Return to articles on cognitive architecture 

16 Autism as Mind-Blindness: an elaboration and partial defence


Peter Carruthers


In this chapter I shall be defending the mind-blindness theory of autism, by showing how it can accommodate data which might otherwise appear problematic for it. Specifically, I shall show how it can explain the fact that autistic children rarely engage in spontaneous pretend-play, and also how it can explain the executive-function deficits which are characteristic of the syndrome. I shall do this by emphasising what I take to be an entailment of the mind-blindness theory, that autistic subjects have difficulties of access to their own mental states, as well as to the mental states of other people.


1. Introduction

In a series of publications since 1985 Alan Leslie, Simon Baron-Cohen and others have argued that autism should be identified with mind-blindness that is, with damage to an innate theory-of-mind module, leading to an inability to understand the mental states of other people. (See Baron-Cohen, Leslie, and Frith, 1985; Leslie, 1987, 1988, 1991; Leslie and Roth, 1993; Baron-Cohen, 1989b, 1990, 1991, 1993; and Baron-Cohen and Ring, 1994.) I shall be concerned to elaborate and defend this proposal, showing that it has the resources to handle rather more of the relevant data, and rather more elegantly, than even its originators have realised.

There is widespread agreement that autism at least involves a kind of mind-blindness. That is, it is generally agreed that autistic subjects have considerable difficulty in appreciating the mental states of other people well-documented in their difficulties with false-belief tasks, for example resulting in impaired social interaction and poor communicative skills. Where there is very considerable disagreement, concerns the explanation of this phenomenon, and also the question of how central it is in the aetiology of the syndrome as a whole.

Both Leslie and Baron-Cohen believe that mind-blindness lies at the very heart of the autistic syndrome. They maintain that autism results from damage to a specialised theory-of-mind module, which underlies the mind-reading abilities of normal subjects. This module is held to contain an implicit theory of the structure and functioning of the human mind, which is accessed whenever a normal subject ascribes a mental state to another person, or seeks a mentalistic explanation of their behaviour. It is possible that this module is organised into a number of distinct sub-systems (see Baron-Cohen and Ring, 1994; Baron-Cohen and Swettenham, this volume), and that it may develop in the normal individual through a number of different stages, perhaps corresponding to the different theory-stages postulated by some developmental psychologists (e.g. simple desire psychology; perception-desire psychology; belief-desire psychology; as postulated by Wellman, 1990 see Segal, this volume). But the two crucial claims made by those adopting this position are firstly, that the theory-of-mind module is an innate, isolable, component of the mind which embodies a theory of the nature and mode of operation of minds. (This is then a version of the so-called 'theory-theory' of our understanding of other minds, defended by many writers, including Lewis, 1966; Churchland, 1981; Stich, 1983; Fodor, 1987; Wellman, 1990; and others.) And secondly, that it is this module which is distinctively damaged in the case of autism.

Others (for example Frith, 1989; Harris, 1989, 1991, 1993; Hobson, 1993; and Melzoff and Gopnik, 1993) take a different view, arguing, in various different ways and for various different reasons, that the mind-blindness of autistic subjects is a consequence of some more basic deficit. I shall make no attempt at a systematic survey of these competitor theories here, focusing only on one which seems to have been gaining ground lately. This is the proposal put forward by Harris (1989, 1991, 1993) and elaborated more recently by Gordon and Barker (1994) and by Currie (this volume), that the mind-blindness of autistic subjects is an effect of a deeper inability to engage in imaginative thinking.

According to this alternative proposal, the fundamental deficit involved in autism is an inability (or at least a reduced ability) to engage in imaginative, counterfactual, suppositional thinking. It is for this reason, it is supposed, that autistic children rarely engage in spontaneous pretend play, and tend to display behaviours that are stereotyped and rigidly routinized. It is also held that the difficulties autistic people have in reading the minds of others results from this same underlying deficit, since mind-reading abilities are claimed to require the ability to identify oneself imaginatively with the other person. (This is the so-called 'simulation-theory' of our understanding of the minds of others, notably defended by Gordon, 1986, 1992, 1994; Harris, 1989; and Goldman, 1989, 1992, 1993.)

These competing explanations of autism involve us in wider disputes about the nature and origins of our conception of the mental states of other people in the normal case. I believe that in general there are powerful reasons for preferring the theory-theory to simulation-theory as an account of these matters (see Carruthers, this volume). I believe that there are convincing reasons, also, for preferring a modularised, nativistic, version of the former to the child-as-little-scientist versions of theory-theory proposed by Gopnik and Wellman (Wellman, 1990; Gopnik and Wellman, 1992; Gopnik, 1993 see Carruthers, this volume and 1992 ch. 8; and Segal, this volume). But it has to be admitted that simulation-theory currently provides a rather more convincing account of some aspects of the autistic syndrome specifically, the absence of pretend-play, and the inflexibility and lack of creativity of autistic thought. To the extent that this is so, to that extent we have some reason to prefer simulationism to any form of theory-theory.

I should stress at this point that in order to provide a viable alternative to the mind-blindness account of autism, the simulationist explanation must involve commitment to the sorts of radical simulationism defended by Gordon and Goldman. This is because the more limited form of content-simulation canvassed by a number of writers in this volume (see Carruthers; Heal; Perner; Botterill) is incapable of accounting for the data. For autistic subjects have problems across a whole range of theory-of-mind tasks, not just with those that involve predicting people's thoughts on the basis of other things that they are already known to think. When I speak of 'simulation' in this chapter, therefore, I should be understood as referring to its more radical variants.

What I propose to do, is not to mount any general defence of nativistic theory-theory, but only to show that an account of autism as mind-blindness can surmount the particular hurdle sketched above, being capable of providing explanations of the data that are just as convincing and elegant as can be given by simulationism. The core of my proposal will be that the mind-blindness theory has only appeared to be losing out in the above respects, because its proponents have paid insufficient attention to the consequences of their view for the access (or rather lack of access) that autistic subjects will have to their own mental states. It is here that I shall begin.


2. Mind-blindness and blindness to self

What account is the theory-theorist to provide of our knowledge of our own mental states? Here I shall help myself to the view elaborated briefly in Carruthers (this volume). (See also Gopnik, 1993, for a variant on the approach.) I claim that a theory-theorist should regard self-knowledge as analogous to the theory-laden perception of theoretical entities in science. Just as a physicist can sometimes (in context, and given a background of theoretical knowledge) see that electrons are being emitted by the substance under study; and just as a diagnostician can sometimes see a cancer in the blur of an x-ray photograph; so, too, we can each of us sometimes see (that is, know intuitively and non-inferentially) that we are in a state accorded such-and-such a role by folk-psychological theory.

It is, I claim, part of the normal functioning of the human mind that a mental state, M, if conscious, will automatically give rise to the belief that one has M where what one will recognise M as, is a state having a particular folk-psychological characterisation. This is not to say that all of the principles of folk-psychology which play a part in generating that belief will necessarily be accessible to us. But although the process of acquiring self-knowledge may involve theories that are only implicitly (and innately) known by the subject, still the upshot of that process the knowledge that I am in M will nevertheless be theory-involving. On the theory-theory account, what I recognise my own mental states as, are states having a particular folk-psychological role, even if I am unable to provide, consciously, a complete characterisation of that role.

Now, what would this account predict, concerning the self-knowledge of someone who has suffered damage to their theory-of-mind module? The answer is plain either such a person will be incapable of recognising their own mental states as such at all, or they will, at best, only be able to do so laboriously and unreliably. For, by hypothesis, it will be an innate theory-of-mind module which provides, in the normal case, the network of theoretical concepts and principles which enables us to individuate our mental states as such, as and when they occur. Anyone who lacks such a module, or in whom such a module is damaged, will either lack those concepts and principles altogether (and so be incapable of self-knowledge), or will have only a fragmentary grasp of them (in which case knowledge of self will be equally fragmentary), or will perhaps have acquired those concepts and principles laboriously, through general learning mechanisms (in which case self-attribution will, almost certainly, be slow and laborious also).

How do these predictions match up against the few empirical findings which are available? They are certainly consistent with the data reported by Baron-Cohen (1989a, 1991), who found that autistic subjects have as much trouble remembering their own recent false beliefs as they do in attributing false beliefs to other people, and that autistic subjects also have trouble drawing the appearance-reality distinction. The former result is exactly what the mind-blindness theory would predict: if autistic subjects have difficulty in understanding the notion of belief, then they should be incapable of (or at least poor at) ascribing false beliefs at all, whether to another agent or to themselves. Similarly, the mind-blindness theory would predict that autistic subjects will lack adequate access to their own experiences, as such (as opposed to access to the states which their experiences are of), and hence that they should have difficulty in negotiating the contrast between experience (appearance) and what it is an experience of (reality).

The above predictions are also consistent with the data recently reported by Hurlbert et al (1994), who tested three Asperger syndrome adults using the descriptive experience sampling method. This technique involves the subject wearing a small device that produces a beep at random intervals through the day, which the subject hears through an earphone. Subjects are instructed that their task is to 'freeze the contents of their awareness' at the moment when the beep began, and then to write down some notes about the details of the experience. Normal subjects report inner experiences in four major categories: inner verbalisation, visual images, unsymbolised thinking, and emotional feelings. All three of the subjects tested by Hurlbert et al using this method were high-functioning autistics, who appeared able to understand the experimental instructions two were capable of passing second-order false-belief tasks, and the third could pass first-order tasks, but not second-order tasks. Hurlbert et al found that the first two subjects reported visual images only no inner verbalisation, no unsymbolised thinking, and no emotional feelings. The third subject could report no inner experience at all.

Of course this is a very small sample of subjects, and the results are perhaps not easy to interpret. But they do suggest that autistic subjects might have severe difficulties of access to their own occurrent thought processes and emotions. For of course no one would want to deny that these subjects have thoughts and emotions. While autistic subjects surely must have propositional thoughts and emotional feelings, they do seem to have difficulty in knowing introspectively what their current thoughts and emotions are. (Why visual images should be any easier to self-attribute is something of a puzzle. Perhaps because perception-desire psychology is easier to acquire than belief-desire psychology see Wellman 1990 and because visual images are closely related to visual perception.)

Data apparently problematic for the predictions of the mind-blindness theory of autism are presented by Naito et al (1995), who claim to find that autistic subjects who are incapable of attributing false beliefs to other people, in standard false-belief tasks, can nevertheless remember their own recent false beliefs without difficulty. (See also Leslie and Thaiss, 1992.) But in fact this data is easily explained away. For the experimental set-up was such that subjects were asked to remember what they had earlier said an object was (Naito et al used deceptive-appearance tasks), and likewise to predict what another person would say that the object was. It is no surprise at all that autistic subjects should pass the first task but fail in the second. For, in fact, no theory-of-mind abilities are required in order for you to remember what you have just said. Whereas in order to predict what someone else will say, you first have to predict what they will think, and then generate a sentence appropriate to express that thought; which of course requires theory-of-mind ability.

If the mind-blindness account of autism is correct, then, we should expect, and we appear to find, that autistic subjects are as blind to their own mental states as they are to the mental states of other people. It is this consequence of the theory which will shortly be put to good work in the main body of this chapter: explaining why autistic children do not engage in spontaneous pretend-play, and explaining why the thought and behaviour of autistic subjects should be so rigid and inflexible. But first, it may be worth noting briefly, here, that such an account seems capable of explaining, also, the many reports of disordered and fragmentary sensations amongst autistic subjects (see Frith, 1989). For one of the normal functions of the theory-of-mind module will be to classify and identify our own sensations for us. If that module is damaged, precisely what one would expect is that the subject's inner life would seem chaotic, fragmentary, and confused.


3. The problem of pretence

The mind-blindness hypothesis has no difficulty in accounting for two of the triad of autistic impairments identified by Wing and Gould in their classic Camberwell study (1979) namely, impairment in social relationships, and in verbal and non-verbal communication. For these deficits are exactly what one would predict of someone who has severe difficulty in reading the minds of others. On the other hand, the third element of the triad the absence of pretend-play is, on the face of it, much more problematic. For why should mind-blindness interfere with pretence? One would expect it to interfere with social, or shared, pretence, of course, since this may require children to read the minds of their co-pretenders. And one would also expect a mind-blind child not to engage in the sort of play which involves attributing mental states to dolls or other pretend agents. But it is far from clear why mind-blindness should lead to any general deficit in pretend-play, such as is found in autism.

Leslie has boldly grasped this nettle, proposing that the very same cognitive mechanisms which are involved in theory-of-mind tasks also underlie the child's capacity for pretence (1987, 1988). He postulates the existence, in normal subjects, of a special-purpose mechanism, the de-coupler, whose function is to uncouple a given representation 'banana', say from its normal input-output relations, so as to enable that representation to be manipulated freely without affecting those relations (as when the child pretends that the banana is a telephone while retaining the knowledge that it is still a banana really). Crucial to this account, for present purposes, is Leslie's claim that the de-coupler functions by forming a second-order representation of the de-coupled representation. For he can then go on to claim (with some plausibility, given that assumption) that this very same mechanism is employed when the child turns to mind-reading tasks, forming a representation of the mental state of another person (also a second-order representation). If such a de-coupler were to form a crucial component of the theory-of-mind module, then it is only to be expected that mind-blind subjects would also display deficits in pretend-play.

Leslie's proposal is deeply unsatisfying, however. For there is nothing to motivate the claim that solitary pretence requires the capacity to form second-order representations of one's own representations, beyond the need to save the mind-blindness theory of autism. Rather, what is required for pretence is the capacity to entertain a representation in a different mode, or as the content of a different mental attitude (as different from belief as belief is from desire) namely, in the mode of supposition, or of imagination. To pretend that the banana is a telephone, the child does not have to represent its own representation of the banana, it just has to suppose that the banana is a telephone, and then think and act on that supposition. (See Perner, 1991, where this criticism is developed at length; see also Jarrold et al, 1994; Currie, 1994.)

In addition to the major criticism just sketched, Leslie's proposal faces a number of other problems, in response to which his account has been forced to become increasingly baroque (see Leslie, 1993). One is to explain why, if the very same de-coupling mechanism is employed in both pretence and theory-of-mind tasks, normal children should show competence in pretending so much earlier (age 2) than they are able to pass false-belief tasks (age 4). Another is to explain why properly social, co-operative, pretence should emerge so much later than solitary pretence (see Jarrold et al, 1994). And yet another problem is raised by the data, replicated in a number of different studies, suggesting that while autistic children do not often pretend spontaneously, they do have the capacity for pretence if prompted (see, for example, Lewis and Boucher, 1988). I shall not pursue these problems here. Instead, I shall show how the mind-blindness theory of autism can explain the absence of spontaneous pretence in autistic subjects without having to have recourse to the hypothesis of the de-coupler.


4. Resolution of the problem of pretence

In a nutshell, my suggestion is that because autistic children are, through mind-blindness, deprived of ready access to their own mental states, they are at the same time deprived of the main source of enjoyment present in normal pretending. So the reason why autistic subjects do not engage in spontaneous pretence is not because they cannot do so, but rather because they do not find the activity rewarding. The problem is one of motivation, not of incapacity. However, this idea will require some setting up.

Why do children pretend? What is enjoyable about the activity of pretending? Plainly what is pleasurable are not, in general, the physical actions by means of which the child carries out the pretence. True enough, the enjoyment of pretend-fighting, and of the sorts of rough-and-tumble wrestling engaged in by the young of most other species of mammal as well as human children, probably does lie in the physical exertion involved, and in the attempt to dominate an opponent. But the same can hardly be true of the kinds of symbolic pretending distinctive of human beings. There can be nothing enjoyable about the activities of putting a banana to one's ear and talking to it as such. Rather, what is enjoyable about pretence, I suggest, is basically the sense of being able to manipulate one's own mental representations in imagination; which then requires, of course, that one should have ready access to the states containing those representations.

The young of all species of mammal engage in play of some sort, the function of which seems to be to prepare them for adult activities. (See Smith, 1982.) Thus young springbok will practice the leaps which will one day be necessary to keep them out of reach of predators; young stag deer will engage in the kind of pretend-fighting and head-butting which will later be employed in the competition for mates in the rut; and the kittens of all species of cat will, in the course of their wresting with their siblings, practice just the kinds of stalking, leaping, biting, and holding which will form an essential part of adult hunting. It seems reasonable to suppose that the young of each species are programmed to find intrinsically rewarding, in play, just those activities (or those that are sufficiently similar to them, at least) which will form crucial components of their adult behavioural repertoire.

Now, for which adult activities are young children practising, when they engage in pretend-play? At one level you might be tempted to say: for those very activities that they are pretending to perform. The little girl pretending to be a mother bathing a baby is practising to become a mother who will bath a baby, the little boy pretending to be an airline pilot is practising to become an airline pilot, and so on. But it would be absurd to suggest that children are programmed to find these activities, described at this level, intrinsically rewarding. For most of the activities that children pretend to perform would not have existed at the time when human cognitive and motivational systems were evolving (mothering, of course, is one of the exceptions). Indeed, many of the things that children may pretend to do or be may actually be impossible to do or be in reality. The child who is pretending to turn objects to gold through a Midas-touch can hardly be practising to turn objects to gold as an adult! And the child who is pretending to be an aeroplane can hardly be practising to become an aeroplane in adulthood!

It is much more plausible to claim that children find rewarding that feature which is common to all forms of pretend-play, namely the manipulation of the child's own mental states, through supposing or imagining. (Hence, of course, the interest that normal children also take in fiction and story-telling an interest notably absent amongst autistics.) Then, just as you cannot enjoy running or jumping without being conscious of (or being aware that you are) running and jumping, so, too, I suggest, you cannot enjoy supposing or imagining without being conscious of your (mental) activity. In general, enjoying Xing presupposes awareness of Xing which is why you cannot enjoy digestion, sleepwalking, or subliminal perception.

It is surely incontrovertible that supposing, or imagining, is one of the distinguishing marks of the human species. It is the human capacity to suppose that such-and-such were the case and reason from there, or to imagine performing some activity and work out the consequences in advance, that underlies much of the success of our species. It has often been said that humans are distinctively rational animals. It may be closer to the truth to say that they are uniquely imaginative animals, since many other species seem to share our capacity to act intelligently in the light of desires, but none (excepting perhaps chimpanzees) share our ability to reason from supposed premises or to explore the consequences of imagined scenarios and certainly none has this ability to such a high degree. Small wonder, then, if much of human childhood should be devoted to forms of play whose function is to practice this very activity.

However, if the enjoyment of pretence requires a child to have access to its own mental state of pretending, then am I not committed, just as Leslie is, to saying that even a two-year-old must be capable of meta-representing its own mental representations? There are two factors distinguishing my proposal from Leslie's, however, in a way that renders it substantially more plausible. First, on my account, unlike Leslie's, there need be no necessity for the child who is pretending that the banana is a telephone to meta-represent its own representation of the banana. Rather, it need only at most meta-represent that it is now pretending. That is, the meta-representation involved may only extend to the attitude of pretending, without also embracing the content of what is pretended. Put differently, one might suggest that children are wired up to detect and represent, and find intrinsically rewarding, the mental state of pretending, without having to form a meta-representation of any other aspects (including the content) of that state. The second factor follows on naturally from the first. It is that young children may only have meta-awareness of their states of pretence, without yet being capable of meta-representing that is, conceptualising or thinking about those states. It may be that young children can detect their own pretence, and find it rewarding, without yet having the capacity to think that they are pretending.

A comparison with another proposed modular system may be of help at this point. Consider the face-recognition module, which has been hypothesised by many different researchers. It is highly plausible that it is this very module (at an early stage of growth and development) which is implicated in the neonate's well-documented discriminations of, and responses to, faces and face-like shapes. But it is highly implausible, of course, to maintain that the infant is, at that stage, capable of entertaining thoughts about faces. Rather, the early development of the face-recognition module enables an infant to detect faces without yet conceptualising them as faces. This early non-conceptualised awareness of faces may be one of the crucial inputs upon which later development of the module depends. So, too, then, in the case of the mind-reading module it may be that in early stages of its development it enables a young child to detect, and be aware of, its own mental state of pretending, among others (which is necessary if the child is to find that state rewarding), but without the child being capable, as yet, of entertaining thoughts about the pretence of itself, or of other people. Here, too, I suggest, the child's unconceptualised (introspective) awareness of its own mental states may be one of the crucial inputs to the normal growth and development of the theory-of-mind module.

I have suggested that what is basically enjoyable about pretend-play is the activity of supposing or imagining itself, which then presupposes that the child has ready access to its own mental states, particularly its own acts of imagining. This should not be taken as denying that there can be some forms of satisfaction to be derived from imagination which do not presuppose meta-awareness of one's own mental activity (or not immediately, anyway). In particular, it seems to be part of the functioning of the imagination that imagined scenarios should engage directly with the appetitive system, and with the emotions and this can be fun. Imagining a juicy steak can make you hungry, imagining an act of sexual intercourse can make you sexually aroused, imagining a free-fall from an aeroplane can make you frightened, and so on. It is easy to see why this should be so, if the main function of the imagination lies, not in fantasy, but in practical reasoning. (Here I am in agreement with Currie, 1994 and this volume; and Harris, 1991, 1993.) If the imagined result of the plan under consideration, of going to the kitchen and doing some cooking namely, a juicy steak had no tendency to engage with the appetitive system, it could have no tendency to set me in motion, either.

Although sometimes it is the imagined object, rather than the activity of imagining itself, which gives rise to enjoyment, it is arguable that even this enjoyment presupposes meta-awareness of one's own mental states. It is not (unassuaged) sexual arousal in itself which is enjoyable, but rather the bodily sensations distinctive of that state, of which one is aware. Similarly, it is not the fear of falling from an aeroplane which is itself enjoyable, but rather the thrill of this combined with the knowledge that I am only imagining, and really sitting safe in my arm-chair at home. So the conclusion stands undamaged: the enjoyment of pretence presupposes that subjects have ready introspective access to their own mental states. Small wonder, then if it is true that autism is a form of mind-blindness that autistic people should rarely be found to engage in pretence.


5. Stereotyped behaviour, inflexible thought

Properly understood, then as entailing its introspective corollary the mind-blindness account of autism can provide a smooth and elegant explanation for the fact that autistic children are distinguished by their absence of pretend-play. The explanation is that such children can pretend but do not particularly enjoy it. The question remains, however, as to why, when autistic children are prompted to pretend, their pretence is often so stereotyped and unimaginative. For example, if asked to pretend, with or without the aid of props, to do as many different things as they can think of (e.g. wear a hat, read a book, etc.) autistic children generate far fewer activities than do normal controls (see Jarrold et al, in preparation). Why should this be so, if autistic children really do have an undamaged capacity for imagination, as the mind-blindness theory of autism would predict, and contrary to what the simulationist account would propose?

The answer, in this instance, is easy it may be because they have had less practice at imagining. If the function of pretend-play is to exercise the imagination, then it is small wonder that those who have exercised little should perform less well. The autistic child who is prompted to engage in pretence is like a bird capable of flight who has never or rarely flown. Although the innate cognitive basis for pretence is present, just as the innate basis of flight is present in the bird, one would expect that it would be rusty and slow through ill-use.

In reality, however, the body of data provided by Jarrold et al (in preparation) is only the tip of a very large and well-established ice-berg, which may threaten yet to sink the mind-blindness theory of autism. This is the aspect of autism which is perhaps the best well-known, namely the tendency of autistic people to engage in repetitive activity, to be obsessed with order and ritual, to have very narrowly-focused interests, and to be generally very uncreative and unimaginative in their thought and behaviour. How is this to be explained, if autism is to be identified with mind-blindness? Are these features not better explained by the simulationist account? My strategy, here, will be to divide and conquer, offering differing explanations of different aspects of the phenomena.

The obsessive side of autism seems best explained as a by-product of social alienation, as has traditionally been proposed. Autistic children live in a world that is at once puzzling and threatening. To see this, think how much of the child's environment is social in nature, and how much of the time of a normal child will be occupied with social interactions of one sort or another interactions with peers or siblings, story-tellings with parents, negotiations with carers over foods and bed-times, and so on. Much of this would seem utterly opaque to a mind-blind child, and most of the behaviour of the people around such a child would seem wholly unpredictable. It would not be entirely surprising, then, if some autistic children should respond by isolating themselves still further from the puzzling social world around them, seeking refuge, out of loneliness and distress, in repetitive activity. And small wonder, also, if some autistic children should try to gain a measure of control over their world by imposing arbitrary, but orderly and predictable, rituals.

The narrowly focused interests of many autistic people, too, may be explained as resulting, partly from loneliness, partly as a further reaction to the opaque nature of the social world. (Yet another proposal will be made in the section 6 below.) I can well remember, myself, as an unusually lonely adolescent, spending hours absorbed in the play of light in the dew-drops on a bud or leaf, trying to persuade myself that this was a matter of the deepest metaphysical significance. And when one reflects on just how many of the normal objects of childhood and adolescent interest fiction, films, sex, and competitive sports, for example presuppose a good deal in the way of social awareness and mind-reading ability, it is not so surprising that autistic people might be happy to discover an area of interest albeit bus time-tables, or the calendar dates of days of the week which they can understand, and at which they can excel.

There remains, however, a good deal of evidence of lack of flexibility of thought amongst autistic individuals, which is not so easily explained as a mere by-product of social alienation and loneliness. In particular, a number of studies have tested autistic people on problem-solving and planning tasks, and found that they perform much less well than controls. For example, Ozonoff et al (1991a) tested autistic individuals on the Tower of Hanoi problem, finding that success or failure in this task successfully classified 80% of the children in both the normal and the autistic groups. The autistic children tended to persevere with unsuccessful strategies, despite repeated failure; whereas normal children were much more ready to try new strategies as old ones failed.

Such findings have led both Leslie and Roth (1993) and Baron-Cohen and Ring (1994) to accept that autistic subjects, in addition to mind-blindness, also suffer from a separate executive-function deficit, caused by collateral damage to areas in the frontal cortex close to those that appear to be involved in theory-of-mind tasks. While this by no means refutes their position after all, every theory of autism must accept that there are some phenomena (for example, the excessive thirst experienced by many autistic children) that are caused, not by the syndrome itself, but by collateral brain-damage nevertheless, it is good scientific practice to try to minimise accidents. I believe it is possible for the mind-blindness theory of autism to do better.


6. Practical reasoning and second-order thought

In philosophical circles it has been widely accepted, at least since Frankfurt's classic 1971 paper, that the normal operation of the human practical reasoning system routinely involves second-order evaluations of first-order thoughts and desires. In deciding what to do on a free afternoon while the family are away whether to stay in with a novel or go out into the garden to cut the grass I may consider such things as: how much I want that the grass should be cut, as opposed to how much I want to read my novel; whether there will be later opportunities to satisfy these desires if they are not satisfied now; how likely it is that I will be able to read free of interruptions if I stay in, and how likely it is that it will remain dry enough to garden if I go out; and so on. In all this I am both evaluating my own desires and determining the degree of credence which I am prepared to place in my own beliefs which presupposes (in many cases, at least) that I have access to my desires and beliefs.

Admittedly, it may sometimes be possible to weigh up one desire against another without raising, explicitly, any meta-representational question. I may, for example, ask myself simply how good the novel is, or how important reading would be as against neatly mown grass. In this I should be reflecting on the goals of my prospective actions, rather than on my desires for those goals as such. But even here, it is hard to see how such a process of reasoning could work, except through introspective access to my appetitive response to the thought of those goals. Similarly in the case of belief I may, sometimes, ask myself simply how likely it is that it will rain, without explicitly evaluating (that is, meta-representing) my belief that it will. But in many cases, at least, degrees of credence cannot rationally be assigned without an element of second-order reflection. For example, in answering the question concerning the likelihood of rain, I may have to recall the source of my belief (whether a weather-forecast, or "red sky at morning", or whatever) and evaluate its reliability. Such a procedure can only be possible if I know that I have (and so meta-represent) that belief.

Not only does human practical reasoning routinely involve access to our own beliefs and desires, it also involves something much stronger, I believe. For it is obvious, on reflection, that our practical reasoning systems routinely employ a kind of reflexive, introspective, access to our own recent sequences of occurrent conscious thinkings. (See my 1996 where this idea is developed in some detail.) Having recently been thinking about a problem in a particular way, I can then think about the thoughts that I have just entertained, and think about the problem-solving strategy I have just been employing. Or having been weighing up carefully the pros and cons of a number of different courses of action, I can then think, 'I shouldn't be so cautious and rational in my thinking spontaneity has its value too!'. All of which again presupposes, of course, that I have regular second-order access to my own occurrent thought-processes.

According to the modular hypothesis being defended here, the capacity for these sorts of swift and reliable forms of meta-access to our own beliefs, desires, and sequences of thinking and reasoning will be mediated, in the normal case, by the operation of the theory-of-mind module. It is therefore to be predicted that someone who is mind-blind, or whose theory-of-mind module is damaged, will experience considerable difficulty in tasks which involve the more complex (second-order) forms of practical reasoning. This is because such a subject's access to their own mental states will be relatively difficult, slow, and unreliable. We should therefore expect such a person to perform poorly on tasks that require them to evaluate their own desires or beliefs. And we should also expect them to perform equally poorly in tasks that require them to evaluate their own recent problem-solving strategies.

So here we have an explanation of the poor performance of autistic subjects in the Tower of Hanoi problem. (This is in addition to the partial explanation already sketched in section 5 above, that the suppositional reasoning of autistic subjects may suffer for lack of practice, because of the rarity of pretend-play in childhood.) The explanation is: they do poorly because solving such problems requires reflection on one's own problem-solving strategies. One must try out a strategy, either in imagination or on the board, and then when it fails, think about that strategy itself and how it might be modified and improved. It is no surprise, then, to find that autistic subjects mostly fail by perseverance, continuing with unsuccessful strategies despite repeated failures. For the mind-blindness theory of autism predicts that such subjects would experience difficulties of access to their own sequences of thinking and reasoning.

Here we also have a further explanation, I think, of the narrow and often idiosyncratic interests of many autistic adolescents and adults. (Again, this is in addition to the explanation given in 5 above, that autistic people will seek areas of interest which do not require mind-reading abilities, and where they can find, or impose, a measure of order and control.) For it is a reasonable hypothesis that regular second-order evaluation of our first-order desires and interests is one of the major engines driving the diversification of our value-systems. It is (at least in part) by reflecting on our desires and interests, and comparing them in imagination with alternatives, that our values diversify. We ask ourselves, 'Is my interest in stamp-collecting really that rewarding? Would it be a good idea if I got myself interested in nineteenth century novels instead?'. And it is by regular reflection on our projects and values that we attain, and retain, a sense of proportion. We ask ourselves, 'Is this project really worth all the time I am devoting to it? Do I care about it that much? Would I not rather be doing something else instead?'. By hypothesis, autistic subjects will find such second-order reflection difficult. So it is only to be expected that their range of interests might remain narrow, and that they should invest a degree of commitment in their projects which appears, to an outsider, to be out of all proportion to the true value of those activities.

My proposal has been that the executive-function deficits of autistic individuals are a consequence of their mind-blindness. An apparent problem for this proposal, however, is the finding that Asperger's syndrome subjects pass false-belief tasks while still failing at problem-solving tasks such as the Tower of Hanoi. (See Ozonoff et al, 1991b.) Does this not seem to show that it is problems of reasoning or imagining which are the more fundamental? Does the data not lend support, indeed, to the simulationists' view that it is, rather, a defective imagination which causes mind-blindness when sufficiently severe? (See Harris, 1993.) I do not believe, in fact, that this data raises any particular problem for the explanations offered above. For the proposal is not that the blindness of autistic individuals to the minds of themselves and others will necessarily be total. On the contrary, the proposal is that the theory-of-mind module may be partially intact in some cases of autism; and in other cases, that subjects may use alternative general learning strategies to gain at least a rudimentary grasp on the theory of mind. The prediction is only that all autistic people will find mind-reading, of themselves or others, relatively difficult, slow, and unreliable.

Now, the point to notice about the false-belief tasks is that they only require, for their solution, a limited number of mental-state attributions. The subject must recognise what the other person has and has not seen, for example, and use this to underpin the attribution of a false belief, before putting this together with the other's supposed desire, to generate a prediction about what they will do. And this is a problem that the subject is normally given, altogether, some minutes to solve. In the case of the sorts of problem-solving tasks that require regular second-order monitoring of one's own problem-solving strategies, in contrast, a whole myriad of mental states will need to be self-ascribed in a matter of seconds. Small wonder, then, that there might be particularly able autistic subjects whose grip on theory-of-mind is good enough to enable them to succeed in false-belief tasks, but still not good enough for them to keep reliable track of their own problem-solving strategies, and so still not good enough for them to succeed in tasks like the Tower of Hanoi.

(Note that it is an empirical prediction of the line being pushed here, that those autistic subjects who succeed in false-belief tasks will nevertheless take considerably longer to solve them than do controls. I find it surprising that amongst all the studies that have been conducted on the theory-of-mind abilities of autistics and normals, no one seems to have thought of gathering data on speed of theory-of-mind problem-solving.)

I have claimed that the mind-blindness theory of autism can explain the executive-function deficits common amongst autistic individuals. But let me stress that this is not, in any sense, to identify autism with failure of executive function. I have only claimed that an intact theory-of-mind module, giving the subject swift and reliable access to their own thought-processes, may be a necessary condition for success in complex practical reasoning tasks. There is, of course, a good deal more to practical reasoning than mere success in accessing one's own mental states. So it is only to be expected that there may be executive-function deficits which do not involve failure in theory-of-mind tasks. And, indeed, that is exactly what we find. Executive-function deficits occur in many clinical populations for example, obsessive-compulsive disorder, schizophrenia, hyperactivity, etc. who do not have theory-of-mind deficits. (I owe this point to Simon Baron-Cohen.)

In this connection, too, it may be worth noting some recent data which causes severe problems for the simulationist explanation of autism (or which will do so, at least, if it proves to be replicable and robust). This data relates to Williams syndrome subjects, who suffer from a rare genetic disorder giving them a distinctive, and highly uneven, cognitive profile. The crucial finding is that Williams syndrome subjects suffer severe difficulties in practical reasoning and problem solving tasks (for example, performing very poorly in a modified version of the Tower of Hanoi modified to become a verbal rather than a visual task, since Williams syndrome subjects have notorious problems with visuo-spatial reasoning); but that they have no difficulty whatever with theory-of-mind tasks on the contrary, their social and communication skills are precocious (Annette Karmiloff-Smith, personal communication).

This data may prove extremely important. For it is very hard to see how hypothetical, imaginative, thinking can be a presupposition of theory-of-mind ability, as simulationists maintain, if there exist subjects who lack the former but possess the latter. According to the mind-blindness proposal, in contrast, there can be (at least) two quite distinct ways in which problem solving abilities can be damaged: one is by damage to the hypothetical reasoning faculty itself (which, one might suppose, is what happens in the case of Williams syndrome); the other is by damage to the theory-of-mind module which gives normal subjects regular access to their own problem-solving thinking, serving to make the latter more efficient and reliable.



I have argued that the crucial implication of the autism-as-mind-blindness hypothesis, hitherto not sufficiently noticed, is that autistic subjects will not only experience difficulties in ascribing mental states to other people, but will equally have problems in achieving second-order awareness of their own mental processes. This is because the theory-of-mind module will be just as much implicated in self-attribution as it is in other-attribution. This implication then enables the mind-blindness hypothesis to provide a simple and elegant explanation of the fact that autistic children rarely engage in spontaneous pretend-play it is because, lacking easy self-awareness of their own mental states, they do not find the sorts of intrinsic satisfaction in pretence that normal children do. The same implication also enables us to explain the narrow interests and problem-solving deficits associated with autism, again because autistic subjects lack regular second-order awareness of their own desires and thought-processes.




I am grateful to the following for many valuable comments on earlier drafts of this chapter: Simon Baron-Cohen, George Botterill, Jill Boucher, Jack Copeland, Paul Harris, Chris Jarrold, Shaun Nichols, Peter J Smith, and Peter K Smith.