Schedule of Events | Symposia

Working Memory Demands Influence the Balance between Detailed and Gist-Level Representations during Planning

Poster Session B - Sunday, March 30, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Zhuojun Ying¹ (z5ying@ucsd.edu), Anastasia Kiyonaga¹; ¹University of California, San Diego

Working memory (WM) is thought to support planning by maintaining representations of future actions and rewards. Our recent work examined WM for reward information during planning using an information-theoretic model approximated by a variational recurrent neural network. Participants chose the best path in a decision tree with nodes revealed sequentially and varying in value. Both model and behavior showed better memory for nodes most informative for path selection, suggesting dynamic WM allocation based on choice relevance. According to the model, this dynamic allocation arises from limited WM resources, predicting that higher WM demands would shift priority toward the overall path quality over detailed node values. Here we manipulated WM demands to examine their impact on path selection and memory. Using a decision task with node rewards indicated by color, we varied the tree size and the color space resolution. In the larger tree condition, participants recalled node values less accurately and selected the best path less frequently. In the higher-resolution condition, participants recalled node values faster and consistently selected high-value alternative paths, even when the best path was missed, suggesting a satisficing strategy. WM demands also interacted with reward magnitude and path quality: individual node values influenced recall most in smaller trees, while overall path value influenced recall most in larger, higher-resolution trees. This suggests that cognitive systems dynamically shift from encoding individual rewards to overall path qualities under increased WM demands, illuminating how neural circuits might adaptively allocate WM resources to balance action- and plan-level representations during planning.

Topic Area: EXECUTIVE PROCESSES: Working memory