Abstract
Recent evidence suggests that working memories (WM) are encoded in qualitatively different states depending on their momentary task-relevance. Items that are used for current behaviour are thought to in an active state that is encoded in content-specific spiking, which is primed to drive behaviour. By contrast, currently irrelevant items can be held in a latent state that is not reflected in activity, but remains accessible for later use. It remains unknown, however, how latent working memories are transformed into active decision circuits when behavioural priorities change. We used time-resolved decoding of WM items using electroencephalography (EEG) in a task that required cued priority switches between decision boundaries, permitting independent decoding of active and latent boundaries on trials when their priority status switched (requiring transfer from a latent to an active state) vs. when the priority status was repeated. WM switches created transient performance costs that recovered after a single trial. EEG revealed that this behavioural cost is driven by lingering over-representation of the latent item. On priority switch trials, both the newly active and the previously active item could be decoded. Intriguingly, active and latent items were represented in distinct neural patterns: training a decoder on the active item did not permit decoding of the latent one, and vice versa. Importantly, the magnitude of latent item decoding tracked participants’ performance cost after switches. On priority repeat trials, only the active item could be recovered from EEG activity, with decoding of the latent item returning to chance. These findings suggest that priority shifts incur transient competition between items for active representation.
Acknowledgement: Research Foundation Flanders, Wellcome Trust, McDonnell FoundationÂ