Abstract
There is a debate surrounding whether visual working memory is continuous (Ma, Husain & Bays, 2014) or discrete (Zhang & Luck, 2008). The former theory assumes that people maintain some information about all items in a display, whereas the latter proposes that discrete loss of item information can occur. The goal of our study was to determine whether people actually guess in working memory tasks. To address this question, we leveraged confidence ratings and presented 960 intermixed trials of a continuous report task and a same-different discrimination task (0˚ or 90˚ from sample orientation). On each trial, participants were briefly presented with 1, 3, or 6 randomly oriented Gabor patches. Next, they were cued to rate their memory quality for a specific item (scale 0-3), after which they had to complete one of the two orientation tasks. Given that people have quite accurate metacognitive awareness (Rademaker et al., 2012), we anticipated that the zero-confidence condition would provide greater sensitivity to test whether any quantifiable amount of information remained present in working memory. Analyses focused on set size 6, as reports of zero confidence rarely occurred at smaller set sizes. As predicted, discrimination accuracy and memory precision increased as a function of reported confidence. In the same-different task, mean accuracy was 51%, 55%, 67%, and 86% for confidence levels 0 through 3, respectively, and 9/10 participants performed no greater than chance level at confidence level 0 (p > 0.1 uncorrected). In the continuous report task, the mean absolute error at confidence level 0 was 44.67˚, which did not reliably differ from an expected value of 45˚ for randomly distributed responses (p=0.76). Our findings provide compelling evidence that discrete loss of information from visual working memory can indeed occur, and that reports of zero confidence are commonly accompanied by random guesses.