Four different accounts of the relationship between third-person mindreading and first-person metacognition are compared and evaluated. While three of them endorse the existence of introspection for propositional attitudes, the fourth (defended here) claims that our knowledge of our own attitudes results from turning our mindreading capacities upon ourselves. Section 1 of this target article introduces the four accounts. Section 2 develops the “mindreading is prior” model in more detail, showing how it predicts introspection for perceptual and quasi-perceptual (e.g., imagistic) mental events while claiming that metacognitive access to our own attitudes always results from swift unconscious self-interpretation. This section also considers the model's relationship to the expression of attitudes in speech. Section 3 argues that the commonsense belief in the existence of introspection should be given no weight. Section 4 argues briefly that data from childhood development are of no help in resolving this debate. Section 5 considers the evolutionary claims to which the different accounts are committed, and argues that the three introspective views make predictions that are not borne out by the data. Section 6 examines the extensive evidence that people often confabulate when self-attributing attitudes. Section 7 considers “two systems” accounts of human thinking and reasoning, arguing that although there are introspectable events within System 2, there are no introspectable attitudes. Section 8 examines alleged evidence of “unsymbolized thinking”. Section 9 considers the claim that schizophrenia exhibits a dissociation between mindreading and metacognition. Finally, section 10 evaluates the claim that autism presents a dissociation in the opposite direction, of metacognition without mindreading.

This commentary suggests an alternate definition for metacognition, as well as an alternate basis for the “aboutness” relation in representation. These together open the way for an understanding of mindreading that is significantly different from the one advocated by Carruthers.

Carruthers claims that “our knowledge of our own attitudes results from turning our mindreading capacities upon ourselves” (target article, Abstract). This may be true in many cases. But like other constructivist claims, it fails to explain occasions when constructed knowledge is accurate, like a well-supported scientific theory. People can know their surrounding world and to some extent themselves. Accurate self-knowledge is firmly established for both somatosensory and social pain.

Carruthers argues that an integrated faculty of metarepresentation evolved for mindreading and was later exapted for metacognition. A more consistent application of his approach would regard metarepresentation in mindreading with the same skeptical rigor, concluding that the “faculty” may have been entirely exapted. Given this result, the usefulness of Carruthers' line-drawing exercise is called into question.

Fifty years or so after the cognitive revolution, some cognitive accounts seem to be converging on treatments of how we come to know about ourselves and others that have much in common with behavior analytic accounts. Among the factors that keep the accounts separate is that behavioral accounts take a much broader view of what counts as behavior.

We agree with Carruthers that evidence for metacognition in species lacking mindreading provides dramatic evidence in favor of the metacognition-is-prior account and against the mindreading-is-prior account. We discuss this existing evidence and explain why an evolutionary perspective favors the former account and poses serious problems for the latter account.

This excellent target article helps to resolve a problem for dual-process theories of higher cognition. Theorists posit two systems, one of which appears to be conscious and volitional. It seems to control some behaviours but to confabulate explanations for others. I argue that this system is only conscious in an illusory sense and that all self-explanations are confabulatory, as Carruthers suggests.

Inner experience is proposed as a basis for self-interpretation in both children and adults, but young children's inner experience may not be comparable to our own. I consider evidence on children's attribution of inner experience, experience sampling, and the development of inner speech, concluding that Carruthers' theory should predict a developmental lag between mindreading and metacognition.

Carruthers' arguments depend on a tenuous interpretation of cases from the confabulation literature. Specifically, Carruthers maintains that cases of confabulation are “subjectively indistinguishable” from cases of alleged introspection. However, in typical cases of confabulation, the self-attributions are characterized by low confidence, in contrast to cases of alleged introspection.

Carruthers considers and rejects a mixed position according to which we have interpretative access to unconscious thoughts, but introspective access to conscious ones. I argue that this is too hasty. Given a two-level view of the mind, we can, and should, accept the mixed position, and we can do so without positing additional introspective mechanisms beyond those Carruthers already recognizes.

Mindreading often requires access to beliefs, so the mindreading system should be able to self-attribute beliefs, even without self-interpretation. This proposal is consistent with Carruthers' claim that mindreading and metacognition depend on the same cognitive system and the same information as one another; and it may be more consistent with this claim than is Carruthers' account of metacognition.

We argue that while it is a valuable contribution, Carruthers' model may be too restrictive to elaborate our understanding of the development of mindreading and metacognition, or to enrich our knowledge of individual differences and psychopathology. To illustrate, we describe pertinent examples where there may be a critical interplay between primitive social-cognitive processes and emerging self-attributions.

Carruthers offers a promising model for how “we” know the propositional contents of “our” own minds. Unfortunately, in retaining talk of first-person access to mental states, his suggestions assume that a higher-order self is already “in the loop.” We invite Carruthers to eliminate the first-person from his model and to develop a more thoroughly third-person model of metacognition.

Carruthers views unsymbolized thinking as “purely propositional” and, therefore, as a potential threat to his mindreading-is-prior position. I argue that unsymbolized thinking may involve (non-symbolic) sensory aspects; it is therefore not purely propositional, and therefore poses no threat to mindreading-is-prior. Furthermore, Descriptive Experience Sampling lends empirical support to the view that access to our own propositional attitudes is interpretative, not introspective.

Thinkers in related fields such as philosophy, psychology, and education define metacognition in a variety of different ways. Based on an emerging standard definition in psychology, we present evidence for metacognition in animals, and argue that mindreading and metacognition are largely orthogonal.

While Carruthers denies that humans have introspective access to cognitive attitudes such as belief, he allows introspective access to perceptual and quasi-perceptual mental states. Yet, despite his own reservations, the basic architecture he describes for third-person mindreading can accommodate first-person mindreading without need to posit a distinct “introspective” mode of access to any of one's own mental states.

Carruthers presents an interesting analysis of confabulation and a clear attack on introspection. Yet his theory-based alternative is a mechanistic view of “mindreading” which neglects the fact that social understanding occurs within a network of social relationships. In particular, the role of language in his model is too simple.

Carruthers' “mindreading is prior” model postulates one unitary mindreading mechanism working identically for self and other. While we agree about shared mindreading mechanisms, there is also evidence from neuroimaging and mentalizing about dissimilar others that suggest factors that differentially affect self-versus-other mentalizing. Such dissociations suggest greater complexity than the mindreading is prior model allows.

I argue that the very reasons Carruthers gives for why the “mindreading is prior” account should allow introspective access to perceptual/quasi-perceptual states, can be given for thought, as well. I also argue that we have good subjectively accessible grounds for the intuition in introspective thoughts, notwithstanding Carruthers' argument to the contrary and his attempt to explain the intuition away.

Carruthers rejects developmental evidence based primarily on an argument regarding one skill in particular: understanding false beliefs. We propose that this rejection is premature; and that identifying and examining the development of other subcomponent skills important for metacognition and mindreading, such as the ability to assess levels of knowledge, will in fact be useful in resolving this debate.