There seems to be a growing divide between two major lines of research regarding the use of Bayesian models of human inference. On the one hand, judgment and decision researchers have conducted a program of research that might be described as 'deconstructing Bayes.' The basic idea is to design clever experiments that show how people systematically violate basic probability rules when a novel problem is described and they are explicitly asked to judge a probability. A classic example is the 'Linda problem' used to demonstrate the conjunction fallacy by Tversky and Kahneman. On the other hand, cognitive scientists have conducted a research program which demonstrates the power of Bayesian principles. These researchers assume that Bayesian principles are used implicitly to perform many common cogntive tasks such as categorization, memory retreival, perceptual identification, ect. A good example is the REM memory model used by Shiffrin. How can we account for this major gap in conclusions about the applicability of Bayesian principles to human cognition? This presentation will review the evidence, propose a hypothesis (formulated in conjunction with Rich Shriffrin) and briefly present some new experimental results that test the hypothesis (from a study conducted by Andrew Cohen).
Back to Graduate Summer School: Probabilistic Models of Cognition: The Mathematics of Mind