Moral Exercises for Human Oversight of Algorithmic Decision-Making

In this paper we present an ethical framework aimed at supporting human agency and oversight throughout the life-cycle of algorithmic decision-making systems. Drawing upon classical philosophical traditions, we shift the focus from ethical solutions to the role of Artificial Intelligence (AI) actors in a quality decision process. To this aim the framework highlights the dynamic nature of people’s moral decisions, and the “ethical tools” that are inherent within the human being. The primary objective is not to enforce morality within the machine itself but to cultivate moral agency in the human. This offers the conceptual coordinates to put forward a set of “moral exercises”, practical activities that can be used for the moral training of human actors involved in the life process of AI-based decision systems. Rather than being algorithmic procedures or workflows for ensuring “moral outcomes”, these exercises are flexible instruments to shape the human processes underlying the oversight of AI systems. We illustrate the practical implications of our framework by showing potential cases of application of the exercises, and by creating connections with existing AI ethics methodologies.