Bryan Daniels (Arizona State University)
Abstract. What design principles underlie efficient distributed decision making in the brain? We propose two temporal phases of collective computation: slow, distributed information gathering followed by fast, redundant information sharing. This dynamic can be seen in a visual discrimination task by analyzing recordings from 169 neural channels in the prefrontal cortex of macaque monkeys. Early in each experimental trial, uncertainty is substantially reduced only by pooling information from many cells, which gain this information slowly in time. Late in the trial, numerous single cells quickly gain all the information previously present at the population level. We demonstrate that these two distinct behaviors are parsimoniously described in terms of critical slowing down and varying distance from a symmetry breaking transition.