Abstract
To make adaptive decisions in natural environments, we must not only commit to an action but, more primarily, determine which information to gather to best guide that action. Traditional studies of selective attention explicitly instruct participants about where or to what to attend but do not examine how people endogenously select decision-relevant information. We developed a task that operationalizes this process as hierarchical Bayesian reasoning. According to Bayesian logic, the brain would ideally prioritize information dimensions (e.g., locations in the visual field) based on the predictive validity (PV) of the stimuli expected to appear within the dimension, and the decision maker’s prior uncertainty. To determine if people use this attentional strategy, we used a 2-AFC task in which participants guessed the identity of a latent state based on combinations of high-PV and low-PV information regarding the state. During a critical anticipation epoch, people were free to allocate covert attention to two competing locations expected to produce high or low-PV stimuli, in conditions of low or high prior uncertainty. Preliminary behavioral measures show that participants had a higher d’ for discriminating the stimuli at high-PV versus low-PV locations. This difference was generally enhanced by prior uncertainty, although the strength of the uncertainty effect showed individual variation. Moreover, an ROIs within the intraparietal sulcus (IPS) responded more strongly when high-PV relative to low-PV stimuli were expected in the contralateral visual field, and this bias increased under uncertainty. Additional ROIs within the frontal eye field (FEF) and locus ceruleus (LC) were globally enhanced by uncertainty. The results show that people allocate covert spatial attention approximately consistent with Bayesian expected information gains, and begin to outline the system-wide mechanisms that control attention to obtain decision-relevant information.