Impaired metacognition in autism

We agree with Carruthers' claim that both mindreading and metacognition are impaired in autism, and that this speaks against models 1 and 3. However, we wish to provide more decisive evidence for impaired metacognition in autism, given that the evidence cited by Carruthers is problematic. For example, contrary to Carruthers' suggestion (sect. 10, para. 10), Kazak et al. (Reference Kazak, Collis and Lewis1997) did not find statistically significant differences between participants with and without autism in either mindreading or metacognition (see Kazak et al., p.1005).

Clearer evidence for metacognitive deficits among children with autism emerges from Williams and Happé (in press a), who assessed awareness of own intentions. Compared to age- and ability-matched comparison children, children with autism were significantly less likely to (a) correctly report their reflex movements as unintentional, and (b) correctly recognise their own mistaken actions (drawing a picture different to that intended, through experimenter manipulation) as unintended. The performance of children with autism on these measures was significantly associated with performance on traditional false belief tasks, independent of verbal ability. These findings suggest that children with autism have a limited ability to represent their own intentions and that these difficulties are fundamentally associated with established difficulties in representing others' mental states (in this case, false beliefs). These findings provide robust evidence against both models 1 and 3, but do not differentiate models 2 and 4.

Impaired episodic memory in autism implies impaired metacognition

Several researchers (e.g., McGeer Reference McGeer2004; Nichols & Stich Reference Nichols and Stich2003; Raffman Reference Raffman1999) have claimed that the autobiographical reports of individuals with autism show that metacognition is intact in autism (supporting models 1 and 3). However, none of these authors, nor Carruthers in his target article, distinguishes between semantic and episodic memory.

Episodic memory is associated with consciously remembering personally experienced events, whereas semantic memory is concerned with factual information. Caution must be exercised when attributing to people with autism memory processes of the episodic kind. I may know that a particular event has happened to me in the past, and hence report details of the event quite accurately, without actually remembering the event. Only this latter kind of “remembering” is thought to rely on metacognition (e.g., Perner Reference Perner, Tulving and Craik2000).

Contrary to models 1 and 3, it may be that people with autism do not engage the same metacognitive processes as typical individuals do when reporting events from their past (Lind & Bowler Reference Lind, Bowler, Boucher and Bowler2008). Lind and Bowler (under revised review; see also Lind Reference Lind2008) found that children with autism (n=53) were as able as age- and ability-matched comparison participants to distinguish events that had occurred from events that had not (whether or not a picture had been picked up and named earlier in the test session). However, participants with autism were significantly impaired at remembering the source of such events (i.e., who – themselves or the experimenter – had picked up the picture and named it). That is, the participants knew that X event had occurred, but they had difficulty in remembering the spatio-temporal context in which X occurred. Therefore, the metacognitive status of the personal memories reported by individuals with autism might justifiably be questioned, and cannot be taken as support for models 1 and 3.

A case of impaired metacognition but intact mindreading? Distinguishing model 2 from model 4

The autism research cited earlier supports equally models 2 and 4. As far as we can tell, once Carruthers concedes (sect. 3) that the mindreading system has different information available to it for the cases of self and others, respectively, the only way in which model 2 differs theoretically from model 4 is with respect to the role of introspection of own propositional attitudes. Other commentators will no doubt debate whether it is possible to introspect our propositional attitudes directly. Here we concentrate on Carruthers' suggestion that different predictions emerge from models 2 and 4.

Williams and Happé (in press b; see also Williams Reference Williams2008) addressed a potential confound within the classic “Smarties” unexpected contents task (Hogrefe et al. Reference Hogrefe, Wimmer and Perner1986), the task used most widely to assess awareness of false beliefs in self and others. In the Smarties task, participants state what they (falsely) believe is inside a Smarties box before they are asked the critical false-belief test questions. As such, it may be possible to answer the Self test question (‘What did you think was inside the box, before you looked?’) simply by remembering what one said and not necessarily what one believed. Although parallel performance across the Self and Other test questions of the task is usually observed among children with autism (e.g., Fisher et al. Reference Fisher, Happé and Dunn2005), this potential confound may have led to an over-estimation of the ability of children with autism to reflect on their own false beliefs.

To test this possibility, we rigged a situation in which participants were asked by the experimenter (who feigned mild injury) to “get me a plaster,” and had ready access to three different boxes: a plasters box, a crisps tube, and a sweets box. By opening the plasters box (which actually contained candles), participants demonstrated their (false) belief that the box contained plasters. However, having never verbalised their belief, success of the Self test question of this task (“What did you think was in the box, before you looked?”) must reflect participants' recall of their false belief rather than of any prior statement.

We found that participants with autism were unique in finding the Self test question significantly harder than the Other-person question (“What will x think is inside the box, before s/he looks inside?”). Almost a quarter (21%) of our sample of 52 children with autism failed the Self question but passed the Other-person question. In contrast, less than 4% of participants with autism showed the opposite pattern of performance. Parallel performance across the test questions was observed in age- and ability-matched comparison participants, and in typically developing 3- to 5-year-olds. These results seem to show the kind of dissociation between mindreading and metacognition that Carruthers suggests would follow from model 2, but not from model 4.

In conclusion, we believe the data so far from autism support model 2 over model 4. But we are grateful for Carruthers' bold and exciting analysis, which helps to shape a new research agenda to answer the fascinating question: How well can people with autism “read their own minds”?