Learning to Pursue Goals that Conflict with Prior Experience Relies on Fast Revaluation and Slow Value Updating

Humans can flexibly pursue novel goals, learning to reach them from goal-dependent teaching signals. However, the breadth of this flexibility and its underlying mechanisms remain unclear. For example, can humans flexibly learn to reach a typically avoided outcome (e.g., +0) while avoiding a typically sought one (+1)? We hypothesized that humans are sufficiently flexible to learn from counterintuitive goals, but at a cost, and this ability is supported by complementary temporal dynamics of rapid cognitive control and slower, iterative outcome value updating. In two experiments, we confirmed that participants learned when pursuing counterintuitive goals, but less effectively than standard goals. Furthermore, extended practice reduced performance gaps. To characterize the underlying mechanisms, we developed a novel computational model incorporating fast goal switching and slow outcome value updating. This model best captured group-level behavior; individuals differed in the efficiency of these processes, capturing differences in the cost initially incurred by counterintuitive goals. Crucially, a stronger initial cost did not permanently limit performance: slower value updating compensated for early inefficiencies in rapid adaptation. Furthermore, comparing early and late goal-switch learning revealed that greater goal alignment was associated with faster outcome processing, suggesting a transition from effortful control to efficient outcome evaluation. Thus, despite individual differences in initial flexibility, repeated attempts to reach counterintuitive goals decreased initial biases. More broadly, our findings show that goal-dependent outcome valuation leverages complementary fast and slow processes to enable flexible and efficient goal pursuit over time.