"The source of the difficulty here is obvious: standard models for explaining such correlations proceed via the idea that facts cause beliefs, or that beliefs cause facts, or that there was a common cause." . . . .