A hierarchy of functional states in working memory

Abstract

Extensive research has examined how information is maintained in working memory (WM), but it remains unknown how WM is used to guide behaviour. We addressed this question by combining human electrophysiology with pattern analyses, cognitive modelling, and a task requiring maintenance of two WM items and priority shifts between them. This enabled us to discern neural states coding for immediately and prospectively task-relevant items, and to examine their contribution to WM-based decisions. We identified two qualitatively different states: a functionally active state encoded only immediately task-relevant items and closely tracked the quality of evidence integration on the current trial. In contrast, prospectively relevant items were encoded in a functionally latent state that did not engage with ongoing processing but tracked memory precision at longer time scales. These results delineate a hierarchy of functional states, whereby latent memories supporting general maintenance are transformed into active decision-circuits to guide flexible behaviour.