Working memory capacity is linked to directed exploration: behavioral, computational, and electrophysiological evidence

Abstract

Balancing the exploitation of known rewards against the exploration of uncertain options is central to adaptive decision making, yet the cognitive limits on strategic exploration are poorly understood. Directed exploration, the deliberate sampling of uncertain options for information, may depend on working memory (WM), but whether WM capacity (WMC) is selectively related to it, and through which processes, is unknown. 52 participants from the upper and lower tails of an operation-span distribution (26 high-WMC, 26 low-WMC) performed an adapted Horizon Task during EEG recording. We combined behavioral strategy classification, hierarchical Bayesian reinforcement learning, stimulus-locked event-related potentials (ERPs), and an exploratory Bayesian mediation analysis. High-WMC participants showed more directed exploration and less exploitation, with no group difference in random exploration. A model in which accumulated sampling history discounts option value best explained choices in both groups, and the discounting parameter {omega} was larger in the high-WMC group. Statistically, {omega} accounted for the association between WMC and directed exploration (indirect effect ab = 0.64, 95% CrI [0.28, 1.05]). Frontocentral P2 amplitude (180 to 280 ms) during directed exploration was larger in the high-WMC group, and the P2 difference between directed exploration and exploitation correlated with behavioral preference for directed exploration (r = .29, p = .040). These findings suggest that individual differences in exploration may reflect not only the willingness to seek information but also the capacity to use what has been sampled. The WMC-related difference lay in information use rather than choice noise and was already visible in early option evaluation.