Humans have the extraordinary ability to flexibly learn, and switch between, different tasks, based on prior experience or simple instructions. However, in doing so, they need to make overarching decisions about whether they want to rely on shared representations (at the risk of more task set interference) or separate, parallel representations (at the cost of more intensive learning). This project will explore how people can efficiently use contextual features in their environment, or their reinforcement learning history, to decide what is their best strategy when learning new tasks. Specifically, we will investigate this using contemporary measures of cognitive flexibility such as voluntary task switch rates, and study the effects of different training schemes thereon. Second, by using both behavioral and functional magnetic resonance imaging studies, we will study not only the performance cost of task interference, but also the putative underlying shared versus parallel representations in the brain, as assessed by voxel pattern similarity analyses. In doing so, this project aims to address both longstanding questions in how people decide to engage in flexible versus stable behavior, as well as recent empirical contradictions on the role of reward in task representations in the brain.