Interview by Richard Marshall.

Jonathan Cohenis the philosopher elaborating and defending a relationalist account of color properties, a view on which colors are constituted in terms of a relation between perceiving subjects and objects. He also thinks about interactions between and within perceptual modalities, and the implications of such interactions for our understanding of a range of issues including perceptual architecture, synesthesia, modularity, and sensory substitution. In language, he works on the semantics and pragmatics of context-sensitive expressions (i.e., expressions such as 'I' or 'this' that express different things depending on the contexts in which they are uttered), and on extrasemantic expansion (i.e., the ways in which we use words to convey more than what they literally mean, such as when we say 'Jonathan ate some of the cookies' and convey that Jonathan didn't eat all the cookies, or when we ask 'can you pass the salt?' to convey a request that our hearer should pass the salt). Here he takes us through those ideas about colour and the rivals, the influence CL Hardin, eliminativism, realism, introspection, differences between audition and vision, self-location, functionalism, special laws of nature for the working of the mind, his better best system, computational modelling of the mind, the puzzling duality of perceptual system response and how to understand the place of philosophy in the interdisciplinary work of philosophy of mind. This one eats all the cookies, and then some....

3:AM:What made you become a philosopher?

Jonathan Cohen:I was a latecomer to philosophy. I had nearly finished my undergraduate degree in math without having taken a single philosophy course, but for some reason thought to enrol in a couple of electives in the history of philosophy purely as a horizon-broadening
exercise. I was immediately attracted to the problems and ways of approaching them I encountered there, and so filled up the rest of my undergraduate time learning as much philosophy as I could, and then applied to graduate school so I could keep going. In retrospect, I suppose by the time I came to philosophy I had already found myself drawn to the more foundational parts of math (and other subjects I studied), so perhaps I was already headed in the philosophy direction. In any case, since I found philosophical problems so intensely gripping as soon as I came to them, it seemed crazy to me to resist thinking about them in favor of doing something else with my time. Very soon I stopped resisting.

I would say that, though my principal areas of specialization these days are the philosophy of mind, perception, and language, I think of myself as a philosophical generalist, and enjoy working in several other areas, including metaphysics, epistemology, philosophy of science, and aesthetics.

3:AM:You're a leading expert in philosophy of mind and colourhas been a key focus of your thinking in this area, using it to draw out philosophically interesting questions and answers. You cite CL Hardin as bringing about a sea change in the philosophy of colour: was it because he insisted that philosophers took on board the results of science that make him so important for the field? Have the armchair philosophers of mind been banished, and is this a good thing?

JC:Hardin's 1988 book, Color for Philosophers, was transformative because it showed how advances in color science (which, I think it's fair to say, hadn't been centrally on the radar of philosophers of mind and perception at the time) had the capacity to reshape longstanding philosophical disputes about color. To a crude first approximation, the landscape of philosophical options, and the kinds of considerations used to choose between those options one saw in philosophical writing about color in the decades before Hardin looked surprisingly similar to those present in philosophical writing about color in the 16th and 17th centuries, in great modern philosophers like Boyle, Galileo, and Locke. Hardin pointed out that the science of color had moved considerably forward since the modern period, and showed by example that its discoveries could constrain philosophical debates about the nature of color and color perception in new and unanticipated ways.

It is too strong to say that the armchair philosophers of mind have now been banished -- there are still many good ways to do important and interesting philosophy (about color and other topics) that are motivated by different mixes of empirical and non-empirical considerations. But I do think it is an adequacy constraint on responsible philosophical theorizing that one's views must ultimately mesh in some way with our best theories of the world. And in cases where one is arguing about a subject, such as color, language, or mind, about which there is a reasonably well-established body of empirical knowledge, this requires respecting -- a fortiori knowing something about -- the relevant facts. I take this not to be any special constraint on philosophical theorizing in particular, but simply a matter of being a responsible inquirer in any field. In my view, philosophical work on color scores better on this yardstick now than it did before Hardin shook the field from its dogmatic slumber.

3:AM:Hardin argued that objects weren't coloured didn't he, so he wasn't a colour realist but some kind of eliminatist? Why do you disagree with him?

JC:Hardin thinks not only that philosophers should pay attention to the findings of the contemporary color sciences, but that these findings collectively amount to an exceedingly stringent set of constraints on what colors could be -- indeed, that these constraints are so stringent that nothing that satisfies all of them, i.e., that there are no colors in the actual world.

To give the flavor, he argues from a number of findings about color psychophysics that colors have to stand essentially in a structure of similarity and exclusion relations --- that red is necessarily more similar to orange than it is to yellow, that blue is necessarily more similar to purple than it is to red, that red is necessarily incompossible with green, that blue is necessarily incompossible with yellow, and so on. Similarly he finds evidence for the idea that there are precisely four chromatically unmixed colors (the "unique hues": red, green, blue, and yellow), whereas all of the other colors are in some sense necessarily mixed. And then he argues that the best candidate mind-independent physical properties for being the colors do not satisfy these essential constraints on being colors, hence are not the colors. Similarly, he considers views that identify colors with high-level dispositions to look certain ways to normal perceivers; but he argues from empirical evidence of the widespread interpersonal variations in color perception that there is no principled way to specify what counts as a normal perceiver, hence that there are no coherent dispositions of the sort considered, so these couldn't be the colors either.

After considering a number of such proposals about what colors are and arguing that they don't meet relevant constraints, he concludes that there are no properties that are suited to being the colors. Of course he thinks that things (falsely) appear to us to bear colors. But since he thinks there are no colors, he holds that these appearances are false. He is, therefore, a color eliminativist/irrealist.

I regard this sort of color eliminativism as a coherent but drastic position of last resort because, on its face, it involves rejecting such a huge number of our beliefs about the world. Obviously it means rejecting lots of first order beliefs about the colors of objects (that ripe tomatoes are red, that ripe lemons are yellow, that my coffee cup is white, that the coffee inside the cup is brown, that the stop sign is red). But beyond these first order beliefs, it would also require treating perception and thought as guilty of systematic and extremely widespread error, and would mean that, at a quite basic level, we're not in the kind of epistemic contact with the world that we take ourselves to be in. And this, in turn, would undermine our usual conception of the world itself and the place we occupy in it. Again, I don't think it is incoherent to imagine ourselves arriving at this position; but it would be quite a radical and unexpected epistemic situation to end up in. Since this would be such a radical and revisionary outcome, it seems to me that this is not the place to start inquiry at the beginning of the day --- rather, it is a place to which we might be forced at the end of the day, and only if we are driven to it by the failure of all less radical alternatives.

That said, I think Hardin's individual challenges against the non-eliminativist theories he considers are quite important. These are legitimate hurdles that proponents of the theories in question must surmount in some way so as to avoid being driven to his radical eliminativism.

3:AM:A problem for colour realists is that different people seem to perceive colours differently. I see something as blue without any tints, you see it as blue with a green tint. So introspection seems to show that colour realism is wrong. You argue for a version of perceptual variation which relativises colours to perceivers don't you. Can you sketch how this works and why you think it is a good position for colour realists?

JC:First, I don't think this is an introspective consideration: it depends crucially on the empirical finding (extensively documented and measured by perceptual psychophysics) that there is considerable interpersonal variation between the perceptual responses of distinct color perceivers to one and the same color stimulus. It turns out that there is also considerable intrapersonal variation between the perceptual responses one color perceiver has to one and the same color stimulus as the perceptual circumstances vary. For example, one and the same uniformly colored patch ordinarily looks very different to a single observer when it is set against different backgrounds (see figure one for an example of this kind of perceptual variation), or illuminated differently, or perceptually grouped together with different subsets of the simultaneously perceived objects, or when the single perceiver is in a different state of visual adaptation, and on and on.

[An instance of simultaneous color contrast. The central patch appears more yellow/less blue when set against the surround on the left, and appears less yellow/more blue when set against the surround on the right.]

So in both the interpersonal and intrapersonal cases, there can be a wide range of perceptual effects of one and the same color stimulus. We can think about these effects as reports that tell the perceiver what color the stimulus is. If so, then we can ask: which, if any of them speak truly about the color of the stimulus? We can use your example for concreteness: suppose x looks greenish blue to me, and unique blue (i.e., blue without looking greenish and without looking reddish) to you. Who is right? The answer must be one of the following: (i) we're both wrong; (ii) you are right, I am wrong; (iii) I am right, you are wrong; (iv) we're both right.

Option (i) leads to a view like Hardin's on which color perception suffers from a widespread and systematic error; as I say, this position would be so costly to our view of perception, minds, and the world that in my view it is best avoided if there are viable alternatives. Options (ii) and (iii) (one of us is right to the exclusion of the other) are certainly coherent, but are objectionably arbitrary: though there are surely some perceptual reactions that can be classified as erroneous for independently motivated reasons, there remains enormous variation between reactions which there is no independently motivated reason --- barring ad hoc stipulation --- to regard as erroneous. I assume we should prefer if possible to avoid both extreme skepticism and ad hoc stipulation; if so, this gives us a reason for wanting to avoid answers (i), (ii), and (iii), and instead for favoring the ecumenical answer (iv), on which both variants are right.

But this leaves us with a further question: what understanding of color properties would allow for that sort of ecumenical answer? The "color relationalist" view I argue for in The Red and the Realis intended to answer this further question. On this view, colors should be thought of as relations, on the model of, say, sisterhood. It's a familiar thought that one single person can (simultaneously) be a sister to you and not a sister to me, just because what it takes to be a sister is a matter of standing in the right kind of relation, and because that single person could stand in the right relation to you and not to me. The relationalist view of color holds that, similarly, bearing a color is a matter of standing in the right kind of relations to perceivers and circumstances, so that a single thing could be (simultaneously) unique green to you in circumstance C1 and not unique green to me in circumstance C2.

This is a species of color realism because it allows that colors are perfectly real and instantiated properties. No one would infer from the relationality of sisterhood to the unreality of sisters. Likewise, color relationalism offers a way of accepting color realism, and so avoiding the eliminativist's extreme skepticism -- but without the unprincipled, ad hoc stipulations required by the many other forms of realism that insist on an exclusively veridical variant in cases of perceptual variation.

3:AM:Does this position undermine the idea that introspection is the final word on not just colour but many of the things going on in our minds? Does it generalize, or must we be careful to draw a conclusion from one of the senses and assume it's the same for the rest?

JC:I find it hard to accept that introspection could be the final word on understanding much of anything about the mind (or anything else). Sure, it's one potential source of information, perhaps better suited to some things than others, but whose reliability and import must, like any other putative informational source, be justified rather than assumed. It's a feature of rational belief acquisition (and not only about minds) that we don't know in advance where relevant evidence might come from. A fortiori, I can't see how it would ever be justifiable to build into our methodology for studying the mind (etc.) that we will limit ourselves to any single type of evidence -- be it introspection or anything else.

Of course, that's a perfectly general conclusion, and one that doesn't derive from my (idiosyncratic) relationalist views about color. There is, of course, a completely separate question about whether a relationalist view like the one I hold about color should be generalized to perceptual qualities other than color -- both in vision and in other senses. To answer this I think we have no choice but to consider cases one at a time.

Thus, for example, I don't think the argumentative strategy used to motivate relationalism about color transfers smoothly to the case of shape. This is because, though there is plausibly intrapersonal variation in the perception of shape -- a square looks different when viewed from an oblique viewing angle than it does when viewed head on, there *is* here a non-stipulative way of breaking the symmetry between variants. Namely, the way we think about shape properties, as spelled out in plane geometry (the science that we think gives the essential nature of shape), tells us that the perceptual reaction we get when we view from head on is the one that counts for the veridical perceptual representation of squares. This seems to me to be an important disanalogy to the color case, and one that plausibly blocks an analog of my argument from perceptual variation to relationalism about shape.

On the other hand, I think a relationalist story about flavor might be right. For one thing, there is plenty of evidence about perceptual variation with respect to flavor that would ground a version of the kind of argument I gave above for color. For another, there doesn't seem to be anything to play an essence-specifying role for flavor (analogous to what plane geometry did in the shape case) that would always break symmetries between competing perceptual variants with respect to flavor. Now, some writers, such as Barry Smith, have held that flavor has to be constitutively independent of our psychological/perceptual responses in order to account for agreement and disagreement about flavor, errors of flavor perception, and the critical appraisal of flavor perception. I agree that these are desiderata worth having, and that the very simple model one sometimes encounters of flavor as merely a reaction in us can't easily supply them. But I think a relational view of flavor can deliver the benefits in question much better than that sort of very simple subjective reaction view, and so is a more sophisticated, and more plausible, version of the general idea (which many have found attractive) that flavors in some crucial way are constituted partly by our perceptual responses.

Obviously there's much more to say about the individual cases mentioned above, but the general lesson about generalizability seems to be that relationalism about different perceptible qualities has to be evaluated case by case.

3:AM:A challenge to unifying theories of perception comes from the issue of whether sound and time are related. If they were this would make sound different from the other senses and dash hopes of unification. So is there evidence one way or the other about this?

JC:As I've just indicated, even within vision I think we have reason for denying that one uniform model applies straightforwardly to all of the relevant perceptible qualities: some may be relational (like, perhaps, color) while others are not (like, perhaps, shape). Hence I'm not myself clinging to any hopes of a very straightforward unification. That said, and as you point out, Casey O’Callaghan and others have indeed held that sound perception is different from vision in the distinct and perhaps more fundamental sense that the things audition relates us to are in some ways more temporal, while the things that vision relates us to are in some ways more spatial.

It does seem true that we think and talk about visual qualities like color in a way that is generally more tied to spatial locations and intervals, while we think and talk about sounds in a way that is generally more tied to temporal locations and intervals. But I think this asymmetry in our thought and talk is better explained in terms of contingent differences in the spatial/temporal resolution of the kinds of visual and auditory systems with which we happened to be endowed than by any deep metaphysical cleavage between the objects of vision and those of audition.

Here's what I mean about the resolution differences between vision and audition. On the visual side, we seem to be more sensitive to spatial than temporal inhomogeneity. For example, we seem to be far better at discerning color differences in patches presented simultaneously at different regions of the visual field than color differences in patches presented successively in the same region of the visual field --- this is why our performance in simultaneous color matching tasks is superior to our performance in successive color matching tasks. But we seem to have the opposite bias in auditory discrimination: we are more sensitive to temporal than to spatial auditory inhomogeneities. One demonstration of this difference comes from the discrepancy between our abilities to discriminate melodies on the one hand, and our abilities to discriminate chords on the other. Even musically untrained subjects are very good at distinguishing one four note melody (i.e., temporal discontinuity in sound) from another when these are played by a single trumpet (as it might be) from a single location. In contrast, distinguishing corresponding spatial discontinuities in sound is much more difficult. If four trumpet players stand arrayed before the subject and simultaneously each plays a single note, they will thereby produce a chord that constitutes a sonic spatial discontinuity. Discriminating one such spatial discontinuity from another is not impossible, but takes considerable effort and practice (music students have to work hard at this kind of ear training).

Given, then, that our perceptual systems seem to be more sensitive to spatial than to temporal inhomogeneities in visual attributes like color, but more sensitive to temporal than to spatial inhomogeneities in sound, it is predictable that we should more easily notice, think about, and talk about the spatial distribution of colors and the temporal distribution of sounds rather than vice versa. But I suggest that we should be on guard against taking this predictable asymmetry in the way we notice, think, and talk about colors and sounds as evidence of an underlying metaphysical asymmetry in the nature of the two sorts of perceptual qualities.

3:AM:An alternative to the relationist theory of colour answer to the variation problem is the self-locating property theory of colour. How does this work and what's it to do with having your cake and eating it too?

JC:I think you're alluding to my paper, "Ecumenicism, comparability, and color, or: How to have your cake and eat it, too," where I discuss a tension between two empirically motivated desiderata for a metaphysics of color. On the one hand, and as sketched above, considerations about perceptual variation give us a reason for wanting a theory on which distinct perceivers can ascribe different colors to a single object (e.g., you say x is unique green, I say x is bluish green) without either being wrong -- in other words, we want the colors you ascribe and the colors I ascribe to be in some ways incomparable and not mutually constraining. Call this the ecumenicism desideratum. And on the other hand, it certainly seems like you and I can agree or disagree about the color of x, which suggests that the colors you ascribe and the colors I ascribe can be directly comparable and mutually constraining. Call this the comparability desideratum. Considered on their own, there's a good empirical case to be made for both the ecumenicism desideratum and the comparability desideratum. But the two appear to come into direct conflict. So the question becomes: on the (defeasible) assumption that these desiderata are both worth respecting, what theoretical options are there that would allow us to accommodate both?

Color relationalism is one such option. It is pretty obvious that relationalism is in a good position to secure ecumenicism, but less obvious that it can handle comparability. But I argue that there are independent reasons for favoring a version of relationalism that adds a secondary (non-perceptual) level of representation at which colors are represented in a less fine-grained way, and that the relationalist can indeed account for comparability in terms of this second, independently motivated level.

Another (slightly more technical) theoretical option is the self-locating property theory, defended by Andy Egan and BritBrogaard. To motivate this view, imagine that there are two clowns -- Bozo and Krusty -- such that Bozo believes his pants are on fire, and Krusty believes his pants are on fire. Intuitively, there's some good sense in which Bozo's belief and Krusty's belief are about the same property --- that their beliefs are about some (deeply unpleasant) common aspect that their two situations share. But on a traditional way of thinking of properties as features of the whole world, their beliefs aren't about any common property: Bozo's believes he's in a world where *Bozo's* pants are on fire, and Krusty's believes that he's in a world in which *Krusty's* pants are on fire. The self-locating property theorist has a solution to this problem: stop thinking about properties as characterizing the world as such, and instead think of properties as characterizing the world-as-centered-on-a-time-and-individual. Then we can say that Bozo's and Krusty's beliefs are about a common property: they characterize the world-as-centered as one in which the pants worn by the center-individual (at the center-time) are on fire. And we can simultaneously recognize that there's a good but different respect in which Bozo's and Krusty's beliefs characterize the world differently because (despite what they share) they pick out different centers: Bozo's is centered on Bozo, while Krusty's is centered on Krusty. When you evaluate the common characterization with Bozo at the center, you get Bozo's belief that Bozo's pants are on
fire, while when you evaluate the common characterization with Krusty at the center, you get Krusty's belief that Krusty's pants are on fire. In this way, the self-locating property apparatus allows us to say that there is a respect in which Bozo's and Krusty's beliefs are alike, and a respect in which they are different.

The self-locating property theorist proposes that, if colors are self-locating properties in the way that the property of having one's own pants on fire is plausibly self-locating, then this would allow us to respect both the ecumenicism and the comparability desiderata for colors in just the way we found with Bozo's and Krusty's beliefs. Start with comparability. When you perceive x as unique green, you represent x as looking unique green with respect to a world-time-individual triple (w,t,i); when I perceive x as bluish green, I represent x as looking bluish green with respect to a triple (w,t,i). But this can count as a genuine disagreement between us because there is no one triple <w,t,i> with respect to which both of our representations can be true: there is no (w,t,i) such that x looks unique green with respect to (w,t,i) and x looks bluish green with respect to (w,t,i). Hence, we have a way of thinking of your and my representations as being in agreement/disagreement. Moreover, and just as we saw with Bozo and Krusty, there's a distinct sense in which we can secure ecumenicism for the two representations. For, though there's no single (w,t,i) with respect to which both can be true, there's no problem about your ascription being true with respect to a world centered on you (i.e., (w,t,you)) while mine is true with respect to a world centered on me (i.e., (w,t,me)). Thus, the self-locating property theory gives us another way, distinct from color relationalism, of respecting both the apparently clashing desiderata.

That both of these theories of color -- unlike many others on offer -- at least attempt to meet the two empirically motivated desiderata at issue is a reason to like them and take them seriously. And it's not easy to choose between them; for me, at least, it comes down to a fairly delicate cost/benefit analysis on which reasonable interlocutors can disagree. But at least one thing that I think is a important cost of the self-locating property theory is its lack of systematic explanation. The view makes room for a subject-involving element in colors by saying that colors pick out extensions in a way that depends non-trivially on the center individual. But it doesn't have much to say about why that is true about colors (and not all other properties), and it doesn't say anything that predicts that that should be the case for each of the colors: not just green, but also blue, and yellow, and red, and so on. From the self-locating property theorist's point of view, it is more or less an unexplained accident that the different colors receive the same theoretical treatment by the view. Whereas, in contrast, the different species of color relationalism will all have an explanation of why each of the color properties constitutively depends on subjects. To the extent that having an explanation is better than not having one, I think this gives us a reason for preferring the relationalist's way of having your cake and eating it, too.

3:AM:Another alternative to the relationist account of colour is a functionalist account isn't it? What are the main differences and which account do you think works best to account for the empirical data we have? You seem to have defended both approaches, so are they compatible?

JC:They are indeed intended to fit together compatibly. To see what's going on here, let's return to the property of sisterhood. If we favor a relationalist view about this property, we're saying we think that being a sister is a matter of (not intrinsic makeup, but) bearing the right sort of relation to a person. That's an advance on our quest to understand sisterhood because it rules out non-relationalist theories. But it nonetheless leaves open the question of *which* sort of relation to a person something has to have if it is to count as a sister of that person. Similarly, if we decide for whatever reason that we favor relationalism about redness, we're saying that being red is a matter of (not intrinsic makeup, but) bearing the right sort of relation to a viewer and a perceptual circumstance. Again, that's an advance on our quest to understand redness because it rules out non-relationalist theories. But it nonetheless leaves open the question of *which* sort of relation to a viewer and a perceptual circumstance something has to have it is to count as red to that perceiver and that perceptual circumstance.

Color functionalism --- specifically, role functionalism about color --- is intended as an answer to this last question. On this answer, colors are functional roles. Specifically, they are the functional roles of disposing their bearers to look certain ways to certain perceivers in certain circumstances. So, for example, red to perceiver S in circumstance C is the functional role of disposing its bearers to look red to S in C (and likewise for the other colors).

Among its other benefits, role functionalism offers a particularly satisfying explanation of the banal observation that the colors and the dispositions to look colored apply to the very same
things: exactly those things that are red are also the things disposed to look red, exactly those things that are green are also the things disposed to look green, etc. Role functionalism explains this banal observation tidily: it entails that exemplifying a color just is exemplifying some or other property that realizes the disposition to look colored, and so exemplifying the disposition to look colored.

Of course, there are forms of color relationalism other than role functionalism. And, at least for those sympathetic to relationalism, these other views are clearly worthy of consideration. However, I believe that each of these other views is ultimately unsatisfactory, hence that role functionalism is the most attractive form of color relationalism --- and, given my commitment to relationalism, that it is therefore the most attractive metaphysics of color full stop.

You might wonder why I bother to defend relationalism, the more general view, if I ultimately endorse role functionalism about color, the more specific view. The most important reason is that there is much that can be said about the pros and cons of the more general view that is independent about the more specific commitment. I want to allow for partial alliances (I need all the help I can get!) with those who might sign on to my arguments for relationalism even if they part ways with me over role functionalism.

3:AM:Are there special laws of nature for the working of the mind. How do you understand the special sciences and their laws – do laws of our mental life introduce any special problems for our understanding of laws of nature? After all, isn't this the claim of someone like Chalmers who argues that the hard problem eludes such laws?

JC:In a couple of papers written jointly with my UC San Diego colleague, Craig Callender, I've defended a way of thinking about laws of both basic and special sciences. Our view, which we immodestly call the Better Best System (BBS) view, is a variation of the classical Best System view sometimes attributed to Mill, Ramsey, and Lewis (MRL).

Classical MRL starts by conceiving of our knowledge of the world as a deductive system containing axioms and results derived from those axioms. Of course, there will be different ways of arranging our knowledge of the world into such a deductive system. Some such systems will be stronger than others in the sense that they contain (among their axioms and derived consequences) more truths about the world than others. Other true deductive systems will be simpler than others in the senses that they include a smaller set of independent axioms or are syntactically less complex. Significantly, strength and simplicity seem to be competing virtues: adding more axioms to a system increases a system's strength at the cost of simplicity, while taking away axioms increases simplicity at the cost of strength. The heart of the MRL approach to lawhood is to say that a true generalization is a law if and only if it is an axiom of all the "Best Systems" --- axiomatic systematizations that best balance strength and simplicity.

So, a generalization like Schrodinger's equation might plausibly count as a law on this account because any deductive system lacking it would be inferior in strength (by leaving out many truths about the world) or simplicity (by recapturing the otherwise left-out truths only by including a huge list of underived axioms about the locations of particles at each moment). In contrast, the true generalization that all your children have brown hair is plausibly something that could be captured in a Best System without making it a law (viz., it wouldn't be too costly in terms of simplicity to include axioms listing all your children and their hair color, from which we could derive the needed result as a consequence).

While we find a lot to like about the classical MRL view, Callender and I have two significant worries for it. The first is that classical MRL has a hard time making room for laws in the special sciences (i.e., sciences other than basic physics). After all, MRL is usually taken to require that laws relate fundamental natural properties, and this means that higher level generalizations in the special sciences could only count as laws if they were translated into statements about long, complicated disjunctions of the fundamental kinds --- but these
are unlikely to be simple on any reasonable metric. Moreover, because special science generalizations tolerate exceptions and are incomplete descriptions of reality, they are unlikely to beat out candidate fundamental laws in terms of strength either. But, even if special science generalizations are neither simple nor strong in the sense that classical MRL requires of laws, it seems unattractive to deny them this status; after all, they have just exactly the role within the special sciences that physical laws have within physical science (they explain, they support counterfactuals, and are simple and strong generalizations about their own domains, etc.). The other problem we take seriously is that classical MRL requires us to make comparisons of the strength and simplicity of generalizations formulated in terms of very different basic kinds (viz., those used in different axiomatizations). But it's unclear how to make such cross-system
comparisons. Famously (and as Goodman's grueexample makes vivid), the standards for the strength and simplicity of a generalization themselves depend crucially on the inventory of basic kinds -- there is no system-independent metric for strength and simplicity that we can use to compare generalizations taken from different axiomatizations.

The BBS response to this situation is to embrace more fully the observed diversity of basic kinds used by different sciences, and think of lawhood as relativized to a choice of basic kinds. Thus, while, say, Malthus's exponential law of population dynamics might not be an axiom of all the simplest and strongest systematizations formulated in terms of the basic kinds of physics (hence not a law of physics) it nonetheless might be an axiom of all the simplest and strongest systematizations formulated in terms of the basic kinds of population ecology. On this picture, we won't be able to say that there is The One True Best System; but we can say that there are lots of different Best Systems of different sciences, and that each such science (special or not) has its own metaphysically autonomous set of laws.

I take it to be an attractive feature (among others) of BBS that it is in principle no less applicable to minds than to any other systems of interest. BBS says that a generalization about a domain D counts as a law relative to that D just in case it is an axiom of all the on balance simplest and strongest axiomatic systematization of D (measured relative to the basic kinds of D). We can, of course, axiomatize whatever domains we favor in terms of whatever basic kinds turn out to be explanatorily fruitful (what could stop us?). In particular, we can do this for minds (we call the resulting science psychology), just as we can for population ecology or statistical mechanics or whatever else we like. And, if we can then make sense of simplicity, strength, and balance with respect to the kinds we choose, we can measure candidate axiomatizations with respect to these virtues, and can decide which generalizations (if any) appear in all the systematizations that best balance simplicity and strength. These will be the laws of the domain, relative to our choice of basic kinds.

I don't think Chalmers would necessarily object to any of the above as far as it goes. He certainly accepts that there are true lawful generalizations about the mind such as form the subject matter of psychology. Indeed, he even holds that there are nomic generalizations connecting non-phenomenal kinds to phenomenal kinds, though he regards the latter as metaphysically contingent rather than necessary (this interacts with his views about the viability of a posteriori/B-type materialism). But I don't think his views about the metaphysics of consciousness are in conflict with the BBS conception of scientific lawhood as such.

3:AM:Does computational modeling offer powerful insights into how to understand the mind, in particular in terms of cognitive control?

JC:It's hard to deny that computational modeling has been an enormously fruitful tool for theory building at all kinds of levels and in all kinds of areas of the cognitive and perceptual sciences. In my own work on perception I've been interested in computation not so much as a modeling tool, but as a way of thinking about a puzzling duality in
the way that perceptual systems respond to the world.

To see the kind of duality I have in mind, suppose you're viewing two contiguous and same-sized, same-shaped patches of a uniformly painted wall that are each (more or less) illuminated uniformly, but such that there is a difference in the illumination between the two --- say, one is in shadow and one is in direct sunlight. If you ask normally sighted subjects to sort the patches by their color appearance, they can and do behave in two distinct ways. Sometimes they sort the patches together, which suggests that their sorting behavior is responsive to a dimension of color appearance on which the patches are alike --- there's some sense in which they regard the patches as appearing the same in color despite the clearly visible difference between the two. On the other hand, they sometimes sort the patches into different classes, which suggests that their sorting behavior is responsive to a dimension of color appearance on which the patches differ --- there's some sense in which they regard the patches as appearing different in color despite having been painted with paint from the same can. And, interestingly, ordinary subjects can switch back and forth between these two ways of sorting (either spontaneously or as a consequence of changing the task instructions). The duality of the behavior here gives us a reason for thinking that perception represents distinct things about the stimulus: it represents an aspect of similarity (driving one response) and it represents an aspect of dissimilarity (driving anotherresponse).


There's a similar situation concerning the perception of size. If a subject views two telephone poles at different distances, she can judge of the different poles both that they are, in one way, alike in their size appearance and that they are, in another way, not alike in their size appearance. Likewise for shape perception. If a subject sees a round dinner plate from two different angles (one head on and one tilted relative to the line of sight), she can report on (or respond in a quantitatively measurable way that is sensitive to) either a respect in which the apparent shape of the plate is similar between the two presentations, or a respect in which the apparent shape of the plate is dissimilar between the two presentations. In these cases (and others like them), it is natural to regard the bimodal pattern of behavioral response as reflecting different contents that perception represents --- one content about an aspect of similarity, and a different content about an aspect of dissimilarity.

The reason this situation can seem puzzling is that the two contents in question are not simply different from one another, but in some sense (that cries out for explanation) contrary, or opposed, to one another. They say that the stimuli are in some way both similar and dissimilar in color/size/shape. At the same time, the distinct contents don't seem like they are inconsistent with one another in the way that (say) the things we represent about impossible Escher figures are jointly inconsistent. So our puzzle is: how can we make sense of the idea that perception represents the telephone poles (without contradiction) as both similar in size and sort the same telephone poles as dissimilar in size, given that two objects cannot be of the
same size and different size?

Now, among the many answers to this puzzle that have been offered by philosophers and psychologists, one of the most popular is a kind of denialism that downplays or disregards one of the two observed behavioral reactions. It's true that if we could downplay/banish the dissimilarity representing content (perhaps regarding it as generated by inference rather than perception, and part of an optional "painterly" form of perception, as some have maintained) then we could eliminate, and so free ourselves from the burden of having to explain the otherwise puzzling clash. The problem is that the dissimilarity reaction seems a perfectly real perceptual phenomenon; and while rejecting the facts may be a good way of avoiding problems, it is not a good way of solving them.

A better way, it seems to me, rests on accepting the dual representations that we have reason to believe are present, and providing a computational conception of how they are related to one another and distinguished by the perceptual system. In broad outline, the idea is that the two representations at issue are treated differently by playing different roles in a certain
functional/computational relationship that perceptual systems compute. More specifically, the suggestion is that one of our properties is the input to a computation carried out by perceptual
systems, and the other property is the output of that computation.

To see what this looks like, return to the example of the tilted dinner plate. I claim that one of the things the perceptual system represents about the plate is that it is similar in appearance to (not circles but) ellipses seen straight on. This representation drives the response (which we really do observe) wherein subjects sort tilted dinner plates with 2d ellipses rather than 2d circles. But the perceptual system can also use this representation, together with a representation of facts about the perceptual circumstance (such as the angle formed by the plate and the line of sight), to compute a further representation of what the plate would look like if seen head on. And this second representation can drive the response (which we also
really do observe) wherein subjects sort tilted dinner plates with 2d circles rather than 2d ellipses. So we have two representations of the plate -- playing different roles within a computational relationship -- corresponding to two distinct observed behaviors.

Predictably, this computational relationship is generally invertible. When the perceptual system already represents the dinner plate's property of roundness (say, when seeing it head on rather than tilted), it can use that representation together with information about the change in the angle between the plate and the line of sight to compute a downstream secondary representation of what the plate will look like at the new angle. Indeed, the perceptual system can even use the two representations of the plate (one that assimilates it to ellipses seen head on, one that assimilates it to circles seen head on) to compute a value for the relevant perspectival parameter (here the angle the plate moves through between the two perceptual presentations). So the most general way to think of the process here is as a computation defined over three variables: the occurrent appearance, the perspective-independent property, and the relevant perspectival property of the perceptual circumstance. The idea is that, roughly and within some interesting limits, perception can compute any one of these given the other two.


3:AM:Do things like change blindness and inattentional blindness show that we can't trust our eyes like we thought we could?

JC:The phenomenon of change blindness (CB) is the inability of a ordinary subjects to report or identify large scale changes in the visual scene; inattentional blindness (IB) is the inability of such subjects to report or identify even very large unattended objects/events in the visual scene.

In the 1990s and early 2000s a number of researchers appealed to CB and IB to argue that ordinary perceivers are laboring under a "grand illusion" about the nature of their perceptual interactions with the world --- usually, the idea was that ordinary perceivers wrongly take themselves to be generating from perception rich and detailed internal representations of the world, and that CB and IB reveal these impressions to be false. In my view, the claim that CB and IB reveal some widespread grand illusion about perception is quite implausible (hence the title of my paper, "The Grand Grand Illusion Illusion"). Of course, the effects are interesting and important, but I think what they reveal about perception is somewhat more modest.

First, it seems dubiously overintellectualizing to think of the effects as falsifying ordinary perceivers' worked-out views about the nature of perceptual representations, simply because ordinary perceivers don't typically hold worked-out views on such matters (or else they would be perceptual scientists).

Quite apart from that, however, there are reasons for doubting that CB and IB threaten the idea that perceivers form rich and detailed internal representations of the world. It's true that, in a case of CB or IB, there's something that transpires in the visual scene (either a change, in the case of CB, or an object/event, in the case of IB) such that the perceiver is unable to report it verbally. But that's consistent with two possible descriptions. On the first description, offered by proponents of the grand illusion, the perceiver is unable to report the missing item (i.e., object/event/change) because she/her visual system never perceptually represents that missing item. The second description holds that the perceiver's visual system DOES perceptually represent the missing item, but that she is unable to report that item because whatever system is responsible for the report is for some reason unable to access fully the perceptual representations that have been formed. So, to draw the lessons they want to draw from CB and IB, proponents of the grand illusion view would owe us an argument that would rule out the second description.

But in fact, there's evidence that supports the second description over the first. In particular, researchers have found that, even when subjects are unable to report explicitly on the missing items in CB and IB tasks, subjects' behavior is systematically affected by the missing items -- suggesting that representations of the missing items are present in the system and having downstream effects despite not being available for access by explicit report. For example, it turns out that, even when subjects are unable to explicitly report whichregion of the scenechanges in cases of CB, their visual fixation times are significantly longer in regions where the change is occurring. Similarly, when asked to identify visually presenteditems, subjects were faster in identifying items that had gone missing in an earlier CB/IB experiment than items that had not been presented to them at all. These and similar effects are hard to explain on the supposition that perceivers never represent the missing items, and therefore tell against the more dramatic first interpretation of CB and IB favored by proponents of the grand illusion view.

Of course, this leaves us with the task of saying why, if missing items in CB and IB are represented, subjects are nonetheless unable to access these representations for explicit report. Luckily, there's been progress in answering this question. Thus, for example, we've learned that subsequent access to representations can be interrupted by (e.g.) saccades, film splices, mudsplashes, real world occlusion, or alternation with a neutral mask, often because these features prevent the kind of visual pop-out that ordinarily alerts us to changes in the perceived scene. In my view, the real interest and importance of CB and IB lies in prompting researchers to think hard about answers to this question and what they can teach us about perceptual systems.

3:AM:Philosophy of mind is very much an interdisciplinary field these days as we've noted earlier – but I guess some might ask what the role of the philosopher is in this mix? Can't it just be left to the scientists now?

JC:Much depends on just what 'it' we're talking about leaving to the scientists. I think there are some genuinely valuable intellectual projects in philosophy that don't interact significantly with the empirical --- say, much of the history of philosophy, large chunks of value theory, certain parts of epistemology, and on and on, and I wouldn't want to leave these entirely to the scientists (who, presumably, would promptly abandon them). Restricting ourselves to projects that have some significant empirical component, of course I accept that there are many that fit squarely within the intellectual umbrella of some first order science; and it would be insane to insist that philosophers (as opposed to the scientists, who have task-appropriate training, tools, and talents) have to be involved with the first order science in every such case. In this sense there certainly are some kinds of intellectual projects that are better off left to scientists.

The proviso about the latter type of case is that the resulting scientific work is fair game for reflection, consideration, and application by thinkers inside and outside of that field -- so, inter alia, by philosophers. An example, already mentioned above, is Hardin's use of pyschophysical work on the similarity structure between colors to argue against various ontological views about the nature of color properties. That's a contribution that undoubtedly pushed debate forward. But it's a contribution that probably wouldn't have been made without the involvement of philosophers: it took Hardin to see the relevance of the psychophysical work to existing ontological controversies.

It's true that the topics that I personally have found most exciting in my own career generally have tended to involve interactions between philosophical and scientific ideas and methods. But within this territory, I would distinguish two different types of interaction, corresponding to different directions of influence. First, there are cases, like the Hardin example just mentioned, where philosophers pick up more or less off-the-shelf scientific findings of some kind and show how they can be used to argue for or against positions in antecedent philosophical disputes. Second, there are cases where philosophers (in some cases, working shoulder to shoulder with foundationally-attuned scientists) take up foundational questions arising within a science, and that haven't been part of any antecedent philosophical dispute. (Think here of philosophers of physics who have fruitfully asked questions about how to understand quantum mechanics; philosophers of biology who have disputed what units natural selection acts on; philosophers of psychology who have wondered about the relation between intermodal interaction in synesthetic perception and in normal perception; etc.)

Of course, it is in the nature of such cases that finding solutions -- and even seeing the problems -- requires achieving a non-trivial level of understanding of the science in question. This makes interdisciplinary work hard, but also rewarding.

3:AM:And for the readers here at 3:AM, are there five books you could recommend other than your own that would take us further into your philosophical world?

JC:I'll begin with a book I already mentioned, Hardin's Color for Philosophers, which has been largely responsible for bringing philosophical work on color to a much higher level of empirical sophistication. However one ultimately ends up assessing Hardin's own substantive views, this book is an excellent place to begin one's philosophical education about color.

One of the authors most influential on my thinking about the mind is Jerry Fodor. I'm by no means alone in this: he has uncontroversially set the agenda in philosophy of mind for decades. Matters of influence aside, Fodor is a thunderingly good philosopher: he's someone who asks the big questions, sees the dialectic faster and more deeply than anyone else in the room, and argues trenchantly and often persuasively for highly controversial views. While many of his books are recommendable, perhaps my personal favorite is Psychosemantics: The Problem of Meaning in the Philosophyof Mind, which usefully draws together several of the problems he has worked on over his long and influential career.

My next choice is Lewis's Collected Papers. This is cheating not only because it is an anthology of separately published papers rather than a monograph, but also because it is published as two volumes. But I'm sticking with my choice because it includes so many of Lewis's philosophical gems in so many different areas (e.g., mind, language, decision theory, probability, causation). Lewis's ingenious and resourceful papers in each of these areas were groundbreaking, and continue to shape discussion throughout virtually every subdiscipline
in philosophy.

The central topic of Burge's Origins of Objectivity is what it takes for minds to represent the world objectively. He criticizes huge swathes of twentieth century philosophy for making the requirements on objective representation either overintellectualized (Strawson, Evans, Quine, Davidson) or too thin to distinguish perception from "mere" sensory registration (Dretske, Millikan, Fodor), and instead proposes an account of objective representation grounded in in the phenomenon of perceptual constancy. Though there's much in this book with which I disagree (among many other things, I disagree with the way Burge
thinks about perceptual constancy), there's no question that it is a commanding and impressive example of what empirically-informed philosophy of mind/perception looks like today.

My final choice is Siegel's The Contents of Visual Experience. In it Siegel offers and defends a method for adjudicating disputes about what is perceptually represented (her "method of phenomenal contrast"), and deploys this method to argue that the representational contents of visual states include not just low-level properties like color, form, and motion, but also high-level properties including kind properties (being a dog/person/bicycle/Scottish Pine, being a word of Russian), as well as action properties (carrying a dog/person/bicycle/Scottish Pine), causation, and particular objects (John Malkovich). Again, while there are many places where I am inclined to disagree with Siegel, this book is beautifully argued, excitingly controversial, and a lovely example of contemporary philosophical thinking about perception.

Richard Marshallis still biding his time.

Buy his book hereto keep him biding!