Location: Room 1504, NYU Shanghai | 1555 Century Avenue, Pudong New Area, Shanghai
Abstract: Decisions are based on predictions of the value of stimuli and actions. Value has commonly been viewed as a single dimension that varies from bad to good (punishment to reward), in analogy to light intensity varying from dark to light. I will present evidence that dopamine neurons in primate midbrain signal evidence for reward, but they are entirely insensitive to aversiveness (punishment). These experiments suggest that ‘reward’ and ‘aversiveness’ are two separate dimensions of value, each represented by two types of reinforcement corresponding to evidence ‘for’ and ‘against.’ I will propose that these four reinforcement signals are summed via G-protein-coupled receptors in one of eight ways, depending on receptor expression, to drive Hebbian learning and shape a neuron’s receptive fields to be one of eight value types. Reward and aversive information is mixed in early sensory neurons, then segregated into parallel paths, then divided into ‘good’ and ‘bad’ along each dimension. At the late motor stage, neurons receive evidence for reward or against aversiveness to mediate either approach or avoidance.