By contrast, no known language has even one standardly used word for a colour between yellow and green: terms such as ‘chartreuse’, ‘lime’ or ‘canary’, for example, are not in general use. Purple is a highly salient colour in its own right in a way in which yellowish green is not.15 Now, this may well no longer be the case if yellow is supersaturated and red is desaturated. It is hard to be certain here, but it is not unlikely, especially since a d-invert will also need to desaturate blue and supersaturate green, albeit only slightly, if relative brightness and chroma levels are to remain unchanged right across the new spectrum (such a constraint is necessary if we are to ensure that d-inverts differ from normal percipients only in respect of the hues they perceive). There are surely likely to be more perceptible shades of difference within a given arc of the colour circle if chroma is increased, and fewer if it is decreased. Other potential asymmetries, such as the greater salience of pink and brown over cream and dark red, might also invert with chroma changes. To repeat, we cannot be sure here; but it is nevertheless evident that until d-inversion is properly addressed, it cannot plausibly be claimed that all hue distinctions are grounded in physiology in an intelligible way.
In fact, we shall still need to address this type of inversion even if it did not exactly preserve all intra-spectral relations. As Levine (1991) points out, merely demonstrating asymmetry will not itself close the explanatory gap. A fundamental asymmetry between red and yellow would do little to explain why green looks like green as opposed to blue, for example. What we also need to find are some further inter-modal features analogous to warmth and coolness which do manage to differentiate green and blue, and which can also be explained physiologically.
How, then, should we set about doing this? I suggest that, in the first instance, we should simply d-invert what we see and then try to devise a phenomenal language which best characterizes what has changed. We can then look for some illuminating connections with the physiological changes that we should need to undergo if we were actually to perceive the world in that way. Some d-inverting spectacles would be helpful here, but they are obviously not to be had just yet. Nevertheless, recent technology has achieved a good deal in this respect. There are many commercially available software packages which may be used to alter the hues within digital photographs, so it is much easier now to represent, and in some detail, what a d-invert’s world looks like. We cannot produce supersaturated yellow, of course, but we can instead desaturate all non-yellows to ensure that relative saturation and brightness levels across the spectrum are maintained, and this should give us some grasp of the idea. (The edited photographs will have a rather washed out look, but since hue-inversions tend to look rather garish, at least at first sight, this is actually something of an advantage.) As far as I am aware, there are not yet any packages available for editing videos or DVDs in this way, so dynamic transitions between colours cannot so easily be represented. Immediacy and miniaturization are likewise still a problem, but the ease of taking digital photographs (most new mobile phones have this feature, for example) ensures that a day to day, regular and systematic comparison between ordinary colour perception and its d-inversion is something that many of us can now realistically achieve.16
4 The ‘D-Project’
However, is such a collective project (let us call it the ‘d-project’) really worth engaging in? It may sound no more than a self-indulgent fantasy, but in fact there are a number of very good reasons for taking it seriously. Firstly, it helps us to sharpen our original question. Even if ‘colours’ are restricted to ‘hues’, the question of why they look as they do notoriously depends on a very wide range of factors. For example, it is just a mistake to suppose, as many do, that hue is rigidly determined by dominant wavelength. Contrastive effects are highly relevant, and the perceived hue of an object can change dramatically if it is isolated from its surroundings (by simply looking at it through a cardboard tube, for instance). Likewise, luminance can alter hue independently of wavelength. For example, if orange light of fixed wavelength is dimmed it will look redder, and if brightened then it will look yellower; likewise, if turquoise light of fixed wavelength is dimmed it will look greener, and if brightened it will look bluer. This ‘Bezold-Brücke shift’ occurs because the r–g channel operates more effectively at lower intensities, whereas the y–b channel operates more effectively at higher intensities. We know something of the physiology that underlies these effects. However, expectation and acclimatization will also play their part in determining appearances, and we are clearly dealing with a fairly high level of cognitive activity here, activity whose physiological grounding is only very poorly understood at present. In general, there is a clear danger that we shall simply be overwhelmed if we do not narrow our investigation, and the d-project helps us here. This is because it is essentially holistic. We are not concerned with why precisely this object in this particular situation has the perceived hue that it has, but rather with why things in general look as they do. The complications mentioned above will all apply mutatis mutandis within a d-inverted world: we shall get analogous contrastive effects, for example; and we can program the software to reverse the Bezold-Brücke shift, if desired. Thus, since they are on both sides of the equation, so to speak, they need not seriously intrude into our inquiry. The point is that (in a slightly eerie echo of opponent processing itself!) it is only the differences between the ordinary world and its d-inversion that should engage our attention. What is common to both need not concern us.
The issue of acclimatization relates to another point. Most philosophical studies of colour perception focus on colour qualia, yet the concept of a quale is notoriously open to many objections. For example, Dennett’s (1991, p. 625) celebrated ‘Chase and Sanborn’ example exposes the difficulty in distinguishing between acquiring different qualia and merely changing one’s attitude towards one’s original qualia. Dennett recommends that we abandon the concept altogether as irretrievably confused, though many suspect that this would require us to disregard subjective, conscious experience itself. Much has been written on this, but the point is that we are not required to take any particular line here. Our topic is hue not qualia, and nobody need dispute that there are hues.17 Moreover, intractable problems about whether acclimatization leads to new hue-perceptions, or just new attitudes towards old hue-perceptions, can likewise be kept at arm’s length, since they apply as much to d-inverts as to normal percipients. The general question of why things look like this (i.e. as now) as opposed to that (i.e. when d-inverted) remain broadly unaffected.
This needs to be qualified, of course, for it may be insisted that habituation will inevitably ensure that someone d-inverted from birth, or even just for a reasonably long time, must see things very differently from a normal percipient exposed for the first time to a few digitally edited photographs. After all, we adjust very quickly to coloured sunglasses, and the case of up–down inverting spectacles may well be relevant here. Experiments have shown that when subjects wear spectacles that include prisms that invert up and down then, although they are highly disoriented to begin with, they gradually acclimatize, and the world eventually starts to look ‘normal’. Tasks such as pouring a cup of tea or even riding a bicycle, which would be virtually impossible to begin with, become feasible. The fact that this happens gradually is itself a remarkable phenomenon, and does much to undermine the ‘pictures in the head’ model of visual perception.18 Now, might it not be that if subjects were made to wear d-inverting spectacles then we would get a similar result? Perhaps: though it would be extraordinary, I think, if all perceptual differences were to be erased after acclimatization. After all, phenomenal redness, for example, appears to have a much more definite, intrinsic quality than phenomenal ‘upness’ (‘upness’ seems to have too direct a connection with gravity to be purely phenomenal). However, the point, surely, is that this hypothesis can only really be decided empirically; and until the d-project is well under way, we cannot do this.
There are actually some live issues here. There is evidence that some people really do have red–green hue-inversion as a result of a rare genetic condition.19 Red–green colour-blindness results from two possible causes corresponding to two genetic features: in the one, the photopigments in the L cones are replaced with those that normally appear in the M cones; in the other, conversely, the photopigments in the M cones are replaced with those normally found in the L cones. In both cases, the L and M cones function indistinguishably, and this ensures that the L – M opponent channel is inoperative. However, it is possible that some people have both genetic features (why not?) which would presumably mean that the L and M photopigments are simply interchanged. The L – M channel is therefore operative, and the subjects appear normal. But, crucially, this channel is excited when it would ordinarily be inhibited and vice versa, thereby ensuring that all neural processes downstream from the retina are more or less the same as those of people with normal retinal chemistry but who have always worn red–green inverting spectacles.
Do such ‘pseudonormal’ people really see red where we see green, and vice versa? Or does habituation (a complex combination of acculturation and neurological adjustment) ensure that they see things pretty much as we do?20 We can, at least, think of a few elementary tests. We can ask such people, for example, whether (what they call) ‘red’ looks to them warmer than ‘green’ or vice versa (let us assume that they do not also have warm–cool sensory inversion!) At a more advanced level, we can also test whether their vital signs behave as ours do when they perceive various colour-transitions, such as grey to red versus grey to green (and there are the more elaborate experiments mentioned earlier). As far as I am aware, nobody has ever performed such tests, or even attempted to locate people with the relevant condition. However, genuine red–green inversion is surely a possibility that deserves to be taken seriously, and it cannot be ruled out solely by a priori philosophical argument. It is an empirical hypothesis, and not merely a piece of arbitrary scepticism.
What, then, of d-inversion? Here, matters are more difficult still, and for two reasons. Firstly, it is much harder to see how to interchange two channels, as opposed to merely reversing the polarity of a single channel. We know a fair bit about the differences between the r–g and y–b processes. What seem to be crucial, at least in the earlier stages, are differences between various cells in the lateral geniculate nucleus, a part of the thalamus which acts as a relay station between the eye and the visual cortex. Some such cells are red–green sensitive, others are yellow–blue sensitive.21 However, although the channels in general have a fairly well understood physiology, this does not translate into more precise anatomical or biochemical detail, so we are still far from being able to identify and distinguish all the underlying neural routes, let alone see how to ‘re-wire’ the connections appropriately. Actual d-inversion therefore remains little more than a philosophical thought-experiment, and cannot be understood as yet in convincing scientific terms.
Even more seriously, we still lack a phenomenal vocabulary to characterize d-inversion. Pace Levine, we can characterize the difference (or, at least, a difference) between green and red by talking of coolness or warmth, for example, but what is the phenomenal difference between green and blue? We can all see the difference, of course, but how do we put it into words? If we declare that such qualities are just ineffable, that they must inevitably lie beyond communicable language, then any scientific explanation of this aspect of hue-perception must fail at the outset, for we shall simply lack the means to formulate whatever it is that we are trying to explain.
Yet this may be unduly pessimistic. True, we currently lack a relevant vocabulary, but then nobody, as far as I know, has ever seriously tried to create one. Our d-project is exactly what is needed here. It is a little hard to say what the development of such new terminology would look like, especially in its early stages, but the natural policy would be to borrow terms drawn from other sensory modes or more abstract fields, and see whether any of the metaphors start to gel. These could be supplemented by words invented from scratch if this seems helpful (perhaps such words have a suggestive sound). The mind can be concentrated by toggling between d-inverted photographs and their (relevantly desaturated) originals, which ensures that perceived hues become the sole focus of attention, and regardless of the types of object that typically have them.
The latter point is evidently important, for the vocabulary becomes useless if it is tainted with ordinary physical associations. To explain why orange is a ‘warm’ colour is an interesting physiological task, for example; but merely to explain why it is a ‘fiery’ colour is obviously not what is wanted at all! It may be feared, however, that if such public anchors are disregarded then the whole enterprise will fall foul of the private language argument. There is perhaps a risk here, but it is surely not inevitable. Most importantly, we must remember that the d-project is essentially a communal project. If early phenomenal discourse sounds no better than mere babble, then it is at least a collective babble! It is also one which genuinely aims to follow public rules even if the rules have not yet solidified. Moreover, such terms as ‘warm’, ‘cool’, ‘advancing’, ‘receding’, ‘positive’ and ‘negative’, which already have fairly well defined phenomenal roles, certainly do not have merely private meanings in a Wittgensteinian sense, even though they do not relate to public objects in any relevant way. Of course, the rules are rather fluid, but that is true with all metaphorical meanings.
Moreover, any such linguistic development is subject to some very powerful objective constraints. The most important is that genuine d-inverts, should there be any, must start to feel at certain stages of the process that other people seem to be getting things the wrong way round. Since it is unclear that this happens even with the actual, and much simpler case of red–green inversion (or, less tendentiously, L/M photopigment inversion), and since it is also unclear that we could ever find any genuine d-inverts to work with, it may be rather difficult to get this constraint to work in practice. Nevertheless, if it were to happen that a small minority of contributors to the d-project persistently say that the terminology seems the wrong way round to them, and if it is subsequently revealed that they really do have appropriately inverted neural circuitry, then we would now have a very powerful reason for supposing that our new language is developing along the right lines. Indeed, we would be in an excellent position to insist that it is no longer babble, even if it still sounds very strange, but genuine scientific discourse.
It may be protested that the commonality of language is too powerful an influence to allow such minority protests to emerge, but this is unobvious. People born with unusual colour-perceptual conditions, such as synaesthesia or tetrachromacy, usually realize that they see things rather differently from other people—even if they prefer to keep quiet about the fact for most of the time (especially before diagnosis). D-invertedness is just a more subtle and less easily identified condition.22
All this may sound hopelessly optimistic, of course. Yet there are clear and constant differences between green and blue, and likewise between yellow and red (or, more relevantly, between yellow and pink), and so there must presumably be a reason for this. If the reason is not to be found in the brain, then where else is it to be found? It might still be insisted, stubbornly, that the physical and the phenomenal are just too far apart for there to be any illuminating explanatory links between them. To this, one may safely retort that over a century and a half of colour vision science has shown otherwise. The case of unique versus binary hues seems clear enough, for example. It is, moreover, not unreasonable to suppose that there are further natural links to be discovered. To this extent, we agree with Hardin. Yet even if hue-phenomenology can one day be completely explained in terms of physiology (and the possibility of seriously alien hues ensures that we have only just scratched the surface of what needs to be done here), we still need to decide what follows metaphysically. Hardin supposes physicalism. But the more natural conclusion to be drawn, surely, is that phenomenology itself will have become scientifically respectable. Moreover, if we can really develop a purely phenomenal language with the constraints sketched above, then it will be respectable in its own terms. This suggests a more interesting and unusual conclusion, namely that a kind of dualism can also be made scientifically respectable, and that we therefore do not need to be physicalists of any sort. On this view, there are indeed powerful explanatory links between the mental and the physical (as even Descartes, for example, recognized), but nothing metaphysically stronger. However, in order to see how far such links can help us with the ‘explanatory gap’ that plays such a central role in our current understanding of the mind–body problem, we need to look more carefully at the concept of explanation itself.
What Should Explanations Look Like?
I shall not attempt a full analysis of the notion, but simply focus on a few key features. The first point is that explanations are typically partial, not complete. They are geared to answer only certain questions, and only to the extent that the questioner requires. Anything else would lead to an overload of information, and the interesting points would be buried amongst a mass of irrelevant detail. This is important, because there is a temptation, especially in the case of the mind–body problem, to suppose that if something fails to explain everything then it does not really explain anything. The ‘explanatory gap’ is frequently imagined to be a deep chasm of some kind, and we are all familiar with the proverb that says that anyone who wishes to leap across a chasm would be ill-advised to take more than one step in doing so. Yet the analogy is not convincing. The results from colour vision science are individually very limited, to be sure, but there is no reason to suppose that a sequence of such small explanatory steps must inevitably fail to get us across the full ‘explanatory gap’. For example, ‘Why does green look as it does?’ is a large question. But it can be broken down into smaller questions such as, ‘Why does green look like a unique hue?’, ‘Why does green look like a cool hue’, and ‘Why does green look like green as opposed to blue?’, and so forth.
This last example points to another general feature, which is that explanations typically have a differential quality about them. Rather than simply asking, ‘Why do we have X?’, we are usually better off asking, ‘Why do we have X as opposed to Y?’, or ‘Why do we have X as opposed to Z?’, and we should bear in mind that the answers given may be very different. Unless the context makes it clear what sort of difference we are interested in, a blunt question of the form, ‘Why this?’, is unlikely to have a satisfactory answer, and not because there are some fundamental mysteries at work. Moreover, as we noted when discussing Levine’s alien hues, we should not suppose that because we are unable to answer all such differential questions, then we cannot really answer any of them, or that such answers that we can give must inevitably be defective in some way.
This is significant because it may be felt that our treatment has simply ignored the real issue here. The really puzzling fact, it may be insisted impatiently, is not that green looks like green as opposed to blue or red or any other particular hue (familiar or alien), but rather that it looks like anything at all. In other words, it is not green-versus-blue that should engage our attention, but green-versus-nothing. The really ‘hard problem’ of consciousness, in Chalmers’ phrase, is not why it should have this rather than that sort of subjective character, but simply why it should exist in the first place. Unless we can explain that, then it might well seem that we have achieved nothing of any real philosophical importance; and colour vision science is unhelpful here.
The immediate response to this should be that, although the ‘hard problem’ is indeed a hard problem, it is not the only problem, and that we lack a reason for supposing that it is the only one that matters. There are other differential explanations concerning why green looks as it does that will appear to the unbiased investigator to be very interesting indeed. Still, an answer to the ‘hard question’ is called for, and it may simply be, as Chalmers suggests, that there is a law of nature that says that whenever matter is organized in an appropriately complex way, then consciousness is generated. This will sound wholly unsatisfactory to many people, partly because it appears to imply dualism rather than physicalism, and partly because it is the presence of any such law that surely cries out for explanation. This leads us to another general feature of explanations, namely the role of laws within them.
According to the Hempel covering law, or ‘deductive-nomological’ model, the fact to be explained is deduced from a combination of two sets of statements, the first of which concern ‘initial conditions’, or particular facts about the circumstances in question, and second of which concern laws of nature. Both these sets are essential. In particular, the laws are essential, for the initial conditions would clearly cease to be explanatory if similar conditions could give rise to quite different results even though there is no further relevant difference involved. Such explanations can be partial or complete, and one could, if necessary, obtain similar explanations of why the cited initial conditions should hold, why the new initial conditions should themselves hold, and so on indefinitely. However, it is not supposed that the laws themselves can be explained. Laws generate explanations, but are not themselves the objects of explanation. Of course, this should be qualified, for we can explain secondary laws in terms of more basic laws, as with the derivation of Kepler’s laws from Newton’s laws, or with the reduction of the Gas Laws to the laws of particle mechanics via Boltzmann’s Kinetic Theory, and such explanations can indeed be made to approximate to the Hempel pattern. But ultimately, laws are just laws and they are simply there. They cannot be eliminated from the explanans altogether, which ensures that they themselves cannot ultimately be explained. This may sound unsatisfactory, and there are certainly a number of detailed flaws in the Hempel model. But we have learnt from Hume that laws of nature do not have to obtain in any very deep sense. Ultimately, there is a strong element of brute contingency in the fact that we have, for example, an inverse square law of gravity as opposed to an inverse fourth power law. We cannot explain these matters beyond a certain point, and we should not attempt to do so.
So why should psychophysical laws be considered to be so much more problematic than ordinary physico-physical laws? The answer, I suppose, is that they are required to connect items that look so fundamentally different, which is why we appear to have more of a mystery with the neurophysiological generation of consciousness than with the fact that, say, one billiard ball can cause another to move. The question, though, is whether they really are so fundamentally different, and this leads to some crucial aspects of the mind–body problem which colour vision science can successfully illuminate.
Physicalism versus Dualism
One of the reasons that the mind-body problem looks so impossible to solve is that we appear to have only very few options. In particular, it can sometimes seem as if we have to choose between a physicalism which insists that there is absolutely no difference at all between conscious states and their neural correlates, on the one hand; and a dualism which insists that there is a huge and untraversable difference between them, on the other. The former seems to deny genuine subjective consciousness altogether; but the latter seems to leave us with an intolerable mystery. Should we be allowed an intermediate position, where mental and physical states are allowed to be genuinely and irreducibly different, but naturally connected and not wildly dissimilar, then perhaps we could make some progress. But this option is apparently not available to us.
This should be qualified, of course. It might be protested that influential positions such as Putnam’s Turing machine functionalism and Davidson’s anomalous monism precisely do occupy such middle territory in so far as they both deny, albeit for different reasons, a rigid correlation between the mental and the neurological. Yet such positions are still uncompromisingly physicalist in so far as they demand that the mental logically supervene on the physical. That is to say, they demand that it be not merely contrary to the laws of nature, but actually logically impossible to have two creatures that differ mentally and yet do not differ physically.23 This is still a massively powerful constraint, and it does not seem right. It may indeed be impossible to have ‘zombies’, that is to say, creatures physically indistinguishable from ourselves but for whom it is ‘all silent and dark within’. Such creatures would be downright bizarre! But they are surely only impossible in the way in which flying pigs are impossible, and not in the way in which square circles are impossible. The imaginability of supervenience-failures here is hard to ignore, as is the link between imaginability and logical possibility.
Likewise, we agree that physically indistinguishable people who are genuinely conscious are not logically required to see colours in the same way. It might be thought that our whole programme assumes otherwise, but all we actually need is the weaker claim that such undetectable inversions are causally impossible. The latter is not unreasonable, for it is surely the simplest hypothesis to suppose perceptual similarity where there are no relevant physical differences (think of the complications that language-inversion alone would bring); and if the similarity is systematic, then there must presumably have to be a law of nature to sustain it. Of course, we can worry about the role of simplicity here, and why it should be a reliable guide to truth, but this is a quite general problem in the theory of knowledge that arises from the underdetermination of theories by data.24 It surely does not give rise to any special difficulties with the problem of other minds.
In general, inverted-hue thought-experiments do not support an arbitrary scepticism. Nor, contrary to received wisdom, do they argue against functionalism in any very interesting way. After all, hue differences seem to have a physiological grounding, and physiology just is the study of bodily functioning, and it differs precisely from anatomy or biochemistry, for example, in so far as it is not much concerned with how the functions are physically realized. It is therefore unclear what sense of ‘functional’ the functionalist has in mind if it is to exclude the physiological. No, the really valuable purpose of these thought-experiments is scientific: they help us to formulate more clearly just what needs to be explained, as is illustrated in the ‘d-project’. Specifically, they enable us to formulate an alternative which differs from the original at precisely the point where an explanation is being sought. As noted, serious explanations typically have a differential quality.
The crucial point, though, is that we do not need to assume logical supervenience in developing these explanations. For example, although Hardin styles his position as physicalist, he never has occasion to address the matter at all: it seems to be no more of an issue for him than it is for me.25
Yet dualism is usually understood on the Cartesian model which ensures that interaction is hard to understand. Indeed, the position itself is radically incoherent in so far as it insists that the physical world is subject only to soulless, mechanical principles, and yet makes an exception for the pineal gland. Now, it might seem that any dualism must likewise be incoherent if we are not to abandon the well-entrenched principle that physical phenomena have only physical causes (at least, at the most fundamental level). However, if there were a two-way nomological parallelism between types of mental state and certain types of brain state, then the autonomy of the physical realm will not be threatened. It might seem that we are now committed to a highly implausible epiphenomenalism; but if there are strong enough two-way nomological connections then that should be sufficient to justify the kind of counterfactual links between mental states and behaviour that we need in order to avoid the worst of the implausibility. Chalmers (1996, pp. 150–61), for example, has suggested a position like this. There remains the problem of seeing how conscious states can arise from complex physical phenomena if there is nothing in the basic constituents of the latter to explain this. Nagel (1979), in his defence of panpsychism, insists on a principle of non-emergence which ensures that the properties of complex systems must derive from the properties of their constituents and their effects on each other. However, his defence of this principle, intuitive though it is, depends on a highly necessitarian, non-Humean conception of causation. If we are allowed to suppose, by contrast, that nomological connections are genuinely contingent, and amount, in the final analysis, to little more than ‘constant conjunctions’, then we can live with emergence: and Hume’s arguments are very powerful. We are not forced to accept everything that Hume says, of course, but all that really matters here is the thesis that breaches of causation (and hence of causal explanation) are logically possible. Hume was surely right on this point.
As mentioned, though, the problem is that connections between mental and physical phenomena seem to involve more radical differences to those within the physical realm themselves, and that there is an intelligible ‘feel’ to the latter that is missing with the former. Yet the case of colour vision science shows that the scale of the difference may be an illusion. Many modern neurophysiologists bewail the mechanization of their discipline, the fact that, around the beginning of the twentieth century, it seemed to split irrevocably into a ‘bodiless psychology and a soulless neurology’, in Oliver Sacks’s words.26 Yet there remains the earlier nineteenth century physiological tradition of treating the physical and the phenomenological as a harmonious whole—the tradition of Ewald Hering, which was disregarded for many years, but which is now regaining favour. Here, it is entirely in order to use phenomenological facts (the perceived distinction between unique and binary hues, for example) to justify physiological hypotheses about colour vision processing; and likewise, for the latter to provide explanations of the former. The two form a natural unity, for nobody who studies the matter should doubt that the connections here between the physical and the phenomenal actually make sense. They are genuinely intelligible, and not at all arbitrary or baffling; and we might even find eventually that the link between hue and its physical basis is as natural as that between brightness and luminance, an association so compelling that it is hard to distinguish them in the first place, as we noted earlier. Of course, the connections are not so strong as to sustain deductively valid inferences; but, outside the tradition of Spinoza and Leibniz, few seriously expect otherwise. Scientific explanations should not be assimilated to mathematical proofs. So it remains logically possible to have the physical without the mental. Yet this residual dualism is sufficiently anodyne as to be unworrying. And it is only philosophers, not scientists, who ever seem to worry about logical supervenience.27
It may be protested that this too schematic a solution, and that it would need massive further development before it could gain any real credibility. No doubt this is so, but the ‘d-project’ indicates how we might proceed here. At any rate, it suggests a fresh, alternative way of looking at certain problems about consciousness. Notably, we no longer feel that our primary duty is to reconcile physicalism with the existence of an explanatory gap between the physical and the phenomenal—a bizarre if not impossible task, though how the mind–body problem is now generally understood. Our position, by contrast, is that there is no special explanatory gap here; and this is exactly why we do not need to be physicalists.
Billock, V. A., G.A. Gleason and B.H. Tsou (2001), ‘Perception of forbidden colors in retinally stabilized equiluminant images: an indication of softwired cortical color opponency?’, in Journal of the Optical Society of America A 18, pp. 2398–403.
Byrne, Alex (2007), ‘Inverted Qualia’, The Stanford Encyclopedia of Philosophy (Summer 2007 Edition), Edward N. Zalta (ed.), URL =