Theories of memory consolidation suggest that initially rich, vivid memories become more gist-like over time. However, it is unclear whether gist-like representations reflect a loss of detail through degradation or the blending of experiences into statistical averages, and whether the strength of these representations increases, decreases, or remains stable over time. We report three behavioral experiments that address these questions by examining distributional learning during spatial navigation. In Experiment 1, human participants navigated a virtual maze to find hidden objects with locations varying according to spatial distributions. After 15 minutes, 1 day, 7 days, or 28 days, we tested their navigation performance and explicit memory. In Experiment 2, we created spatial distributions with no object at their mean locations, thereby disentangling learned object exemplars from statistical averages. In Experiment 3, we created only a single, bimodal distribution to avoid possible confusion between distributions and administered tests after 15 minutes or 28 days. Across all experiments, and for both navigation and explicit tests, representations of the spatial distributions were present soon after exposure, but then receded over time. These findings help clarify the temporal dynamics of consolidation in human learning and memory.
A computational model of systems memory consolidation and reconsolidation
