We assessed the effects of the same ACC manipulation used in the two-step task on a probabilistic reversal learning task. In this task both model-free and model-based RL are expected to generate qualitatively similar influence of trial events on subsequent choice, i.e., rewarded choices will be reinforced, though there may be quantitative differences if the model-based system is able to learn the block structure and infer block transitions rather than relying on TD value updates.
