Description:

Professor Kevin Dorst is an Assistant Professor in the Department of Linguistics and Philosophy at MIT. He works at the intersection between philosophy and social science, focusing on rationality.

Title: Bayesian Bias

Abstract: Standard Bayesianism says that rational opinions must both be probabilistic and update in a way that satisfies a standard “Reflection” (or “martingale”) principle. This theory is not only dominant in formal epistemology, but embedded in the practice of social science: it is the basis on which economists predict market behavior, cognitive scientists explain inference, and psychologists evaluate human rationality.  But Standard Bayesianism is wrong, for it entails Access Internalism: that rational people must always be certain of exactly what rationality requires of them.  Without Access Internalism, the foundational arguments for Bayesianism (Dutch books, accuracy, etc.) establish a weaker theory that makes qualitatively different predictions.  I’ll illustrate this by focusing on calibration: the constraint that degrees of belief should match objective frequencies.  Though Standard Bayesianism predicts that rational people will usually be calibrated, real people often are not calibrated. It's often inferred from this that real people are irrationally overconfident.  But this is too quick. Without Access Internalism, rational (Bayesian) opinions will often be predictably miscalibrated—the connection between rationality and truth is looser than most have assumed