Assigning Moral Responsibility for AI-Derived Errors in Healthcare: Shared Responsibilization Without Responsibility Gaps

As artificial intelligence becomes increasingly advanced, its usage in high-stakes environments such as healthcare is bound to become widespread. The concern thus arises that when erroneous AI outputs result in clinical mistakes that cause harm to patients, there is a need to attribute moral responsibility for the harm that has occurred. In this article, we aim to provide a commentary on a recent publication by Lang et al., in which the authors expound a framework of shared responsibilization for the commission of such clinical errors. Lang et al. argue that human stakeholders such as physicians, hospital administrators, AI programmers, and technology companies must voluntarily take on moral responsibility in an act of moral heroism to overcome the "responsibility gaps" created by AI-enabled medical errors. These gaps arise when the moral responsibility of a clinical harm must be assigned to morally inculpable parties: human stakeholders who are not to blame for the faulty output of an AI, and an AI which is not sufficiently agential to warrant moral blameworthiness. We agree with the authors in that the moral responsibility for AI-enabled medical errors is one that should be shared across relevant stakeholders. However, our commentary challenges the notion that AI-enabled medical errors exhibit responsibility gaps, and as a corollary, we find that the shared responsibilization is something that is not taken on, but rather, something that human stakeholders are unambiguously morally responsible for when it results in patient harm.