In the last of his three Royce Lectures called "Self‑Knowledge and 'Inner Sense'", Sydney Shoemaker attempts to reconcile two commitments: (1) that experiences have "qualia", nonrepresentational features that constitute what it is like to have the experiences, and (2) that perceptual experiences seem "diaphanous", yielding to introspection only the way they represent the environment, not intrinsic or otherwise nonrepresentational qualia. On the idea that we internally sense qualia--that we sense what our experiences are like--one way to explain apparent diaphanousness is to maintain that these qualia are mistakenly "projected" onto the environment, that in perception we erroneously sense qualia as belonging to environmental objects. Shoemaker rejects both the projection view and the existence of inner sense, and develops an alternative reconciliation. I will describe reasons to doubt his positive proposal, and ways to save projection and inner sense from his criticisms.
I Perceptual content
Shoemaker develops his reconciliation of qualia and diaphanousness in most detail for the case of perceiving color, although the solution is supposed to apply more widely. Suppose a perceptual experience represents an apple as red. To satisfy the commitment to qualia, Shoemaker maintains that the experience has a color‑related quale, call it "R". To avoid convicting perceptual experience of widespread error, he denies that the perceiver projects R‑ness onto the apple. In fact, to satisfy the commitment to diaphanousness, he maintains that the perceiver does not "in the first instance" introspect the R‑ness of the experience at all. Instead, on his view the perceiver correctly perceives the apple as having a different property R*, roughly, the property of causing an R‑experience.<1> He calls such causal properties "phenomenal properties", to be distinguished both from qualia and from familiar secondary or primary sensory qualities (e.g., color and shape). Since the term "phenomenal properties" is often used for qualia themselves, I will refer to R* and the like instead as "phenomenal‑causal relations". Shoemaker's term seems designed to highlight the fact that we perceive these relations as monadic properties: "R* ... is not represented as a relational property, and no reference to R enters into the content of the experience" (1996, p. 256).
This is an ingenious reconciliation, and it may turn out to be correct. But there are difficulties. Let's begin with some questions about how perceptual experiences could come to represent phenomenal‑causal relations. The idea that we perceive these relations coheres poorly with more general naturalist theories of (referential) content in philosophy of mind.<2> For example, on causal theories, perceptual states represent certain of their normal or standard causes.<3> Perception of phenomenal‑causal relations doesn't fit with causal theories, since no (instantiations of) phenomenal‑causal relations help cause relevant experiences: R‑experiences are not caused by environmental objects' having the property of causing R‑experiences. So on a causal theory of perception, it is mysterious how an experience could represent the property of causing it.
Sometimes causal theories of referential content are meant to apply only to semantically simple representations, with semantically complex representations deriving their content from the simples they are structured out of, not from the conditions that (normally or standardly) cause them as wholes. For instance, a complex perceptual representation of golden mountains can inherit this content from the causation‑based content of simpler perceptions of gold and of mountains, independently of whether golden mountains ever (normally or standardly) trigger this complex perception. If the perceptual representation of R* is a complex of simpler perceptual representations, there might be a similarly indirect explanation of its content. But Shoemaker denies that perceptions of phenomenal‑causal relations are structured out of representations of causation, of qualia, and of experience. In the absence of any suggestion from Shoemaker about how these perceptions might be semantically structured, it seems that a plausible causal theory of perceptual content should apply to them directly. But no causal theory can yield the result that they represent phenomenal‑causal relations.
Perception of phenomenal‑causal relations might seem to be explained by a theory of representation other than a causal theory, such as a correlational theory, according to which perceptual states represent certain conditions that they normally or standardly correlate with. Leaving aside cases of hallucination, perceptual experiences correlate perfectly with (instantiations of) phenomenal‑causal relations: given that there is a perceptual R‑experience, some environmental object has the property of causing an R‑experience, and vice versa. The problem is that this correlation has nothing specifically to do with R's being a quale. For each feature F of a perceptual experience, some environmental object has the property of causing an F‑experience. F could be the general property of being realized in molecules, or the general property of being a perception, or the more specific property of being realized in such‑and‑such molecules, or the relational property of occurring in New Orleans, or even the property of being caused by an environmental object that causes an R‑experience. Each type of perceptual experience correlates perfectly with the presence of an environmental object causing that type of experience. Why would an experience with all these properties, as well as quale R, represent specifically the property of causing an R‑experience? Why would an experience represent phenomenal‑causal relations rather than a multitude of nonphenomenal‑causal relations? What singles out R as the representation‑generating feature of the experience?
A natural suggestion is that since one does not internally detect an experience's being realized in molecules or its being in New Orleans, one is not in a position externally to detect the property of causing a molecule‑based or New-Orleans-located experience. This would provide the needed contrast with qualia if R, as opposed to these nonphenomenal features of the experience, were itself first detected (e.g., in inner sense, or in some other form of introspection). In part by representing an experience's R‑ness, one could represent an environmental object's causing an R‑experience. But this explanation is precisely what Shoemaker rules out in order to satisfy his commitment to diaphanousness, to the idea that "in the first instance" we introspect only representational features of experience, and not qualia (on his conception). Shoemaker concludes that "we know the qualia themselves only 'by description' ... as those properties experiences must have if external objects are to have [phenomenal‑causal relations]". His idea seems to be that one can introspect that one's experience represents an apple as R*, and then, if one "reflects on the matter" (1996, p. 255), one can figure out that one's experience must have R for it to represent the apple as having R*. So for Shoemaker qualia are hidden from direct introspection, just as molecules and New Orleans are. The problem remains of why phenomenal‑causal relations, as opposed to nonphenomenal‑causal relations, would be represented in perception.
This is not a problem Shoemaker can avoid merely by rejecting a correlational theory of perceptual content. He would need some positive suggestions about content that would result in the representation of phenomenal‑causal relations, without presupposing representation of qualia. And at first few glances this seems mysterious. We should wonder: how can one perceive (instantiation of) a causal relation without perceiving--or in any other way representing--the relevant effect? I have no argument that this is impossible, but it would be helpful to know whether Shoemaker's suggestion commits us to a sui generis form of representation. If it does, that is a potential drawback, a complexity that a rival theory could exploit.
The other main gap in Shoemaker's account concerns how one is supposed to reach knowledge of qualia "by description". By what steps should one conclude that one's experience has quale R, starting from the introspective premise that the experience represents an apple's R*‑ness? If one could perceive explicitly that an apple is causing an R‑experience--if one could perceive the apple as causing an R‑experience--then one might have a fighting chance of performing the needed inference. But as mentioned above, Shoemaker concedes that "no reference to R enters into the content of the experience" (1996, p. 256). Presumably, then, no reference to R is made in the content of the introspections of the experience's content. So how is one to infer "by reflecting on the matter" anything about the R‑ness of one's experience?
Even if perception did explicitly represent R* as a relation to R experiences, it would still be mysterious how one comes to represent R. Suppose one could introspect that an experience represents an apple as causing an R‑experience. This tells one nothing about the accuracy of the experience: perhaps the experience misrepresents the apple as causing an R‑experience, when instead the apple causes a Q‑experience.<4> It also tells one nothing about the source of the experience's content. Suppose Shoemaker is right (though for reasons I do not yet see), that R‑experiences represent things as causing R‑experiences. Even so, why should one assume that only R‑experiences represent things as causing R‑experiences? Yet it is the latter, extremely incautious, assumption that one would need to infer R‑ness from R*‑representation.<5>
These challenges should be enough to warrant another look at the view Shoemaker rejects, according to which we internally sense features of experience and mistakenly project them onto environmental objects. Shoemaker makes this task seem very daunting. Of all the various critics of inner sense--including for starters Dennett, Dretske, Harman, Husserl, McGinn, Rosenthal, Ryle, Sartre, Searle, Sellars, Tye, and Wittgenstein--I consider him the clearest, the hardest working, and the most challenging. In the first Royce Lecture he objects that there are no sense organs, that there are no inner experiences of experiences, that there are no sense data, and that introspectible representational properties are not "inner". The second Lecture is devoted to a complicated argument that there can be no such thing as inner blindness, comparable to outer blindness. In the third Lecture he argues that even if there were inner sense, literal projection would be unintelligible. My aim here is defensive, to develop initial responses to each of these challenges, and a few more, rather than to build a positive argument for the view.<6> In the process I will also consider further difficulties with Shoemaker's positive proposals.
II Inner sense without sense organs
Shoemaker argues first that although "sense perception involves the operation of an organ of perception whose disposition is to some extent under the voluntary control of the subject", "[t]here is no organ of introspection that the self directs either to itself or to mental entities of any kind" (1996, pp. 204‑205, 207). This should be no surprise. The main role of sense organs is to convert or "transduce" nonneural stimuli into neural signals, but if there are inner "stimuli" for inner sense, they are presumably already neural (or functional, but realized neurally), so there is no point to inner transducers. One way to see the threat to inner sense is this: since candidate inner stimuli are neural, they almost certainly produce their neural effects by familiar neural processes such as activation and inhibition, and not by any processes more perception‑like than those. So probably no process deserves the name "inner sense" by virtue of analogies to processes of outer sense.<7>
Even in the absence of inner sense organs and attendant inner‑sensory processes, some inner‑directed representations could qualify as "sensory" (or more specifically as "visual", "auditory", etc.). Notice that we already use these adjectives, quite literally, to describe states of imagining, dreaming, and hallucinating that are not caused by sense organs. Thinking about what puts the "visual" in "visual imagining" could give us a guide to what could put the "vision" in "inner vision". (This would neither commit us to asserting nor to denying that imaginings involve inner sensations.) The key similarity is between the products of visual perception and visual imagination--certain representations--rather than between the processes of visual perception and visual imagination. If there are inner-directed representations with the same or greater degree of similarity to outer-directed visual perceptions, we might justify calling them "inner visual" or "inner sensory".
On plausible psychological theories (cf. Fodor, 1983), ordinary sense perception helps to produce judgments (or beliefs), but it does so by producing representations that differ from judgments, and that have other roles as well (such as guiding visuomotor skills). I will call these representations "outer sensations"--or "outer perceptions", interchangeably--since they are products of outer sense representing environmental (and bodily) conditions. Even in the absence of inner‑sensory processes, perhaps there are representations of the mental that differ from judgments about the mental, and deserve the name "inner sensations", in virtue of their similarity to outer sensations: a similarity in form or content or use, even if not in manner of production.
For illustration purposes only, here are a few guesses about relevant detailed similarities we might look for between inner and outer sensations. Compared with judgments, outer sensations (i) are largely under stimulus control, (ii) are in control of reflexes and motor skills, independently of countervailing desires, (iii) are resistant to rational inference and will, (iv) are restricted in content, and (v) are not easily matched in content by desires, memories, communication‑intentions, etc. Similarly, if there are inner sensations we might expect them (i) to be largely under control of their mental subject matter, (ii) to guide reflexes and skills, (iii) to be difficult to form by will or reason, (iv) to involve a limited range of contents, and (v) to be difficult to remember and express in public language. These expectations about alleged inner sensations carry no commitments about the internal structure of a process of inner sense. Nevertheless, they make available a strong literal sense in which inner‑directed representations could count as visual, auditory or (in general) sensory, with minimal violence to the ordinary use of these adjectives.<8> On this view, once inner sensations are produced, in whatever way, they act like outer sensations, as further states in particular sense modalities. For example, inner‑sensory states in vision would help to produce visual beliefs, would help to control visuomotor skills, and might not even be introspectively distinguishable in kind from outer‑visual states. A mental state will count as internally sensed if it is represented by such an inner sensation, regardless of the process by which the inner sensation is produced.
III Inner sense within experience
Shoemaker's second objection is that inner sense would involve ongoing experiences of ongoing experiences, implausibly. He presses the accusation as follows:
This objection rests on a controversial claim about ordinary perception--that every perception "involves" a sense‑experience. On the contrary, plausibly, one can have states that are perceptual even though there is nothing it is like for one to have them. (I assume that for a mental state to qualify as an "experience" in the sense Shoemaker intends, there must be something it it is like for one to have it. Otherwise, we should not expect introspection to be reliable about whether we have experiences of experiences.) In cases of subliminal visual perception (Dixon, 1987) and blindsight (Weiskrantz, 1988), subjects act on the basis of information about the visual features of objects, despite denying--sincerely and without hypochondria--that they have relevant visual experiences. It is plausible, though somewhat controversial, that in at least some cases there is nothing it is like to have subliminal and blindsight states. If so, these are perceptions that do not involve sense-experiences.<10>
Something needs to be added to nonphenomenal sensing to reach phenomenal experience, but what? An inner‑sense theory can, and I think should, be cast as a theory of phenomenal consciousness, holding that experience arises only when states are internally sensed, states that otherwise (on their own) would be nonphenomenal. The idea is that although there are both outer and inner sensations, neither of these is, taken alone, a phenomenal experience. Some people may have qualms about calling nonphenomenal states "sensations", so even though I do not share these qualms it may be best for me to introduce a less loaded terminology. Call the (hypothetical) sensory representations produced by inner sense "I‑states," since they are typically inner‑directed (or about one's own mental states). Similarly, call the sensory representations produced by outer sense "O‑states," since they are typically outer‑directed (or about things other than one's own mental states, such as environmental or bodily entities). Sensory representations produced in imagination (and in dreams and hallucinations) also count as O‑states due to their outer‑directed, though typically false, content. On an inner‑sense theory of phenomenal experience, a perceptual or imaginative experience as of some entity E (an object, event, etc.) essentially involves both O‑states and I‑states of O‑states, as illustrated here (where the dashed arrows signify causation that is typically present but not strictly required, for reasons given in the previous section):
The important point is that on a properly formulated inner‑sense theory experience is compositional. An experience as a whole need not be internally sensed. Rather, a component of an experience (an I‑state) represents another component of the experience (an O‑state). In a derivative way, the experience itself may be said to be internally sensed in virtue of having a part that is internally sensed, just as a house may be sensed in virtue of having a part that is sensed. And since the inner sensing is itself part of the experience, the experience may be said to represent itself, "reflexively," in addition to representing features of E.
So I agree that when we introspect experiences, we don't get new experiences of the experiences. On this inner‑sense account an experience represents itself, in virtue of having component I‑states that represent component O‑states. The two states (O‑ and I‑) combine to form a single perceptual or imaginative experience.
IV Inner sense with varying perspectives
Shoemaker envisions another alleged disanalogy between introspection and outer sense:
[T]here are not different introspective perspectives on the same mental entities, in the way there are different perceptual perspectives on the same physical ones. ... [I]t seems plausible that a nonfactual object of a perception‑like epistemic access should be something on which different perspectives are possible. (1996, p. 208)
These different perspectives may be reflected in the often‑remarked changes in experience that come from concentrated exposure in a perceptual domain such as wine‑tasting or music‑listening. Such changes can occur quickly, as in the following case of Dennett's:
We would like an account of what constitutes such changes in experience: in what way does the experience (or the qualia) get more complex? And how do these changes result from long‑term or short‑term "training"? It seems implausible to suppose that the listener in Dennett's example needs to acquire a new concept, rather than some kind of temporary auditory modification. But although training may build more complex outer‑sensory discriminatory abilities--bigger and better O‑states--this does not seem to account for all such changes. E.g., in Dennett's example, the sound's "complexity" is discriminated "all along"--the O‑states do not become more complex in content. If the perceptual modifications are not (always, only) due to changed concepts and O‑states, what other kind of modification could there be?
The inner‑sense account provides a modifiable level of representation besides concepts and O‑states: namely, I‑states. Training in wine‑tasting or music‑listening may change experiences, not merely by developing one's theoretical concepts, and not necessarily by changing one's outer‑sensory discriminatory abilities, but by changing one's inner‑sensory sensitivities to one's outer‑sensory states. I agree with Dennett that "the complexity apprehended in the third plucking was there all along (being responded to or discriminated)," but this is only to say that the O‑states--not the I‑states--on the first and third plucking are constant and complex. On an inner‑sense account, the change in qualia could be due to a change in I‑states. The I‑states on the first plucking may represent the complex O‑states as simple, while the I‑states on the third plucking represent these same complex O‑states as complex. They amount to different perspectives on the same (type of) mental state, and consequently they amount to different experiences.
It is more plausible that inner sense lacks varying spatial perspectives. This is unsurprising, because inner sense does not represent spatial properties at all. Strictly speaking, spatial perspectives are not required for outer sensing, especially in outer‑sense modalities that do not represent spatial properties. Taste--as distinguished from tactile sensation via the tongue--is a good example. We cannot taste where something is, and equally we cannot adopt different spatial perspectives in taste. (I leave aside the fact that different spatial regions of the tongue respond selectively to different flavors. For simplicity, imagine a tongue with varying taste buds intermingled uniformly, or imagine a tongue containing only sweetness receptors.) If Shoemaker is thinking of spatial perspectives, then the difference he identifies between inner sense and (some kinds of) outer sense is, at root, a difference in subject matter. But of course, inner‑sense theorists are interested in pursuing an analogy with outer sense only modulo differences in subject matter. We would turn a deaf ear to a critic who argued that inner sense has an inner subject matter and so--surprise, surprise--lacks a crucial feature of outer sense.
V Inner sense without inner attention
One can sense things without attending to them; similarly, inner‑directed sense does not imply inner‑directed attention. Drawing this distinction is compatible, however, with maintaining that inner‑directed attention is at least sometimes a kind of inner‑directed sensing. Perhaps one way of attending to perceptual states is by having attentive inner sensings of these states. Sometimes when we speak of looking at something and "attending to" what we're doing, we mean looking carefully at the thing, especially shifting our gaze to it. But this cannot handle all cases of attending to vision. Normally one is unaware of the fuzziness of one's visual periphery, but one can willfully "attend to" the periphery and thereby notice the fuzziness with which one sees things there. This must be different from shifting one's gaze to the periphery and so removing the fuzziness. Similarly, in degraded cases of vision, such as crossed‑eye vision, it does not seem difficult to attend to the degradation--but it is the experience, not the environment, that seems degraded. Nevertheless, in general it is at least very difficult to redirect attention from environmental objects to experiences of them.
Since on the present view I-states are directed at O‑states, not (except derivatively) at full-blown experiences, one need not be able to attend to full-blown experiences, much less be able to do so regularly, easily, or on command. This yields a response to the following informal experiment proposed by Shoemaker:
The present view is not committed to attentive inner sense of full-blown experience. Even if it were, it would not be committed to the idea that attention can be redirected at will. Outer attention is not always at our command. Raise both of your hands before you, and have a friend set your left one on fire. Now shift your attention to your right hand. Can you do so? Probably not, because your representation of your left hand captures your attention at the expense of your right hand, for excellent natural‑selective reasons. Similarly, for excellent natural‑selective reasons, environmental objects capture attention, at the expense of mental states. This does not show that mental states are (inattentively) unsensed, any more than the burning-hand case shows that one's right hand is (inattentively) unsensed.
I think a process can be "sensory" even if it is incapable of supporting attention, but others may disagree. If they are right, then my view does have a commitment to the claim that one can attend to O-states as well as to environmental objects. Whether there is such attention is not a matter to be settled purely introspectively, but by a combination of introspection and psychological theory. Dennett's guitar case described in the previous section provides some evidence that there can be attentive inner sense of O-states, distinct from attentive outer sense of environmental objects, because one can attend to the qualitative changes that are arguably due to I‑states.<11>
VI Inner sense without sense data
On widely rejected "sense‑data" or "veil" theories of perception, one perceives environmental objects indirectly, by inwardly perceiving interposed mental objects. Critics including Shoemaker have suggested a connection between inner sense and sense data.<12> But a properly formulated inner‑sense model of experience is not committed to sense data. Inner sense as I have described it requires two things: O‑states of environmental objects, and I‑states of the O‑states. O‑states aren't interposed between environmental objects and our perceptions of them--they are our perceptions of environmental objects. And I‑states certainly aren't interposed between objects and our perceptions (O‑states) of them; they come "after" O‑states in the cognitive flowchart. Inner sensings are directed not at sense data mediating between objects and one's perceptions of them, but at one's perceptions themselves. The causal chain in perceiving an apple needn't proceed from the apple to an inner sensing and then to a sensing of the apple. Rather, on a more natural view, the causal chain goes directly from the apple to a sensing of the apple (an O‑state), and then (in cases in which the apple‑sensing forms part of a phenomenal experience) to an inner sensing (an I‑state) of the sensing of the apple. Both outer sense and inner sense are "direct" in that neither requires mediation by further sensings.
So I do not agree with Shoemaker's suggestion that "reasons for rejecting the sense‑datum theory are at the same time reasons for rejecting the act‑object conception [of inner sense] as applied to pains and after‑images" (1996, p. 215). By the "act‑object conception" he means the idea that one primarily perceives nonfactlike objects, e.g., that one senses pain‑objects or after‑image‑objects rather than merely sensing that one is pained or that one is afterimaging. But we can reject interposed sense data, and we can agree that outer sensing is direct, without threatening the idea that one inwardly senses the products of outer sensing (O‑states). I see no considerations here that weigh against the existence of inner sense, or even against the idea that inner sense is primarily directed toward nonfactlike inner objects (e.g., the neural objects in virtue of which one has O‑states).
VII The content of inner sense
An inner‑sense theory needs an account determining which properties of mental states are internally sensed. Shoemaker emphasizes the implausibility of internally sensing the contents of perceptual states:
I agree that inner sense cannot (simply and directly) account for introspective knowledge of content properties.<13> I advance inner sense not as a general theory of introspection but as a theory of one specific kind of introspection, the kind (I would argue) that helps explains phenomenal consciousness. I‑states represent features of O‑states, but they do not represent the representational relations of O‑states. These are not plausible candidates for being internally sensed, since they are relations between O‑states and other things that are not internally sensed--e.g., environmental objects.
What other features of O‑states might I‑states represent? Well, what other features do O‑states have? Instead of O‑state content, I‑states might represent intrinsic, physical properties of O‑states. These could include abstract "syntactic" features of O‑states, and even features of O‑states specific to their "hardware" realization, such as the rough number of neurons that realize an O‑state, or their rough average rates of firing. (I do not assume that I‑states involve complex descriptive concepts of neural structure or of psychological function; another possibility is that they respond to specific neural or psychological features in more primitive ways, just as in outer sense one may be sensitive to wavelengths or molecular motion without having concepts of waves or molecules.) As an illustration of how this might help shape experience, consider Christopher Peacocke's much-discussed example of seeing two same‑sized trees, at varying distances from one:
My aim here is to describe an inner‑sense account of this example, not to evaluate rival proposals. Peacocke argues that "you simply enjoy an experience which has the feature" of different sizes‑in‑the‑visual‑field. It is not simply that the experience "has" this visual‑field feature, in the way it might "have" the feature of being realized in, say, molecules: in addition, one is normally sensitive to an experience's visual‑field features (in a way one is not normally sensitive to its being realized in molecules). An inner‑sense account can explain the visual‑field differences in Peacocke's two experiences, as well as one's sensitivity to these features, as follows. Compared with O‑states about the distant tree, O‑states about the nearer tree are realized by (or causally connected to) many more O‑states in retinotopic maps in the early visual system. This is just the sort of property to which inner sense may make one sensitive.
I‑states may also represent interrelations among O‑states, relations that cannot be reduced to their intrinsic, monadic properties. These interrelations are better candidates for inner sense than O‑state content, because their relata are all candidate objects of inner sense (i.e., are all O‑states, rather than environmental objects). As illustrations, consider cases of double vision and blurred vision. Typically we are sensitive to the doubleness or blurriness of such experiences. In double vision, we may internally sense the distinctness of two O‑states (say, two matching perceptions of an edge). In blurred vision, we may internally sense the causal interrelation of a certain O‑state (say, a perception of an edge) with an unusual set of other O‑states (say, states that involve neurons unusually and poorly "lined up" in the retinotopic maps in primary visual cortex). In normal focal vision, by contrast, we are typically sensitive to the nondoubleness and nonblurriness of our experience. This could be explained by our internally sensing related structural features of O‑states. We detect the absence of distinct O‑states matching an O‑state of an edge, and we detect the O‑state's causal interrelation to a standardly "lined‑up" set of other O‑states.
There are other nonrepresentational features of O-states that might potentially be represented by I‑states, such as their degree of attentiveness, or their modality (visual, auditory, etc.), or the degrees of similarity among them obtaining in virtue of their intrinsic physical properties or functional interrelations. In short, I accept Shoemaker's point that representational features of outer-directed perceptions (O‑states) are beyond the reach of inner sense. But O-states have other properties that may reasonably be considered subject to inner sense, and that may reasonably be claimed to be relevant to what the resulting experiences are like.
Perceptual experience seems diaphanous to the extent that we do not introspect features of experience as intrinsic to it, or otherwise as nonrepresentational. But for all this shows, introspectible features of experience may be intrinsic to experience, or otherwise nonrepresentational. It may be that one (internally) senses an O‑state's having some intrinsic property Q, but senses it as an environmental object's having Q. Just as one can perceive an object as a different object, so one can perceive an object's having a property as a different object's having the property. I hear the ventriloquist's speaking, but I hear it as the dummy's speaking. Likewise, in watching a big‑screen, fast‑paced, high‑visual‑fidelity film consisting of life‑size images, it is surprisingly hard to see colors and shapes as being on the screen rather than in the remote scene. If one blocks out the surrounding room (perhaps by looking through a tube), it does not seem that one is looking at a screen at all. Nevertheless, what one is seeing is a screen, and the colors and shapes one sees are on it. One merely projects them onto the represented remote scene. Similarly, perhaps one projects features detected in inner sense onto the outer environment.
Shoemaker admits to being "in the uncomfortable position of finding the [projection] view both plausible and unintelligible" (1996, p. 139):
Shoemaker would presumably add that the examples of the ventriloquist and the film screen do not make sense of experience-to-environment projection. It is perfectly conceivable for ventriloquists' dummies to make noise, and perfectly conceivable for remote-scene colors to be screen colors, while it is inconceivable for environmental properties to be experiential properties.
There are a few responses worth trying to Shoemaker's inconceivability claim. Either in isolation or in some combination, they may be adequate.
First, although Shoemaker speaks of "the" property represented when one sees a surface as red, a projection theorist can speak of two represented properties, the ones represented by O‑states and the ones represented by I-states. Only the latter are misrepresented in projection. A projection theorist need not, and I think should not, claim that color properties are projected. Color can be thought of as an objective feature typically represented accurately by O-states (perhaps a complicated wavelength-transmission property). Thus a projection theorist need not consider perceptual experience mistaken about where the colors are, but only about where the properties internally sensed are. Shoemaker cannot object to bifurcation of perceived properties without undermining his own view that perception represents both color and phenomenal-causal relations. He might instead point at his shiny red book and insist that all of those colorish properties are only conceivable as belonging to spatially extended things. But (to argue ad hominem) this would also undermine his positive proposal, for his phenomenal-causal properties are not only conceivable as belonging to spatially extended things. An object occupying a spatially unextended point could conceivably cause an R‑experience, as could a wholly nonspatial god.
As a second response, consider that experiences may have highly abstract features that could accurately be ascribed literally to spatially extended objects. Multiple experiences can stand in similarity relations to one another, and if these similarities are internally sensed and projected as a group to environmental objects, the result would be that the environmental objects would be represented as standing in isomorphic similarity relations. The same is true for degrees along abstract dimensions. For example, in Peacocke's two-trees case, the O-state for the nearer tree could be internally sensed as "greater than" than the O-state for the farther tree (in, say, number of component O-states), and this could be projected so that the nearer tree itself seems "greater than" the farther tree (along a dimension one has trouble naming, but that is easily confused with size). Examples such as these could be multiplied, but a more generally applicable third response would be useful.
Shoemaker rightly insists, in his treatment of phenomenal-causal relations, that perception need not yield an explicit conception of the essential nature of perceived properties. Likewise, inner sense need not deliver explicit concepts of the essential nature of the projected features of experience. But even if a property represented in inner sense is something that, in its nature, could not plausibly or even conceivably be instantiated in environmental objects--such as the neural property of firing at a certain rate, or the property of being in one's visual system--so long as one is not aware of this nature explicitly, one will not be aware of the implausibility or inconceivability of environmental instantiation, and there will be no available barrier to projecting. In short, even if a projection could not plausibly or conceivably be accurate, it could still conceivably and plausibly be actual, and that is all the projection theorist needs to avoid Shoemaker's charge that projection is unintelligible.<14>
A more detailed projection theory should explain why projection occurs. I have attempted to explain this elsewhere, and unfortunately it's too long and complicated a story to relate here.<15>
IX Inner sense without the possibility of inner blindness
Shoemaker attempts to clinch the argument against inner sense by appeal to the following feature of outer perception:
According to Shoemaker, if mental states are subject to inner sense, they should similarly be independent of it. However:
His strategy is to show that "there is a conceptual, constitutive, connection between the existence of certain sorts of mental entities and their introspective accessibility," at least for creatures that have "the ability to conceive" of those entities (1996, 272‑273). The disanalogy with outer sense is that one cannot be completely introspectively blind (or "self‑blind") to these mental entities, while one can be completely perceptually blind to trees, mountains, etc.
His argument for this conclusion is long and complex. Even without considering the details, though, it seems that the overall argument would prove too much. Just as there is an "independence condition" for outer‑directed perception, so there is one for outer‑directed belief. Trees, mountains, houses, automobiles, and human bodies, etc. can all in principle exist without being believed in, and even without there being creatures capable of believing in them. If the argument yields evidence that we do not have inner‑directed sensings, it yields equally good evidence that we do not have inner‑directed beliefs--inner sense would be "disanalogous" to outer sense only to the same degree that inner belief would be disanalogous to outer belief. If Shoemaker is right about the conceptual relation between introspection and introspected, this would be a "fundamental difference" between outer-directed belief and introspection! But of course that should not incline us to doubt the existence of inner‑directed beliefs. Nor should Shoemaker's argument, whatever the details, incline us to doubt the existence of inner‑directed sensings.
Shoemaker's proceeds to argue that introspectibility is necessary for the existence of pain feelings and perceptual experiences. This provides another simple way to see that the argument presents no threat to the inner‑sense theory I am proposing: the (fundamental) objects of inner sense are nonphenomenal O‑states, rather than phenomenal pain feelings or perceptual experiences. I agree that pain feelings and perceptual experiences are dependent on introspection, and in addition my view explains this: the inner sensings of parts of these experiences are themselves parts of the experiences.
X Introspection of attitudes
In this paper I have focused on inner sense of sensory states, as opposed to inner sense of propositional attitudes or selves, because I agree with Shoemaker that inner sense does not apply to these. But I would like to finish with some comments on his treatment of attitude self‑knowledge.
Shoemaker argues that "[n]o one thinks that one is aware of beliefs and thoughts by having sensations or quasi‑sense‑experiences of them" (1996, p. 207).<16> I agree that there are no such sensations, but consider this compatible with and even congenial to an inner‑sense theory aimed at explaining phenomenal consciousness. To explain why, I need to treat thoughts separately from beliefs and other propositional attitudes.
By thoughts, I mean first and foremost the fairly slow, roughly serial, and typically deliberate phenomena we try to describe as "talking to oneself" or "thinking in words". These seem to be constituted (at least as far as phenomenal consciousness goes) by imaginings of words or speech acts, typically auditory or visual ones. There are also nonverbal thoughts, which we are tempted to describe as "thinking in images", and these also seem constituted by imaginings. An inner‑sense theory should treat all such experiences within the stream of thought in the same way it treats other imaginative experiences. Thoughts involve imaginative O‑states representing speech acts or nonverbal things, together with I‑states representing features of these O‑states. Inner sense is part of what explains why something is a thought‑experience in the first place, not something that kicks in once a thought experience is already formed. So Shoemaker's plausible claim that we do not have separate experiences of thought experiences is compatible with my claim that inner sense is involved in thought experiences (and in at least one kind of introspective awareness of them).
By contrast, beliefs and other attitudes do not seem to be constituted even in part by sensory representations, and I agree with Shoemaker that they are not subject to inner sense at all. If beliefs and other attitudes were phenomenal experiences, this would undermine an inner‑sense theory of phenomenal consciousness. But on independent grounds it is plausible that attitudes are not phenomenal (see Lormand, 1996). There is nothing conscious attitudes themselves are like, although there is often something that accompanying states are like, especially perceptual experiences, bodily‑sensational experiences, imaginative experiences, and experiences in the stream of thought (I give some examples of such accompanying states in the next paragraph). The absence of inner sense for attitudes is not a weakness of inner‑sense as a theory of phenomenal consciousness, but one of its greatest strengths. A serious difficulty for theories of phenomenal consciousness is to explain why attitudes are not phenomenal, even when they are conscious in some other sense. Of course, attitudes can be, and often are, excluded by fiat. But to my knowledge only inner‑sense theories of phenomenality provide an explanation of the nonphenomenality of (conscious or unconscious) attitudes.
If we do not internally perceive our attitudes, how do we know about them? Is inner sense completely irrelevant to attitude self‑knowledge? Shoemaker turns the screws:
On the contrary, I think it is utterly normal, in becoming conscious of one's attitudes, to have accompanying phenomenally conscious "symptoms" of the attitudes (see Lormand, 1996). For example, one's standing, unconscious belief that snow is white may cause one to form an auditory image of quickly saying the words "snow is white" (or "I believe snow is white" or "Mon Dieu! La neige! Blanche!" or ...). There are normally more aspects to this verbal imagery, which help one to determine which kind and strength of attitude (belief, desire, suspicion, etc.) is revealed via the thought. In cases of belief, for instance, one may imagine saying the words in an assertive tone of voice, and without any concomitant proprioceptive sensations of suppressed giggling, or auditory images as of appending "NOT!," etc. In addition to such verbal imaginings, there are often nonverbal imaginings, e.g., of a white expanse of snow, and perhaps visual imaginings of words. And consider William James' interesting introspective report:
It is no part of my proposal that each belief or desire has a canonical phenomenal symptom, and I do not here assert (or deny) that every bit of attitude self‑knowledge proceeds via phenomenal symptoms. But to say the least, we should be very suspicious of any view of introspection that marginalizes such an important and nearly ubiquitous entryway into knowledge of attitudes.
Shoemaker's own positive account seems to leave no room for knowledge of attitudes via phenomenal evidence:
I have no disagreement with the idea that introspection (at least often) produces beliefs. I do claim that such beliefs are often caused via inner‑directed sensations, as opposed to Shoemaker's bare "wired" connections. I do not see how Shoemaker's view can explain lawful and systematic patterns among introspectible and nonintrospectible states, without an ad hoc assumption, for each pattern, that there happens to be the right "wiring" to higher‑order beliefs. For example, why does it seem easier to introspect one's fleeting thoughts than one's deeply held beliefs? After all, deeply held beliefs would have more time to grow wired connections to beliefs about them. On my inner‑sense account, fleeting thoughts, but not deeply held beliefs, are realized in O‑states (typically auditory imaginations of speech). These are subject to inner sense, and thus are easier to introspect than beliefs. Likewise, why on Shoemaker's account aren't brain states governing autonomic bodily functions (easily) introspectible? There could have been wired connections from these to beliefs, so why aren't there? I claim that we lack (easily attained) beliefs about these states because we lack inner sensings of these states--they are motor or glandular "commands" rather than sensory O‑states.
I am in agreement with Shoemaker on most of his main points: I agree that there is no inner sense of the self, and no inner sense of propositional attitudes. I even agree that there is no inner sense of independently existing sensory experiences. I have tried merely to carve out a space, compatibly with most of his insights, for inner sense of something else, namely, of certain pre-phenomenal components of sensory and imaginative experiences. This has been a defensive task; I have been able to argue here only that such inner sense might exist, but not positively that it does exist. It seems an interesting possibility, because it may cohere better with general theories of representation than does Shoemaker's positive account of perceptual content, and because it might be part of the elusive explanation of phenomenal experience itself. But I have also not been able to argue here that inner sense is part of the correct explanation of experience. It is a testimony to the power and diversity of Shoemaker's arguments--and there are several I have not even mentioned--that he has put inner-sense theories on the defensive to such a degree.<17>
Ayer, Alfred (1956) The Problem of Knowledge, Harmondsworth UK: Pelican Books.
Dennett, Daniel (1988) "Quining Qualia", in A. Marcel and E. Bisiatch, eds., Consciousness in Contemporary Science, Oxford: Oxford University Press: 42‑77.
Dixon, N. (1987) "Subliminal perception" in R. Gregory, ed. The Oxford Companion to the Mind, Oxford: Oxford University Press: 752‑755.
Dretske, Fred (1995) Naturalizing the Mind, Cambridge: MIT Press.
Fodor, Jerry (1983) The Modularity of Mind, Cambridge: MIT Press.
Güzeldere, Güven (1995) "Is consciousness the perception of what passes in one's own mind?," in T. Metzinger, ed., Conscious Experience, Thorverton, UK: Imprint Academic: 335‑357.
Harman, Gilbert (1990) "The intrinsic qualities of experience," Philosophical Perspectives, 4: 31‑52.
Hill, Christopher (1991) Sensations, Cambridge: Cambridge University Press.
James, William (1890) The Principles of Psychology, reprinted 1993, Cambridge MA: Harvard University Press.
Lyons, William (1986) The Disappearance of Introspection, Cambridge: MIT Press.
McGinn, Colin (1982) The Character of Mind, Oxford: Oxford University Press.
Peacocke, Christopher (1983) Sense and Content, Oxford: Clarendon Press.
Rosenthal, David (1990) "A theory of consciousness," ZIF Report 40, Bielefeld: Center for Interdisciplinary Research.
Shoemaker, Sydney (1996) The First Person Perspective and Other Essays, Cambridge: Cambridge University Press. The Royce lectures were first published in Philosophy and Phenomenological Research, 54 (1994): 249‑314.
Stich, Stephen and Ted Warfield, eds. (1994) Mental Representation, Oxford: Blackwell.
Weiskrantz, Lawrence (1988) "Some contributions of neuropsychology of vision and memory to the problem of consciousness" in A. Marcel and E. Bisiatch, eds., Consciousness in Contemporary Science, Oxford: Oxford University Press: 183‑199.
 Less roughly, Shoemaker characterizes R* as "the property something has just in case it is currently producing an R‑experience in someone related to it in a certain way, namely someone viewing it under normal lighting conditions" (1996, p. 254). I don't understand why he speaks indirectly--"the property something has just in case it is currently producing ..." instead of "the property of currently producing ...". I also don't understand why he prefers to speak of "an R‑experience in someone" rather than "an R‑experience in me" or even "this R‑experience". And I don't understand why he builds in reference to "normal lighting conditions", since the theory should also apply to perceptual experiences produced in abnormal conditions. So I will try to keep my discussion clear of these intricacies.
 These accounts are in terms of some mix of causation, correlation, information, inferential role, conceptual structure, verification, evidence, teleological function, psychological explanation, similarity, etc. (some influential papers are collected in Stich and Warfield, 1994).
 Causal theories have not gained wide acceptance as applied to mental states generally (since, for example, beliefs may be about conditions other than their causes, such as events in the future), but they have seemed especially appropriate in explaining perceptual representation (which is much more restricted).
 Another version of this problem arises with extending Shoemaker's account to imaginative experiences. If imagining a red apple includes merely imagining an apple's causing an experience with a certain quale, and if in introspection one realizes one is merely imagining, how could this license the conclusion that one does have an experience with the quale?
 One more potential gap arises from the fact that R* is the property of causing an experience in someone, rather than a specified experience in oneself. Even if one's way of representing R* were explicit, one would not be able to deduce that R is a quale of one's own experience, nor would one be able to deduce which of one's experiences had R. But since I don't understand the motivation for this aspect of Shoemaker's proposal (see note 1), and it seems easily modifiable, I don't wish to focus on it. And anyway, I grant that one might be able to guess, fairly reliably, that the R‑experience is happening in one, and is the same experience that represents the apple as R*.
 I present more positive arguments in manuscripts such as "The Explanatory Stopgap" and "Phenomenal Illusions", available online at http://www‑personal.umich.edu/~lormand/phil/cons.
 For similar objections to inner sense, see: Lyons, 1986, p. 96; Rosenthal, 1990, p. 35; Güzeldere, 1995, p. 346.
 We have no qualms about calling imaginings "visual" even though in salient respects they are less similar to outer sensations than I envision inner sensations to be: imaginings are fairly independent of stimulus control, fairly disconnected from reflexes and motor skills, and fairly pliable by the will.
 For similar objections see: McGinn, 1982, p. 50; Dretske, 1995, p. 63.
 Also, the early layers of processing in vision and other sensory modalities, although on most accounts they do involve perceptual representations, seem not to constitute experience; further processing is necessary. Otherwise, it would be difficult to explain why phenomenal visual experiences are not continually like double images--since one has separate left‑eye‑caused and right‑eye‑caused early visual states. It would also be difficult to explain why one seems not to experience arrays of incoming brightnesses, i.e., why phenomenal experiences do not represent distant objects as being "behind" two splotches of light near one's eyes.
 Christopher Hill provides other plausible examples that indicate the possibility of inner‑directed attention. He calls the first kind "volume adjustment," in which, by increasing one's attention to a sensation, one increases either its "intensity" or its "internal articulation." He illustrates these as follows:
As with Dennett's guitar case, inner sense can explain volume adjustment easily. By attending to an O-state, one changes the content of the corresponding I-state (i.e., one "notices" something different about the O-state). Just as increasing outer‑sensory attention to an environmental object can increase the detail or strength of one's O‑states of the object, so increasing inner‑sensory attention to an O‑state can increase the detail or strength of one's I‑states of the O‑state. Since I‑states are parts of experiences, this changes what the experiences are like.
The second effect of inner-directed attention is what Hill calls "activation," in which one seems to bring a sensation into existence, rather than merely modifying one:
I‑states are tailor‑made to bring new experiences (phenomenally conscious sensations) into existence--without I‑states, there are no experiences, but instead only O‑states (or what Hill calls mere unconscious "packets of information", 1991, p. 122). Just as shifting outer‑sensory attention to some location can generate O‑states of formerly unperceived objects at that location, so shifting inner‑sensory attention can generate I‑states of formerly unperceived O‑states. This "activates" a phenomenally conscious experience where there was none before.
 Gilbert Harman attributes the initial plausibility of inner sense to a failure to "distinguish between the properties of a represented object and the properties of a representation of that object," and continues:
William Lyons considers a view very much like my position on inner sense, namely, the view that we introspect "the products of an act of perception" (1986, p. 105). He presents the following argument against this position:
Thus, Lyons suggests that to embrace inner sense of sensory states, one must be a "veil‑of‑perception" theorist.
 For similar use of this point against inner sense, see: Dretske, 1995, pp. 108‑109; Güzeldere, 1995, pp. 349‑352. At the end of this paper I consider complex and indirect ways for inner sense to contribute to knowledge of content.
 In principle, even the property of being a prime number can be ascribed to a material object, so long as one represents the property in an unrevealing enough way (and not explicitly as the property of being a prime number). A being who can reliably and without conscious inference distinguish prime numbers from nonprime numbers, but who is unreflective enough not to realize the nature of what he is "sensing" or the conditions in which he is reliable, could well mistakenly experience or think of rocks as he does primes.
 Here is a fragment of the story, from "Phenomenal Illusions" (see note 6). This need not be read to understand the rest of the paper. In normal outer‑directed perception, small portions of sense organs transduce ambient energy into initial neural "maps", which then produce, in separate areas of the brain, structures bearing information or misinformation about various properties of stimuli. This creates what is known in perception research as the "binding problem": how does a perceptual system keep track of which properties belong to which perceived objects? As usually raised in the cognitive neuroscience of vision, the problem is a bookkeeping one: what neural or functional "stamp" does vision give to these separate structures when the features are seen as coinstantiated? But this leaves the problem: how does a perceptual system determine which features to stamp as coinstantiated? I suggest that one extremely plausible and reliable method for determining coinstantiation is to treat perceptual structures caused at a moment by the same portions of sense organs (or initial neural maps) as applying to the same object. This binding heuristic would be reliable for sensory states, though not for judgments generally (since judgments from a common source may vary arbitrarily in subject matter).
If some of the perceptual structures (or O‑states) are internally sensed, this binding heuristic could also apply to I‑states about them. An I‑state caused by an O‑state is of course indirectly caused by the same transducers as the O‑state, and (unlike a judgment that may happen to be caused by the O‑state) is producible in no other way, barring gross malfunction. If so, it would be natural for the binding heuristic to be extended to the products of inner‑sense, and so the features represented by I‑states and O‑states would be "bound" to the same objects, which in turn would begin to explain projection. In other words, on this account the price of solving the binding problem for perceived properties of physical objects is that internally sensed properties of perceptual structures are bound along with them. Projection could be a side‑effect of a valuable perceptual strategy understandably overextended to inner perception.
 He emphasizes that "the perceptual model of introspective self knowledge has its greatest plausibility" for sensory events, as opposed to propositional attitudes (1996, p. 214). David Rosenthal also argues that there is a special implausibility to inner perception of attitudes, as follows:
Of course, in the end Rosenthal and Shoemaker reject an inner‑sense model for sensory states as well as for attitudes.
 I wrote an earlier version of this paper for Shoemaker's "Author Meets Critics" session at the 1999 Central APA meetings in New Orleans. Thanks to Sydney and the audience for their comments, and to moderator Mark Crimmins for inviting me.