Conference on Cognitive Computational Neuroscience

When we make plans for the future, we often do so in the face of uncertainty. How does the level of uncertainty impact the effort we put into planning? Recent work suggests that people plan in a resource-rational way, considering how deeply to plan based on a cost-benefit analysis. This framework would predict that people plan less deeply in more stochastic environments because the expected marginal benefit of planning decreases with stochasticity while the marginal cost remains the same. Here we test this prediction using a game where participants plan ahead on to accumulate rewards, where the game environment is subject to one of four levels of stochasticity. We fit a model in which agents plan optimally up to a limited depth and are subject to decision noise. This model accounts well for the cumulative reward trajectories across stochasticity levels. Consistent with resource rationality, we find that planning depth decreases monotonically with the level of stochasticity in the environment. Accordingly, first-move response time decreases with stochasticity. Our findings provide new evidence for resource rationality in planning.