Bayesian Criterion Update with Working Memory Explains the Trial-to-trial Variability of Perceptual Decision and Cortical Representation of Decision Uncertainty

Abstract

We change our perspective to the world depending on moment by moment influx of the new information. Because our references of the world are updated every moment, we used to decide on something as a different one under the different context, although the object is same. The aim of this study is to investigate this variable characteristic of human mind. Our question is that why and how reference of the world is restlessly adjusted? By implementing ring-size discrimination task, we reproduced the phenomenon, in which the decision reference of subject was not static but modulated by previous stimulus, in controlled experimental condition. Also, to investigate the phenomenon in model-based approach, we developed the model of trial-to-trial update of decision criterion by using Bayesian inference and generated 3 simpler versions of the model. By applying all the 4 models to behavior and fMRI data and comparing their predictions, we found that decision criterion is not only modulated by sequence of previous stimuli but also anchored by long-term criterion in trial-to-trial manner and also that the modulation was well captured in Bayesian framework. In this regards, we suggest that dynamic nature of world reference is the consequence of human nature in searching for most probable understanding of the reference combining long-term and recent information.