Human minds have two seemingly different memory systems at their disposal: Working Memory (WM) holds currently relevant information, and Long-Term Memory (LTM) stores knowledge about the world at much longer time scales. However, the boundaries of these memory systems and their function are the subject of ongoing debate. In particular, much is unclear about how and when information in LTM can support maintenance in WM. In the context of this debate, recent work has shown that working memory precision of an individual feature (e.g., color) improves when presented as part of a real-world object stimulus, compared to 'blob' -stimuli that hold no other semantic information. It has been proposed that this performance benefit is due either to a 'scaffold' for WM maintenance provided by LTM knowledge, or to deeper processing of meaningful stimuli.
Here, we investigate another explanation: Using meaningless or repeated simple stimuli fosters proactive interference across trials. Indeed, we show that WM benefits for real-world objects are attenuated if the objects are often repeated throughout a block, rather than them being unique on each trial. In a series of experiments, we investigate the nature of such proactive interference, and demonstrate how it affects WM fidelity. Based on our findings, we evaluate the role of knowledge in LTM as a way to assist WM, and discuss the complementary roles of these memory subsystems.