To integrate or not to integrate: Testing degenerate strategies for solving an accumulation of perceptual evidence decision-making task

Abstract

A common approach in the study of cognition is to train subjects to perform a task that requires a particular cognitive process to solve. Analysis of the subjects response behavior while they perform these tasks can offer valuable insight into the underlying mechanisms that give rise to cognition. However, if subjects are able to accurately perform such a task by using a strategy that doesnt involve the targeted cognitive process, data from those experiments becomes more difficult to interpret. A number of perceptual decision-making tasks have been designed to study the accumulation of evidence, i.e. how noisy information presented over time is used to form a decision. Recent work, however, has highlighted how a variety of non-integration strategies can by some measures yield strikingly near-optimal performance on such tasks, raising the possibility that past conclusions from these experiments may be incorrect. Here we assemble the largest data set of animals performing one such task - the "Poisson Clicks" task - which is optimally solved by the gradual integration of pulsatile auditory noise. To investigate whether rats are in fact using this strategy, we compiled data from 515 rats performing over 35 million trials. We compare performance of 3 degenerate strategies (that circumvent the need to integrate evidence) to the optimal (integration) strategy. We demonstrate that the pulsatile nature of the stimuli used in the Poisson Clicks Task makes it possible to distinguish which strategy subjects use. Overwhelmingly, we find the rats are using an integration strategy when performing the Poisson Clicks Task.