Model-free and model-based learning in human fear conditioning

Learning to predict threat based on environmental cues, and to adapt behaviour accordingly, is critical to human survival. This type of learning is governed by distinct systems in the brain. Model-free learning is reflexive and computationally efficient, allowing for rapid responses based on past experience. In contrast, model-based learning is more flexible and relies on an internal model of the environment, which can be updated without direct experience. To enhance the translational value of human fear-conditioning research and facilitate the interpretation of fear-reduction interventions, it is essential to determine what type of learning underlies commonly used measures of conditioned responding. To address this question, we evaluated whether fear potentiated startle (FPS) and skin conductance responses (SCRs) reflected model-free and/or model-based learning in three fear-conditioning experiments (total n = 132). All experiments included a phase in which the value of the US was decreased through instructions, or through removal of the US electrode (US devaluation), enabling us to distinguish between model-based and model-free learning processes. In the first two experiments, FPS responses remained elevated after US devaluation, reflecting model-free learning, while SCRs updated instantly, indicating model-based control. In the third experiment, both measures followed a model-based pattern. We speculate that these differences between experiments are due to decreased uncertainty and overall less differential conditioning (FPS) in the third experiment. In sum, while SCRs seem best explained by a model-based account of learning, FPS responses appear able to index both model-free and model-based learning, with the balance between these two systems potentially depending on the level of uncertainty in the environment.