Neural signatures of model-free learning when avoiding harm to self and other

Moral behaviour requires learning how our actions help or harm others. Theoretical accounts of learning propose a key division between ‘model-free’ algorithms that efficiently cache outcome values in actions and ‘model-based’ algorithms that prospectively map actions to outcomes, a distinction that may be critical for moral learning. Here, we tested the engagement of these learning mechanisms and their neural basis as participants learned to avoid painful electric shocks for themselves and a stranger. We found that model-free learning was prioritized when avoiding harm to others compared to oneself. Model-free prediction errors for others relative to self were tracked in the thalamus/caudate at the time of the outcome. At the time of choice, a signature of model-free moral learning was associated with responses in subgenual anterior cingulate cortex (sgACC), and resisting this model-free influence was predicted by stronger connectivity between sgACC and dorsolateral prefrontal cortex. Finally, multiple behavioural and neural correlates of model-free moral learning varied with individual differences in moral judgment. Our findings suggest moral learning favours efficiency over flexibility and is underpinned by specific neural mechanisms.