12 May 2017
Jan Drugowitsch
Department of Neurobiology
Harvard Medical School
The variability of human and animal decisions based on perceptual evidence is frequently larger than what can be explained by variability in the evidence itself. This additional variability has in the past been ascribed to noisy sensory processing or stochastic action selection, thus placing its origin at the peripheries of the decision-making process. Based on a combination of psychophysical experiments and computational modeling I will discuss evidence that most of the variability instead has its origin in imprecisions in central probabilistic computations. This finding supports the view that real-world decisions are computationally intractable and need to be approximated, causing this additional variability. Indeed, a sizable fraction of the variability we observed in our experiments can be attributed to deterministic deviations from Bayes-optimal decision strategies. The residual unstructured variability seems to arise from perturbed temporal evidence accumulation, but its precise origin still remains to be determined.