Title: Decision-theoretic models of action and their representations: the knowledge problem.
Speaker: Laurence T Maloney, Psychology & Neural Science, New York University
Abstract: Imagine that you are holding a ball. Now survey the room around you. Estimate the probability that you can throw the ball and hit each of the objects in the room – the drinking glass across the table, the doorknob on the other side of the room, your classmates. Are your estimates accurate? Would you bet money on them? Bayesian decision theory (BDT) is the mathematical framework common to many models of biological action under uncertainty. Using BDT as a model of biological action presupposes that the organism has complete knowledge of the probabilities that any specific action will lead to any specific outcome. But do organisms really have this knowledge? I will first describe BDT and its applications to modeling human movements and then describe recent work estimating the organism’s actual representation of the uncertainty inherent in all of its possible actions and testing whether it is accurate.
Joint work with Hang Zhang, Nathaniel D. Daw. Supported by NIH EY019889 and a Guggenheim Fellow Award.