Abstract

Traditional approaches to understanding neural mechanisms of economic choice treat it as a serial and localized process. Such models divide choices into discrete evaluation, comparison, and selection stages, and seek to link these stages with corresponding neuroanatomical loci. However, several recently observed features of single neuron and macroscopic data are difficult to reconcile with this divide-and-conquer approach. In this talk, I will argue that these data can more readily be reconciled with a model in which value-guided decisions are implemented in a distributed fashion across many brain regions performing canonical computations in concert. This account draws on recent ideas in deep learning to emphasize the importance of recurrent over feedforward architectures in solving real-world sequential decision problems. It suggests that value may be an emergent phenomenon, not represented explicitly in any particular part of the brain.