Abstract
Visual working memories are known to be imprecise because perceptual information is corrupted by internal noise during the memory delay. Can memory noise create strong but entirely false memories if no perceptual information is actually present? If so, how do these memories compete with true memories? In a series of two experiments (total N=90), participants were exposed to 2 or 4 colors for a single frame (~16ms), followed by a dynamic noise mask for 200ms and an 800ms delay. Crucially, in half the cases, only 1 item was presented, but the masking and test positions implied the full set of items was present. In Experiment 1, participants were probed about a specific position, which might be one where a color was never presented, and then asked to report their confidence. In Experiment 2, participants were free to choose which position they want to report on half of trials. Replicating prior work, there was a robust confidence-accuracy relationship for items actually shown. However, participants often gave high confidence responses to locations that did not have items, and regularly chose these locations when reporting on their "strongest" memory, indicating a propensity for false memories. Critically, when only a single item was presented, participants’ memory was more precise when they freely chose that item as their strongest than when they were forced to report it, even though in both cases they reported on the only actual item presented. Overall, these results imply that the phantom items are genuinely experienced. Furthermore, an actually shown item wins the "competition" against phantom items more often when this item is strongly represented. This provides evidence for the general view that participants are always reporting based on an underlying noisy sensory signal even when they are reporting an item they have no information about ('guessing').