[Seminar] The neurobiology of perceptual and value based decisions: A memorable connection by Prof. Michael Shadlen, Columbia University
We ought to make decisions by combining evidence with prior knowledge (e.g., base rates) and the anticipated value/cost of deciding correctly. If the evidence arrives sequentially, the decision may take time as we wait for the accumulated evidence to reach a criterion level, or for a sufficiently compelling sample (e.g., an extrema). Both strategies are examples of “sequential sampling with optional stopping” (SSwOS). It is surprisingly difficult to differentiate these strategies based on behavioral data, but I will supply two examples in which nonhuman primates use an accumulate-to-threshold strategy to make decisions. These examples, drawn from perception and reasoning, have helped to elucidate the neural mechanisms of decision making, including speed-accuracy relationships. I will suggest that they implicate the involvement of associative memory. This insight could help resolve a much larger mystery. Many decisions are supported by just a few samples of evidence, supplied at once. Yet these decisions also exhibit speed-accuracy regularities consistent with SSwOS. Value-based decisions, such as a preference between two familiar items, are exemplary. Such decisions take more time and the preferences less consistent when the choice is between items of similar value. This speed-consistency relationship—analogous to speed-accuracy—is also explained by SSwOS. But what constitutes the samples of evidence? And why are they processed sequentially? In other words, where does the time go? I will describe recent findings (collaboration with Daphna Shohamy lab) that suggests the sequential samples are formed though a constructive process of valuation that draws on episodic (hippocampal) memory.