Examining how task abstractions support flexible planning

About This Grant

PROJECT SUMMARY Flexibly adapting to changing circumstances is a hallmark of human intelligence. This adaptation is often formu- lated as resulting from the use of a “model” of the environment that is used to reason about how decisions should change. However, standard approaches make simplifying assumptions about how the environment and actions are represented, leaving them unable to account for how people adaptively represent tasks. Deficits in learning task structure have been observed in psychiatric disorders and can considerably impact the difficulty of model- based reasoning. It is challenging to study this because the representations underlying model-based reasoning are believed to occur on a faster timescale than is amenable to standard neuroimaging methods. In addition, different planning strategies can often result in the same choices, making it difficult to predict planning represen- tations from behavior alone. The proposed research aims to address this gap. We develop two experiments to see how factorized task representations are used to simulate and plan. First, we examine an experiment that requires generating actions based on a well-learned sequence of stimuli and a per-trial cue for a response rule. While several task representations are possible, we hypothesize that people represent tasks in a highly factorized way, to generalize rapidly to future tasks. Second, we examine an experiment that requires planning, extending beyond simulation to generate action plans. We use eye tracking to record participant acquisition of task infor- mation, offering rich insight beyond choice data alone. Our task design also admits several task representations, making it possible to study how task representations influence the planning process, and are transformed into motor plans. We hypothesize that people plan in abstract task spaces, despite this process of transformation, because of the overall potential to simplify planning. In both aims, we will test these representational hypotheses by examining neural signals acquired with MEG, a method that is well-suited to testing the rapid sequences of neural activity associated with sequential simulation and planning. Going beyond existing methods developed to investigate sequential recall, our approach will reveal how sequential simulation interacts with selected task rep- resentations to support flexible behavior. This research will contribute novel understanding of how people select and use task representations for simulation and planning. While this work has no direct clinical aims, it will lay the groundwork for future investigations into the relationship between task abstraction learning and dysfunctional day-to-day decision making in psychiatric populations related to poor structure learning and generalization.