In the theory of consciousness, the term “representationalism” has, aptly but somewhat confusingly from a historical point of view, come to designate any view according to which being phenomenally conscious is equivalent to representing the right sort of things in the right sort of way. There is, of course, much internecine disagreement over these things and ways, but the main idea is simple and attractive enough. If we could understand consciousness in terms of representation and representation in terms of some naturalistically acceptable relations, then we could “naturalize” consciousness. I’ll call representationalisms that are coupled with naturalistic theories of content reductive representationalisms.
Representationalisms are typically divided up into various “orders.” These orders have, in a way, to do with the kind of content (or object) a conscious representational state supposedly must have. For First-Order (F) representationalisms the relevant states are, fundamentally, just directed at worldly objects and properties (typically the sensible properties of tables, chairs, etc., see, e.g., Tye 1995, 2000; Dretske 1995). For Higher-Order (H) representationalisms, the states must be directed at mental states of “lower-order”—possibly but not necessarily first-order (see e.g., Rosenthal 2005; Lycan 1996). For Same-Order (S) representationalisms, the representational state must be directed at itself (or, perhaps, some part of itself, or a whole of which it is a part, or another part of the whole of which it is a part).[6] I also add Privileged-Object (P) representationalisms as a distinct category. For these, the state must be directed at some special type of entity—a model of the organism as a representational or embodied homeostatic system, a “proto-self” or, less naturalistically, perhaps an enduring substantial ego entity.[7]
There is, however, no obvious reason why there could not be unconscious representations with any of these contents. And, generally, it seems implausible that something could be conscious in virtue of representing a certain type of object—this is Alvin Goldman’s so-called “Problem of the Rock” (thinking about or seeing rocks does not make them conscious, so why should it make anything else conscious?), which seems to apply to H, S, and P theories—but see below (see e.g., Goldman 1993; Gennaro 2005; Lycan ms).
For F theories, since it is admitted there can be conscious and unconscious states with the same sort of content, another distinguisher between conscious and unconscious mental states will have to be found. For F theorists, this has typically been a functional constraint placed on the representations (e.g., poise, feeding into the mind-reading system, becoming available to the global workspace, see, e.g., Tye 2000 and relatedly Baars 1997; Dehaene & Naccache 2001), sometimes coupled with the necessary condition that the properties represented must be represented in a “non-conceptual” way (whatever that is taken to amount to).[8] For the H theorists, it has been a somewhat different story.
H theorists are generally motivated by a phenomenological inadequacy they see in F theory. F theorists generally stress the so-called “Transparency Intuition”—the idea, roughly put, that first-order consciousness reveals only properties and objects in the world and nothing directly about consciousness itself, the perceiving mind, the subject, or the vehicles of representation (see e.g., Harman 1990; Tye 2000; Byrne 2001). H theorists, on the other hand, are with varying degrees of explicitness motivated by the equally powerful intuition that consciousness involves some sort of “for-me-ness” or “to-me-ness,” often termed “subjective character” (see e.g., Rosenthal 1986, p. 345 and Gennaro 2006. See also Levine 2001, pp. 104–111). This gets encoded in the H mantra that the conscious states are just those that one is “Aware of Being In”, those that one is aware that one is oneself in (see e.g., the “Introduction” to Kriegel & Williford 2006). The thought is that F theory simply does not capture that intuition. F theorists and their fellow travelers would consider such “essentially indexical” contents or the “sense of self” to be more advanced cognitive products or artifacts of social cognition, certainly not in the very ground floor of consciousness (see e.g., Edelman & Tononi 2000, pp. 103–104 and Macphail 1998, pp. 2–5).
There is here an important bifurcation in intuitions about consciousness. Some significant percentage of us thinks that subjective character (however we ultimately understand it) is essential to consciousness, is in the ground floor. And some significant percentage of us thinks that it is not; that somehow qualitative character (perhaps understood as having the right sort of representational content) is essential but that subjective character is derived, secondary, or tertiary. This bifurcation shows up in neuroscientific and psychological thinking on consciousness as well.[9] We will briefly return to the significance of this bifurcation point in the next section.
The H theorist has a few options about the exact content of the H representation, the higher-order thought (or perception [or global state]).[10] There are serious and well-known problems here. If the represented lower-order state (L state) of, say, visual perceptual awareness were different from the representing H state in terms of relevant content (e.g., if the one represents a phenomenally green ball and the other a phenomenally red one), what would we consciously see? “Red. No, green. No, red…” This is the Problem of the Division of Phenomenal Labor, or mirepresentation problem, as I will sometimes call it, and is related to deep and probably insoluble problems about the epistemology of introspection that are pertinent to such models (of both H and S varieties).[11] If the L state simply did not exist, would your conscious experience in that case be a sort of Meinongian hallucination? This is the Problem of Targetless H States.[12]
To take up the latter problem just a bit, if one takes literally much of the talk one finds in the literature on H theory, the H thought is supposed to make the L state conscious. Being conscious is a kind of extrinsic (external relational) property of the L state, a property it has in virtue of its being represented by the H state. Thus, if there is such an H state, it does confer at least a relational property (the property of “being made conscious by the H state”) on the L state. In the cases in which the L state does not exist but the H state directed at it does, some non-existent object, the L state, is made conscious by an H state. Thus the L state would literally have a relational property; it would stand in a relation, even though it does not exist. This literal interpretation of the view entails some form of Meinongianism (at least about non-existent L states) and that you can seem to yourself to be conscious when you are not. Thus, presumably, it should not be taken so literally.[13]
The non-literal interpretation, however, is inimical to one of the reductive pretensions of the H strategy. It’s not inimical to reductive representationalism as such. But it does draw in to question the idea that a reductive theory must construe the property of being conscious as an external-relational property of otherwise unconscious mental states (see Rosenthal 1997). Thus, it could only be in virtue of the specific content or structure of the H state itself that there is consciousness. One would then be putting forth the presumably phenomenologically motivated a posteriori identity hypothesis that the conscious representational states are just the ones with that content. There may be differences over the specific content (e.g., Is it about some of my other mental states, or is it just about the non-mental objects and properties of the world?) and differences over other criteria (e.g., poise); but otherwise, on the non-literal interpretation, H theory is structurally just like F theory. We can of course wed either of these to a reductive theory of representation, but this will only make “being conscious” into an external-relational property to the extent that the theory of representation adopted makes all representation an external-relational matter.
If one is still conscious when the L state does not exist, then the H state would seem to be doing all the work. And that’s what we should focus our explanatory efforts on. What could be special about it? Again, putting aside other types of external relations (e.g., being available to the global workspace), it must have a special sort of content. But it is not in virtue of being represented that a state could be conscious. Rather, on this non-literal interpretation, it is in virtue of being a representation of X (where X is a special object of some sort, e.g., oneself being in a state) or that p (where p is a proposition with a special content) that the state is conscious; and we can, as with any other sort of contentful state, try to figure out how different naturalistic theories of representation would construe states with that content.
Whatever theory of content we adopt, we’ll want to know what salient or interesting properties, from an explanatory point of view, such representations have. What is it about you that you can represent yourself as being in a state or that a conscious state of yours is occurring now? Find that out, the promise goes, and we will understand consciousness. But, I would argue, none of the theories of representation we have to go on tell us anything very significant about such states. The beaver’s tail splash, says Millikan, to take one sort of example, can represent the very time at which it occurs (among other things; Millikan 1995, p. 98). This does not make it conscious. This particular example applies directly to Same-Order theories, but surely the beaver’s tail splash could have represented a previous tail splash and its content or its simultaneous front paw splash, etc., but that would not in itself make anything conscious either, right?
Naturalistic theories of representation will not themselves tell us anything that interestingly distinguishes H states (or S states) from F states (or P states for that matter). In every case (F, H, S, P), it is just a matter of some physical representational vehicles standing in some set of external (or externally mediated) relations to other physical objects (and sometimes to themselves). From this point of view, we see nothing that interestingly distinguishes the theories.
Moved by these problems, H theorists might try to go the “essential indexical” route (cf. Weisberg 2012). After all, on Rosenthal’s original formulation, the conscious states are those one is aware of oneself as being in. But here they are faced with a difficult choice. If they presuppose a teleosemantic theory, then they have to face the fact that on this theory there are no literally essential indexicals (see e.g., Millikan 1990). Change the relevant history and other external relations and you change the content—now an indexical, now a proper name, now a substance term, etc. If they abandon teleosemantics, they could go back down some Fregean rabbit hole.[14] That way lies murk or perhaps triviality (see Cappelen & Dever 2013). But it seems inadequate just to postulate that the H state contains a definite description that happens to pick oneself out. Thus the H theorist might be led to consider what is in effect a P theory. One then tries to find a suitable entity to play the role of the privileged object (a privileged signified, if you will): the proto-self, the self-model, or what have you.[15]
It is hard to see how any of these possible objects would somehow help us to make sense of subjective character. And it is hard to see how representing some special object could be that in virtue of which something is conscious. If “essentially indexical” content is either explicable in terms of something more basic (as seems to be the case to me), or impossible (as on teleosemantics), or metaphysically fraught in an ultimately un-illuminating way, then it seems like the best bet is to adopt a version of S theory.
For one thing, we can reduce the metaphysical load that threatens to plague the notion of essential indexicality and solve the non-existence problem at once.[16] All we need are token mental states representing themselves. As a corollary, we can give a deflationary account of “essentially indexical” content in token-reflexive terms[17] that is potentially compatible with teleosemantics (or whatever non-Fregean account one prefers) and find some other way to capture the grain of truth reflected in the opacity arguments presented by Castañeda, Lewis, and Perry.[18] In my view, anyone committed to the intuition motivating H theory should become an S theorist, if for no other reason than because of the non-existent L state problem. The other possible solutions (e.g., Gennaro’s “WIV”) introduce a kind of theoretical inelegance that renders them less plausible.
H theories are better than F theories, given my intuitions anyway, because they encode the essentiality of subjective character to consciousness. If that intuition is good then, of the two classes, H theories are the better ones. But H theories face the non-existent L state problem. To solve it, they must either embrace murk or metaphysical baggage (if they go in the direction of some P theories), or embrace the postulation of certain epicycles, or go same order. S, in my view, is evidently the best option for the representationalist.
S theory avoids ad hoc moves, better reflects the clarified phenomenological intuitions that are the real motivation, can ground a theory of indexicals, and does not commit one to an enduring self-entity of any sort; nor does it seem to attempt to get subjective character out of something’s representation of something else that is structurally similar to itself, as this last move runs afoul of the Fichte-Shoemaker Regress.[19] S theory evidently does not fall prey to the non-existent L state problem, even if it does not avoid the misrepresentation problem. In the end, however, it is itself nothing more than a type of P theory. The Privileged Object is just the token mental state (or episode) itself. Clearly, there is no self-evident reason why something’s representing itself should make it conscious, even if it is in fact true that all conscious episodes do represent themselves.
We surely cannot seriously imagine that consciousness emanates from a special object it needs to look at, even if that object is just the current experiential time-slice itself. Further, something’s representation of itself, naturalistically understood, is no more theoretically interesting (or even surprising) than its representation of the world or of one’s other thoughts and perceptions. Thus, that does not, a priori, appear to be the sort of thing that would be more likely to be equivalent to consciousness than something’s representation of something else. Perhaps adding functional constraints would help here but no more than it might help H or F theory.
Even if it is true that all conscious states are self-representational, it is, of course, far from clear how that fact should help us explain consciousness. The same can be said for H theory and other P theories. Rather, in all these accounts, we are merely trying to isolate what we take the unique content of consciousness to be and then to apply our theory of representation to states with such content. Absent some strong phenomenological intuitions to the contrary, the conscious mental states, it seems, might well have been all and only those states in which dogs are represented. In the end, though, all “normal” physicalists (i.e., those who reject Russellian Monism, Panpsychism, and Pan-proto-psychism) are reduced to some such strategy. All “normal” physicalists, representationalist or not, will identify consciousness with something that is not a priori known to be equivalent to it. We return briefly to this familiar problematic at the end.
But, perhaps most alarmingly, reductive S theories (and H theories, and everything in between) are either subject to a version of the old Swampman objection or otherwise untenable.[20] Since the conscious states are, on the theory, just special representational states, they are subject to the constraints of the underlying theory of representation (in this case, teleosemantics). If they don’t have the right history, then they don’t have the right content. And if they don’t have the right content, they are not conscious. Surely there is something simply absurd about the idea that one might or might not be conscious depending on how one’s atoms happened to get into the current arrangement. It is not that one cannot concoct a response to the objection; it is, rather, just the very fact that the view invites such objections in the first place. It demands a rather serious and ugly epicycle; and that counts strongly against it. But if we reject teleosemantics and adopt an internalist theory so as to escape from Swampman, we face equally difficult problems that we cannot, unfortunately, go into here.[21]
The view then is that H theories are better than F theories on phenomenological grounds and that S theories are better than H theories on dialectical and phenomenological grounds. But all versions wedded to naturalistic historico-externalist theories of representation are shipwrecked on the Swampman problem, and internalist versions face other equally difficult problems. What then shall we do?
We might consider trying out a non-reductive representationalist version of S theory. This is a possibility we will return to in section 4. But first we need to reflect a bit on what H, S, and P theories are trying to capture in the first place. What is the phenomenological datum designated by this phrase “subjective character,” and why is it that F (and related) theorists don't see it as essential to consciousness, while H, S, and some other theorists do?