Computational algorithms and neural mechanisms for estimating decision confidence

From Santa Fe Institute Events Wiki

By Adam Kepecs, CSHL.

Decision-making is a fundamental process for adaptive behavior throughout the animal kingdom. In addition to making a choice, estimating the degree of uncertainty or confidence in a decision confers significant benefits for a broad range of activities. Yet, despite its importance, the neural processes for uncertainty estimation and the computational algorithms used during behavior are poorly understood.
My talk will describe a combined computational and experimental approach to investigate the algorithms and neural processes for decision uncertainty. We recorded single neuron activity in rats trained in an olfactory categorization task that allowed us to manipulate the uncertainty of individual discrimination problems. We found that firing rates of many neurons recorded in the orbitofrontal cortex were correlated with the difficulty of a decision during reward anticipation and could be used to predict outcomes on a trial-by-trial basis. By delaying reward delivery, we were able to show that rats can report their confidence by selectively aborting less certain trials. The pattern of results observed could not be accounted for by standard reinforcement learning models for reward prediction. Instead, they could be parsimoniously explained by decision models that can compute an instantaneous measure of confidence in each choice, along with the choice itself.
Together, these results indicate that the neural representation of and access to an explicit, internal measure of decision confidence is fundamental to adaptive behavior. We suggest that decision confidence, like reward value, is likely to be a core variable for decision making and therefore there is a need to investigate models of decision making that incorporate mechanisms for estimating confidence.
Joint work with N. Uchida, H. Zariwala and Z.F. Mainen.

Back to Agenda.