Return to Carruthers' books


9          Dispositionalist higher-order thought theory (2): feel



In this chapter we come to the crux. I shall examine how the competing higher-order accounts can explain the defining feature of phenomenal consciousness – namely its subjective feel, or ‘what-it-is-likeness’ – and I shall give a final adjudication between them on this ground. I shall argue that each of the two forms of higher-order thought (HOT) theory – in contrast with higher-order experience (HOE) theory – can advance essentially the same (fully successful) reductive explanation of phenomenal consciousness. In which case, given the strength of the arguments which we were able to deploy in chapter 8 on behalf of dispositionalist forms of HOT theory, it is the latter which should emerge as the overall winner.


1          HOE theory and feel

I argued at some length in chapter 8:1 that higher-order experience (HOE) theory is highly implausible, from both functional and evolutionary perspectives. In the present section I shall consider whether those disadvantages might nevertheless be outweighed by the capacity of HOE theory to explain the subjective feel of experience. My answer will be negative.


1.1       The transparency objection

An objection which is often made against HOE theory is as follows (e.g. Dretske, 1995). If HOE theory were true, then how is it that there is no phenomenology distinctive of inner sense, in the way that there is a phenomenology associated with each outer sense? Since each of the outer senses gives rise to a distinctive set of phenomenological properties, you might expect that if there were such a thing as inner sense, then there would also be a phenomenology distinctive of its operation. But there does not appear to be so.

            This is the point about the ‘transparency’ of our perceptual experience, again. Concentrate as hard as you like on your ‘outer’ (first-order) experiences – you will not find any further phenomenological properties arising out of the attention you pay to them, beyond those already belonging to the contents of the experiences themselves. Paying close attention to your experience of the colour of the red rose, for example, just produces attention to the redness – a property of the rose.

            Put like this, however, the objection just begs the question in favour of first-order (FOR) theories of phenomenal consciousness. It assumes that first-order – ‘outer’ – experiences already have a phenomenology independent of their targeting by a HOE. But this is just what a HOE theorist will deny. And then in order to explain the absence of any kind of higher-order phenomenology, a HOE theorist only needs to maintain that HOE-states are not themselves ever conscious in the same way.

            While HOE-states may be targeted by HOTs, in introspection (as we noted in chapter 8:1.1), and hence become access-conscious as a result, there is no third-order faculty of inner sense, charged which scanning the outputs of our second-order inner sense, in virtue of which the latter could become phenomenally conscious. We have a second-order faculty of inner sense, which scans the analog outputs of the various outer senses, and in virtue of which the latter are phenomenally conscious; but we have no faculty which is in turn charged with scanning the outputs of our faculty of inner sense. Or so a HOE theorist can (and should) maintain.

            It may well be that a variant of the original objection can return, however, under the guise of a challenge to the HOE theorist to advance any sort of positive explanation of the subjective feel of phenomenally conscious experience. Just how is it that targeting our first-order, non-phenomenal, perceptual states with a set of higher-order scanners can give rise to the distinctive features of phenomenal consciousness – subjective feel, ‘what-it-is-likeness’ and the rest – while preserving transparency? This challenge will be developed in the sub-section which follows.


1.2            Explaining feel, and the transparency objection again

HOE theory can, of course, draw a formal distinction between conscious and non-conscious experience, claiming that analog states are phenomenally conscious when and only when they are targeted by inner sense. But why should states which are targeted by inner sense be like anything to undergo? Why should such states possess any sort of subjective feel? Can we genuinely explain why our conscious experiences have a subjective ‘what-it-is-likeness’ to them, in terms of the operations of inner sense?

            It might seem, on the face of it, that HOE theory is well placed to provide such an explanation. For just as our outer perception gives the world a subjective aspect, then so, too, may inner perception give our experience a subjective aspect. Just as features of the world (and states of the body) take on a subjective aspect when targeted by a set of sensory systems with a particular set of sensitivities and powers of discrimination, so too – it might be said – our own experiences of the world (or of states of our own bodies) take on a subjective aspect when targeted by inner sense.

            And then the phenomenally conscious differences between distinct phenomenally conscious states can be said to reside in the different higher-order analog contents involved in each case. The difference between the feel of a phenomenally conscious percept of red and the feel of a phenomenally conscious percept of green will reside in the differences between the higher-order analog contents which are generated when first-order percepts with the analog contents reda and greena are targeted by inner sense.

            But now the problem of transparency returns with a vengeance. If the sort subjectivity produced by inner sense were really like the sort of subjectivity of the world produced by the operations the first-order senses, then it is mysterious how our phenomenally conscious experiences could have the property of transparency. If inner sense picks up on, and represents in a particular manner, properties of our experiences – in the sort of way that our outer senses pick up on and represent in distinctive ways some of the properties of the world – then surely we would expect there to be a distinct (non-worldly) set of properties of phenomenally conscious experiences on which introspection could concentrate. But there is not. For to repeat, concentrating on your experience of red just is concentrating more closely on the redness represented. The explanatory potential of HOE theory is therefore weak.

            The best way to develop and drive home this objection is to recall a point made briefly in chapter 8:1.2. This was that the set of inner scanners postulated by HOE theory would have to operate by detecting the physical events in our brains which serve to realise our phenomenally conscious experiences. Those scanners would be physical systems charged with detecting physical activity of various sorts, and with computing, on that basis, the properties of the experiences realised in that activity. It is very hard to see how any inner scanner could detect experiences as such, or how it could simply co-opt and re-use the content of a first-order experience into a second-order one.

            This makes it difficult to see how the results of inner scanning could fail to have some sort of ‘excess content’ over and above the contents of the first-order experiences targeted. How could the particular sensitivities of the scanning devices, together with the computational principles on which those devices work, fail to have an impact on the second-order analog contents which result?


2          Actual HOTs and feel

Actualist higher-order thought (HOT) theory, too, is implausible on functional and evolutionary grounds – or so I argued at length in chapter 8:2. Might these difficulties for actualist HOT theory be outweighed by the explanatory power of the latter? In particular, can actualist HOT theory advance any convincing explanation of the subjective ‘what-it-is-likeness’ of experience?

            On the face of it the prospects do not look especially promising. For why should an analog, but non-conscious, perceptual representation suddenly acquire the subjectivity distinctive of phenomenal consciousness merely because it causes a higher-order belief about itself? How can the mere fact that I have non-inferential knowledge of the occurrence of a certain experience make it the case that there is suddenly something which it is like to undergo that experience? And how can the phenomenally conscious differences between distinct phenomenally conscious states be explained, on this account? For it looks as if those differences can only be differences between the contents of the first-order states targeted by HOTs in each case. The difference between having a HOT that I am undergoing a state with the analog content reda, on the one hand, and having a HOT that I am undergoing a state with the content greena, on the other, can only reside in the differences between the first-order contents reda and greena – but these differences are already there in the non-conscious perceptual states which become targeted! How can the attachment of a HOT to each of those contents render them, at the same time, as phenomenally distinct?


2.1            Rosenthal’s response

Rosenthal (1998) considers some of these questions, and responds by means of an indirect argument. He points out that there seem to be cases where an increase in the range of HOTs of which subjects are capable results in a corresponding change in the phenomenal properties of their experiences. In which case it is not unreasonable to suppose that the phenomenal properties of experience, in general, result from targeting by some suitable HOT.

            The example he uses is wine-tasting. To a beginner, all wine just tastes like wine, with a phenomenology not too distinct from that of drinking sweetened vinegar, and with little differentiation between the tastes of different kinds of wine and different vintages. But as one learns more about wine – and in particular, as one acquires a new set of phenomenal concepts with which to categorise one’s experiences – the phenomenal properties of those experiences are transformed, becoming much richer and more varied. This gives us some reason to think, Rosenthal claims, that phenomenal consciousness, in general, is a product of the targeting of experience by higher-order thoughts and concepts.

            I have two points to make about this. The first is that even if Rosenthal’s argument were sound, it would not really succeed in making the subjective properties of phenomenally conscious experience any the less mysterious. For providing us with reason to think that phenomenally conscious properties are caused to exist under certain circumstances is not at all the same thing as reductively explaining them. Remember that our goal is to find a naturalistically acceptable reductive explanation of what it is for a percept to possess those properties. Our task is to explain, not when an experience is caused to become phenomenally conscious, but rather what phenomenal consciousness itself actually is, or what constitutes it. And we want to know in virtue of what the phenomenally conscious differences between phenomenally conscious states are distinct. These questions simply are not, and cannot be, addressed by the sort of indirect argument Rosenthal proposes.

            My second point is that Rosenthal has in any case misdescribed his own example, in a way which renders his argument unsound. For the new concepts acquired by a nascent wine-taster are concepts of properties of the wine itself (smooth, sweet, fruity, etc.), rather than higher-order concepts picking out properties of our experience of the wine – those concepts are applied to the wine, rather than to our experiences thereof. And the phenomenon is a perfectly general one, which we have already had occasion to note in chapter 5:3.1 – namely, that the acquisition of new (first-order) concepts can transform the similarity spaces in perception. When I acquire new concepts, items which previously seemed similar will now strike me as dissimilar; and the phenomenal properties of my experiences will change too, of course. But there is no reason to think that this is primary. Rather, when the similarity spaces in my first-order percepts undergo a shift as a result of new (first-order) concept acquisition, the phenomenal feels of my experiences are changed as a result.


2.2            Coinciding explanations

I do believe, in fact, that actualist HOT theory can advance a far better explanation of the subjective feel of experience than this – indeed, that it can provide an explanation which is actually successful. But this will only be by embracing essentially the same explanation as is offered by dispositionalist HOT theory, grounded in some form of consumer semantics (see sections 3 and 4 below). This will not, then, provide any distinctive advantage of actualist over dispositionalist forms of HOT theory.

            Indeed, I claim something a good deal stronger. Once we realise that consumer semantics only requires dispositions to make judgements or inferences, in order for the content of the consumed state to be determined, then all motive evaporates for insisting that the HOTs which are involved in the explanation of phenomenal consciousness must be actual ones. Consumer semantics fixes contents in terms (inter alia) of relations of availability to consumer systems; so, too, then, dispositionalist HOT theory claims that the availability of the contents of C to a mind-reading, HOT-wielding, system is sufficient to transform those contents, giving them all – and categorically – a subjective dimension (see below).

            I conclude this section, then, with the claim that there are no particular reasons to favour actualist HOT theory over my dispositionalist alternative. But since we do have good evolutionary (and other) grounds to reject actualist HOT theory, it is the dispositionalist form of HOT theory which should be believed if either is.


3            Consumer semantics and feel

In chapter 8:3 I proposed that experiences are phenomenally conscious when they are held in a special-purpose short-term memory store (C) available to a variety of down-stream concept-wielding consumer systems (including a practical reasoning system, and probably also a variety of quasi-modular belief-forming systems). Crucial amongst these consumer systems is a certain sort of mind-reading – or ‘theory of mind’ – faculty, which is capable of generating HOTs in respect of any of the contents of C. In particular, the mind-reading system needs to have an understanding of the is–seems distinction, and/or of experience as a subjective, representational, state of the perceiver.

            Such a system will then easily be capable of generating recognitional concepts of experience, riding piggy-back on the first-order contents of experience and on the subject’s first-order concepts of the objects of experience. So wherever previously the subject could discriminate one colour from another, say, and was capable of thoughts of the form, ‘This is distinct from that’, then the presence of the HOT consumer system renders the subject capable of thoughts of the form, ‘This has a distinctive seeming distinct from the seeming of that’ or ‘This experience is distinct from that’. And wherever previously subjects possessed a recognitional concept for some experienced property – red, say – then the presence of the HOT consumer system renders them capable of recognitional concepts of seems red or experience of red.

            The account is dispositional. It explains phenomenal consciousness in terms of the availability of first-order analog contents to HOT. And then the challenge left over from chapter 8:4.3 is to explain how a perceptual state which is not phenomenally conscious can come to acquire the properties of subjectivity and what it is likeness distinctive of phenomenal consciousness by virtue of such availability. Such properties are, surely, categorical ones. So it might seem puzzling how mere availability to HOT could confer these additional properties on a perceptual state. How can something which hasn’t actually happened to a perceptual state (namely, being targeted by a HOT) confer on it – categorically – the dimension of subjectivity?

            Worse still, indeed: when I do actually entertain a HOT about my experience – thinking, say, ‘What a vivid experience!’ – it is surely because that experience already has the distinctive subjective properties of phenomenal consciousness that I am able to think what I do about it (Robb, 1998; see also Lyvers, 1999; Saidel, 1999). So, once again, how can we legitimately appeal to HOTs in the explanation of those very properties?


3.1       Dual-content representations

The answer to these challenges is simple, in outline at least. (The remainder of this chapter will be devoted to its elaboration and defence.) The answer is that, given the truth of some form of consumer semantics, the contents of C will depend, in part, on what the down-stream consumer systems can do with those contents. And the attachment of a HOT consumer module to an otherwise first-order cognitive system (such as the one depicted in figure 6.4, for example) will transform the intentional contents of the events in C.

            Where before these were first-order analog representations of the environment (and body), following the attachment of a HOT system these events take on an enriched dual content. Each experience of the world–body becomes at the same time a representation that just such an experience is taking place; each experience with the content reda, say, is at the same time an event with the content seems reda or experience of reda. And the events in C have these contents categorically, by virtue of the powers of the HOT consumer system, in advance of any HOT actually being tokened.

            My claim is that the very same perceptual states which represent the world to us (or the conditions of our own bodies) can at the same time represent the fact that those aspects of the world (or of our bodies) are being perceived.[1] It is the fact that the faculties of thinking to which experiences are made available can make use of those experiences in dual mode which turns them into dual-mode representations. This is because, in general, the intentional content of a state will depend upon the nature and powers of the ‘consumer-systems’, as Millikan (1984) would put it. The content possessed by a given state depends, in part, upon the uses which can made of that state by the systems which can consume it or draw inferences from it. And similarly, then, in the case of perceptual representations: it is the fact that perceptual contents are present to a system which is capable of discriminating between, and of making judgements about, those perceptual states as such which constitutes those states as second-order representations of experience, as well as first-order representations of the world (or of states of the body).

            Note that on this account phenomenal consciousness is constituted by higher-order analog representations, or higher-order experiences (HOEs), just as HOE theory – or ‘inner sense’ theory – maintains. So in one sense it is quite right to accuse me of being a closet HOE theorist (Browne, 1999). But there don’t actually need to be two physically distinct sets of representations to carry the two sets of perceptual contents, in the way that HOE theory supposes. Nor do we have to postulate anything like an ‘inner scanner’ to generate the contents seems reda, seems greena and so on. (And nor, therefore, is the account subject to the ‘excess content’ problem we raised against HOE theory in section 1.2 above.)

            Rather, dual content comes for free with the availability of perceptual contents to the mind-reading faculty, or with the availability of those contents to HOT. It is in virtue of the availability of first-order perceptual contents to a mind-reading system which understands the is–seems distinction and/or contains recognitional concepts of experience, that all of those first-order contents are, at the same time, higher-order ones. If this makes me a ‘closet introspectionist’ (Browne, 1999) then I am happy to concur; but it is introspectionism (i.e. a form of HOE theory) without costs.

            As for the phenomenally conscious differences between distinct types of phenomenally conscious experience, these can be said to reside in differences in the higher-order analog contents which those experiences possess. The difference between the feel of a phenomenally conscious experience with the analog content reda, on the one hand, and the feel of a phenomenally conscious experience with the analog content greena, on the other, is that the former also has the higher-order content seems reda whereas the latter also has the higher-order content seems greena. So each phenomenally conscious experience has its distinctive form of subjectivity by virtue of acquiring a higher-order analog content which precisely mirrors, and represents as subjective, its first-order content.

            What we have here is, I claim, a good and sufficient explanation of the defining feature of phenomenal consciousness – its subjectivity, or ‘what-it-is-likeness’; and it is, moreover, an explanation which is fully acceptable from a naturalistic perspective. That feature gets explained in terms of the dual representational content possessed by all phenomenally conscious states (they have both ‘objective’, or world–body-representing content and ‘subjective’, or experience-representing content), by virtue of their availability to both first-order and higher-order consumer systems.


3.2       How consumers can transform contents

The claim made by dispositionalist HOT theory is that the attachment of a new consumer system to a set of states with given contents can transform the contents of those very states. It is therefore important that this claim should be established – or at least rendered plausible – independently of any considerations to do with the nature of phenomenal consciousness.

            Consider a developmental example. Theorists such as Wellman (1990) and Perner (1991) have argued that the mind-reading abilities of young children pass through a number of more-or-less discrete stages. Here let me focus on just two of these stages. Between the ages of about 18 months and 3 years, it is said that young children do have a conception of belief and of various forms of perception (vision, hearing, and so on). This is grounded in the child’s abilities to do such things as track another person’s line of sight, and draw inferences concerning what that person will or will not know depending upon what is in their line of sight – so children can at this stage pass tests of ignorance, for example. But the child does not yet have any understanding of belief and vision as subjective states of the agent, which may be false or illusory as well as true and veridical, and which can represent some aspects of the perceived object/event but not others. Rather, they have what Wellman calls a ‘copy theory’ of belief and perception, according to which seeing is conceived to be an ordinary (non-intentional) relation between perceiver and perceived, and according to which perceptual contact with an object is sufficient for an observer to know everything about the object which the child knows.[2]

            Then between the ages of 3 and 4 (on this account – recall from chapter 7:3.6 that there may in fact be good reason to push these ages considerably lower), children enrich their understanding to acquire what Perner calls ‘a representational theory of mind’. (This can occur either through learning, or, as I suppose, through maturation of the mind-reading module in interaction with a normal environment.) By the end of this period, most children will be passing the so-called ‘false–belief’ and ‘appearance–reality’ tasks, which test for an understanding of belief and perception as representational, intentional, states of the agent.

            Consider, then, the internal representation a sees b, first as entertained by a two-year-old child and then as entertained by a four-year-old.[3] Isn’t it plausible that this representation will have undergone a significant enrichment of its content in the interim, resulting from the enhanced inferential powers of the consumer system? Surely for the two-year-old a sees b will mean something like A is in visual contact with B; whereas for the four-year-old it means something like A is in a subjective state resulting from visual contact with B which represents just the visible aspects of B.

            Here we have two tokens of one type of representation, each of which is caused in the same sort of way by external stimuli, but which differ in intentional content. In which case it is the intervening change in the consumer system – the ‘theory of mind’ faculty – which has made the difference. And note, by the way, that the four-year-old’s token has not lost its previous meaning. Rather, the idea of visual contact has been incorporated into the more sophisticated notion. And note, too, that the child does not actually have to be drawing any inferences from a sees b for it to be entertaining a thought with that enriched content; it is the fact that it can draw those inferences which matters.

            Admittedly, there is an alternative way in which we can tell this developmental story (Woodfield, 1996). We can say that the representation a sees b means A sees B throughout (the full-blown adult content, that is), while insisting that the younger children lack a great many of the beliefs about the nature and significance of seeing which the older children possess. In particular, the younger children, while meaning the same by a sees b, do not know that seeing is a subjective state of the agent, and do not know that it is possible to see some aspects of an object but not others.

            But theorists who adopt this option will, of course, be rejecting any form of consumer semantics, and embracing some version of informational or causal co-variance semantics in its place. And then the approach will be subject to all the weaknesses inherent in causal co-variance theories, some of which were identified briefly in chapter 5:4. (See also my 1996a, ch.3; and Botterill and Carruthers, 1999, ch.7.) There is, however, an additional criticism which can be presented in the present context.

            The objection is that the consumer-semantics way of individuating concepts and contents is actually the mode of individuation employed in the relevant scientific discipline, namely developmental psychology. These psychologists are in the business of charting the way in which children develop their concepts, and formulating a set of laws or nomic tendencies relating some of the resulting intentional contents to one another at the various developmental stages. They seek to formulate principles which relate the intentional contents of children’s beliefs and other attitudes to one another, in such a way that the intentional contents governed by distinct nomic tendencies are distinct, even when they concern the same subject-matter. (See, e.g., Carey, 1985; Wellman, 1990; Perner, 1991; Gopnik and Wellman, 1992; Carey and Spelke, 1994; Gopnik, 1996.)

            And then the criticism goes: if scientific psychology individuates contents in line with some form of consumer semantics, then some sort of consumer semantics is what individuates the relevant real-world intentional properties. Recall from chapter 2:2.2 that the most plausible position on the metaphysics of properties maintains that the real-world properties which there are – which really exist independently of us – are the properties which would be picked out by the predicate terms of a completed science. If we take developmental psychology seriously as a science, then we should take seriously the idea that it is consumer semantics which serves to pick out intentional contents as a natural kind.[4]


3.3            Inferential role semantics and categorical feel

The explanation of phenomenal consciousness which I am putting forward, then, claims that it is because the contents of C are available to two sets of consumers – first-order conceptual reasoning systems, as well as a higher-order mind-reading faculty – that those contents actually (categorically) have dual representational status; acquiring a dimension of subjectivity. In framing this explanation I have been careful to speak of ‘consumer semantics’ in general, understood as embracing various forms of teleosemantics and also various forms of functional and inferential role semantics. For all of these approaches can advance essentially the same solution. I am inclined to think, however, that some form of inferential role semantics is probably more plausible than any form of teleosemantics (Botterill and Carruthers, 1999, ch.7). So in making the points which follow I shall assume the correctness of this more restricted type of semantic theory.

            According to inferential role semantics, it will not especially be the evolved uses which a given representation has come to have within a cognitive system which confers on it its content. Rather, what distinctively determines content are the inferences the system can engage in, the further judgements which the representation may lead to, and so on – whether those inferences and judgements were selected for or not. Nor will the content of a representation depend upon all aspects of its causal role. Rather, it is especially the beliefs which we are apt to form on the basis of our experiences, and the further inferences which we are disposed to draw, which are partly determinative of the content of those experiences.

            Inferential role semantics admits of a variety of different alternatives, in its turn – ranging from total holism about representational content (where any belief you may be apt to form, no matter how remote, contributes to the content of a given state), to various forms of semantic localism (according to which only some belief-forming processes figure in the individuation of content).[5]

            Everyone will allow, however, that the immediate inferential connections of a state – connections which are unmediated by any further belief or inferential capacity – are particularly plausible candidates to be partially determinative of content. For example, a disposition to infer P from P & Q – but not necessarily a disposition to infer ~(~P v ~Q), which may require a number of distinct steps – is partially constitutive of the content of ‘&’. And what we have in the case of the architecture depicted in figure 8.1, are a set of recognitional concepts of experience – concepts, that is, which are higher-order ones – being deployed in relation to the contents of C. It therefore seems especially plausible that these should have an important impact upon the nature of those contents.

            I claim, then, that once first-order perceptual representations are present to a consumer-system which can deploy a representational theory of mind, and which contains recognitional concepts of experience, then this is sufficient to render those representations at the same time as higher-order ones. This is what confers on our phenomenally conscious experiences the dimension of subjectivity, and makes it the case that there is something which it is like to undergo them.

            Each experience is at the same time (while also representing some state of the world, or some state of our own bodies) a representation that we are undergoing just such an experience, by virtue of the powers of the mind-reading consumer-system. Each percept of green, for example, is at one and the same time an analog representation of greena and a representation of seems greena or experience of greena. In fact, the attachment of a mind-reading faculty as one of the consumers of the outputs of (some of) our perceptual systems completely transforms the contents of those outputs, and transforms them categorically.


3.4       Setting inferential role semantics to work

Recall that in chapter 5:2.4 we drew a distinction between worldly-subjectivity and experiential-subjectivity. Any first-order experience will involve a kind of subjectivity, in the sense that it will involve a subjective take on some aspects of the environment (or body) but not others, depending upon the nature and discriminatory powers of the creature’s perceptual systems. This is the worldly-subjectivity of experience. But it is also natural to think that our experiences themselves have a subjective aspect, captured in the idea that there is something which they (and not just the world) are like for the subject who undergoes them. On the present account, this is explained as the dimension of seeming or subjective appearance which attaches to all of our phenomenally conscious states, in virtue of their availability to second-order recognitional concepts of experience. Since our conscious perceptions involve, not just a subjective take on aspects of the world, but also a representation of their own subjective, or ‘seeming’, status, they have both forms of subjectivity at once.

            Recall, too, the so-called ‘transparency’ of perceptual experience, which has previously been introduced into our discussion on a number of occasions. This is now easily and satisfyingly explained. The reason why you do not discover any additional properties of your experience when you concentrate your attention on it – in addition, that is, to the properties of the world (or body) represented – is that there are no such properties. All that happens when you focus your attention on your experience of the ripe red tomato is that you attend to a state with an analog content representing rednessa, which also represents seeming rednessa. And either way, to focus your attention on this state is to focus on the redness represented. So transparency is preserved. Yet the dual content of the representation gives us all the materials we need to draw the distinctions between conscious and non-conscious experience, and between worldly and experiential forms of subjectivity.

            Notice that according to the current use being made of inferential role semantics, experiences will already have their dual content in advance of any actual targeting by a HOT. Consider a propositional example first. Suppose that I hear the weather-forecaster say that the afternoon will be cold and wet; and suppose that I believe her. Then I have a belief about the weather with a conjunctive content, in advance of deriving either one of the conjuncts from that belief. What I believe is that it will be cold and wet, in advance of deriving the belief that it will be wet. But for all that, I am only capable of entertaining conjunctive thoughts at all by virtue of having such an inferential disposition. It is only because I am disposed to derive, ‘It will be wet’ from, ‘It will be cold and wet’ that the latter can count as having a conjunctive content. Then so, too, in the case of experience. My experience will have the dual contents reda and seems reda in advance of me entertaining any HOT about it. But it is only because I am capable of recognitional judgements of experience, as such, that my experience has such a dual content.

            Notice, indeed, that it is an implication of the current use being made of inferential role semantics that the subjective content which attaches to the experience can figure in a causal explanation of the particular HOT entertained. Consider a propositional example again: when I do derive, ‘It will be wet’ from, ‘It will be cold and wet’, we can explain how I come to have this new belief in terms of my possession of the old one. It is because I first believed that it will be cold and wet that I then came to believe that it will be wet. But for all that, a capacity to engage in just such inferences, in general, is partly constitutive of my possession of any conjunctive belief.

            In the same way, then, it can be because my experience has the dual content which it has, with a particular subjective phenomenology, that I then go on to think, ‘What an interesting experience!’. But for all that, the latter is an actual manifestation of the very capacity which confers dual content on my experiences. It is also because I am disposed to entertain HOTs of just this sort that my experience has its dual content, and so has a subjective phenomenology in the first place. (I shall return to elaborate and defend this point in section 4.3 below.)

            We can easily explain, too, how our higher-order recognitional concepts of experience can ‘break free’ of their first-order counterparts, in such a way as to permit thoughts about the possibility of experiential inversion and such like. Here is how the story should go. We begin – both in evolutionary terms and in normal child development – with a set of first-order analog contents available to a variety of down-stream consumer systems. These systems will include a number of dedicated belief-forming modules, as well as a practical reasoning faculty for figuring out what to do in the light of the perceived environment together with background beliefs and desires. One of these belief-forming systems will be a developing mind-reading module.

            When the mind-reading module has reached the stage at which it confers on us an understanding of the subjective nature of experience, and/or a grasp of the is–seems distinction, then we will easily – indeed, trivially – become capable of second-order recognitional judgements of experience, with these judgements riding piggy-back on our first-order recognitional concepts. So if subjects had a recognitional concept red, they will now acquire the concept seems red, or experience of red, knowing (a) that whenever a judgement of ‘red’ is evoked by experience, a judgement of ‘seems red’ is also appropriate on the very same grounds; and (b) that a judgement of ‘seems red’ is still appropriate whenever a disposition to judge ‘red’ has been blocked by considerations to do with abnormal lighting, or whatever. Note that at this stage the higher-order concept in question is still a theoretically embedded one, with conceptual connections to worldly redness (it is, after all, a seeming of red). What one recognises the state as is a state whose normal cause is worldly redness, and so on.

            This change in the down-stream mind-reading consumer system is sufficient to transform all of the contents of experience (and imagination), rendering them at the same time as higher-order ones. So our perceptual states will not only have the first order analog contents reda, greena, louda, smootha, and so on, but also and at the same time the higher-order analog contents seems reda, seems greena, seems louda, seems smootha, and so on. The subject will then be in a position to form recognitional concepts targeted via just these higher-order contents, free of any conceptual ties with worldly redness, greenness, loudness, and smoothness. (This can either be done by fiat, by dropping any connection with redness from the recognitional concept seems red, or by introducing new concepts of the form this experience.) And once possessed of such concepts, it is possible for the subject to wonder whether other people have experiences of seems red or of this sort when they look at a ripe tomato, to conceive of worlds in which zombies perceive red without undergoing this experience, and so on.[6]

            Note how this account makes our possession of purely recognitional concepts of experience – Cartesian concepts, that is – parasitic upon a theoretical understanding of the idea of subjectivity. For it is only by virtue of our achieving such an understanding of the nature of experience that there come to exist any higher-order analog contents to drive the recognition-process. These higher-order contents are caused to exist through the availability of their first-order counterparts to a recognitional application of our theoretically-imbued concepts of experience. I can thus agree with Descartes, Kripke (1972) and Goldman (1993) that we do possess some purely recognitional concepts of experience, while claiming that theory-theory is still essentially correct as regards the nature of our mental-state concepts (see chapter 1:2 above). For our purely recognitional concepts of experience presuppose theoretically-embedded ones, but not vice versa.

            Does the availability of second-order analog contents to a set of recognitional concepts whose application can be driven by just those contents, then give rise to a set of third-order analog contents? Indeed, what is to prevent a whole hierarchy of higher-order analog contents from spiralling upwards, all realised in one and the same set of perceptual states? There is nothing to prevent this happening, in principle, except that in my view we don’t actually possess the requisite recognitional concepts of third-order, fourth-order, and so on.

            Notice, first, that the mere fact that some of our recognitional concepts are applied via exposure to second-order experiences doesn’t yet mean that those concepts are themselves third-order ones, any more than applying a recognitional concept red in the presence of a first-order experience of red means that the concept in question is really second-order. In fact our purely-recognitional concepts of the form this experience, although applied in the presence of second-order analog experiential contents, are themselves just second-order ones. It is the first-order experience which is actually recognised, even though the application of the recognitional concept may be driven solely by the higher-order analog content which represents that experience.[7]

            Whether or not a concept is a higher-order one depends, not on the order of the experience through which it is applied, but rather on what can then be done with that concept – on its inferential role. And although we can, in principle, make sense of sentences such as, ‘I am aware of the way this experience of red seems to me’ (where the awareness attributed might then appear to be third-order, in the light of its application to a seeming-content which is second-order), we actually lack any further use for the notion of an awareness-of-a-seeming, in addition to the notion of a seeming itself. Certainly we lack any recognitional concept expressible in terms such as ‘This awareness of a seeming of red’ or ‘This seeming of seeming red’, not constructed by inference from our belief in the second-order nature of the seeming we are aware of. So there is no reason to think that we are actually ever subject to third-order analog perceptual contents. (But even if we were, there would be no particular problem for dispositionalist HOT theory here, given my story about the ease with which higher-order analog contents can be generated.)


3.5            Perceptual contents versus belief contents

The examples considered above make clear that if perceptual contents are like belief contents in depending partly upon the powers of down-stream consumer systems, then our perceptual contents will (actually and categorically) possess dual representational content (both is and seeming) by virtue of their availability to both first-order and higher-order consumer systems. But it might be doubted whether perceptual states and belief states acquire their contents in the same sort of way. Someone might hold that a pure form of causal co-variance semantics is true of perceptual contents, while allowing that some kind of consumer semantics is true of belief contents.

            Note, however, that the objections to causal co-variance semantics which we raised in chapter 5:4 focused on perceptual content in particular. Recall the length-of-causal-chain problem, for example. The perceptual state which carries information about worldly redness equally carries information about, and causally co-varies with, a certain complex property of the human retina – call this R. So it seems unlikely that we can distinguish between the putative contents reda and Ra in information-carrying terms. Rather, my percept has the content reda, rather than Ra, by virtue of my capacities to focus thoughts and actions guided by that percept on the surface of the object, rather than on the surface of my retina. It is what the consumers of a given perceptual state can do with it that determines which, out of all the various sorts of information carried by that state, are genuinely represented.

            There is, moreover, independent reason to think that changes in consumer-systems can transform perceptual contents, and with them phenomenal consciousness. (See Hurley, 1998, for presentation and discussion of a range of different cases.) Consider the effects of spatially-inverting lenses, for example (Welch, 1978), which have already been mentioned in chapter 3:4.3. Initially, subjects wearing such lenses see everything upside-down, and their attempts at action are halting and confused. But in time – provided that they are allowed to move around and act while wearing their spectacles – the visual field rights itself. Here everything on the input side remains the same as it was when they first put on the spectacles; but the cognitive action-controlling and planning systems have learned to interpret those states inversely. And intentional visual contents become re-reversed as a result.

            Notice, moreover, that someone who is normalised to inverting spectacles doesn’t actually have to be acting on or making plans about the environment in order to see things right-side-up. She can be sitting quietly and thinking about something else entirely. But still the spatial content of her perceptual states is fixed, in part, by her dispositions to think and move in relation to the spatial environment.


3.6       Is–seems and bodily sensations

The dispositionalist HOT account, as developed thus far, might be thought to imply that phenomenal consciousness depends upon experiential (and imagistic and emotional) states being made available to a mind-reading system capable of deploying the is–seems distinction, in particular – in such a way that it is by virtue of our capacity to apply this distinction that those states acquire dual representational content (both first-order and second-order). A problem for any such account, however, is that the is–seems distinction appears to find no application in connection with bodily sensations like itches and pains. If there is no distinction between being in pain and seeming to be in pain, then it cannot be by virtue of our capacity to deploy such a distinction that our pains become phenomenally conscious.

            One way of responding to this challenge would be to insist that the is–seems distinction is applicable to itches and pains. The first step would be to reiterate the point made in chapter 5:1.3, that pains and other bodily sensations are best understood on the model of perceptions of secondary qualities like colour. Just as to be in a state of seeing reda is to be in a state which represents a certain analog property – rednessa – as distributed over a given area or surface, so to be in a state of feeling paina is to be in a state which represents a certain analog property – paina – as distributed through a certain region of one’s body. And the second step would be to appeal to occurrences of referred pain and phantom limb pain to argue that such states can represent incorrectly, or represent what is not there, just as representations of colour can. In which case an is–seems distinction can find application here after all.

            Although tempting, and correct as far as it goes, this response to the problem is not really very plausible. This is because information about referred and phantom limb pains is not an enduring and ubiquitous feature of all human folk psychologies. There may be many people who never learn of the existence of such oddities, and many children will not learn about them until quite late – perhaps only at some point in adulthood. But it would hardly be very plausible to claim that phenomenally conscious pains are lacking in the case of such people.

            A more effective way of responding involves noticing that the subjectivity of bodily sensations is better explicated in terms of some sort of privileged access, rather than in terms of the is–seems distinction. What children have to learn is that properties such as hurt and itch – when instantiated in themselves – are perceptually accessible to them in a way that those instances are not accessible to others.

            Initially, children will have first-order recognitional concepts of hurt and itch, applicable to presented properties of their body, in much the same sort of way that they have recognitional concepts of red and smooth, applicable to presented properties of objects in the world. At this stage, one may suppose, young children will have no understanding that pains are fundamentally different from colours, in that recognitional concepts for them can only ever be applied to a given instance by one person. These children do not yet understand that a given coloured surface can be observed by many, whereas a pain can only be observed (in the sense of felt) by one. But when they do come to understand this, they come to realise that they stand in a relation to their own pains and itches (the feeling-relation) which other people don’t.

            Corresponding to the is–seems distinction as it applies to colours, then, will be a pain–feels-pain distinction. This is the distinction between pain as a represented (first-order) property of our bodies, and feeling pain as the privileged representational relation in which one stands to instances of that property in one’s own case (in which case representations of this relation are second-order ones). And just as children can first acquire recognitional concepts of colour before acquiring second-order recognitional concepts of seeming-colour, so in the case of bodily sensations they can first acquire concepts for the various represented properties (pain, itch, tickle and so on) before then acquiring second-order recognitional concepts of feeling pain, feeling itchy and so on. And it is at this stage that the analog representations involved acquire a dual intentional content (both first-order and second-order), and hence become phenomenally conscious, according to a dispositionalist HOT account.[8]

            I suggest that the basic notion which children have to grasp, then, is that of the subjectivity of all forms of experience. This manifests itself, in the case of outer experience, in an understanding of the representational character of such experience, and in a corresponding grasp of the is–seems distinction. But in the case of bodily sensations, to understand their subjectivity is to understand the privileged access which we each us have to our own sensations. I am not aware of any data on the question of which, if either, of these forms of understanding comes first in child development. That is, whether it is an understanding of privileged access which bootstraps the is–seems distinction (since seemings are also states to which we have such access), or rather whether it is an understanding of the is–seems distinction which helps us to grasp the idea of privileged access. But then I don’t think dispositionalist HOT theory needs to take a stand on this issue.


3.7       The unity of subjectivity

It may be of some importance, however, for dispositionalist HOT theory to find a single characterisation of the notion of ‘subjectivity’ which can cover both outer perceptions and bodily sensations equally; for otherwise our theory might appear to lack the kind of unity which is so often the mark of a successful theory. So: what is it that is common to the subjectivity of is–seems and the subjectivity of privileged access?

            My answer comes in two parts. Recall from chapter 1:3.4 that the relevant kind of subjectivity attaching to all phenomenally conscious states, which constitutes them as phenomenally conscious, is that they should possess properties of a sort which can be available for immediate introspective recognition – the feel of a mental state is that property in virtue of which we can recognise it when we have it. Then according to dispositionalist HOT theory this sort of subjectivity consists in the possession of higher-order analog contents, as explained above. In which case the first part of my answer to the unity-challenge is to say that the subjectivity of all phenomenally conscious states (whether they be experiences of red or feelings of pain) consists in the higher-order analog contents of those states. So, what is common to the subjectivity of is–seems and the subjectivity of privileged access is that an understanding of either one of these is sufficient to create the sort of subjectivity which is definitive of phenomenal consciousness, and which consists in higher-order analog intentional contents.

            The real challenge, however, concerns the question of unity in the consumer-semantic mechanism which creates the form of subjectivity constituted by higher-order analog contents. Here we should distinguish between the unity (or rather lack of unity) in the child’s developing understanding of the subjectivity of its mental states, on the one hand, and the sort of understanding which we theorists can attain, on the other. I see no problem about claiming that children’s outer experiences become phenomenally conscious by virtue of their understanding of the is–seems distinction, whereas their bodily sensations become phenomenally conscious by virtue of their understanding of privileged access. Indeed, I see no problem in claiming that these forms of understanding may make their appearance at somewhat different developmental stages, in such a way that the child’s outer experiences may come to be phenomenally conscious before its bodily sensations are, or vice versa. Any pre-theoretical intuition to the contrary – to the effect that all forms of experience must become phenomenally conscious together at the same developmental stage – is unlikely to be worth very much, in my view; for how is such an intuition to be grounded?[9]

            What is important, however, is that we, as theorists, should be able to see that it is no accident that both of these forms of understanding deserve to be classed as insights into forms of subjectivity. And it appears plain that the basic kind of subjectivity consists in the subjectivity of our access to (some of) our own mental states. The is–seems distinction is really just another way of marking this sort of access in the case of outer experience, given an understanding of the causal processes which operate – and are apt to fail – between our experiences and their objects.

            Since our common-sense psychology includes a conception of outer perception as a causally-mediated and fallible process, we know that a recognitional judgement that we are in one of these mental states (a seeming) can be correct, when the relevant first-order judgement grounded in the occurrence of that state (a judgement of what is) isn’t correct. In contrast, our common-sense psychology doesn’t contain any worked-out conception of bodily sensation as a causally mediated process; and consequently we have no use for a distinction between is and seems in connection with such sensations. But still we know that we have non-inferential access to our states of awareness of these sensations, just as we have non-inferential access to our perceptual seemings.

            As theorists, therefore, we can say this: it is when we acquire an understanding of the nature of percepts of red and feelings of pain sufficient to ground the idea that we each of us have a kind of access to our own perceptual states which others do not – and consequently come to construct some recognitional concepts for those states to ride piggy-back on our first-order recognitional capacities – that the states in question acquire their dual analog intentional contents (both first-order and second-order). What subjectivity fundamentally amounts to, here, is that these states and their contents are ones to which we can have immediate, non-inferential, access. Thus one sort of subjectivity (the subjectivity of feel or of what-it-is-likeness) gets created through our understanding of another sort (the subjectivity our access to our own mental states).


3.8            Metaphysical status

What is the modal and metaphysical status of dispositionalist HOT theory? First, it claims to provide a reductive explanation of phenomenal consciousness. This means that the theory describes a way of linking together cognitive structures and contents (in terms which do not themselves presuppose phenomenal consciousness), any instantiation of which is supposed to be metaphysically sufficient for phenomenal consciousness to occur. So there are no possible worlds where the appropriate HOT structures and dual-content analog representations are instantiated, but where phenomenal consciousness is absent.

            Of course we can imagine or conceive of such worlds, because our recognitional concepts of feel are not definitionally tied to the rest of the system (see chapter 2:4 and section 3.4 above). But HOT theory claims that such worlds are not genuinely possible, because the properties which our recognitional concepts pick out are in fact none other than (or are constituted by, at least) the properties involved in being an analog perceptual content (narrowly individuated) present to an appropriate HOT faculty (and hence possessing dual intentional content).

            Does dispositionalist HOT theory propose to identify the nature of phenomenal consciousness? Should it be construed as making an identity claim? There are two grounds on which such a claim might be challenged. First, it might be said that identities must be necessary if true; yet we have conceded that there are possible worlds in which irreducible qualia exist, or worlds in which there are phenomenally conscious states in the absence of a capacity for HOT (see chapter 2:3.5). But as we also saw in chapter 2:3.5, identities are only necessary when both terms are used as rigid designators, being employed with the intention of designating an underlying nature. And terms such as ‘feel of pain’ and ‘manifest-water’ are not used with any such intention, but are rather tied to the manifest – ‘thin’ – properties in question. In cases such as this, statements of the form, ‘Manifest-water = H2O’, or ‘The feel of pain = such-and-such a higher-order analog content’ can be both true and contingent.

            The second ground on which an identity-claim might be challenged, is that property identities require us to look at all worlds in which the actual laws of nature obtain (at least where those properties are natural ones, individuated ‘thickly’; see chapter 2:2.2). Yet we haven’t shown that either HOE theory or actualist HOT theory are naturally impossible – just that it is very unlikely that systems instantiating them should evolve. So there may be naturally possible worlds in which there exist creatures who entertain an actual HOT in respect of each and every phenomenally conscious content which they have, for example. But who has the burden of proof, here? Perhaps either HOE theory or actualist HOT theory is naturally possible; but perhaps they are not. In the absence of a demonstration that either one is naturally possible, it might be reasonable to claim that dispositionalist HOT theory identifies the actual – thickly individuated – property of phenomenal consciousness.

            I don’t actually need to defend an identity-claim, however. Since we have no reason to believe that there really are any creatures of which actualist HOT theory is true (or that there really are any creatures with ‘inner sense’); and since, indeed, we have little reason to think that any creatures besides human beings are really phenomenally conscious; we can claim that dispositionalist HOT theory gives us the actual natural constitution of phenomenal consciousness, in this world. We can claim that analog contents available to HOT are what the (worldly, thickly individuated) property of phenomenal consciousness consists in, as manifested in the actual world. And that, surely, is as much as we need in order to claim that phenomenal consciousness has been successfully reductively explained.

            Compare the following example. As is well known, the semi-precious stone jade admits of two distinct naturally occurring varieties, with different chemical compositions – jadeite (a silicate of aluminium and sodium) and nephrite (a silicate of calcium and magnesium). Now suppose that the world had, as a matter of contingent fact, failed to contain any nephrite (perhaps because magnesium is very much rarer in that world than in the actual world). And suppose that the scientists in that world had been trying to discover the explanation for the distinctive properties of jade. After much work, they propose that those properties can be explained as emerging from a certain sort of silicate compound of aluminium and sodium. Would the success of this as a reductive natural explanation of the properties of jade as it actually occurs be in the slightest way undermined by the fact that there is an alternative naturally-possible compound which would also present those properties? Surely not. Then nor, too, in the case of dispositionalist HOT theory’s proposed reductive explanation of phenomenal consciousness, in the form that such consciousness actually occurs.[10]


4            Elucidations and replies

In this section I shall say some more by way of elucidating the explanatory potential of dispositionalist HOT theory, and reply to some objections.


4.1       Is this a reductive explanation?

Is this really a successful reductive explanation? It is easy to get oneself to feel that it is not. Focus, again, on that ubiquitous red tomato and ask yourself, ‘Is the nature of this experience, which I am now undergoing, completely exhausted by it being an analog perceptual state with the contents reda and seems reda? Does this proposal about its nature explain what it is like for me to have it?’

            There is certainly a natural temptation to answer these questions in the negative. One is inclined to respond, ‘Surely it is possible that I might undergo this type of experience even though it did not seem to me that I was presented with something red’. And I agree that this is conceptually possible. This is because I accept that we can form purely recognitional concepts of experience which lack any conceptual ties with the first-order recognitional concepts which we deploy in experience. So I can form a concept expressible only as ‘This (type of) experience’, applicable to experiences of red, but which lack any conceptual connection with my first-order concept red – it is not even the concept of a seeming red, it is just a concept of a this experience. Then given such a concept, I can of course conceive of a world in which I have this experience, but in which I am not undergoing an experience of seeming red.

            Recall from chapters 2 and 3, however, that mere conceivability experiments are not to the point. I allow, of course, that we have concepts of experience which do not conceptualise those experiences as analog intentional contents made available to a HOT faculty. Indeed, it is precisely because we have recognitional concepts of the sort characterised above that thought experiments involving inverted and absent qualia strike us as so inevitable. But then dispositionalist HOT theory is not proposed as a piece of conceptual analysis. It is rather put forward as a naturalistic theory of the nature of the properties (‘worldly’, or thickly individuated) which we actually pick out when we deploy such recognitional concepts.

            Recall from chapter 3:2.3, too, that the existence of purely recognitional concepts of experience means that we cannot hope for a reductive explanation of phenomenal consciousness which will be what I called ‘immediately cognitively satisfying’, in the sense of dove-tailing with the manner in which we pre-theoretically characterise our conscious states. Once these points are brought into proper focus, we should see that there is nothing left out by the proposed reductive explanation of the feel of an experience of red as an analog representation with the (narrow) contents reda and seems reda.

            An alternative way of generating a naturally-felt suspicion of the proposed theory is to attend to the phenomenally conscious differences between experiences of colour – look, say, at a ripe tomato alongside a fresh green leaf. Can dispositionalist HOT theory explain the ways in which our experiences of these colours differ? Or can it explain why the one sort of experience should be caused by worldly redness while the other is caused by worldly greenness, rather than vice versa? The answer, as you might expect, is: ‘In some respects, yes; in some respects, no’. Let me elaborate; taking the negatives first.

            HOT theory cannot hope to explain the various colour asymmetries. It cannot explain, for example, why red strikes us as warm while green strikes us as cool. But my guess is that we should still look towards some sort of consumer semantics – this time consumer semantics of a first-order variety – for the explanation. The relative warmth of red may derive from hard-wired inferential connections with fire and with heat, resulting from our hominid fire-using past. Similarly, the relative coolness of green may derive from hard-wired inferential connections with shade (a premium commodity on the savannahs of Africa where we evolved). Such explanations are certainly not entailed by dispositionalist HOT theory. But they are consistent with it. And their plausibility does serve to support the idea that the inferential powers of consumer systems can alter the contents of perceptual experience.

            On the other hand, HOT theory can explain why the experience I enjoy when looking at the tomato is caused by worldly redness rather than greenness – it is because that experience intrinsically has the contents reda and seems reda. And the individuation conditions for those contents, in their turn, will make reference to causation by redness. The mere fact that I have endorsed consumer semantics certainly does not rule this out. On the contrary, teleosemantics will individuate content in terms of those aspects of the causal information carried by a state which the consumer systems have evolved to pick up on. And inferential role semantics, too, comes in so-called ‘long-arm’ varieties, where causal relationships with the world are considered to form part of the inferential role of a given state. Nor does my endorsement of narrow content raise any difficulty here. For an experience with a narrow content as of reda, for example, will be that narrow content which is normally caused by red in actual circumstances. It is merely that we individuate it in such a way that the very same content can be present in someone born with natural colour-inverting corneas, say, who is then looking at green grass.

            Much does still remain to be explained, of course. It remains to be explained how perceptual contents are realised in the computational and neurological processes in the human brain, and how the various distinctive profiles of the intentional contents generated by our different sense modalities are determined. And it remains to be explained how and in what manner the perceptual systems make their outputs available to down-stream executive systems of thought and reasoning. That is to say, there are many hard problems remaining to be solved in our search for an account of content-involving cognition (although much progress has been made in recent decades). The problem of reductively explaining phenomenal consciousness is (contra Chalmers, 1996) the easy problem; and that has now been successfully resolved, I claim.


4.2       The heat watchers

Let me now present a rather different sort of objection, which focuses on and criticises the idea that down-stream inferential roles are an important determinant of perceptual content. We can imagine fitting someone up with a machine which transduces radiant heat into shades of black and white – the hotter the object, the whiter it will appear (c.f. Churchland, 1979). This machine might be built into the corneas in someone’s eyes, so that they thereafter perceive in shades of grey and white, where those percepts carry information about heat, not reflective properties of surfaces. And so far as I can tell, this story is not just imaginable, but physically possible.

            Thus far the story raises no problem, even for any kind of broadly functionalist account of phenomenal consciousness. For the percepts of white in question are states which would normally be caused by the presence of a white object, not a hot one. And if we try to imagine a creature whose perceptual system is built to detect heat, but who nevertheless enjoys experiences as of whitea when looking at a hot object, then we can legitimately reply that this is merely imaginable – there is no reason to think that such a creature is naturally possible. So we can allow that the example is conceivable, but deny that it shows anything about the real-world property of being a percept as of whitea.

            But (the objection might go) there is a problem here for those – such as myself – who want to claim that the content of a state can depend on the down-stream effects of that state. For what are we to say when the person who has been fitted up with the heat detector learns what has happened, and subsequently acquires dispositions to form beliefs about heat when perceiving an object as white? Intuition would suggest that his percepts would still have the phenomenally conscious properties of a percept of white, whereas an inferential role account of semantic content would suggest that they have become transformed to acquire the content hota. This would then imply that phenomenal consciousness cannot be reductively explained in terms of intentional contents (not even when individuated narrowly).

            Now there is no particular problem here, in fact, so long as the person in question retains his concepts of white and grey, and continues to be capable of applying these recognitionally in the face of his experiences. If, as I do, you think that it is the immediate semantic effects of a state which have a crucial role in determining its content, then the fact that someone who has recognised white is disposed to go on to infer hot from that, need have no particular impact on the content of the perceptual state which underpins the initial recognition. But what if we imagine that he loses his colour-concepts through selective aphasia or amnesia? Or what if we imagine someone brought up from birth wearing one of these transforming machines, whose only visually-based recognitional concepts are those pertaining to heat rather than colour?

            In such a case I no longer have the intuition that the person’s phenomenally conscious experience would be the same. For sure, the nature of their experience might be in some respects similar to the experiences of a normal person perceiving a white surface. This is because all of the input-side of their perceptual system, from the cornea inwards, will be performing in the same way. Merely changing recognitional concepts of white and grey for recognitional concepts of hot and warm is not going to transform the contents of someone’s visual experiences completely, making them as different from one another as a percept of white is from a tactile experience of heat. (So visual percepts of hot can still be dazzling, for example, just as bright white is.) But it might very well make a significant difference. And that is all I need, in order to be able to tell my sort of story about the nature of phenomenal consciousness.

            For example, imagine that the transforming-machine were removed again without the subject’s knowledge, and that he then looks at someone who is dressed all in white – it would surely seem to him that the person is incandescent, and that they are glowing from within. These are significant differences in content from normal, and they are differences which would appear to have a significant impact upon phenomenology.

            More importantly, perhaps, what one says about the heat-watcher case may well depend on the details of how intentional content is individuated, and on the truth of some hard-to-evaluate counter-factuals. In particular, suppose that the heat-watcher’s experience of a hot object is such that it would still evoke a recognitional judgement of ‘seems white’, if he had not suffered his amnesia. Then we have enough here to claim that the narrow content of his percept is still the same as it was, just as intuition suggests. (C.f. the discussion of intra-personal experience inversion in chapter 4:4.1 above.) Similarly in the case of someone born wearing a heat–colour transformer: if it is true that the state he is in when looking at heat is such that it would evoke a recognitional judgement of ‘seems white’ in a normally sighted person, then again we can claim that his percept has the narrow content white.

            Whether or not these counter-factuals can be shown to be true is beyond the scope of this book; for I have sworn off providing a reductive account of intentional content, and yet this is probably what would be required to generate an answer. At issue, is the extent to which perceptual content is determined by input-relations, and the extent to which it is determined by down-stream inferential role, or by ‘consumer-relations’. All that I need for my reductive explanation of phenomenal consciousness to work is that some aspects of perceptual content depend upon consumer-relations; I do not need to claim that they all do.


4.3            Dormative virtues

Someone might object that the account provided here has all of the hallmarks of an explanation in terms of dormative virtue – that is to say, all the hallmarks of no explanation at all. For recall the line taken in section 3.4 above: it is because my experience already has a given higher-order analog content that I think, ‘What an interesting experience!’; but it can also be because that state is of a kind which is disposed to cause HOTs of just this sort that it possesses a higher-order content in the first place. The account then seems formally analogous to this: if I fall asleep after drinking a soporific cocktail, it can be because that drink is already a soporific that I come to lose consciousness; but it can also be by virtue of my disposition to lose consciousness in just this way that the cocktail is a soporific in the first place.

            The first point to make by way of reply is that explanations of the ‘dormative virtue’ sort are perfectly appropriate in their place. It can be both true and explanatory to say that I fell asleep because I drank a liquid containing a soporific. This is to explain one particular event (me falling asleep) in terms of another which is its cause, and to indicate that there is some property (not further specified) of the cause such that events of that kind are correlated with sleep in a law-like way. And it can be both true and explanatory to say of the liquid in question – opium, as it might be – that it is a soporific. This is to provide a partial functional specification of its properties. Where dormative virtues definitely become non-explanatory is if we appeal to them in trying to answer the question, ‘Why does opium put people to sleep?’ (Bad answer: ‘Because it is a soporific’.) For this question is a request to specify the underlying mechanism, not just to be told that some such mechanism exists. (That is, we don’t just want to be told, ‘Because it has some property which tends to cause sleep’ – we knew that already.)

            In the same way, it can be both true and explanatory to say that I came to have a belief with the content that it will be wet because I already had a belief with the content that it will be cold and wet. This is to explain one event in terms of another with which it is connected in a law-like manner. And it can be both true and explanatory to say that and-beliefs tend to cause beliefs in their individual conjuncts. This is to provide a partial functional specification of the nature of conjunctive content. Where explanation by content runs out, is when we ask the question, ‘Why do people with conjunctive beliefs tend to believe the individual conjuncts?’ For this, too, is a request to specify the underlying mechanism, needing to be answered by appeal to some sort of computational account, for example, and not by an appeal to content. Likewise, then, for the relations between higher-order analog contents and higher-order recognitional judgements: appeals to them are only non-explanatory if our question is why such contents give rise to such judgements at all.

            Notice, too, that in one respect saying that I came to believe P because I already believed P&Q is quite unlike saying that I fell asleep because I took a soporific. For to say the latter is just to say that I fell asleep because I drank something which tends to make people sleep, since a soporific is nothing other than a substance which causes sleep. Conjunctive beliefs, in contrast, aren’t identical with beliefs which cause belief in the individual conjuncts, since introduction-rules are just as important as elimination-rules in specifying the contents of the logical connectives. The functional specification of conjunction by its elimination-rule is only a partial one. So to explain my belief that P in terms of my belief that P&Q is to give a good deal more information about the cause, of a functional sort, than merely to say that it has some property which tends to cause P-beliefs.

            Likewise for higher-order analog contents; only more so. To say that someone is in a perceptual state with the analog higher-order content seems reda is not just to say that they are in a state which tends to make them judge that they are experiencing red. This may be a partial characterisation of the content of the state, but it is only partial. In addition we need to say that the state has an analog content; that it is also an analog representation of red, normally caused by exposure to red; and so on. So here, too, the explanation of my higher-order judgement is a good deal more informative than a mere ‘dormative virtue’ one.

            It is particularly important to stress the analog nature of the higher-order contents in question. For this means that there is no end of possible higher-order judgements, each employing one of an unlimited range of potentially-available higher-order recognitional concepts, to which those contents could give rise. On the present account, it only requires the subject to have an understanding of the subjectivity of experience in general, and to possess some higher-order recognitional concepts in each modality, for all of the subject’s perceptual (and imagistic) states to acquire a dimension of subjectivity. This means that there is a richness of content to higher-order experience which goes far beyond a mere disposition to make a few types of higher-order judgement.[11]

            In general, then, my answer to the challenge is this: higher-order analog contents are just as real, and just as categorical in nature, as are any other species of intentional content; and causal explanations by appeal to them can be explanatory. But just as with other types of content, their nature is determined, in part, by their effects on the down-stream consumer systems – in this case subjects’ capacities to make higher-order recognitional judgements about their experiences. So the one question which this account cannot (and is not designed to) answer, is why people tend to make such higher-order judgements at all. Here the answer, ‘Because they undergo higher-order analog contents’ – although it does give a good deal of additional information – is not really an explanatory one.[12]


4.4            Freudian phenomenologies

Would a Freudian unconscious, if it existed, contain phenomenally conscious states? Most people have a strong intuition that it would not. (It is, after all, supposed to be the unconscious mind.) But dispositionalist HOT theory probably entails that it would. It seems quite likely that experiential states would have to be available to the systems which constitute the Freudian unconscious, in order that revealing slips of the tongue or obsessive behaviours should be manifested in the right circumstances to subserve the unconscious’ purposes. It also seems plausible that the unconscious would be capable of higher-order thought, and of operating with the is–seems distinction. For it will need to have thoughts about how things will seem to the conscious mind if the latter undergoes a dream with a certain content, for example. In which case it looks likely that the Freudian unconscious would contain phenomenally conscious experiences, on a dispositionalist HOT account. Is this a problem for dispositionalist HOT theory?

            I am happy to bite this particular bullet. I think that Freudian theories of the unconscious are, in fact, almost completely false. But I am happy to allow that if they were true, then each person might be subject to two distinct and mutually inaccessible sets of phenomenally conscious experiences – the experiences of the conscious mind, and the experiences of the unconscious. Each of us would contain two distinct centres of subjectivity. Indeed, since it is often remarked that the Freudian unconscious has many of the properties of a person – it has its own goals, beliefs, and limited powers of agency – it is hardly surprising that it might turn out to have phenomenally conscious experience as well. Our pre-theoretical intuition to the contrary is not really worth very much, in my view, and is easily explained as arising from a conflation of access-consciousness with phenomenal consciousness.[13]

            Recall from chapter 1:3.5 that there are (at least) two distinct forms of access-consciousness. There is a form of access-consciousness which is defined quite generally in terms of (first-order) inferential promiscuity – where for a state to be access-conscious is for its content to be accessible to a wide range of other systems, including practical reasoning and verbal expression. And then there is a form of access-consciousness which is particularly tied to the capacity for higher-order thought (HOT).

            In the actual architecture of our cognition, I believe, these two forms of consciousness are co-instantiated. For it is the same set of contents which is both inferentially promiscuous and available to HOT (see figure 8.1; more on this in chapter 11). But it did not have to be so. For if Freudian theories had turned out to be true, then there would have been a set of contents which were available to HOT (and so phenomenally conscious, possessing dual content and all the properties of subjectivity) without being fully inferentially promiscuous, or available to (direct) verbal expression. This seems sufficient to explain the intuition that these experiences would be unconscious ones – because not accessible to the language-using subject – consistent with allowing that the experiences in question would, nevertheless, be phenomenally conscious.


4.5       Intra-modular HOTs

Another objection to the dispositionalist HOT approach is that phenomenally conscious states are in danger of proliferating in intuitively unacceptable ways. Suppose it turned out, for example, that there are HOTs internal to the visual system(s), helping to process visual input. This might work somewhat as follows. The visual system needs to be able to distinguish between changes in the visual field which are induced by movements of the head and eyes of the perceiver, and those which result from motion in the environment, in order to build a reliable picture of that environment and its properties. This might be done by allowing the visual system(s) to access the agent’s intentions or high-level motor instructions, say, which could then be used as internally-generated inputs to modulate perceptual processing. But if so, then it seems there will be HOTs internal to the visual module, and the result could be phenomenally conscious contents which are inaccessible to their subjects.

            It is actually very unlikely that anything like this is a genuine possibility. For, first, HOTs do not come cheap. To the best of our knowledge, systems which are capable of generating HOTs have emerged only once in the evolution of life on Earth – at some point in the great-ape–hominid lineage, in order to deal with the vital and complex exigencies of social intercourse.[14] So it seems unlikely that such systems could have evolved independently, internal to the visual systems of (presumably) a great many creatures.

            Second, it is far from clear that anything as rich and complex as HOTs about intentions-to-move would be required for the purpose. Simpler representations of initiations of activity in the relevant motor-systems would seem adequate for the job. (And recall that a state’s carrying information about intentions, because being reliably caused by them, is not sufficient for that state to have a higher-order content. It also has to be used as a HOT – and it doesn’t look as if modulating the processing of the visual system(s) is the right kind of use to be constitutive of HOT.)

            Moreover – and third – HOTs about intentions or about agency would not be sufficient for phenomenal consciousness in any case, on my account. The relevant HOTs have to be thoughts about experiences considered as subjective states of the perceiver, reflecting an understanding that experience may be private, partial, misleading or illusory. And the HOT system has to be capable of forming recognitional concepts of experience (seems red, seems green, and so on) in order for those states to acquire the requisite sort of dual content. None of this is plausibly present in the described intra-modular mechanism; nor is it easy to envisage one which would contain it.

            I am, however, prepared to bite this bullet in principle as well. If it were to turn out that recognitional concepts of experience are deployed internally within the visual system(s), then I would be committed to allowing that there are phenomenally conscious states which are inaccessible to the subject, in the sense of being unavailable to guide inferences or generate beliefs, or to be reported on. While I don’t see this as a real possibility, what the objection does bring out is that higher-order access and first-order access, while co-instantiated in the structure of human phenomenal consciousness (and probably for good reason – see chapter 11), are in principle dissociable. And according to dispositionalist HOT theory it is higher-order access which does the work in explaining phenomenality. So it is possible to imagine creatures having phenomenally conscious states which lack first-order accessibility.

            Why can’t I insist that phenomenal consciousness is constituted by a conjunction of first-order and higher-order access? Why can’t I respond to the difficulty by tapping into the intuition that states have to be fully accessible to their subjects in order to be phenomenally conscious? The answer is: for essentially the same reason that first-order (FOR) theorists cannot respond to the evidence of multi-layered perceptual cognition by insisting that only those first-order analog states which are available to the highest-level executive are phenomenally conscious (see chapter 6:3). The point is that these definitional moves add nothing to the proposed explanations. Our target is to explain the subjective feel of experience. And all the work in that explanation has now been done by our appeal to higher-order analog contents. To insist that these contents must, in addition, be available to the systems which engage in practical reasoning, construct memories, and/or generate verbal reports adds nothing.

            Is there any real problem for dispositionalist HOT theory here? I think not. Any intuition that the envisaged sort of inaccessible phenomenality is inconceivable seems easily explicable. We can appeal to the universal co-instantiation of first-order and higher-order access within our experience, together with the intuition that anything which was not first-order accessible would not be an experience. Well, so be it – I claim that it is possible to imagine states which are not experiences (nor images, nor belonging to any other familiar folk-psychological category) which are phenomenally conscious while being inaccessible to their subjects. There is no real objection to be mounted from here.



I have argued that dispositionalist HOT theory provides us with a successful reductive explanation of the subjective ‘what-it-is-likeness’ of phenomenally conscious experience. The account relies crucially on a form of consumer semantics to explain how the attachment of a HOT system to our perceptual contents can radically transform those contents, giving them a higher-order dimension of seeming or subjectivity. HOE theory, in contrast, has trouble explaining the ‘transparency’ of experience. And while actualist HOT theory could in principle advance essentially the same sort of consumer-semantics-based explanation, its insistence on the actual presence of HOTs would then be quite unwarranted, since consumer-semantic theories are themselves dispositionalist in nature.


Return to Carruthers' books

[1] Dretske (1995, ch.2) develops a somewhat similar position; but it is presented as an account of introspection, rather than as an account of phenomenal consciousness as such.

[2] In an elegant series of experiments, Taylor (1988) presented young children with a largely occluded picture of a familiar object – say an elephant of which only a portion of the toe-nail was visible. The children were asked to guess what the object was; naturally they got it wrong. The covering was then removed and the true nature of the object revealed. When the covering was replaced, the children were asked what someone else (who had been out of the room during this time) would think was in the picture, or would see in the picture. Younger children replied, ‘An elephant’, apparently reasoning that perception of any part of an object is sufficient to confer knowledge of it. (All the usual controls of memory, etc., were in place.)

[3] I assume for these purposes that the internal representation in question is a sentence of Mentalese, rather than an activity pattern in a connectionist network. This assumption is for convenience only at this point in the argument, provided connectionism is construed in such a way as to make room for internal representations – and if it isn’t, then it shouldn’t be taken seriously, in my view. I return briefly to an aspect of the Mentalese–Connectionism debate in chapter 10. See also Botterill and Carruthers, 1999, ch.8.

[4] See also the arguments of chapter 4:3.2 above and Botterill and Carruthers, 1999, ch.6, to the effect that scientific psychology needs to work with a notion of content which is narrow. For the only plausible way of fleshing out an account of narrow content will be in terms of some or other version of consumer semantics, I believe.

[5] Fodor famously alleges that there cannot be a localist version of inferential role semantics – any such view must be completely holistic, he thinks. See his 1987, and Fodor and Lepore, 1992. For defences of inferential role semantics against the charge of holism, see Carruthers, 1996a, and Botterill and Carruthers, 1999, ch.7. For defence of a rather different form of localism, see Devitt, 1994.

[6] Of course Wittgenstein (1953) famously argued that the very idea of private concepts of experience of this sort is impossible or conceptually incoherent. In due deference to Wittgenstein, I spent a good many years of my life trying to find a viable version of the Private Language Argument, one manifestation of which was my Oxford DPhil thesis (1979; see also my 1986, ch.6). I ultimately came to the conclusion that there is no good argument in this area which doesn’t presuppose some form of verificationism about meaning or quasi-behaviourism about the mind. But I don’t need to argue for this here. I doubt whether Wittgensteinians will be reading this book in any case; they are not exactly notorious for their openness to substantive enquiry.

[7] Note that it is because our application of the recognitional concept is driven just by the second-order analog content, that the concept in question can break free of any conceptual connections with the objects of first-order experience, and also of any theoretical beliefs about the causal role of such experience – hence becoming purely recognitional (i.e. becoming a ‘Cartesian concept’).

[8] In fact, since our common-sense psychology finds little place for an is–seems distinction in connection with pain, it may not overtly mark any difference between the concepts pain and feeling pain, either. So I do not mean to claim that our second-order recognitional concepts of sensation will necessarily receive separate expression. Rather, the best way to picture what happens is that our first-order concept pain acquires a second-order dimension when we come to understand that pains are items to which we stand in relations of privileged access. We ordinary folk surely do have a firm theoretical grasp of the subjectivity of pain experiences, even if our understanding of their representational nature is insecure; and this is sufficient for our concept pain to acquire a second-order aspect, on the present account.

[9] Indeed, it might appear to be an advantage in the present account that it can at least find a limited place for the notion of degrees of phenomenal consciousness, since the different forms of experience may become phenomenally conscious at different stages. But this is, of course, a very limited place. It remains unclear at what point in the child’s developing understanding of (some particular form of) subjectivity phenomenal consciousness will first appear, or whether we can make sense of the idea that such forms of higher-order analog content might appear by degrees, corresponding to the incremental nature of the child’s understanding.

      Answering these questions would require a worked-out reductive theory of intentional content, I suspect. But I have no such account to offer, of course. My lack of such an account is no real embarrassment at this point, however, since most other theories face essentially the same problem – for on the one hand (as we noted in chapter 1:3.8), it seems as if phenomenal consciousness must be all-or-nothing, and yet on the other, the terms employed in any proposed explanatory account will normally apply to a creature more-or-less. Indeed, even in advance of any explanatory proposal (but as soon as we begin to reflect), it looks most implausible that full-blown phenomenal consciousness could spring into existence between one moment and the next in the life of a developing organism.

[10] It might be objected that in a world where the only naturally occurring form of jade is jadeite, users of the term ‘jade’ would refer to jadeite rather than to jade in general (as we do). In which case it is hardly surprising that when they discover the constitution of jadeite they would take themselves to have explained jade – this is because they would have explained the substance which they refer to by ‘jade’. But in fact the same point can go through even if these people don’t use ‘jade’ as a natural kind term, or if they are explicitly working with a concept of ‘manifest-jade’. (Compare the discussion of manifest-water in chapter 2:3.5.) When they discover the constitution of jadeite they will have reductively explained manifest-jade in the form that it actually occurs, even if it is true that there are naturally-possible forms of manifest-jade for which they would have to seek a separate explanation, were they to occur.

[11] Compare: it is only necessary for someone to have some first-order concepts applying to surfaces of objects, and/or some behavioural dispositions which are focused on those objects, in order for all of their first-order visual states to represent properties of external objects, rather than complex properties of the retina. See the discussion in chapter 5:4.3.

[12] Of course it would be explanatory to answer this question by saying, ‘Because there are intrinsic properties of people’s experiences of which they are aware’, if such properties existed. So in this respect qualia-freaks can claim some explanatory advantage over dispositionalist HOT theory. But as we have seen in chapters 2, 3 and 7 above, (a) there are no good reasons to believe in the existence of any intrinsic, non-intentional, properties of experience (qualia), and (b) it is easy for a higher-order theorist to explain why people are so naturally tempted to believe in such properties.

[13] Might it not be said that our argument against Tye’s form of first-order (FOR) theory in chapter 6:4 is similarly worthless, depending on the same conflation? For part of that argument turned on the intuitive implausibility of the idea that there might be phenomenally conscious experiences to which the subjects of those experiences are blind. But the difference is that HOT theory can explain in what circumstances such inaccessible phenomenality might conceivably occur, whereas FOR theory was left struggling to find some principled way of explaining what kinds of first-order access might give rise to phenomenal feel.

[14] In fact there is some preliminary evidence that dolphins, too, may have theory-of-mind abilities (Dunbar, personal communication).