Abstract: Rejecting the claim that human reasoning can approximate generally NP-hard Bayesian models—in the sense that the mind’s actual “computations” come close to, or be like, the inferences that Bayesian models dictate—this paper addresses whether a Bayesian model can have normative pull on human reasoners. For such normative pull to arise, we argue, a well-defined and empirically supported approximation relation is required—but broadly absent—between (i) human reasoning on the ground and (ii) the behavior of a non-NP-hard model. We discuss a responsible stance on this issue.

A long abstract can be found here.