Carruthers argues that passivity symptoms (e.g., thought insertion) in schizophrenia result not from a special metacognitive deficit, but from “faulty data being presented to the mindreading system” (sect. 9, para. 2). Although I endorse Carruthers' Frith-inspired (Frith et al. Reference Frith, Blakemore and Wolpert2000a; Reference Frith, Blakemore and Wolpert2000b) appeal to efference-copy deficits in the explanation of passivity symptoms, his claim that the mindreading faculty itself is undamaged raises questions. First, any attribution of one's own thoughts to another is equally a mistake in first- and third-person mindreading (false positives count as errors just as much as false negatives do). Carruthers should therefore hold that mindreading – first- and third-person – is deficient in these forms of schizophrenia; this still allows him to deny any dissociation between mindreading and metacognitive abilities, in line with what his theory predicts. It also avoids his having to make the hard-to-test claim that it is intermittently faulty data and not an intermittently faulty mechanism that is to blame for passivity symptoms.
Second, Carruthers holds that humans have introspective access to some mental states (e.g., perceptual states, imagery, and inner speech), but not to cognitive attitudes such as belief. But if information is extracted from globally broadcast perceptual states in third-person mindreading without introspection occurring, why think that the extraction of information from inner speech and visual imagery during first-person mindreading involves an introspective process different “in kind” from the way we form beliefs about the mental states of others? If, as Carruthers argues, passivity symptoms result from faulty data being input to the mindreading system (data that should have been interpreted as internally generated is interpreted as externally generated), then it seems the very determination of whether an input is self or other-generated – and thus whether one is seeing or visualizing, hearing or sub-vocalizing – requires an inferential or interpretative step (Langland-Hassan Reference Langland-Hassan2008).
Carruthers would likely respond that this inner-or-outer inferential step involves nothing more than the “trivial” form of inference that occurs in any layered representational scheme, where representations at one level can, in a “supervisory” role, intervene on those at another. However, many instances of third-person mindreading are equally fast and automatic, and they are implicit in the very cases of metacognition that, on Carruthers' theory, would be achieved through the “encapsulated” process of introspection. Consider a visual representation had by someone who looks up and sees another person staring at him. Suppose this visual perceptual state is accessed by the mindreading system, which issues in the introspective judgment: “I see a man seeing me.” This judgment contains within it a judgment that another person is having a visual experience of a certain kind (cf. Jeannerod & Pacherie's [Reference Jeannerod and Pacherie2004] “naked intentions”). So, unless the mindreading faculty in its introspective mode lacks the concepts needed for this judgment (unlikely, since it must have the concepts of self and of sight in order to issue any introspective judgments about visual experience), third-person mindreading can occur through the encapsulated “introspective” process that Carruthers describes. Yes, some cases of third-person mindreading require much more sophisticated feats of interpretation, but so too do many cases of first-person mindreading, as revealed by the confabulation data Carruthers discusses (Gazzaniga Reference Gazzaniga and Gazzaniga1995).
Thus, even if it is possible to draw a line between mindreading that is informationally encapsulated and that which is not, it will not cut cleanly across cases of first- and third-person mindreading. Nor is the existence of such domain-specific mechanisms supported by recent neuroimaging studies (Decety & Lamm Reference Decety and Lamm2007). What we have instead are inferences, concerning both first- and third-person mental states, that require greater or lesser degrees of supporting information; none of this implies a special mode of access to facts about one's own mental states. This is obscured by the tendency of researchers to compare easy cases of metacognition (e.g., inferring one's intentions from one's own inner speech) with difficult cases of third-person mindreading (e.g., inferring what someone thinks based solely on their posture and facial expression) – for it creates the impression that first-person mindreading occurs through some more “direct” process. But if we instead compare the third-person mindreading that occurs when we judge that a person believes what we hear her saying, to the first-person mindreading that draws on “listening” to one's own inner speech, there is less intuitive pressure to posit a difference in the kind of inference. Of course, if there were genuine dissociations revealed between third- and first-person mindreading abilities, as Nichols and Stich (Reference Nichols and Stich2003) and Goldman (Reference Goldman2006) claim, then we would have reason to posit differences in the kinds of mechanisms and inferences involved in each; but Carruthers is at pains to deny any such dissociations, and his alternative explanations are plausible enough.
The issue can be reframed in terms of the larger evidence base we have for first-person rather than third-person mindreading. Carruthers notes that the resources available to first-person mindreading are different because, “unless subjects choose to tell me, I never have access to what they are imagining or feeling” (sect. 2, para. 8). This is potentially misleading; the situation is rather that the single mindreading system, as he describes it, only ever has access to globally broadcast perceptual and quasi-perceptual representations (and memory), and, with this single source of information, must accomplish both its first- and third-person mindreading tasks – one of which is to determine whether the signal counts as a case of imagining or perceiving in the first place.
The fact that we have so much more “evidence” for first-person mindreading than third-person may still tempt some to posit a special form of access. Yet, if humans always audibly narrated their inner speech and expressed the contents of their visual imagery, the evidence bases for first- and third-person mindreading would be comparable. So it may be a contingent fact about how humans behave that accounts for the difference in evidence bases, not a special mode of access.
I therefore urge Carruthers to adopt a more thoroughgoing anti-introspectionism. Not only can first-person mindreading be explained without appeal to the introspection of propositional attitudes, it can be explained without granting a distinct introspective form of access to any of one's own mental states.
