Submission 391
Competition During Working Memory Enhances Long-Term Memory
SymposiumTalk-04
Presented by: Anastasia Kiyonaga
The thoughts and goals we hold in working memory (WM) interact continuously with perception and long-term knowledge. Competition between WM and perception has been a central testbed for neural theories of WM storage, yet the principles that govern this competition remain unresolved. Here, we test how the visual feature overlap between sensory input and WM content influences memory retention. We used artificial neural networks to generate and validate a stimulus set of naturalistic images that varied systematically in high- and low-level visual correlations. These images were used as both WM samples and task-irrelevant sensory inputs in a WM task. Following the WM task, participants completed a surprise LTM test for a subset of the images. After piloting the design, we pre-registered the study and collected two independent cohorts, replicating results with separate stimulus sets (n= 134; n=136). We found that sensory input had a non-monotonic effect on WM performance, whereby strongly corelated input progressively reinforced WM, until it crossed a threshold and became interfering. The behavioral impact of sensory input also depended on whether it overlapped with prioritized or deprioritized WM content, as well as the feature distinctiveness of the WM content. Intriguingly, we found that stronger interference during WM predicted better long-term memory. Competition between internal and external representations may therefore promote differentiation processes that ultimately promote durable encoding. Together, these findings help reconcile perplexing discrepancies in the effects of distraction on WM, revealing how the same input can produce distinct behavioral outcomes depending on task context.