Chapter 21 - Bayesian decision making in two-alternative forced choices

Abstract

One of the fundamental questions in the study of decision-making is whether or not the strategies used by humans or animals are optimal, for example, if these strategies lead to the highest amount of collected reward. Recent experimental and theoretical results suggest that humans use Bayesian optimal strategies in a wide variety of tasks. In simple experimental settings, such as two-alternative forced choice (2AFC) tasks, the optimal decision strategy can be described quantitatively as an integration to threshold. Optimal decisions in 2AFCs can be separated into an inference stage, computing the probability for each choice, and a decision criteria stage, setting the time/accuracy trade-off. This chapter considers a Bayesian decision model that infers both the choice probability and the reliability of the sensory input, within a single trial, and based solely on the responses from a population of sensory neurons. This results in a “modified” diffusion model updating the impact of a sensory spike and the decision threshold online. As a consequence, sensory spikes early in the trial have typically a stronger impact on the final decision, and decisions are made with lower accuracy in harder trials. It is shown that this Bayesian decision model can account for recent findings in primates trained at a motion discrimination task.