Hana Yabuki¹ (), Caitlin Tozios¹, Susanne Ferber¹, Keisuke Fukuda^1,2; ¹University of Toronto, ²University of Toronto Mississauga

Not only is visual working memory (VWM) limited in capacity, some VWM representations may be maintained inaccurately even though we feel confident about them. Given that such confident errors can cause severe costs (e.g., traffic accidents), we tested a new approach to improve observers’ insights into the accuracy of their representations. Previously, we successfully reduced confident errors through a 1.5-hour-long VWM training during which participants received performance feedback based on the accuracy of metacognitive assessment of their VWM representations. Participants remembered a briefly presented array of six colored squares, then reported each item with their confidence in the accuracy of their report. Critically, they received 10, 5, or 0 points for an accurate VWM report coupled with high, low, or no confidence, respectively. Conversely, they lost 10, 5, or 0 points for inaccurate responses coupled with high, low, or no confidence. This training reduced the occurrence of confident errors (i.e., errors coupled with high confidence). Interestingly, training benefits emerged in the first several minutes of training. Thus, in the current study, we tested whether shorter training (10 mins) was sufficient to produce a training benefit that also generalizes across different stimulus types. Specifically, participants (n = 78) performed two VWM tasks (10 mins each) where they remembered an array of six colored squares or oriented bars and reported each item with their confidence (high, low, or no confidence). After measuring baseline performance, participants repeated the color or orientation VWM task with feedback. Participants then repeated the two VWM tasks without feedback to assess the training benefit and its stimulus generalizability. Here, despite a shorter training session, participants reduced confident errors during and after training relative to baseline. This benefit also generalized to the untrained stimulus. Taken together, our results demonstrate rapid and stimulus-general improvement of metacognitive assessment of VWM representations.