Trade-offs between visual sampling and memory in stable and changing worlds

In natural behavior, humans typically do not fully utilize their memory capacity but often choose to sample information from the external world to alleviate the burden on memory. This trade-off between memory and sampling has been demonstrated in previous studies by a stronger reliance on memory when sampling costs increase (e.g., longer waiting time for accessing information). These studies have mainly focused on working memory, which is resource-costly, while ignoring another memory system - long-term memory (LTM) - that can store vast amounts of information relatively effortlessly. Here, we investigated whether recruiting LTM resolves the trade-off between external sampling and internal memory, by reducing the cost of memory use. We employed a copying task in which participants reproduce an example display containing (six or eight) colored polygons that were either repeated over consecutive trials or not. Experiment 1 showed that, when the same example display was repeated, the sampling frequency and sampling durations decayed exponentially, to the point that most participants eventually no longer sampled the example display. In Experiment 2, we replicated the greater reduction in sampling frequency when only half of the polygons were repeated (versus no repetition). Importantly, a surprise memory test showed that the repeated items were encoded into LTM, and recall performance in this surprise memory test predicted the reduction in sampling frequency. Taken together, our findings reveal how memory is used in stable visual environments, where long-term memory can complement resource-costly working memory use and resolve the trade-off between sampling and memory.