The Causal and the Epistemic Conditions for Moral Responsibility

Moral responsibility for an outcome of an agent who performs some action is commonly taken to involve both a causal condition and an epistemic condition: the action should cause the outcome, and the agent should have been aware – in some form or other – of the possible moral consequences of their action. In this paper I offer a formalization of both conditions within the framework of causal models. Since causal models are widely used within Artificial Intelligence, the proposed definition is also suitable for evaluating the responsibility of AI systems. I defend my approach by contrasting it to two existing approaches: that of Braham and van Hees (BvH), and that of Halpern and Kleiman-Weiner (HK). First, I argue in favor of formulating the causal condition using the recently proposed Counterfactual NESS definition of causation, as opposed to the direct NESS definition (BvH) or the HP definition (HK). Second, I argue in favor of formulating the epistemic condition by combining the epistemic conditions of the two other approaches: that the action minimizes the probability of the outcome occurring (HK), and that the action minimizes the probability of causing the outcome (BvH). My condition gives priority to the former, but also incorporates the latter. Third, I sketch how my approach opens the way to a compromise regarding the much debated control condition on responsibility. Fourth, I use the epistemic condition to define the degree of blameworthiness and suggest that it helps explain recent empirical results on responsibility judgments