In the last 15 years, there has been a flourishing of research into the neural basis of reinforcement learning, drawing together insights and findings from psychology, computer science, and neuroscience. This remarkable confluence of three fields has yielded a growing framework that begins to explain how animals and humans learn to make decisions in real time. The confluence was catalyzed by the discovery of a close correspondence between the behavior of dopamine neurons in classical conditioning tasks and the prediction error in the temporal-difference (TD) algorithm from reinforcement learning.
The RLAI project is focused primarily on the artificial-intelligence aspects of reinforcement learning, but we have participated in some of the related fields as well. One way we have done this in past years is by developing models of classical conditioning that match the neuroscientific and behavioral data better, and by generating additional targeted behavioral data in collaboration with Prof. Kehoe of the University of New South Wales in Australia. In June 2009 we organized (with Prof. Precup of McGill) a special Multidisciplinary Symposium on Reinforcement Learning, held in conjunction with several international computer science conferences in Montreal. We were able to attract top scientists from neuroscience, operations research, and industry as invited speakers and about 150 scientists as participants. The program, videos, and information on this event are available at http://msrl09.rl-community.org. Finally, we have written an introduction to the psychology and neuroscience of reinforcement learning targeted at computer scientists and other people without a prior background in these fields.