Mindreading often requires access to one's own beliefs.Footnote ¹ Consider the following mental state attributions: Bill believes a first-aid kit contains bandages, though the kit actually contains feathers; Louise is an expert in British history, so she knows that the Battle of Hastings occurred in 1066; and Sally, age 2, desires candy when offered a choice between this and sushi as a snack. These mental state attributions do not depend on the interpretation of others' speech or behavior. Instead, they primarily depend on your beliefs (i.e., first-aid kits normally contain bandages; the Battle of Hastings occurred in 1066; children typically prefer candy over unfamiliar foods) in combination with other principles (e.g., experts in British history know a lot about British history).

The need to access beliefs is not restricted to just a few cases of mindreading. Instead, such access may be the rule in belief attribution: Most beliefs are true, and so one's own beliefs are indicative of what others believe. Because of this, people may have a default tendency to attribute their “true” beliefs to others (Fodor Reference Fodor1992; Leslie & Thaiss Reference Leslie and Thaiss1992; see Leslie et al. [Reference Leslie, Friedman and German2004] for a review of much evidence favoring an account making this claim). To operate according to this default tendency, the mindreading system requires access to beliefs.

The mindreading system's access to beliefs is problematic for Carruthers' account of metacognition, which denies such access (target article, sect. 2, para. 6).Footnote ² For if the system accesses beliefs when attributing mental states to others, then it should also access them when attributing mental states to the self. For instance, if the mindreading system accesses the belief “the Battle of Hastings occurred in 1066” when attributing it to Louise the historian, then the system should also be able to attribute this belief to the self. The mindreading system's access to beliefs allows people to engage in non-interpretative metacognition.

This proposal does not necessarily imply non-interpretative access to other mental states, such as intentions, desires, and past (currently false) beliefs. Unlike currently held beliefs, these other mental states are typically uninformative about the world and about others' mental states. One's intention to drink coffee says little about the world except perhaps that people sometimes drink coffee; and it says little about other people because relatively few share this intention at any time, meaning that it will seldom be useful to quickly extend this intention to others. So mindreading may not require access to such mental states. If the mindreading system lacks this access, it will also be lacking for metacognition.

Against our proposal, it might be claimed that the mindreading system does not access beliefs, but only inner speech and mental imagery that express beliefs. But this claim requires people to know which fragments of inner speech to use when attributing mental states to others. This claim also contradicts the view that people have a default tendency to attribute true beliefs. And given that inner speech and mental imagery are not required when answering questions about when the Battle of Hastings occurred (sect 2.1, para. 1), it seems doubtful that either is needed when answering when Louise thinks it occurred. Put more baldly, it is difficult to believe that attributing a desire for candy to Sally requires one to express in inner speech the belief “young children typically like candy.”

Our proposal is not strongly challenged by evidence that people sometimes confabulate when reporting beliefs. Confabulation is only problematic to the extent that it involves metacognitive errors in which people misreport beliefs. But such errors are difficult to distinguish from accurate reporting of irrational beliefs. When subjects reported that the rightmost of four identical pantyhose was softest (Nisbett & Wilson Reference Nisbett and Wilson1977), they might have been misreporting a belief (i.e., reporting a belief they did not have), but they also might have been faithfully reporting a false belief formed while deciding which item was softest. Also, that people sometimes err in reporting beliefs does not imply that they never have non-interpretative access to their beliefs. Self-interpretation and metacognitive errors may be particularly common for certain sorts of beliefs, and perhaps they are particularly common when people are motivated to report beliefs they do not actually have. In the pantyhose experiment, subjects might have had no belief about which item was softest, but still might have felt compelled to answer. Coming to this answer might open the way for metacognitive errors. But this does not imply that self-interpretation would be needed if subjects were instead asked about something they already believed, such as whether they thought the pantyhose samples were soft at all.

One might also challenge our proposal by conceding that the mindreading system accesses beliefs when making attributions about others, but then denying that it has this access for self-attributions. This defense makes little sense in light of the most detailed account of how beliefs are actually attributed (Leslie et al. Reference Leslie, Friedman and German2004). According to this account, the mindreading system operates according to the default assumption that beliefs are true, but sometimes overrides this assumption, as when reasoning about beliefs that are false. This account makes little distinction about whether beliefs are attributed to others or to oneself.

Carruthers' “mindreading is prior” model claims that mindreading and metacognition depend on the same cognitive system and on the same information. Our proposal is consistent with this claim and seems more consistent with it than is Carruthers' account of metacognition. Mindreading requires access to beliefs. Carruthers denies that such access is available in metacognition, which implies that the two processes draw on different information. The account we propose claims that access to beliefs occurs in both mindreading and metacognition, and this implies non-interpretative self-attribution of true belief.