Model-based Analysis of the Tower of London Task

The planning process is central to goal-directed behaviour in any task that requires the organization of a series of actions aimed at achieving a goal. Although the planning process has been investigated thoroughly, relatively little is known about how this process emerges and evolves during childhood. In this paper we describe three reinforcement learning models of planning, in the Tower of London (ToL) task, and use Bayesian analysis to fit each model to pre-existing data from 3-4 year-old and 5-6 year-old children performing the task. The models all capture the increased organisation seen in the older children’s performance. It is also shown that, at least for this dataset, the most complex model – that with discounting of future rewards and pruning of highly aversive states – provides no additional explanatory power beyond a simpler discounting-only model. Insights into developmental aspects of the planning process are discussed.