Working Memory Recruits Long-Term Memory When It Is Beneficial: Evidence From the Hebb Effect

When encoding task-relevant information in working memory (WM), we can use prior knowledge to facilitate task performance. For instance, when memorizing a phone number, we can benefit from recognizing some parts as known chunks (e.g., 911) and focus on memorizing the novel parts. Prior knowledge from long-term memory (LTM), however, can also proactively interfere with WM contents. Here, we show that WM selectively recruits information from LTM only when it is helpful, not when it would interfere. We used variants of the Hebb paradigm in which WM is tested through immediate serial recall of lists. Some lists were repeated frequently across trials, so they were acquired in LTM, as reflected in increasing serial-recall performance across repetitions. We compared interference conditions in which that LTM knowledge could interfere with holding another list in WM to a neutral condition in which that knowledge could be neither beneficial nor harmful. In Experiments 1-3, lists in the interference conditions shared their items with the learned lists but not their order. We observed no proactive interference. In Experiments 4 and 5, the interference lists' first three items overlapped exactly with the learned lists, and only the remaining items had a new order. This made LTM knowledge partially beneficial and partially harmful. Participants could use LTM flexibly to improve performance for the first part of the list without experiencing interference on the second half. LTM-mediated learning of the first part even boosted memory for the unknown second part. We conclude that there is a flexible gate controlling the flow of information from LTM and WM so that LTM knowledge is recruited only when helpful.