Conference on Cognitive Computational Neuroscience

Planning, the process of evaluating the future consequences of actions, is typically formalized as search over a decision tree. More search means higher expected rewards, but tree search is computationally expensive. One open question is how people are able to approximate the values associated with potential plans while they are planning. In this work, we propose a normative framework for meta-planning via Bayesian inference. Specifically, we model a metacognitive process where tree search is viewed as gaining noisy measurements of the value of a given action. This statistical estimate is then combined with prior experience to decide whether and in which direction to plan. Our model produces intuitive simulation results across a range of parameters and acts as a more valuable, informed method for guiding planning when compared to best-first and breadth-first search. Additionally, we show that our meta-planner has the capacity to qualitatively predict response time trends in a complex planning task.