Children leverage predictive representations for flexible, value-guided choice

• DOI: 10.31234/osf.io/y3dzn_v2
• Publication Date: 2025-01-29T11:16:00Z
• Authors: Alice Zhang; Ari Kahn; Nathaniel Daw; Kate Nussenbaum; Catherine A. Hartley
• Keywords: Leverage (statistics)

By building a mental model of how the world works and using it to forecast outcomes different actions, learner can make flexible choices in changing environments. However, while children adolescents readily acquire structured knowledge about their environments, relative adults, they tend demonstrate weaker signatures leveraging this plan actions. One explanation for these developmental differences is that prospectively simulate potential computationally costly, taxing cognitive control working memory mechanisms continue develop into adulthood. Here, we ask whether might effectively leverage by relying on two alternative strategies do not require costly simulation at choice time. First, through offline replanning, models be queried before time generate possible scenarios update values Second, an abstracted predictive model, known as Successor Representation, built harnessed enable simplified computation long-run reward candidate without requiring iterative multiple steps. To assess children, adolescents, adults aged 7 - 23 years similarly harness learning strategies, ran three experiments. In Experiments 1 2, used revaluation task which manipulated opportunity replanning during rest, found flexibly updated behavior adult-like manner. Surprisingly, across age, rest did mediate raising possibility participants may have behaved adaptively harnessing representations online. Experiment 3, directly tested use representations. observed early-emerging SR, providing mechanistic account guide detailed model-based simulation.