Memory Constraints in Uncertainty Misestimation: A Computational Model of Working Memory and Environmental Change Detection

Reinforcement learning models often rely on uncertainty estimation to guide decision-making in dynamic environments. However, the role of memory limitations in representing statistical regularities in the environment is less understood. This study investigated how limited memory capacity influence uncertainty estimation, potentially leading to misestimations of outcomes and environmental statistics. We developed a computational model incorporating active working memory processes and lateral inhibition to demonstrate how relevant information is selected, stored, and used to estimate uncertainty. The model allows for the detection of contextual changes by estimating expected uncertainty and perceived volatility. Two experiments were conducted to investigate limitations in information availability and uncertainty estimation. The first experiment explored the effect of cognitive load on memory reliance for uncertainty estimation. The results show that cognitive load diminished reliance on memory, lowered expected uncertainty, and increased perceptions of environmental volatility. The second experiment assessed how outcome exposure conditions affect the ability to detect environmental changes, revealing differences in the mechanisms used for environmental change detection. The findings emphasize the importance of memory constraints in uncertainty estimation, highlighting how misestimation of uncertainties is influenced by individual experiences and the capacity of working memory (WM) to store relevant information. These insights contribute to understanding the role of WM in decision-making under uncertainty and provide a framework for exploring the dynamics of reinforcement learning in memory-limited systems.