Working memory as persistent neural activity

Working memory (WM) is an online memory system that allows us to hold information “in mind” in service of ongoing cognitive processing. Here, we emphasize that short-term retention of information typically involves an interplay between WM and long-term memory (LTM), especially when task demands or interruptions divert our focus from remembered items. We suggest that active neural representation may distinguish between “online” representations in WM and “offline” representations in LTM. This perspective is at odds with “activity-silent” models of WM, which hold that WM representations can be sustained without persistent neural activity. We suggest that activity-silent representations might be more productively conceptualized as offline representations in LTM because accessing these representations shows multiple signatures of retrieval from LTM. Moreover, active neural traces track WM load, predict individual differences in performance, and respect sharp item limits in WM storage. Thus, we argue that using neural activity as an operational definition of WM may provide strong traction for studying the dynamic collaboration between WM and LTM that is critical for intelligent behavior.