Angela Yu
(Department of Computer Sciences University of California)
Thursday 14th June 2012
15:30 - 16.30
B10 Seminar Room, Basement,
Alexandra House, 17 Queen Square, London, WC1N 3AR
Oft Wrong, but Never in Doubt: Pascal's Wager in Everyday Cognition
Humans often exhibit a stubborn tendency to detect patterns where patterns do not exist, and predict future outcomes with great confidence based on flimsy evidence. I present a normative account of this bias, as a rational consequence of extrapolating from the past to the future based on noisy or incomplete information. Just as Pascal pointed out that it is rational for a skeptic to wager that God exists, since the reward of being right far exceeds the cost of being wrong, I will show that it is Bayes-optimal to err on the side of over-confidence in divining the future. I will also present data from a diverse array of behavioral experiments, in which a large amount of the trial-to-trial variability in subjects' response choice and delay can be attributed to subjects' making implicit Bayesian predictions about future stimuli based on a particular assumption (faulty, in these cases) about the statistical relationship between the past and the future. Finally, I discuss some neural evidence of how these statistical predictions are computed and represented in the brain.
http://www.cogsci.ucsd.edu/~ajyu/