Social dysfunction is a prominent and disabling aspect of borderline personality disorder. We reconsider traditional explanations for this problem, especially early disruption in the way an infant feels physical care from its mother, in terms of recent developments in computational psychiatry. In particular, social learning may depend on reinforcement learning though embodied simulations. Such modeling involves calculations based on structures outside the brain such as face and hands, calculations on one's own body that are used to make inferences about others. We discuss ways to test the role of embodied simulation in BPD and potential implications for treatment.