Developmental differences in exploration reveal underlying differences in structure inference

Across development, we balance exploring to refine our causal models of the world with exploiting what we already know. Children and adolescents often explore more than adults, a tendency commonly attributed to greater decision-making noise or stronger motivation to learn. Here, we propose developmental changes in structure learning as an alternative driver of exploration. Through exploration, we uncover relevant statistical relationships in our environment that can be leveraged to obtain rewards. To test this proposal, 252 8-to-25-year-old participants completed a patch-foraging task indexing individual differences in structure learning. Younger participants explored more, leaving patches sooner than adults. Computational modeling revealed that their early departures stemmed from their use of simpler and easier-to-plan-over structure representations to guide their decisions. Our findings go beyond previous algorithmic accounts of developmental change in exploration, suggesting that heightened exploration can also arise from differences in how children learn the structure of their environments.