How does working memory stabilize perceptual biases?
Poster Presentation: Sunday, May 18, 2025, 2:45 – 6:45 pm, Pavilion
Session: Visual Memory: Capacity and encoding of working memory
Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions
Kuo-Wei Chen1 (), Scott Janetsky1, Mckay Turner-Colling1, Keimya Pennington1, Gi-Yeul Bae1; 1Arizona State University
While perception and working memory (WM) are fundamental functions for visually guided behaviors, perception often exhibits systematic biases, and these biases are exaggerated in WM. For example, the cardinal bias in orientation perception increases when the report is made after a delay. This increase in bias in WM suggests that the visual system recruits correction mechanisms to mitigate neural noise accumulated over time by drifting the representation toward more stable states (i.e., attractors). However, WM involves both encoding and maintenance processes, and how these two processes coordinate to mitigate neural noise is not well understood. To address this question, we investigated how cardinal bias varies over time by measuring the bias during perception (i.e., when the stimulus is in view), WM encoding (i.e., a 0-second delay between the stimulus and response), and variable WM delays (2, 4, and 6 seconds) in orientation estimation tasks. We found that the magnitude of cardinal bias was significantly greater in the longest delay condition (6 seconds) compared to the perception condition, consistent with findings from previous studies. Critically, however, the majority of the increase in bias magnitude (69% of the total increase) occurred between the perception and WM encoding conditions, rather than gradually increasing over time. In follow-up experiments, we replicated this effect with another perceptual bias (i.e., the visual tilt-illusion), demonstrating the generalizability of this effect. We confirmed that this effect persisted even when different delay conditions were randomly mixed within a block of trials, ruling out the possibility that it was driven by task-dependent strategies. Together, our results provide converging evidence that perceptual biases are stabilized in WM in two discrete steps: the WM encoding process rapidly pushes the representation toward an attractor with greater magnitude, followed by a slower, smaller drift toward the attractor during WM maintenance.