Probabilistic retro-cues do not determine state in visual working memory

The effective use of our capacity-limited visual working memory (VWM) requires mechanisms that govern how it represents information. Validly cueing an item in VWM after encoding, for instance, enhances memory performance for that item and biases its state in VWM, bringing its representation to an active state such that attentional selection is biased towards perceptually similar inputs. Critically, when the retro-cue is less than 100% valid (i.e., probabilistic rather than deterministic), the effect of the cue on memory performance varies. Here we investigated whether deterministic and probabilistic retro-cues also differ in their influence over item state in VWM. Participants encoded two colored squares, and a retro-cue indicated which item was most likely to be probed in a subsequent memory test. Across blocks, we manipulated cue validity to be deterministic (100% valid) or probabilistic (70% valid). On a subset of trials, no memory probe was presented and the trial ended with a visual search task in which a colored distractor –matching the cued memory item, the non-cued item, or neither – was presented. Predictably, in the deterministic condition, the presence of a singleton distractor matching the cued item reliably slowed reaction times during visual search. In the probabilistic condition, however, there were no differences in reaction times when the singleton matched the cued item or the non-cued item, despite a reliable benefit to memory performance on valid memory trials. We suggest that, while probabilistic retro-cues improve memory of the cued item, they are not sufficient to bias its state in VWM.