The Explanatory Stopgap

Eric Lormand
University of Michigan

Forthcoming 2005 in Philosophical Review

             Is there an explanatory gap between raw feels and raw material?  Some philosophers argue, and many other people believe, that scientific explanations of conscious experience cannot be as satisfying as typical scientific explanations elsewhere, even in our wildest dreams.  The underlying philosophical claims are:

   (i)      typically when satisfactory explanations are available in other domains of science, what is explained is conceptually necessitated by what explains it, although

   (ii)     the existence of conscious experience is not conceptually necessitated by any possible combination of premises about the nonconscious or the nonexperiential. 

This requirement of conceptual (epistemological, a priori) necessitation is, intentionally, demanding in the extreme.  Claim (i) is usually argued for by giving examples from science that seem to fulfill it.  More generally, it seems to be motivated by the idea that explanation answers questions, and that a fully satisfying answer to a question should not raise a new question in its place.  For example, a fully satisfying answer to the question 'what is necessary and sufficient for consciousness [or life, water, rocks, and so on]?' should not in turn raise the question 'why is that necessary and sufficient for consciousness [life, water, rocks]?'  And, it must be admitted, nothing closes off further questions quite like revealing them to presuppose conceptual contradictions.  While I think there is much to be said against this stringent requirement on fully satisfactory explanation--both by arguing against the examples adduced in support of claim (i), and by arguing that questions may be rendered pointless without resting on conceptual contradictions--my aim in this article is to argue against (ii), and indeed to show in detail, step by conceptually necessary step, how to deduce phenomenal feels from purely nonphenomenal material.   

1     The gap

            Since the appearance of Thomas Nagel's "What is it like to be a bat?" defenders of the explanatory gap have been getting bolder with each passing decade.  Nagel claims in that article that there is a "gap between subjective and objective," but he does not wield his bat against the metaphysical doctrine that everything mental is physical.  Rather, he argues that the gap is in our current understanding of this doctrine: "we do not at present have any conception of how it might be true" (1974, 176, my emphasis).  Joseph Levine, the originator of the phrase 'explanatory gap,' concurs that the gap is epistemological rather than metaphysical, but strengthens the epistemological upshot: "there is a significant problem about our ever coming to know that statements like ['pain is the firing of C‑fibers'] are true" (1983, 359, my emphasis).  Most recently, David Chalmers argues from the explanatory gap not only to epistemological conclusions but also to a wide range of squarely metaphysical positions, including property dualism, epiphenomenalism, and even panpsychism (1996; for a more modest panpsychism, see Nagel 1979).  My concern in this article is not with these family squabbles about the ramifications of the gap, but with its alleged source. 

            Fortunately, gap‑theorists speak in unison here; all emphasize that the explanatorily recalcitrant aspect of conscious experiences is that there is something it is like to have them.  Nagel explains the troublesome sense of 'conscious' as follows:

[F]undamentally an organism has conscious mental states if and only if there is something it is like to be that organism--something it is like for the organism.  We may call this the subjective character of experience.  It is not captured by any of the familiar, recently devised reductive analyses of the mental, for all of them are logically compatible with its absence.  (1974, 166)

Levine similarly insists that "what is at issue is the ability to explain qualitative character itself; why it is like what it is like to see red or feel pain" (1993, 128), and maintains: 

No matter how rich the information processing or the neurophysiological story gets, it still seems quite coherent to imagine that all that should be going on without there being anything it's like to undergo the states in question.  (Levine 1993, 129)

Chalmers characterizes the problem most fundamentally in the same 'something it is like' way, then adds more jargon‑laden characterizations:

We can say that a being is conscious if there is something it is like to be that being ....  Similarly, a mental state is conscious if there is something it is like to be in that mental state.  To put it another way, we can say that a mental state is conscious if it has a qualitative feel--an associated quality of experience.  These qualitative feels are also known as phenomenal qualities, or qualia for short.  The problem of explaining these phenomenal qualities is just the problem of explaining consciousness.  This is the really hard part of the mind‑body problem.  (1996, 4)

Why is it so hard, according to Chalmers?   For exactly the reasons specified by Nagel and Levine:

[N]o matter what functional account of cognition one gives, it seems logically possible that that account could be instantiated without any accompanying consciousness.  It may be naturally impossible--consciousness may in fact arise from that functional organization in the actual world--but the important thing is that the notion is logically coherent.  If this is indeed logically possible, then any functional and indeed any physical account of mental phenomena will be fundamentally incomplete.  (1996, 47)

Helpfully, Chalmers hones the challenge as follows:

If proponents of reductive explanation are to have any hope of defeating the arguments above, they will have to give us some idea of how the existence of consciousness might be entailed by physical facts.  While it is not fair to expect all the details, one at least needs an account of how such an entailment might possibly go.  But any attempt to demonstrate such an entailment is doomed to failure.  For consciousness to be entailed by a set of physical facts, one would need some kind of analysis of the notion of consciousness--the kind of analysis whose satisfaction physical states could imply--and there is no such analysis to be had.  (1996, 104)

Okay, then: I choose both "truth" and "dare."  I will try to give defenders of the explanatory gap exactly what they say they want.   

2     The stopgap 

            Clearly, to avoid charges of missing the point, my attempt to stop up the explanatory gap must come face‑to‑face with the concept of there being something it is like for a creature to have a feature.  The feature might be a nonmental kind-property (being a bat, being human, being the creature itself, ...), or a nonmental bodily or environmental property (flying, going to the store, being bitten, ...), or a mental property (seeing a bat, imagining flying, thinking about going to the store, feeling bitten, ...).  In the sense intended by defenders of the gap, I think, the last of these is conceptually fundamental.  As we might say, there isn't anything it is like simply for a creature to be a bat, or human, or itself.  To say the least, we ordinarily think it possible for a bat or a human being to spend its entire life in a deep coma, with only autonomic brain activity and no consciousness of any sort, while still being a bat or a human being.  Similarly, there isn't anything it is like for a creature simply to fly, to go to the store, or to be bitten--where these are construed as purely bodily and environmental features.  A deeply comatose bat can be flung through the air, and its teeth can be sunk into a deeply comatose human being who is being carried to the store.  When there is something it is like to have a kind property or a bodily or environmental property, this is explained by there being something it is like to have a mental property--indeed, something it is like simply to have the mental property.  Perhaps there is something it is like simply to see a bat, to imagine flying, to think about going to the store, or to feel bitten--at least when these are active and introspectible mental features.  Creatures without such mental properties--perhaps the deeply comatose--need not apply. 

            Suppose then that a creature c has a mental property M.[1]  How can we get at the concept of conscious experience that defenders of the explanatory gap seek to isolate?  I will begin (but not end) by investigating why the following (somewhat) ordinary expression strikes the right conceptual chord in so many people:
   (1)     There is something it is like to have M.
Although I do not assume (or deny) that linguistic analysis is the best inroad into concepts, this methodology seems especially relevant here, since there must be some explanation for the striking fact that defenders of the gap rely almost entirely upon (1) and its slight variants at almost every crucial turn.[2]  To determine what concepts this phrase expresses, it would help to determine the contributions of the component phrases 'there is,' 'something,' 'it,' 'is like,' and 'to have M.'  What conceptual roles do these play?  I will reserve the fishy 'something' and 'is like' for last, and consider the other phrases as preliminaries.  First, however, I would like to register two stipulations about the force of claim (1), in order to ensure that it is taken at full strength. 

Two overarching requirements 

            To my ear, (1) can be heard either as entailing that M is actually and currently possessed--perhaps with the emphasis 'there is something it IS (now) like to have M'--or as making a claim about possible cases of having M--perhaps with the emphasis 'there is something it is like (simply) TO have M.'  Since my aim is to take the explanatory‑gap challenge as seriously as possible, I want to ferret out any and all content that may be suggested by the pet phrase of gap‑defenders.  So I will construe (1) as making a claim both about the actual (present) case and more broadly about possible cases.  The broader claim is not merely that there is a correlation between there being something it is like, on the one hand, and the possession of M, on the other.  It is that there is something it is like simply to have M--that there must be something it is like in order to have M.  Consider again the hesitation or looseness we sense about 'it is like something to be a bat' or 'it is like something to go to the store'--as opposed to 'it is like something to feel pain'--as described in the first paragraph of this section.  This is a sign that a necessity claim is suggested by (1).  The hesitation or looseness might be explained as follows: we deny 'in order to be a bat, there must be something it is like,' and 'in order to go to the store, there must be something it is like,' but perhaps we affirm 'in order to feel pain, there must be something it is like.'  This necessitation requirement is never far from the surface, since we can favor it merely by emphasizing the word 'to' in (1), as above.  On a natural reading this is elliptical for 'in order to.'  On this reading, for (1) to be true:
   (α)     M must be (actually, currently) possessed, and
   (β)     as a necessary condition for (α) there must be something it is like.
I will not try to make these overarching requirements explicit by expanding (1), but I will appeal to them from time to time, and they are to govern every expansion of (1) below.[3] 

'There is', 'it', and 'to have M' 

            Let us turn to the component phrases of (1).  First we have to note that (1) is clearly elliptically equivalent to (2):
   (1)     There is something it is like to have M.
   (2)     There is something it is like for c to have M.[4]
The 'for c' plays multiple roles simultaneously.  The two most obvious roles depend on whether 'for c' forms a unit with the phrase following it or the phrase preceding it.  We can read (2) both as 'it is like something for‑c‑to‑have‑M' and as 'it is‑like‑something‑for‑c to have M.'[5]  In the former case, 'for c' is redundant given that c is M's bearer; on this reading it can be omitted without noticeable semantic loss--as in the Levine and Chalmers quotes above, yielding (1).  But it also plays the more substantive latter role; on this reading, it can be stressed without noticeable semantic gain--as in Nagel's emphatic "something it is like for the organism" in the quote above.  (The stress encourages the second reading, but this reading is available without the stress.)  The best way to make these two roles explicit is to duplicate 'for c,' construing (2) as equivalent to (3):
   (2)     There is something it is like for c to have M.
   (3)     There is something it is like for c, for c to have M.
Next, consider two possibilities for the role of 'it.'  Sometimes a phrase‑initial 'it' is merely for grammatical show, as in 'it is raining' or 'it stinks.'  We typically use this bare 'it' when we refer to a property‑instantiation without having identified a bearer of the property--for instance, when we don't yet have a view about what stinks, or when there isn't anything (that needs mentioning) that is raining.  But the 'it' in (1)-(3) seems quite unlike these--it purports to refer to the bearer of the property of being like something for c.  One symptom of this more substantive use of 'it' is that the predicate 'is like something' can be completed with 'to ...,' where the ellipsis completes a verb phrase, as in (1)--'it is like something to have M.'  We cannot say 'it is raining (for one) to [verb]'.  We can say 'it stinks (for one) to [verb],' but only when we mean that (one's) [verb]ing stinks.  As a comparison, the sentence 'it is fun (for one) to teach' is equivalent to '(one's) teaching is fun (for one).'  So (3) is equivalent to (4):
   (3)     There is something it is like for c, for c to have M.
   (4)     There is something c's having M is like for c.
As a final preliminary, note that there is nothing that 'there is' adds to the zing of (1)-(4), since (4) is clearly equivalent to (5):
   (5)     c's having M is like something for c.
Here is what we can glean from these preliminaries and the overarching requirements (α) and (β): whatever property (if any) being like something for c is, (1)-(5) express that it is possessed by c's having M, and necessarily so (given that c has M at all).  One of (1)'s attractions is that it  expresses this conceptual material so compactly and fluidly. 

'Is like' and 'something' 

            Let us turn finally to the peculiar phrase 'is like something for c.'  The 'something' plays an especially odd role.  It clearly functions as a variable, a placeholder, but over what does it generalize?  Not primarily "things" designated by noun phrases, but features specified by predicative phrases.  If asked what it is like to wrestle with a riddle, the adjectives 'interesting' or 'fatiguing' are better answers than the nouns 'interest' or 'fatigue.'  This use of 'like' in (1)-(5) mirrors a more widespread use described as follows in the 1971 Oxford English Dictionary (O.E.D.):

Some phrasal uses of the adj[ective] ['like'] in this construction ['is like'] have a special idiomatic force.  The question What is he (or it) like? means 'What sort of a man is he?', 'What sort of a thing is it?', the expected answer being a description, and not at all the mention of a resembling person or thing.  (Like, adj., A.1.b., L‑283)

If we were to try to express (5) in something more like logical notation than grammatical English, we would have to write 'is like some F'--using a predicate variable 'F'-- rather than 'is like some x'--using a term variable 'x.'  But it is only in rare cases that we can complete 'is like ...' with a predicate rather than a term, without losing grammaticality.  We can say a mental state 'is like interest and fatigue' but not that it 'is like interesting and fatiguing,' except when 'like' is, like, a mere interjection, as it clearly is not in (1)-(5).  However, we can say, with only a slight awkwardness, 'it is like something--interesting and fatiguing' or 'interesting and fatiguing are some things it is like.'  Let us record as follows the very odd pair of grammatical properties displayed by 'is like' and 'something' in (1)-(5):
   (a)     'something' is best specified by predicates, not terms, yet
   (b)     predicates typically cannot grammatically follow 'is like.'
I will return to the rare exceptions to (b) later, exceptions such as 'is like mad' and 'is like new.'  A further piece of evidence for (a) stems from the philosophical jargon of 'qualia' and 'phenomenal properties,' which defenders of the explanatory gap (and others) introduce by the following stipulation:
   (c)     qualia and phenomenal properties are what in particular it is like to have mental states.
Grammar aside, qualia and phenomenal properties are the particular "things" it is like to have certain mental states.  But qualia and phenomenal properties are, of course, properties--if they are anything at all.  These "things" are ways of being--values of 'F'--as opposed to ordinary beings--values of 'x'.  Somehow, despite the ungrammaticality, 'is like something' comes to 'is like some way' rather than 'is like some thing.'[6] 

            I believe that (a) and (c) are genuine semantic features that must not be lost in interpreting (5), but that (b) is a relatively unimportant accident of modern English syntax.  Although with the exception of 'is like mad' and 'is like new' we do not often say 'is like F'--not even 'is like sane' or 'is like old'--such phrases were once more common in English.[7]  It is no surprise that they would be rendered obsolete, given the dominant competing use of 'is like' for 'is similar to,' which demands completion by terms (for instance, nouns) rather than predicates (for instance, adjectives).  Yet while 'is like [adjective]' has lost its head‑on competition with 'is like [noun],' it lives on in the simple variant 'is [adjective]-like.'  The O.E.D. describes modern Scotch usage in ways that seem tantalizingly relevant to the present quest:

In Sc[otch] the suffix ['-like'] is added freely to almost any descriptive adj[ective], esp[ecially] those relating to mental qualities, conditions of temper, or the like; the general sense of the compounds is 'having the appearance of being --'.  (-like, suffix, 2.a., L‑287)

Modern (19^th‑ and 20^th‑century) examples given include "greedy-like," "grim-like smile," "square‑like room," "herbaceous-like shrub," "sublime-like beauty," "gluey-like material," and so on.  This usage is not only a survival of earlier Scotch usage,[8] but of a much more extended usage in English: in fact, the ubiquitous use of the suffix '-ly' for adverbs derives from the Middle English suffixes '-lik' and '-like,' as in modern English 'greedily' from Middle English 'gredilike.'[9]  Modern English also has a small number of survivors such as 'genteel-like' and 'humanlike.'  Frequency and breadth of use aside, the important point is that these constructions are all easy for the ordinary speaker to understand.  I believe that the best way to make sense of grammatical features (a)-(c) is to interpret (5) as (6):
   (5)     c's having M is like something for c.
   (6)     For some F, c's having M is F-like for c.
As quoted above, the O.E.D. identifies the general meaning of 'is [adjective]-like' as '[has] the appearance of being [adjective]' (-like, suffix, 2.a., L‑287).  Given this definition, (6) in turn is equivalent to (7):
   (7)     For some F, c's having M has the appearance of being F for c.[10]
To counter the application of the O.E.D.'s definition to (6), what is required is some better way of understanding '[adjective]‑like.'[11]  I think it is difficult to motivate a plausible rival interpretation.  Being F‑like presumably does not amount simply to being F; if it did, '‑like' would be idle and (6) would be trivially true ('for some F').  Nor does it seem to require being F; if it did, '‑like' would be entirely misleading.  Certainly there are uses of '‑like' that are not to be explained in terms of "appearances" but in terms of "similarity" more generally construed--for instance, for imperceptible entities such as "electron-like particles" and "Platonic‑Form‑like universals"--but in these cases '‑like' attaches to nouns rather than to adjectives.  A similarity construal of 'like' does not make good sense for '[adjective]‑like,' since c's having M may appear (to have the property) F but is hardly similar to (the property) F.  I discuss "similarity" interpretations at much greater length in section 3 below. 

            Of course, 'appear' and 'appearance' can have a variety of meanings, and much of the remaining task is to select the relevant one and stick with it without equivocating.  For example, I have not yet discussed whether "having an appearance" in the sense relevant to (1)‑(7) must be a matter of actually appearing or as a matter of being disposed (in some sense) to appear.  Nor have I discussed what kind of appearing is involved--for instance, perceptual appearance versus appearance in judgment, and phenomenal appearance versus nonphenomenal appearance.  Before addressing these issues, I want to support the invocation of (7) with some supplementary grammatical considerations. 

'For' and 'c' 

            A semantic virtue of the "appearance" reading (7) is that it makes sense of the conceptual role of 'for' in 'x is like something for c.'  Out of the 31 major uses of the preposition 'for' listed in the O.E.D., there are only two that make much sense in this construction: "in the presence or sight of" (For, prep., A.I.1.b, F‑409) and "as regards, with regard or respect to, concerning" (For, prep., A.IX.26, F‑412).[12]  There are good reasons for thinking "in the presence or sight of" is especially relevant.  First, if 'for' meant only "with regard to," we should expect that it could be followed by terms other than those for creatures.  A pain in my leg is "with regard to" me, but it is also "with regard to" my leg, my medicine cabinet (since it sends me hobbling there), and so on  Nevertheless, the pain in my leg is like something only for me; it is like something neither for my leg nor for my medicine cabinet.  This is explained on the "in the presence or sight of" reading of 'for': the pain in my leg is like something in the sight of me, but not in the sight of my leg or my medicine cabinet.[13]  Second, I think we can only understand the substantiveness (versus redundancy) of 'for'--the fact that it can call for emphasis, as Nagel shows--on the "in the presence or sight of" reading.  Given that c is the creature that has M, features of c's having M are trivially 'for' c in the "with regard to" sense--they are trivially "with regard or respect to" c.  Whatever being like something is, if one knows that a mental state of c's has that feature, one would learn nothing further by being told that this fact is 'for' c in the weak sense that it is "with regard to" c.  One would learn something very interesting, well worth stressing, if one were told that this fact is 'for' c in the sense that it is "in the presence or sight of" c.  Only in that case, as Nagel would urge, have we characterized c's point of view.  This protects (7) as an interpretation of (6) even if--somehow, contrary to the O.E.D.'s suggestion--'is F‑like' in (6) does not itself mean "has the appearance of being F."  We should still interpret the 'for c' in (6) strongly, as (6a):
   (6)     For some F, c's having M is F-like for c.
   (6a)   For some F, c's having M is F-like in the presence or sight of c.
On its own various readings, this is easily seen to be equivalent to versions of (7). 

            Another virtue of the "appearance" reading (7) is that it explains why there is no significant difference between 'is like' as used in (1)‑(5), and corresponding uses of 'feels like' or 'seems like.'  People do not ordinarily distinguish between "what it is like to go on a safari" and "what it feels like to go on a safari," or between "what it is like to be an accountant" and "what it seems like to be an accountant."  Likewise, it is hard to find a difference in meaning among (1) and two variants, as they would ordinarily be understood:
   (1)     There is something it is like to have M.
            There is something it feels like to have M.
            There is something it seems like to have M.
It might be supposed that the "feels" and "seems" claims favor actual appearing rather than mere dispositions to appear, while the "is" claim is more neutral.  But in fact 'feels' and 'seems' can be used dispositionally--'the dark side of the moon feels [/seems] cold' can be true, even when nobody's there--and in fact (as I will argue below) the "is" claim has further grammatical features that disguise an actuality (versus dispositionality) requirement.  Similarly, there is no difference among (5) and two variants:
   (5)     c's having M is like something for c.
            c's having M feels like something for c.
            c's having M seems like something for c.
In these cases 'feels' is used not specifically for "appears to the touch" but more broadly.  This is what allows Chalmers to characterize conscious experiences as those with a "feel," as quoted above (see section 1) and as in his statement (1996, 11): "On the phenomenal concept, mind is characterized by the way it feels; on the psychological concept, mind is characterized by what it does."  Although there is something loose about saying one "feels" one's mind, 'feels' serves better than 'looks' ('sounds,' 'smells,' and so on) because it is used for proprioception.  Proprioceptively feeling one's body serves as an intuitive analogy for introspectively "feeling" one's mind because neither process detours noticeably through the environment.  Also, in these cases 'seems' is not used broadly for "is judged to be"--perhaps due to testimony or wild speculation.  Both 'feels' and 'seems' are best rendered as "appears," with its general perceptual connotations.  As the O.E.D. explains, 'is like' is on occasion used in place of 'looks like'--more generally, in place of "certain idiomatic uses [of 'like'], chiefly with the [verbs] feel, look, sound," (Like, adj., A.7, L‑284)--and "means 'to have the appearance of being' so‑and‑so" (Like, adj., A.1.b, L‑283).  So for reasons independent of (6) or (6a), in addition to those based on the routes through (6) and (6a), (5) is equivalent to (7):
   (5)     c's having M is [that is, appears] like something for c.
   (7)     For some F, c's having M has the appearance of being F for c. 

            Let us now consider what sense of 'appear' in (7) best accounts for the relevant sense of 'is F-like' in (6).  I think this is a very demanding sense.  Notice, first, that the 'for c' at the end of (6) is detachable in inference.  Just as (2) entails (1), there is an entailment from (6) to (6b):
   (6)     For some F, c's having M is F-like for c.
   (6b)   For some F, c's having M is F-like.
In order for x to be F-like--for a creature to be humanlike, for a smile to be grim-like, for a car to be like new, and so on--it is not enough that there be (dispositions to) judgments that x is F--that the creature is human, that the smile is grim, that the car is new, and so on.  Some irrelevant (dispositions to) judgments are based solely on rumor, lies, or wild guesses about x.  As ordinarily understood, rather, x must be perceivable as F--the creature must be perceivable as human, the smile must be perceivable as grim, the car must be perceivable as new, and so on.  But even this is not demanding enough: x is not F‑like if it is perceivable as F only by people with radically defective or abnormal perception.  Of course, it is not required that x actually has the feature F--humanlike creatures needn't be human, grim‑like smiles needn't be grim, cars that are like new needn't be new, and so on--but in some middling sense the appearance as of F-ness must be objectively x's own.  The F-like thing must have a range of easily perceptible features in common with an F thing, enough so that in conditions fairly good for perceiving whether something is F, someone who (nondefectively and normally) perceives these features would to some degree be disposed to perceive x as F, or at least would to some degree be disposed to believe x to be F if the perceptions were taken at face value.  The requirement is partly counterfactual; x may be F-like even when it is not actually perceived at all.  We can summarize all this by construing 'x is F‑like' as 'due to easily perceptible features of x's own, x is perceptible as F,' or, more briefly, 'x itself has a perceptual appearance of being F.'  In the general case, the corresponding use of 'x is like something' would be equivalent to 'due to easily perceptible features of x's own, x is perceptible as being some particular way' or 'x itself has a perceptual appearance of being some particular way.'[14]  

            Notice, second, that we can easily append to 'is ... like ... for c' in (2)-(6) a restriction such as 'but not for anyone else.'  This is one potentially relevant difference between the use of 'is [adjective]‑like' for conscious experience and the use of it elsewhere--for humanlike creatures, grim‑like smiles, cars that are like new, and so on.  In the latter cases the most natural qualifiers restrict the ease of perception to a group (a car may be like new for consumers but not for mechanics, a smile may be grim‑like for children but not for adults, and so on), while for an experience the most natural qualifier restricts the ease of perception to an individual (the bearer of the experience).  This presents a difficulty of interpretation.  By comparison, what would it mean to say 'the smile is grim-like for c but not for anyone else'?  It can mean that only c can be fooled, because of some defect in c.  Or less plausibly it can also mean that c uses some abnormally keen perceptual faculty that no one else in similar circumstances would share.  But when it does mean either of these things, it is incompatible with the detached claim 'the smile is grim‑like.'  As described in the previous paragraph, the simplified claim requires a kind of objectivity and reference to normal perception.  What would it mean, then, to say 'the smile is grim‑like for c but not for anyone else' in a sense compatible with the objective 'the smile is grim‑like'?  Here, I think, it would be meant that the smile actually perceptually appears only to c as grim (or actually appeared so, with the perceptual memory intact in c), although if others could perceive (the relevant properties of) the smile, they too would perceive it as grim.   

            Likewise, since 'c's having M is like something' is compatible with 'c's having M is like something for c but not for anyone else,' this favors the stronger reading that c's having M actually perceptually appears or (memorably) appeared to c as being some way.  The present‑tense "appears" requirement is favored over the more relaxed "appears or appeared" requirement by an element in (1)‑(7) that is absent in the grim‑like-smile case, namely, the overarching requirements (α) and (β), that c now have M and that the appearance be necessary in order for c to have M.  The persistence of a perceptual memory of having had M, after c no longer has M, could not normally be necessary in order for c to have M.  The requirement is more stringent for experience than for other cases, since a creature's being humanlike is not required for the creature's existence, a smile's being grim-like is not required for the smile's existence, a car's being like new is not required for the car's existence, and so on.  The upshot of all this is that (1)‑(7) are equivalent to (8):
   (7)     For some F, c's having M has the appearance of being F for c.
   (8)     c's having M itself perceptually appears some way to c.
Given (α) and (β), this should be understood as:
   (9)     c's having M itself now perceptually appears some way to c, and this is in order for c to have M. 

            Finally, the 'c' in 'is like ... for c' (and 'appears ... to c') seems to express an additional requirement, which can be reached by emphasizing that it means "c as a whole" rather than "some small part of c."  This sort of requirement is widespread in contemporary discussions, following a suggestion of Gareth Evans that "conscious perceptual experience ... serves as the input to a thinking, concept-applying, and reasoning system," so that "we can say that the person, rather than just some part of his brain, receives and possesses the information" (1982, 158).  Evans' requirement is designed to apply to persons, but perhaps some similar requirement could extend to creatures more generally.  Much more needs to be said about the proper construal of 'as a whole' (see section 5 below), but for now let us record it by construing (9) as equivalent to:
  (10)    c's having M itself now perceptually appears some way to c‑as‑a‑whole, and this is in order for c to have M.
It should be admitted that the appeal to "wholeness" can only be maintained somewhat vaguely.  In biological creatures, probably no mental process, and certainly no particular mental state, is avaliable to the "whole" of a creature.  Some processes and states within a perceptual subsystem may be unavailable to processes of reasoning, and so on, but equally, processes and states within a reasoning subsystem may be unavailable to processes of perception.  It is important to recognize some such vague requirement, however, in the spirit of extracting every ounce of content from the phrase 'something it is like for c.' 

Results 

            (10) is my stopgap analysis of the very concept of there being something it is (now) like for a creature c (simply) to have a feature M.  It gives expression to a number of requirements or "specs" for phenomenality:
   (A)    c has M.
   (B)    c mentally represents c's having M as being some way.  [This is part of "appearing."]
   (C)    This representing is necessary for c to have M rather than added to c's having M.
   (D)    This representing is in perception ["an appearing"] rather than merely in (disposition to) judgment.
   (E)    This representing is actual rather than merely counterfactual.
   (F)    This representing is present‑tensed rather than merely in past‑tensed memory.
   (G)    This representing is due to perceptible features of c's having M ["itself"].
   (H)    This representing is for c as a whole rather than for a small part of c.
I have tried throughout to take (1) and its variants in the very strongest ways suggested by their grammar.  There may be weaker readings of (1) that require a subset of (A)‑(H), but at best these yield unclear cases of phenomenality.  At any rate, (A)-(H) are especially appealing because they give expression to the idea, emphasized without end by defenders of the explanatory gap, that conscious experience is subjective rather than objectively independent of a subject's access to it.  If c's having M satisfies (A)-(H), then its being requires its being perceived, in the strongest sense.  This is further evidence that we have not "missed the point" or "defined consciousness away" in analyzing the concepts expressed by (1).[15]   

            The central task ahead is to investigate whether specs (A)‑(H) can be explained satisfactorily by a scientific model.  A satisfactory explanation of (A)‑(H) would explain the truth of (10), which in turn entails (9), which entails (8) ... and so on counting down to (1) itself:
  (10)    c's having M itself now perceptually appears some way to c‑as‑a‑whole, and this is in order for c to have M.
   (9)     c's having M itself now perceptually appears some way to c, and this is in order for c to have M.
   (8)     c's having M itself perceptually appears some way to c.
   (7)     For some F, c's having M has the appearance of being F for c.
   (6)     For some F, c's having M is F-like for c.
   (5)     c's having M is like something for c.
   (4)     There is something c's having M is like for c.
   (3)     There is something it is like for c, for c to have M.
   (2)     There is something it is like for c to have M.
   (1)     There is something it is like to have M.[16]
Using this stopgap analysis we would be able to deduce that there is something it is like to have M, from whatever scientific premises are needed to explain (A)‑(H).[17]  Of course, if an explanation of (A)‑(H) is to yield a satisfactory explanation of (10)‑(1), the term 'appear' must bear the same sense throughout the explanation and throughout the stopgap analysis (as must its cognates: 'appearing,' 'appearance,' and so on).  I consider this issue in sections 4 and 5, and then procede with the search for a scientific account of (10) in section 6.  But first, in the following section, I try in other ways to motivate (10)-(6) as an interpretation of (5)-(1), by considering and rejecting rival interpretations of 'is like something.'  In the process I will describe several further virtues of my interpretation, but readers who are not tempted by the rivals, or who are averse to the technicalities of section 3, may proceed to section 4 at any point. 

3     Scrap 

            Leaving aside uses of 'like' related to the verb 'to like,' the O.E.D. distinguishes two families of uses: as an adjective modifying terms--where 'x is like y' means 'x is similar to y' or 'x resembles y,' and as an adverb modifying predicates--where 'x has F like y' means 'x has F in the manner of y' or 'x has F as y has F' or 'x has F as if x = y.'  Judging solely from what precedes 'like' in (1)‑(5)--'c's having M is ...' (or 'it is ...')--either of these uses could be in play.  'Is like ...' could be an adjective modifying 'c's having M,' so that 'c's having M is like ...' would come to 'c's having M is similar to ....'  Or 'like ...' could be an adverb modifying the being of c's having M, so that 'c's having M is like ...' would come to 'c's having M exists like ...' and therefore to 'c's having M exists in the manner of ....'  Neither avenue is particularly promising, however.  I will discuss the adjectival "is similar to" proposal; the results would apply straightforwardly to the adverbial "in the manner of" proposal.  

            If instead of moving from (5) to (6) we were to try reading 'is like' literally, as 'is similar to,' we would get (5a), an implausibly weak construal of (5):
   (5)     c's having M is like something for c.
   (5a)   c's having M is similar to something for c.
One sign that this is a bad reading of 'is like' is that it does not make sense of the optional detachability of 'for c' described above. The detached claim (1) can easily be heard as elliptical for (2), as the detached (6b) can easily be heard as elliptical for (6).  By contrast the detached version of (5a) is utterly trivial, and does not sound at all elliptical for (5):
   (5b)   c's having M is similar to something.
Each thing is similar to something or other, and in fact, each thing is similar in some respect or other to each other thing.[18]  So in this weak sense each of one's mental states "is like something," no matter how far it may be removed from conscious experience.   

            Of course, for similarity claims relating two specific kinds of entities, there are pragmatic implicatures about relevant respects of similarity.  But there is no such pragmatic substance under generalization, when one of the "kinds" is "something".  By contrast, the interesting thing about 'is ... like' in (1)-(6) is that it maintains substance despite the generality of 'something,' even in detached claims such as (1) and (6b).  This is explained quite simply on the "perceptually appears" interpretation I have given of (5):
   (8)     c's having M itself perceptually appears some way to c.
On my account (5) remains substantive despite the generality of 'something' because perceptually appearing some way is substantive in a way that being similar to something is not.  

            While this detour through the detached (5b) is telling, it is perhaps more important for an interpretation to succeed with undetached claims such as (5a).   Even though a "similarity" reading is trivial for (5b), perhaps it is not for (5a).  Perhaps some "hidden" mental states of mine--latent memories, tacit knowledge, Freudian hatreds, Heideggerian anxieties, subliminal perceptions--fail to be conscious experiences because they are not similar to something for me.  Presumably to ask whether they are similar to something "for me" is to ask whether they are similar to something to me, that is, whether I believe they are similar to something.  Well, since I believe everything is similar to everything else, as soon as I so much as guess that these states exist--say, by reading Freud, Heidegger, or Chomsky uncritically--I believe they are similar to something.  This does not raise them to the status of conscious experiences, however.[19]  (5) should not be interpreted literally, as (5a).  By contrast, my interpretation (8) can explain why there is nothing it is like for c to have these "hidden" states, even when c believes in them via testimony or wild guessing.  It is not even intuitively tempting to think they are perceptible by c. 

            It is unsurprising that 'is like' in (1)‑(5) should be taken nonliterally, given the grammatical peculiarities described in the previous section:
   (a)     'something' is best specified by predicates, not terms, yet
   (b)     predicates typically cannot grammatically follow 'is like.'
As the O.E.D. recognizes, (a) and (b) are signs that 'is like' is "idiomatic" here, and cannot easily be interpreted as the literal 'is similar to' (Like, adj., A.1.b, L‑283; quoted above).  If we try to respect (a) directly--allowing that 'something' generalizes over ways of being--this literal interpretation yields logical gibberish.  (5) would become the absurd (5c):
   (5)     c's having M is like something for c.
   (5c)   c's having M is similar to some way for c.
A state (event, process, ...) such as c's having M is at best only abstractly and unintuitively "similar" to any feature or "way" itself.  There are no gibberish problems affecting my interpretation (8), since 'x perceptually appears some way' is perfectly intelligible in a way 'x is similar to some way' is not.  A similarity reading can avoid (most of) the gibberish as follows:
   (5d)  c's having M is similar to something that is some way, for c.
However, this falls back into utter triviality, since everything is "similar to something that is some way."   

            Further difficulties arise from the widespread philosophical stipulation mentioned in the previous section:
   (c)     qualia and phenomenal properties are what in particular it is like to have mental states.
Given that the ways specifying 'something'--values that potentially determine what it is like for c to have M--are supposed to be troublesome "qualia," there is no way to rescue a literal "is similar to" reading of (5) as in (5d).  My present visual experience of this page has the feature of being in Ann Arbor.  It is therefore similar to something that is in Ann Arbor--in fact, to a great many such things.  But being in Ann Arbor is not a quale; it is not something that my experience "is like" in the relevant sense.  Even if I know that my experience is in Ann Arbor, this feature is not part of what my experience is like "for me" in the relevant sense.  Not all the ways of a conscious experience are qualia, nor even all the ways that are necessary in order for the experience to exist--being self‑identical, being part of a functioning mind, and so on. 

            We cannot even rescue an "is similar to" reading of (5) by building in a restriction to similarity in phenomenal respects (or in qualia), yielding (5f):
   (5)     c's having M is like something for c.
   (5f)   c's having M is similar in phenomenal respects to something that is some way, for c.
First, of course, this would lead to circularity, since defenders of the gap seek to characterize the philosophical jargon 'phenomenal' and 'qualia' in terms of "what it is like," as in (c) from the previous section.  Even if this difficulty were waived, because 'what it is like' and the philosophical jargon are somehow to be characterized jointly, this reading of 'is like' would sever even a correlation between "what it is like" and the philosophical jargon.  Suppose that Q is a legitimate quale, and that Q is "something" my present visual experience of this page is like.  If (5f) is a good reading of (5), then (5f) should apply to my experience with 'Q' substituted for 'some way.'  This is the case--an experience with quale Q is similar in phenomenal respects to something that is Q.  However, my experience with quale Q is also similar in phenomenal respects to something that is Q and in Ann Arbor (or Q and part of a functioning mind).  Nevertheless, being Q and in Ann Arbor is not something my visual experience is like in the relevant sense, not even if I know my experience has the feature.  So not even (5f) is a good interpretation of (5).  (6)‑(10) by contrast can explain why such features as being in Ann Arbor and being Q and in Ann Arbor are not qualia.  Consider that a car that is like new is not like new and in Ann Arbor, even if it is in Ann Arbor, or is nearly in Ann Arbor, and even if everyone knows this.  This is because the car perceptually appears to be in Ann Arbor not primarily due to easily perceptible features "of its own," but also due to perceptible features of Ann Arbor.[20]  Likewise, if my experience appears to be in Ann Arbor (or in a functioning mind) this is not primarily due to perceptible features of its own. 

            If an "is similar to" reading has any remaining plausibility, I think this is because we so easily construe similarity specifically as perceptual similarity--a construal that is even easier for 'resembles'--as in the following potential interpretations of (5):
   (5g)   c's having M is similar in perceptual appearance to something that is some way, for c.
   (5h)   c's having M perceptually appears similar to something that is some way, for c.
For potential purposes of stopping up the explanatory gap, such readings of (5) would be close enough to my own "perceptually appears" reading (8), which I derived independently via (6)-(7):
   (8)     c's having M itself perceptually appears some way to c.
If anything, (8) sets the standard for scientific explanation higher than (5g)‑(5h), since these can be read as appealing merely to counterfactual appearings, while (9) appeals to actual ones.  Neglecting this difference, the three seem equivalent.  To the extent that similarity is most frequently understood, ordinarily, as perceptual similarity, this gives independent support to my stopgap analysis of (1). 

            The strength of 'for c' (described in the previous section) yields another way to protect (7)‑(10) as an interpretation of (5), even if--contrary to my argument from grammatical peculiarities (a)‑(c)--(5) is equivalent not to (6) but to the literal:
   (5a)   c's having M is similar to something for c.
Even if so, we should still interpret the 'for c' in (5a) strongly, yielding:
   (5i)    c's having M is similar to something, in the presence or sight of c.
Leaving aside the counterfactual/actual contrast, this is equivalent to (8).  On the extremely plausible assumption that entities do not appear "similar" unless they each appear some (shared) way, for (5i) to be true c's having M must appear some way to c, so (5i) entails (8).[21]  (8) in turn entails (5i), on the extremely plausible assumption that if x appears some particular way, it appears similar to other things that would appear that way.[22]  So (8) would be equivalent to (5) even if--contrary to what I maintain--(6) is a red herring. 

4     A trap 

            As I mentioned at the end of section 2, in order to stop up the explanatory gap (10) must involve an unambiguous sense of 'appear,' one that is both relevant to conscious experience and explainable scientifically.  According to defenders of the explanatory gap, ambiguity is often the crucial deficiency in attempts to explain conscious experience.  To be on the safe side, then, let us consider how these objections run. 

            Chalmers says that Daniel Dennett (1991) "claims (in effect) that what needs to be explained is how things seem, and that his theory explains how things seem" (1996, 370).  Chalmers objects as follows:

This is an elegant argument, with a ring of plausibility that many reductionist arguments about consciousness lack.  But its elegance derives from the way it exploits a subtle ambiguity in the notion of 'seeming,' which balances on the knife edge between the phenomenal and psychological realms.  There is a phenomenal sense of 'seem,' in which for things to seem a certain way is just for them to be experienced a certain way.  And there is a psychological sense of 'seem,' in which for things to seem a certain way is for us to be disposed to judge that they are that way.  It is in this first sense that a theory of experience must explain the way things seem.  But it is in the second sense that Dennett's theory explains it.  (1996, 190)

I am not concerned here with whether Chalmers' interpretation and criticism of Dennett is correct.  I wish to distance my approach from Dennett's, and to explain how I do not slip between two senses of 'appear' (or of 'seem'), whether "phenomenal" or "psychological" or otherwise. 

            Although the results I have enshrined in specs (A)‑(H) depend heavily on the use of 'appear,' and although this is similar to Dennett's term "seem," the similarities end very quickly.  First, Dennett seeks to explain how things seem--the things comprising the typically nonmental subject matter of experience--while I wish to explain how experience seems (or appears).  I am promoting a kind of inner "theater" model that Dennett has argued most forcefully against (see my 1994 for a defense of theater models from Dennett's attacks).  Second, as is already emphasized in (D), I do not give pride of place to mere (dispositions to) judgments, in the way that Dennett does.  So I definitely do not trade on Chalmers' "psychological" sense of 'seem' (or 'appear').  I do not even assume that this is a bonafide sense of 'appear'--I may judge that the universe is finite, but it does not in any sense I care about appear finite, and at the moment I am disposed to judge of any ripe banana that it is yellow, but at the moment none appears yellow to me.  My sense of 'appear' is far more demanding than the psychological sense of 'seem' that Chalmers finds in Dennett. 

            Given my reliance on the notion of perceptual appearings, which have subject matter or representational content, my view has some affinities with a view that Chalmers calls "representationalism": 

A recently popular position ... has been that phenomenal properties are just representational properties, so that yellow qualia are just perceptual states that represent [that is, perceptual properties of representing] yellow things, or something similar.  Most often, the suggestion is combined with a reductive account of representation (usually a functional or teleofunctional account) ....  (1996, 377)

One difference between my position and ordinary representationalism is that my crucial focus is not on the (perceptual) representing of nonmental "yellow things" but on the (perceptual) representing of representings (that is, of what I will argue are experiences) of yellow things.  The detailed sense in which my position is representationalist will only become explicit in section 7.  For now, what matters is the challenge of avoiding equivocation.  Chalmers' overall criticism of representationalism echoes his criticism of Dennett:

The surface plausibility of some representationalist accounts may well arise from a slide between inflationary and deflationary readings of 'representation,' where the second is a purely functional (or teleofunctional) notion, but the first is not.  The link between phenomenology and representation is made plausible on the first reading, but the reduction of representation is made plausible on the second.  (1996, 377)

Again, my concern is not with the merit of this diagnosis of previous representationalist theories.  I agree that (A)‑(H) must be explained with 'appear' in precisely the same sense that makes (7)‑(10) justified construals of (1)‑(6), to avoid charges of missing the point: 
   (6)     For some F, c's having M is F-like for c.
   (7)     For some F, c's having M has the appearance of being F for c. 

            What then is the sense of 'appear' that makes (7)‑(10) plausible as a reading of (6)? My answer is: whatever sense generally makes 'x appears F' a plausible reading of 'x is F‑like.'  What sense is that?  It is certainly not Chalmers' alleged psychological sense of 'seem' (or 'appear').  We do not understand 'the creature is humanlike,' 'the smile is grim‑like,' and 'the car is like new' in any literal way we might understand 'there is a disposition to judge that the creature is human' or '... that the smile is grim,' or '... that the car is new.'  A devious or wildly guessing rumor mongerer (or a used‑car dealer on television) can bring about the truth of the latter statements without bringing about the truth of the former statements.  The relevant sense of 'appear' is, as Chalmers would expect, much more "inflationary."  It is demanding in ways expressed in specs (D)‑(F), requiring actual, present‑tensed perception.  Call this the "perceptual" sense of 'appear' (or 'seem').[23] 

            Although the perceptual sense is much more stringent than Chalmers' alleged psychological sense of 'seem,' it does not involve his alleged phenomenal sense of 'seem.'  The concept of there being something it is like is not presupposed (individually) by the concepts of actuality, present‑tensedness, or perception.  I assume this is obvious for the concepts of a state's being actual and being present‑tensed, but perhaps I need to argue that a state can (as a matter of conceptual possibility) be perceptual without being phenomenal.  The intuitive conceivability of perceptual states it is like nothing to have is supported by actual cases of wholly subliminal perception and "blindsight," and of "early" states in processing in the retina, lateral geniculate nucleus, and (perhaps) primary visual cortex.  I do not need to establish here (though in fact I believe) that in some cases there is nothing it is like to have such states, but I do claim that we conceive of them as perceptual without knowing or caring whether there is--and so we speak of subliminal perception, blindsight, primary visual cortex, and so on.[24]  (A critic might emphasize 'blindsight' rather than 'blindsight,' hoping to deflate the idea that this is perceptual.  But in fact I am thrilled to emphasize 'blind.'  Blindsight is not sight in which one is blind--that would be a conceptual contradiction--but sight to which one is blind--this contrasts nicely with phenomenal seeing, if I am right that phenomenal seeing requires perceptually‑apparent‑to‑one seeing.  I discuss blindsight--and the possibility of inner blindsight--more fully in the next section.) 

            The broadly perceptual (rather than any strictly phenomenal) sense of 'appear' is what is relevant to the 'is F‑like' cases: humanlike creatures, grim‑like smiles, like‑new cars, and so on.  For a used car to be like new, it is enough for it perceptually to appear new, whether it does so because of wholly subliminal perception or full‑fledged conscious experience.  As any psychologically savvy used‑car dealer must know, the more subliminal perception involved in the appearance of newness the better (and, indeed, truer) the claim that the car is "like new."  Cars could be like new even for beings as much like us as possible without having phenomenal consciousness--though I do not say they would be philosopher's "zombies" exactly like us in nonphenomenal ways.  The same holds for humanlike creatures, grim‑like smiles, and so on.  At no point along the way to (10) does the stopgap analysis of (1) introduce phenomenal concepts, circularly, and at no point does the power of the arguments depend on implicit assumptions of phenomenality. 

            "But surely there must be a double use of 'appear' somewhere, if it is to forge a link between raw material and raw feels!"  Absolutely.  I do not use 'appear,' fallaciously, with a double sense (meaning, intension, concept, ...)--here with one sense, there with another.  But I do use 'appear,' nonfallaciously, with a double object--here one class of things does the appearing, there another class of things does the same kind of appearing.  First, I appeal to scientific explanations of what it is for arbitrary objects and events (actually, present‑tensedly, and perceptually) to appear.  This appeal certainly does not all by itself explain why there is something it is like to have certain mental states.  Such appearings can and do exist without there being anything they are like.  But second, holding unequivocally to this scientifically explainable sense of 'appear,' I apply it not to arbitrary objects and events but (in ways to be described in section 6) to appearings themselves.  There is only one sense of 'appear' in my account, the perceptual sense.   

            If the sense in which an experience 'appears' to one is to be exactly the same as the nonphenomenal sense in which a car 'appears' to one, without room for a conceptual residue, the two relations had better be commonsensically practically indistinguishable except for the difference in relata.  If we could tell the two relations apart introspectively, or through any other means easily available in commonsense thinking, and if this difference were of any practical concern to us (aside from the difference in relata), then it would be likely that we would have different concepts for them, different senses of the word 'appear.'  To insure that we do not cheat by assuming phenomenality or equivocating in any other fashion, then, let us for emphasis include a conceptual requirement (I) in addition to specs (A)‑(H):
   (A)    c has M.
   (B)    c mentally represents c's having M as being some way.
   (C)    This representing is necessary for c to have M rather than added to c's having M.
   (D)    This representing is in perception rather than merely in (disposition to) judgment.