Talk at Metacognitive science meeting 2026

Abstract

Humans have limited cognitive resources, but collaboration offers a way to surmount these limitations by distributing information across many minds. To do so effectively, people must form conventions about who keeps track of what. Prior work has shown that people can flexibly form such conventions to divide labor in collaborative search and memory tasks, often without explicit communication. However, much of this work has focused on how conventions emerge under external task constraints, leaving open the question of whether people are also sensitive to internal cognitive constraints. We addressed this question in a collaborative visual working memory task. Participants (N = 356) memorized grids of 4, 16, or 36 images, both alone and with a partner. We adapted a visual working memory model to predict how much each dyad would benefit from dividing the grid rather than each encoding it independently. The model predicts the largest gains at intermediate grid sizes (16), where redundant encoding wastes capacity but the task remains tractable. Behavior only partially tracked this prediction. Dyads tacitly adopted spatial conventions to divide medium and large grids — reducing overlap in the tiles they studied — and outperformed solo participants in those conditions, but not for small grids where solo performance was already at ceiling. Additionally, whether participants completed the solo task before or after the dyadic task affected their tendency to collaborate – those who first encountered the task alone were more likely to divide the grid with a partner afterward. This suggests that direct experience with one’s own memory limits shapes subsequent collaborative choices, pointing to a possible role of metacognition in convention formation. Future work needs to investigate how the the adaptivity of the formation of collaborative conventions may depend on the metacognition of one’s own cognitive capacities.