Abstract
Intelligence relies on the capacity to find appropriate sequences of decisions over large decision hierarchies. We do not know yet how decisions are formed at different levels of a hierarchy at the time-scale of a single action or thought. Here we exploit the association between individual perceptual decisions and patterns of activity in the visual cortex to measure the timing of decisions made at different levels of a decision tree.
We trained macaque monkeys to mentally trace a curve through two bifurcation points and presented stochastic evidence at every bifurcation by varying the luminance of the two branches. The animals had to choose the brighter branch at each bifurcation. The stimulus therefore represented an explicit decision tree where the second decision became relevant only after a choice for the first one. We recorded multi-unit activity in primary visual cortex (V1) and area V4, from neurons with receptive fields at different levels of the decision tree.
We used reversed correlation to measure the influence of the luminance fluctuations on the animals’ choices and found that decisions at both levels formed in parallel. The activity of neurons in V1 and V4 reflected the accumulation of evidence: it was enhanced when their receptive fields fell on the relevant branch with a delay >100ms after the presentation of the stimulus. Remarkably, the time course of this decision making signal was similar for the decisions at both levels and evidence even accumulated for the irrelevant lower level decision.
The performance of the monkeys benefitted from their parallel decision making strategy. If the first decision is difficult, it is advantageous to choose the branch with the easier second level decision. Our results, taken together, reveal how parallel evidence accumulation enables animals to optimize complex decisions under uncertainty.
Meeting abstract presented at VSS 2013