What I claim is that we make systematic errors about our own thoughts, and that the pattern of errors reveals something about the mechanisms that normally give us access to those thoughts. (Compare the way in which visual illusions are used by cognitive scientists to give us insight into the mechanisms involved in visual perception.) In particular, I claim that people make errors whenever they are provided with cues that would lead them to make a similar error about the thoughts of a third party. This suggests, I think, that they are using the same mental faculty for both (often now called the ‘mindreading’ faculty), relying upon the same sorts of cues.
[W]hat I argue [in other words] is that there is a single ‘mindreading’ faculty that enables us the perceive our own thoughts as well as the thoughts of other people. This faculty evolved initially for social purposes, enabling us to anticipate (and sometimes to manipulate) the behavior of other people, as well as to better coordinate cooperative activities. But it can likewise be turned on the self, relying on the same channels of information that are used when interpreting the behavior of others. Sometimes we attribute thoughts to ourselves by literally perceiving our overt behavior. But often we rely on sensory cues that utilize the same perceptual channels, such as our own visual imagery, or our own inner speech.
There’s a roughly analogous claim you can make about the content of concepts, I think. Sometimes people proceed as if the content of concepts is “directly accessible” — simply reflecting on a concept is enough to tell you what its content is. (Stock example: if you reflect on the concept bachelor you can see that it is the concept of an unmarried man.) But an alternative picture is that because the content of a concept is fixed by how the concept is used, it cannot be understood by “directly accessing” its contents, but must instead be grasped by looking for use-based evidence of its content. (To make the analogy explicit: some people think we can sometimes directly access the content of our own mind, but Carruthers argues that it must be grasped by synthesizing behavioral evidence.)
Which picture of concept-analysis you ought to adopt may have implications for the debate over compatibilism in free will. Compatibilism typically arises from synthesis of data about how we use the concept of free will; incompatibilism typically arises from direct reflection on what the concept requires. Assuming the usage-data does support compatibilism, resolving the methodological question about concept-analysis will help solve the compatibilism debate.