Ideally, the ideas of reinforcement learning could constitute part of a computational theory of what the brain is doing and why. A number of links have been drawn between reinforcement learning and neuroscience, beginning with early models of classical conditioning based on temporal-difference learning (see  Barto and Sutton, 1982;  Sutton and Barto, 1981,  1990;  Moore et al., 1986), and continuing through work on foraging and prediction learning (see  Montague et al., 1995,  1996), and on dopamine neurons in monkeys as a temporal-difference-error distribution system. A good  survey paper is available. See also  Suri, submitted. A  book collects a number of relevant papers. Doya has extensively developed  RL models of the basal ganglia. Many of these areas are very active at present and changing rapidly.