I just read a paper by Bob Gordon entitled “Simulation Without Introspection or Inference from Me to You” (in Mental Simulation, ed. M. Davies and T. Stone, Blackwell, 1996). Bob argues that introspection (looking inside your mind/perceiving your mental states/observing the qualitative aspect of your mental states) is unnecessary for mental state attributions to oneself (and to others). All you need is “ascent routines” (an idea indebted to Gareth Evans). Here is how the ascent routine for belief self-attribution works: consider whether P, and if so, then conclude that you believe that P. When considering whether someone else, “recenter your egocentric map” and proceed as in your own case, although your conclusion now apply to the other person. Bob also suggests that introspection develops later than the attribution of mental states. I find this compelling.
Here is another reason to believe Bob’s account:
Why should you need to introspect in order to ascribe to yourself mental states while no one thinks you need to introspect in order to act in a nonverbal way on those same mental states (or at any rate, on the mental causes of your behavior)? You can seek water when you are thirsty (“want water”), seek food when hungry (“want food”), pick up a toy when you believe it’s in front of you, etc. No need to introspect there. But isn’t the attribution of mental states just another kind of behavior? And if you can act on your thirst by seeking water without introspecting first, why wouldn’t you be able to act on your thirst, in appropriate circumstances, by saying “I’m thirsty” (which in many circumstances would actually qualify as water-seeking behavior, especially for helpless babies) without introspecting first?
Obviously there is a lot more to say here (including facing some objections), but I’ll stop here for now.