Abstract
In many perceptual tasks, humans near-optimally use sensory uncertainty information in their decisions. It is unknown whether they do so in decisions based on visual working memory (VWM). Some circumstantial evidence is available: humans' confidence reports are positively correlated with their errors in a delayed-estimation task (Rademaker et al., 2012), and humans near-optimally integrate current knowledge of uncertainty with working memories (Keshvari et al., 2012). However, it is unclear whether people accurately store uncertainty information in VWM and use it in a subsequent decision. To investigate this, we collected data in two change localization tasks with variable stimulus reliability. Each trial consisted of a sample array of four Gabors, a delay, and a test array of four Gabors. Participants reported which of the four Gabors changed in orientation. In Task 1, we used two levels of contrast to manipulate memory uncertainty. In the sample array, the stimuli could be all high contrast, all low contrast, or two of each. In the test array, stimuli were either all high or all low contrast. In Task 2, we replicated this result with variable delay times (1 or 3 seconds) instead of variable contrast. We evaluated two models. The Optimal model assumes that observers know their memory uncertainty on a trial-to-trial and item-to-item basis and use this information to maximize performance. The Fixed model assumes observers do not use knowledge of their uncertainty, but assume that stimuli are equally uncertain. In both tasks, the Optimal model outperformed the Fixed model for three of four participants (: Task 1: M = 10.1, SEM = 6.9; Task 2: M = 9.3, SEM = 15.3). Moreover, the Optimal model provides good fits to the psychometric curves. These results provide preliminary evidence that humans maintain uncertainty information in VWM and use it in a subsequent decision.
Meeting abstract presented at VSS 2017