NEURAL BASIS OF CAUSAL LEARNING IN REWARD-GUIDED DECISION MAKING

Animals constantly face the challenge of linking their actions to outcomes that may occur especially after a temporal or spatial delay. In a complex environment, how the individual establishes the causal structure of the environment is still poorly understood. Distal credit assignment should be distinguished from causal inference. An animal may learn that a seemingly neutral action increases the probability of a later reward, without necessarily representing the causal mechanism underlying this relation. In other words, distal credit can emerge from value propagation through reinforcement signals, whereas causal inference implies an explicit model that can be flexibly generalized, manipulated, and applied to counterfactual reasoning. In this work, we try to decipher causal inference and state representation by using a “causal behavioral task”. By combining behavioral analyses, dopaminergic and prefrontal recordings, and computational modeling, we ask whether mice solve this task through distal credit assignment or if their behavior reveals causal inference of hidden states, and how neural circuits encode these contingencies.