Model-based foraging using latent-cause inference

Foraging has been suggested to provide a more ecologicallyvalidcontext for studying decision-making. However, the environmentsused in human foraging tasks fail to capture thestructure of real world environments which contain multiplelevels of spatio-temporal regularities. We ask if foragers detectthese statistical regularities and use them to construct amodel of the environment that guides their patch-leaving decisions.We propose a model of how foragers might accomplishthis, and test its predictions in a foraging task with a structuredenvironment that includes patches of varying quality andpredictable transitions. Here, we show that human foragingdecisions reflect ongoing, statistically-optimal structure learning.Participants modulated decisions based on the current andpotential future context. From model fits to behavior, we canidentify an individual’s structure learning ability and relate itto decision strategy. These findings demonstrate the utility ofleveraging model-based reinforcement learning to understandforaging behavior.