Theory of reinforcement learning and motivation in the basal ganglia

Abstract

This paper proposes how the neural circuits in vertebrates select actions on the basis of past experience and the current motivational state. According to the presented theory, the basal ganglia evaluate the utility of considered actions by combining the positive consequences (e.g. nutrition) scaled by the motivational state (e.g. hunger) with the negative consequences (e.g. effort). The theory suggests how the basal ganglia compute utility by combining the positive and negative consequences encoded in the synaptic weights of striatal Go and No-Go neurons, and the motivational state carried by neuromodulators including dopamine. Furthermore, the theory suggests how the striatal neurons to learn separately about consequences of actions, and how the dopaminergic neurons themselves learn what level of activity they need to produce to optimize behaviour. The theory accounts for the effects of dopaminergic modulation on behaviour, patterns of synaptic plasticity in striatum, and responses of dopaminergic neurons in diverse situations.