Conference on Cognitive Computational Neuroscience

Flexible arbitration between instrumental and Pavlovian learning allows for adaptive behavioral control. In controllable environments, instrumental learning can support the selection of actions that bring about rewards and avoid punishments. In uncontrollable environments, where actions have no causal influence on reward outcomes, individuals may rely on less computationally costly “default” Pavlovian reactions. Prior studies suggest that adolescents may demonstrate reduced Pavlovian biases on instrumental learning in controllable environments. Here, we asked whether greater adaptability of action selection to environmental controllability could explain these adolescent-specific effects. Ninety participants, aged 8-27 performed a Go/No-Go task in both controllable and uncontrollable environments. We implemented a Bayesian reinforcement learning model to evaluate controllability inferences and corresponding changes in the engagement of instrumental and Pavlovian learning systems across age. Relative to children and adults, adolescents exhibited greater flexibility in calibrating expression of Pavlovian bias to the controllability of the environment. Such flexibility may be beneficial for discovering rewards in the new environments adolescents encounter during their transition into more independent adults.