抄録
Animals choose actions that are most likely to yield rewards among alternatives. Standard processes in reward-based decision making is first to evaluate the reward for each action candidate and then to select the highest valued action at the highest probability. Neural correlates of reward expectation have been found in prefrontal cortex, parietal cortex and striatum. The midbrain dopamine neurons code the errors of reward expectation. Reinforcement learning model have been proposed as a possibility of the explanatory model of the neuronal activity and the choice behavior. However, Two critical questions remain to be answered. How is the reward value of each action candidate represented in the neuronal circuitry? How is a particular action selected based on the values assigned to multiple alternatives? To clarify the basal ganglia's role in such processes, we recorded striatal neuronal activities of macaque monkeys performing a free choice task of hand movement with variable combinations of reward probabilities. Majority of reward-predictive neurons encoded values for either one of two actions. Fewer neurons encoded differential values, linked directly to action selection. Our result suggests that striatal neurons represent action-specific reward values allowing reward-based action selection in the downstream of the basal ganglia. [Jpn J Physiol 55 Suppl:S50 (2005)]
