The Two-Dimensional Content of Consciousness [Draft #3] Simon Prosser sjp7@st-andrews.ac.uk 1. Introduction For many years philosophers of mind tended to regard phenomenal consciousness and intentionality as posing separate problems whose solutions could be given independently of one another to quite a large degree. Recently, however, this has begun to change. Over the last decade or so there has been a rapidly developing interest in the relation between phenomenal consciousness and intentionality, with a number of philosophers arguing for close connections between the two. Most conspicuous among the theories that have been put forward is representationalism. Let the phenomenal character of a conscious experience be defined such that two experiences differ in phenomenal character if and only if there is a difference in what it is like to have those experiences. Then representationalism is the view that the phenomenal character of an experience depends entirely on its representational content; experience contains no non-representational features. This contrasts with the idea that there are qualia that may or may not act as intermediaries for represented contents but which also contribute intrinsic phenomenological features that do not represent anything. Representationalism has been most notably defended by Michael Tye (1995, 2000, 2002, 2003) and Fred Dretske (1993),
but similar views or variants have also been put forward by Sydney Shoemaker (2000), David Chalmers (2004b, forthcoming b), and many others. 1 Another family of views combining phenomenal consciousness with intentionality are higher-order representational (HOR) theories. According to these theories phenomenal consciousness can be analysed in terms of higher-order representational states that represent first-order conscious states. HOR theories have been defended in various forms by David Rosenthal (1986, 1997, 2002), Peter Carruthers (1996, 2000), Daniel Dennett (1978, 1991), David Armstrong (1968, 1984) and William Lycan (1996), among others. In this paper I put forward a version of representationalism that differs from standard versions. In brief, the view is that phenomenal character correlates with a content equivalent to the diagonal proposition in Robert Stalnaker s (1978, 2001, 2004) version of two-dimensional modal semantics for utterances. Consequently I shall call the view diagonal representationalism. As I shall explain, diagonal representationalism can be seen as a very broad generalisation of the non-standard version of representationalism put forward by Shoemaker. It is also a close cousin of Chalmers representationalist theory, though I shall note some possible points of disagreement. Showing that representationalism is true, albeit in a non-standard form, should be of interest in itself. What I take to be of additional interest about diagonal representationalism, however, is that it brings out connections between representationalism, HOR theories and functionalism that, as far as I am aware, have not previously been noticed. In effect it combines elements of all three doctrines within a single theory. 1 A full list of those who have defended representationalism or related views would be very long, but the following is a representative sample: Byrne 2001, Clark 2000, Crane 2003, Egan (forthcoming), Harman 1990, Kriegel 2002, Levine 2003, Thau 2002. Not all of these defend reductive claims (see below). 2
The structure of the paper is as follows. I start by describing some motivations for representationalism, some varieties of representationalism and the standard objections to it. I accept that at least one of these objections poses genuine problems for standard versions of representationalism. I then show, moreover, that some minimal functionalist assumptions accepted by most philosophers (including even some property dualists) straightforwardly disprove standard versions of representationalism. I then outline Shoemaker s non-standard representationalism and describe a modification of it that I call perspectival representationalism. This is a step in the right direction but falls prey to an objection put forward by Ned Block. I then outline some key features of Stalnaker s two-dimensional semantics before describing diagonal representationalism in detail. I compare diagonal representationalism with Chalmers Fregean representationalism, concluding that the two are quite closely related. I then explain how to deal with a putative objection regarding experiences whose contents are impossible states of affairs. HOR theories are then discussed and it is shown that diagonal representationalism is a kind of HOR theory. At this point a number of different threads from previous sections are drawn together to give a unified account combining representationalism, higher-order representation and a broadly functionalist theory of consciousness. In the final sections I show that standard versions of representationalism are committed to a questionable view of the mind often known as the Cartesian theatre. Diagonal representationalism, by contrast, makes it easy to account for the indeterminacy of phenomenal character implied by the denial of the Cartesian Theatre because there is a corresponding indeterminacy in the representational content that correlates with phenomenal character. Different yet equally good interpretation schemes can be given for both. Finally I make some brief remarks 3
about externalism and direct realist theories of perception, suggesting that diagonal representationalism is consistent with both. 2. Motivations and standard problems for representationalism G. E. Moore (1903) argued that experience was diaphanous. When one sees a ripe tomato one can focus one s attention on the tomato, or on its redness, but it is not clear that one can focus one s attention on the experience itself without simply focusing once again on the tomato or its redness. Representationalists take this to suggest that all there is to the experience is its content: the ripe red tomato. Whether or not this is a good argument, the relation between content and phenomenology cannot fail to be important in both the philosophy of mind and in epistemology. This relation can be investigated by treating representationalism as a hypothesis, to be rejected only if it is shown to be false. Hence I shall briefly outline some distinctions between different versions of representationalism then I shall begin examining the main objections. For materialists a further reason to try to defend representationalism arises from that fact that some versions are reductive; they hold that phenomenal character can be reduced to representational content. Reductive representationalism holds the appeal of reducing the problem of accounting for phenomenal consciousness to the far more naturalistically tractable problem of accounting for representation. Advocates of reductive representationalism include Tye and Dretske though Tye, at least, includes only what he calls abstract content in his claim. This excludes particular objects or surfaces. This restriction is needed because hallucinatory experiences and experiences of different qualitatively identical objects can all be subjectively indistinguishable while differing in the particulars represented. 4
Those who reject the reducibility of the phenomenal to the physical on independent grounds can, however, accept non-reductive versions of representationalism. Within the non-reductive camp we can distinguish two quite different sorts of view. Firstly there are those who regard phenomenal conscious as involving non-physical properties with no metaphysically necessary connection to the physical world. Let us call this dualist representationalism. According to this view there is a nomological correlation between phenomenal character and representational content. Chalmers is an advocate of non-reductive dualist representationalism of this kind, but there is also a substantial disagreement between Chalmers and Tye and Dretske over which representational contents correlate with phenomenal character. In principle, however, dualist representationalists could agree with Tye and Dretske about the correlation between phenomenal character and representational content in the actual world, disagreeing only with regard to worlds differing nomologically from the actual world. Secondly, one could advocate a non-reductive representationalism consistent with materialism by holding the weakened claim that phenomenal character supervenes on representational content. Let us call this supervenient representationalism. According to this view it is metaphysically necessary that no two experiences differ in phenomenal character without differing in representational content, but the converse would not hold. A dualist could make a weaker nomological claim in order to allow a one-many correlation between phenomenal character and content instead of a one-one correlation. Finally, one further alternative would be to restrict the correlation between phenomenal character and representational content to a population, or perhaps to an individual. This would allow the possibility of two subjects having inverted spectra relative to one another without differing in terms of representation. In 5
principle reductionists and dualists could accept restricted views, though both would have to explain the ground of the difference between different populations or individuals. I lean strongly toward a reductive version of representationalism, but for the most part I shall present my theory as a general framework that is also consistent with some of the non-reductive variants discussed above. By standard representationalism I shall mean the view that the phenomenal character of an experience depends entirely on its direct externalistically individuated representational content, not including modes of presentation or indirectly represented contents of any kind. For reasons that I hope are clear I shall call this kind of content Russellian. 2 So, for example, when one has a visual experience as of a black cat against a green background, and nothing else, the Russellian content of one s experience is black cat against green background. This content is made up entirely from the objects and properties in the world external to the subject, i.e. black, cat, green, etc. (though for theories that incorporate Tye s restriction to abstract content the Russellian content will not include specific objects). According to standard representationalism it is Russellian content alone that determines the phenomenal character of experience. It is natural to think of this as analogous to Russell s view that denotation alone determines meaning. It seems to me that many of the moves that have been made in the debate about representationalism correspond to moves that were made in the debate over linguistic meaning post-russell, substituting linguistic meaning for phenomenal character. I shall mainly use Tye s theory as an example of standard representationalism, though non-reductive versions of representationalism also count as versions of standard representationalism insofar as they deal only with Russellian content. 2 Cf. Chalmers 2004b, p. 167. 6
Now to the objections. If inverted spectra are possible then unrestricted standard representationalism is immediately in trouble. On the face of it, in order to defend a view that was not restricted to individuals one would have to say that when two individuals visual spectra were inverted relative to one another only one of them could have veridical perceptions. This is certainly counterintuitive; if the two individuals were identical in other ways then it is hard to see what could make one perception veridical rather than the other. Moreover it would be hard for a standard representationalist to defend the view that no colour experiences are veridical, given their commitment to the wholly representational nature of experience. The possibility of inverted spectra that make no functional difference is controversial; materialists (myself included) typically do not accept that they are possible. The inverted spectrum objection does not, however, require inverted spectra that make no functional difference. This is best illustrated by Ned Block s (1990, 1995) Inverted Earth thought experiment. Inverted Earth is identical to Earth in every detail except that each Inverted Earth object has the complementary colour to its duplicate on Earth. On Inverted Earth the sky is yellow, ripe tomatoes are green and ripe bananas are blue. If you are wearing a blue hat then you have an Inverted Earth doppelgänger who is wearing a yellow hat, and so on. Now, suppose you were to travel to Inverted Earth but while you travelled colourinverting prisms were placed in your eyes, or perhaps your visual system was rewired such that each colour experience became the complementary colour experience. 3 Suppose that your own bodily colours were also exchanged for their complements. When you arrived on Inverted Earth your experiences would match 3 In later versions of his argument Block (e.g. 1995) has changed some of the details of the scenario in order to avoid some possible objections. For simplicity, since I find Block s argument convincing (with regard to standard representationalism) I shall gloss over this. I 7
the experiences you would have had in the corresponding spatiotemporal location on Earth. If you had been asleep during the journey you would not even be aware of being in a new environment. If this happened then it would seem natural to say that upon waking your perceptions would be illusory; the sky would look blue to you when really it was yellow. But if you remained in your inverted state on Inverted Earth for long enough then according to Block it would no longer seem reasonable to say that you would perceive incorrectly. 4 The content of your experiences would shift just as, arguably, the meaning of a word eventually shifts when it is consistently used in the presence of a new object, unbeknown to the speaker. What it took, in terms of phenomenal character, for you to correctly perceive blue would no longer be what it took for you to perceive blue in the past. Consequently it is possible for there to be experiences that differ in Russellian representational content but not in phenomenal character. Hence phenomenal character at most supervenes on Russellian representational content. Worse still, however, there can be experiences differing in phenomenal character that do not differ in Russellian representational content. Suppose that on Earth in your normal state you saw an unripe, green tomato. Years later on Inverted Earth, where ripe tomatoes are green, you saw a ripe tomato. In both cases your experience correctly represented the colour green. But the phenomenal character differed between the two experiences because of the colour inversion of your visual apparatus in the intervening period. Hence, since Russellian shall also make some small changes of my own for ease of exposition, but as far as I can tell nothing important is changed. 4 Inverted Earth is not at all like Kohler s (1961) experiments with spatially inverting goggles. In the latter case one might argue that when subjects learn to compensate for the goggles they return to their original functional organisation and there is then a question as to whether their experiences change along with their functional organisation. On Inverted Earth, by contrast, there is no change in the subject s functional organisation apart from the change initially effected by inserting colour-inverting prisms. 8
representational content and phenomenal character can vary independently of one another, even supervenience versions of standard representationalism are false. 5 Not all objections to standard representationalism are as effective as Inverted Earth. Tye (2002, 2003) shows how a number of objections regarding blurry images, after images, hallucinations, aspect switches and other phenomena can be dealt with. In my opinion these responses are largely convincing. But Inverted Earth strikes me as fatal to standard representationalism (as does another of Block s objections, discussed below). So if representationalism is to be saved a nonstandard version is required. 3. Functionalism and representation I shall now show why anyone who accepts even a minimally functionalist theory of phenomenal consciousness should reject standard representationalism. By a minimally functionalist theory I mean any view according to which in the actual world every phenomenal state of an individual has a functional correlate within that individual (i.e. the phenomenal state occurs when and only when the same functional state occurs). In subsequent sections I shall sometimes make the slightly stronger assumption that phenomenal character and functional role correlate across all individuals in all worlds nomologically equivalent to the actual world. In 5 Those who are prepared to assume that phenomenal character correlates with physical state can reach the same conclusion by supposing instead that Inverted-Earthlings are congenitally colour-inverted relative to Earthlings because they are born with colourinverting prisms or appropriate neural cross-wiring. One can then compare Earthling and Inverted-Earthling experiences. This avoids any controversy regarding the assumption that the representational content of the transported Earthling s experiences gradually shifts over time. Michael Tye (2000, pp. 136-40) resists Block s objections by trying to make it plausible that the transported Earthling continues to misperceive. I do not find this plausible; but Tye 9
fact I believe something still stronger to be true, but this is not needed here. Minimal functionalism involves no commitment to the view that phenomenal consciousness is reducible to functional role or that a specific functional organisation is metaphysically sufficient for consciousness. Even most property dualists count as functionalists in this minimal sense. Note that the minimal functionalism under consideration says nothing about propositional attitudes. In principle one could reject a functionalist account of propositional attitudes while accepting a close connection between phenomenal character and functional role, though one would of course have to explain how the two kinds of states interact. One standard way to capture the notion of a functional role is to use a formalisation derived from the Ramsey sentence of a psychological theory that relates mental states to one another and to inputs and outputs (Lewis 1972). Consider a subject S who has a set of internal states x1 xn whose causal roles in relation to one another and to sensory inputs I1 Im and motor outputs O1 Ok are captured by a theory T: 6 S is in mental state Mi x1 xn [T(x1 xn, I1 Im, O1 Ok) S( xi )] The strongest, materialist form of functionalism requires a strict biconditional; the most minimal version requires only a material biconditional. Now, consider a phenomenal state such as a visual sensation of the kind produced by the presence of a ripe tomato in front of the subject in normal lighting conditions. The functional correlate of this state is a member, xi, of the n-tuple of would also have to say that the congenitally colour-inverted Inverted Earthlings just described would misperceive colours throughout their lives. This seems even less plausible. 6 S(xi, xj ) should be read as S is in internal states xi, xj and S( xi ) should be read as S is in a set of internal states that includes xi. T is the conjunction of all statements of the form if S(x α, x β ) and input I γ occurs, there is a probability P that motor output O λ will be produced and that S s internal states will change so that S(x µ, x ν ). 10
internal states <x1 xn>. Crudely speaking, the Ramsey-Lewis formalisation above allows us to think of the functional role of this state as involving three components: an in component, an out component and a sideways component. The in component consists of the sensory inputs that typically cause the mental state and out component concerns the motor outputs typically caused by it. The sideways component consists of the typical causal interactions between xi and the other members of <x1 xn>. This is an oversimplification because the three components are not independent of one another; in particular the motor outputs to which an internal state contributes are typically the result of interaction between several different internal states. It is not an oversimplification, however, to say that in humans, at least, there are some internal states that can be thought of as standing at the end of an afferent (upward) pathway from a perceptual organ. In such cases the typical cause of the state can be specified without making reference to the way in which the state interacts with every other internal state in the system. It is not plausible that our brains are so radically Quinean as to make the causation of each internal state by external influences depend on interactions with every other part of the system. 7 The behavioural effects of xi must normally coordinate with sensory inputs in such a way as to allow S s actions to be guided by S s perceptions, but the manner in which this is achieved does not have to be specified in order to describe the afferent pathway leading to a perceptual state xi. 7 Consider for example the pathway from the eyes to the occipital cortex. Relevant pathways might extend beyond this (see below on indeterminacy regarding where the pathway ends). For the use of Quinean above, and for arguments in support of the claims made above see Fodor 1983. One does not have to accept very much of Fodor s modularity hypothesis to accept the very weak point relied upon here, which is just that there could be internal states with the same typical external causes but different internal functional roles. I take it that this would have some plausibility even without the Fodorian arguments. 11
We can now consider the representational content of xi. In what follows I shall assume that representation has close connections with the notion of information associated with Claude Shannon s (1948) work and brought to the attention of philosophers by Fred Dretske (1981). Crudely speaking, events of type A carry information about events of type B if and only if the occurrence of events of type A covaries with the occurrence of events of type B (this typically involves events of type B causing events of type A, though it does not have to). So, for example, the sound of a smoke alarm carries information about the presence of smoke because the alarm sounds when, and only when, there is smoke in the vicinity. The sound of the alarm thus carries the representational content that smoke is present. Attempts to define representation purely in terms of covariance are problematic, but I take it to be common ground among most theories of representation that covariance is at least an important factor in determining what a state represents. This is all I really need to assume. Even theories that deny this, however, such as causal chain theories, will probably turn out to be compatible with most of what I shall claim, though I shall not pursue this here. 8 I assume, then, that each member of <x1 xn> represents whatever it covaries with. Now, insofar as xi covaries with states of affairs external to the subject it will only be the in and out components of the causal role of xi (and only part of the 8 Notoriously a full account of representation in terms of information involves more than simple correlation because, for example, the smoke alarm might be reliably triggered by any number of conditions other than the presence of smoke, but which rarely obtain in the smoke alarm s actual environment. Some theories appeal to teleology to solve this problem (Millikan 1989, Dretske 1994). I hope, however, that I shall be permitted to side step these controversies in what follows. Propositional attitudes probably involve more than mere representation, but I do not offer an account of them here. Insofar as conscious perceptual content is conceptual there may also be further requirements; I make some general remarks below about how the general framework to be developed can incorporate more specific theories on such matters. For the most part, however, my interest here is in accounting for conscious states at a level of generality that captures only what is common to both conceptual and non-conceptual organisms. 12
latter) that determine this. If xi is typically caused by input Ij then xi represents the state of affairs associated with Ij, at least provided that S s behavioural outputs match up adequately with perceptions. So, for example, if xi is typically caused by the presence of a ripe tomato in front of the subject in normal lighting conditions then this is the state of affairs that xi represents, at least provided that S s resulting actions are directed, where appropriate, toward the tomato. But this fixes only part of the functional role of xi. The sideways component of the functional role, and part of the out component, could vary independently of the in component. That is to say, the fact that xi represents the tomato leaves open a great deal with regard to the interactions of xi with other internal states and the nature of the resulting actions. Two individuals could thus have internal states that differed in functional role yet represented the same states of affairs. Presumably it is the functional role of xi as a whole, and not just the in component, that correlates with phenomenal character; consequently phenomenal character can vary independently of Russellian representational content. Hence standard representationalism is false, given minimal functionalist assumptions. 4. Perspectival representationalism Sydney Shoemaker (2000) has defended a version of representationalism that avoids standard objections such as Inverted Earth. While Shoemaker s theory faces certain difficulties I shall suggest that a modified version of it, which I call perspectival representationalism, would be a step in the right direction. Although this theory is vulnerable to a further objection due to Block it is worth briefly examining it because diagonal representationalism can be thought of as a generalisation of it. 13
Shoemaker holds that experiences represent phenomenal properties of objects. A phenomenal property is a disposition to produce an experience of a certain kind in a certain kind of subject. Thus the representational content of a visual experience of a ripe tomato includes the property that the tomato has of being disposed to produce a certain sensation, phenomenal redness, in a certain kind of subject. But the experience of a subject whose visual spectrum is inverted represents a different phenomenal property such as the disposition that the tomato has to produce a different sensation, phenomenal greenness, in a subject of that particular kind. 9 An objection that has been raised to this type of account is that it suggests that colour experiences (for example) attribute properties other than colours to coloured objects (see Chalmers 2004b). An account in the same spirit as Shoemaker s can, however, avoid the problem. Consider Michael Tye s (2002) fairly widely accepted rebuttal of an objection to representationalism raised by Christopher Peacocke. Peacocke (1983) had noted that two qualitatively identical trees seen at different distances are represented as having identical properties yet the experiences of the two trees differ in phenomenal character. Tye points out that while the intrinsic properties of the trees are represented as being the same the trees are nonetheless represented as standing in different spatial relations to the subject. This explains the difference in the phenomenal characters of the two experiences. Tye s claim does not require the subject to be explicitly represented in the experience. The trees may be presented as possessing properties such as near and far which can be ascribed using one-place predicates even though they express two-place relations (the trees are near or far from the subject or the subject s location). 9 Andy Egan (forthcoming) puts forward a view that modifies Shoemaker s proposal by including an indexical element. His account also has a certain amount in common with Chalmers (2004) account (on which see below), though it does not involve modes of presentation. Egan s theory can be thought of as a version of perspectival representationalism, and seems to me to be subject to the problem described below. 14
This allows the subject to be an unarticulated constituent of the content of the experience. 10 The problem with Shoemaker s account is clear in spatial cases. One would have to say, about Peacocke s example, that the relevant phenomenal properties of the two trees were dispositions to produce certain experiences in certain subjects (those at certain relative locations). Thus one would not perceive the size of a tree but merely a phenomenal property that was systematically related to the size of the tree. This seems rather unintuitive. Shoemaker might perhaps argue that the subject would perceive the size of the tree by virtue of perceiving the relevant phenomenal property. But Tye s account of spatial perception, while closely related to Shoemaker s, seems better; the subject perceives the size of the tree but also perceives the spatial relation between the tree and the subject. Both components of the representational content combine to produce an experience with a certain phenomenal character; if either component were to change then the phenomenal character of the experience would change in a corresponding way. The principle behind Tye s view of spatial experience is that the world is viewed from a spatial perspective and this perspective itself contributes a component to the representational content of an experience. This principle could be generalised beyond the spatial case. Each subject has a set of causal powers including the subject s muscular powers and the causal powers of the set of internal (brain) states <x1 xn>. Think of this as locating the subject in a causal space whose axes measure various causal parameters (the precise details of this do not matter; I only intend it as an rough intuitive aid). Each perceptible property also has a location in causal space by virtue of the causal powers it bestows on an 10 For more details of the notion of the subject as an unarticulated constituent in spatial or temporal content see Prosser 2005 and forthcoming. Shoemaker (1994) endorses a similar idea with respect to predicates like heavy, as does John Campbell (1993, 1994). For general 15
object. To distinguish one property from another is to distinguish its location in causal space from those of other properties. An object with several causal properties is thus located along several dimensions of causal space. For simplicity let us consider just one property of an object. A subject whose visual experience represents an object as being red in effect represents the object as located in a certain position in causal space. But the subject also has a location in causal space and hence stands in a certain relation to the object within causal space. We can therefore generalise Tye s spatial account to say that the subject s experience of an object represents both the absolute position of the object within causal space and the relation in which it stands to the subject within causal space. The resulting view, perspectival representationalism, avoids many standard objections to representationalism including Inverted Earth. Consider an Earthling and an Inverted Earthling both observing a red object. I shall suppose from here on that Inverted Earthlings are congenitally colour-inverted relative to Earthlings due to being born with appropriate cross-wiring in their optic nerves. The Earthling and the Inverted Earthling stand in different positions in causal space because of the cross-wiring in the Inverted Earthling. Consequently, although the visual experiences of both subjects correctly represent the location of the red object in causal space the representational contents of their experiences also include their own relations to the object in causal space. These relations differ, and it is because of this difference in representational content that the phenomenal characters of the two experiences of the same red object differ. The situation is analogous to that of two people looking at the same tree from different positions in space. Similarly in the converse situation in which an Earthling looks at the sky on Earth and the Inverted Earthling looks at the sky on Inverted Earth, both beings have experiences apparatus for dealing with unarticulated constituents in thought as well as language see Perry (1986). 16
with the same phenomenal character despite different colours of sky being represented. This is because each of them stands in the same relation to their respective skies even though the absolute positions of the two skies and the two subjects in causal space are different. A further case helps illustrate the advantages of perspectival representationalism over standard representationalism. Consider a person called Maxi looking at a tree twice their height and at a distance equal to three times their height. Consider also a miniaturised doppelganger of Maxi in a miniaturised environment. Call this person Minnie. Minnie is looking at a tree twice Minnie s height and at a distance equal to three times Minnie s height. Now, how do Maxi and Minnie s visual experiences of their respective trees compare? And how would Maxi and Minnie perceive the same object? Would it seem bigger to Minnie than it would seem to Maxi, or would it seem the same size? Tye s view implies that if Maxi and Minnie s perceptions were veridical then if Maxi and Minnie perceived the same object at the same distance it would look the same because sameness of representational content yields sameness of phenomenal character. It would follow that Minnie s miniaturised tree looks smaller to Minnie than Maxi s normal-sized tree looks to Maxi. Yet this seems very counterintuitive. One would think that being miniaturised would make small things seem big and close things seem far away. 11 I think this presents a serious problem for Tye s response to Peacocke, as it stands, and consequently for Tye s view as a whole. One must, however, be a little careful here. Simply scaling everything down would not make Minnie s environment seem the same to Minnie in every respect 11 Hollywood seems to agree. See for example the 1957 film The Incredible Shrinking Man or the 1966 film Fantastic Voyage in which a submarine, complete with crew, is miniaturised and injected into the blood stream of a human being. These illustrate the common intuitions, whether or not those intuitions are true. 17
as Maxi s environment seems to Maxi. At a smaller scale things behave differently; some insects can carry droplets of water, or stand on the surface of a pond without getting wet. This is because of a difference in the causal powers of the environment relative to those of the creatures in question. When we think of Minnie being miniaturised we tend to think of Minnie s causal powers being miniaturised in a corresponding way. Something that is light for Maxi to lift is thus heavy for Minnie to lift. Similarly, something that is within easy reach for Maxi is a long walk away for Minnie. This has to do with causal powers rather than size per se. Consequently how far away something looks, or how big it looks, depends not only on the objective spatial distances involved but also on the subject s causal powers (by objective distance I mean distance as measured in terms of physical constants such as the speed of light). So when Maxi and Minnie look at the same tree from the same distance it looks different to them because it is small and close for Maxi (i.e. relative to Maxi s causal powers) while being big and far away for Minnie. As noted above, though, Maxi and Minnie need not be explicitly represented in their own perceptions in order to be constituents of the representational contents of their perceptions. What all this shows is that perspectival representationalism, which includes the subject s causal powers in the contents of experience, is better able to deal with the phenomenal character of an experience with a spatial component than all versions of standard representationalism including Tye s. The subject s behavioural dispositions are related to the represented causal perspective. The bitter taste of a natural poison includes a representation of the subject s causal relation to the poison; it is appropriate for the subject to avoid ingesting bitter tasting substances precisely because of the causal relation in which these substances stand to the subject. When one enjoys the taste of an artificial substance that is detrimental to one s health one has, in effect, misperceived its 18
causal relation to one s body; one s senses are fooled because they evolved in an environment in which such substances did not occur. A similar story may be true of the perception of colours and combinations of colours, whose natural role in indicating danger, nutrition and other important causal relations is well known. Despite the advantages of perspectival representationalism, however, it is vulnerable to an objection due to Ned Block (1995, p. 25). Block claims that there can be phenomenal character yet no representational content whatsoever. This, if correct, is fatal for both standard and perspectival representationalism even in their weaker forms. Block s scenario combines Davidson s (1987) swampman with the brain in a vat whose entire existence has been spent in the vat. Davidson held that the swampman, created instantaneously by chance from stray molecules, lacked representational content at the instant of creation because of its lack of causal relations to its environment. But this would only be correct given a causal chain theory of content; many other theories would allow that the swampman has representational content because the swampman s internal states could track states of its environment counterfactually. 12 Brains in vats also have representational contents. Most theories of content allow that the internal states of a brain in a vat represent features of the computer environment to which it is connected. But Block invites us to consider a brain created by chance, à la swampman, in a bathtub and not connected to a computer or anything else. Assuming that phenomenal states at least supervene on brain states the brain should have some kind of experience for a few fleeting moments. Yet since it is not and has never been connected to any perceptual apparatus its 12 By analogy, consider radiation alarms in a nuclear power stations. As a guard against malfunction these are designed to emit a regular tone and to fall silent only when there is radiation present. The tone carries the information that there is no radiation present even if the alarm has never been subjected to radiation. 19
internal states do not track states of its environment and consequently represent nothing. 5. Stalnaker s two-dimensional framework I must now digress a little in order to explain some relevant features of twodimensional modal semantics of the kind developed by Robert Stalnaker (1978, 2001, 2004). In this section I shall outline some important features of Stalnaker s theory as it applies to linguistic utterances and to beliefs. Then in the next section I shall show how the same apparatus can be applied to conscious experience. The resulting theory is what I call diagonal representationalism. Let W1 be the actual world, in which Le Corbusier was an architect. Let W2 be a world in which words mean what they do in the actual world but Le Corbusier became a comedian in a double act with Stan Laurel. Let W3 be just like the actual world except that in W3 Le Corbusier names Stan Laurel and architect means comedian. For simplicity, let those be all the worlds. The horizontal proposition expressed by an utterance is the set of worlds in which the utterance would be true given the meaning it has in the world in which it is uttered. This can be thought of as what is said by the utterance. So, for example, when Le Corbusier was an architect is uttered in W1 it is true. When uttered in W1 but evaluated relative to W2 it is false. When uttered in W1 but evaluated relative to W3 it is true because in W3 the very same person, Le Corbusier (or a modal counterpart if you prefer), was an architect despite not being named Le Corbusier. The horizontal proposition expressed by the utterance in W1 is therefore {W1, W3}, as represented by the two T (true) symbols in the first horizontal line of table 1. When the utterance occurs in W2 the same horizontal proposition is expressed but the utterance is false 20
relative to its own world. When the utterance occurs in W3, however, it expresses the same horizontal proposition that would be expressed by an utterance of Stan Laurel was a comedian in W1. In our example this is the necessary proposition {W1, W2, W3}, though the necessity comes about because we have considered only three worlds. World of evaluation W1 W2 W3 World of W1 T F T utterance W2 T F T W3 T T T Table 1 One of Stalnaker s key insights was to recognise the importance of the diagonal proposition. This is the set of worlds that we get if we look down the diagonal of the matrix of truth values from top-left to bottom-right and include all the worlds against which there is a T. The diagonal proposition for our example is therefore {W1, W3}. This is the set of ways that the world could be, given the assumption that the utterance is true as evaluated relative to the world in which it occurs. Given a larger set of worlds this would normally differ from any of the horizontal propositions associated with the utterance. Suppose that you hear an utterance of Le Corbusier was an architect but do not know who Le Corbusier was or what an architect is. There are many ways the world might be, for all you know. But if you 21
know that the utterance is true then you narrow down the possibilities. You might, for example, be in a world in which Le Corbusier refers to Stan Laurel, architect refers to a designer of buildings and Stan Laurel designed buildings but you cannot be in a world in which Le Corbusier refers to Le Corbusier, architect means comedian but Le Corbusier was not a comedian. In many cases, because we are not omniscient, the diagonal proposition captures a psychologically important component of the content of our utterances or beliefs. I shall suggest below that it also determines the phenomenal character of conscious experiences. Now consider utterances of Le Corbusier was an architect and Charles- Edouard Jeanneret was an architect. As a matter of fact Charles-Edouard Jeanneret was Le Corbusier s real name. What is said by both utterances is the same; they both express the same horizontal proposition. But because there are worlds in which the two names do not refer to the same person the diagonal propositions associated with the two utterances differ. The set of ways the world might be from the point of view of someone who knows only the truth of Le Corbusier was an architect is different from the set of ways the world might be from the point of view of someone who knows only the truth of Charles-Edouard Jeanneret was an architect. This explains why the two utterances differ in cognitive significance for someone who does not know that Le Corbusier was Charles-Edouard Jeanneret. Although we are sometimes aware of entertaining explicitly metalinguistic thoughts the explanation of differences of cognitive significance just given does not require this. Consider token beliefs, to which the same account applies. If one believes Le Corbusier was an architect the relevant diagonal proposition is that one s belief is true. But one need not have higher-order beliefs of which one is aware. One s belief concerns Le Corbusier, not the name Le Corbusier ; but this is consistent with the claim that the psychological role of one s belief depends on a 22
diagonal proposition that one only tacitly accepts (or rejects, if one disbelieves the horizontal proposition). The difference between ascribing architect-hood to someone qua Le Corbusier and ascribing it to them qua Charles-Edouard Jeanneret makes no difference to the horizontal proposition that one believes. It does, however, constitute a difference in the way in which one narrows down one s overall epistemic possibilities because there are worlds in which one belief token would be true while the other would be false due to a difference in the references of the names. The mere difference in the names themselves is all that is needed for this. 13 Another key feature of Stalnaker s framework is that in individuating an utterance across worlds no restriction need be placed on the way in which the truth value of the utterance, or the references of the terms that comprise it, are determined within a given world. Suppose that the extension of a term is determined by a descriptive condition (we need make no commitment to descriptive conditions; the assumption is just to illustrate a point). Suppose, for example, that in some worlds the extension architect is determined such that architects are designers of buildings whereas in other worlds its extension is determined such that architects are humorous entertainers. Stalnaker s account allows that utterances of architect in all these worlds can count as utterances of the same term. Only the word itself is held constant. This metasemantic account, as Stalnaker describes it, differs from the semantic accounts of other twodimensionalists including Chalmers. 14 Chalmers, as we shall see, individuates one 13 A more complex account may be needed when there are different tokens of the same generic name referring to different individuals. One way to deal with this would be to treat the tokens as tokens of different names that happen to be orthographically or phonetically identical. For elaboration of this idea see Kaplan 1990. 14 See Stalnaker 2001, 2004 for the semantic/metasemantic distinction. A similar point is made in Stalnaker 1999, pp. 14-16. 23
dimension of content in terms of primary intensions, which can include descriptive contents. Although Stalnaker says relatively little about empty names or other empty terms his theory provides some useful apparatus for dealing with such cases. 15 This will be important when we come to consider hallucinatory conscious experiences. Let us use the symbol to mean no truth value. Consider another set of three worlds W1, W2, and W3, in which the same utterance occurs. Suppose that in world W1 a component of the utterance lacks a reference. When the utterance occurs in W1 it is neither true nor false when evaluated relative to W1 or any other world. The horizontal proposition expressed is the empty set. The set of worlds in which the utterance s negation is true is also the empty set; this distinguishes the utterance from one that expresses a necessary falsehood. The matrix of truth values is shown in table 2. World of evaluation W1 W2 W3 World of W1 utterance W2 T T F W3 F T F Table 2 15 Stalnaker (1999, pp. 92-5) does briefly discuss empty names in a discussion of negative existentials. Negative existential statements may, however, be a special case. It might be held, for example, that negative existential statements containing empty names are true (Stalnaker appears to hold this view). But it is not clear that other statements containing empty names would thereby have truth values. 24
Although an utterance in W1 expresses no horizontal content there is, however, a diagonal proposition provided that the utterance could have been true. If the term that fails to refer in W1 could have referred then the utterance would have been true or false in at least one world, had it occurred in that world. This is again illustrated in table 2; the utterance expresses a truth about W2 when it occurs in W2 and a falsehood about W3 when it occurs in W3. If these were all the worlds then the diagonal proposition would be {W2}. When, in W1, one acts on the belief state expressible by the utterance one manifests a tacit acceptance that the utterance is true. Hence it is a tacit acceptance of the diagonal proposition that accounts for one s actions. One acts, mistakenly, in a way that would be appropriate in any of the worlds in the diagonal proposition but without needing any further commitment to which of these worlds one inhabits. There are, however, some subtleties to be noted, especially when the twodimensional framework is applied to thoughts or, as I intend, conscious experiences. For all I have said it may appear that the diagonal proposition constitutes a kind of narrow content that is available regardless of the layout of the world external to the subject, in contrast to the externalistically individuated horizontal proposition. Notice, however, that the diagonal proposition is a set of worlds, just like the horizontal proposition. So the diagonal proposition really does capture a content in the same sense as the horizontal proposition; they are not different kinds of proposition but merely different propositions. I shall say a little more about narrow content and externalism below. Certain philosophers who hold radically externalist will reject the assumption that content that is available regardless of the existence or non-existence of objects external to the subject can be adequate for the purposes of psychological explanation. They may, for example, hold that psychological explanation makes reference to object-dependent de re senses that comprise object-dependent Fregean 25