Exploring Moral Exercises for Human Oversight of AI systems: Insights from Three Pilot Studies

Moral exercises can cultivate virtues needed for AI oversight: The paper proposes that structured moral exercises—drawing from philosophical and ancient practices—help AI professionals develop the personal qualities and wisdom necessary to effectively oversee AI systems, moving beyond purely technical approaches to governance.

Three core pillars support responsible AI culture: The methodology centers on (1) fostering engaged personal commitment to responsibility, (2) building relational understanding among team members, and (3) developing "technomoral wisdom" that integrates ethical reasoning with technical knowledge—findings supported across pilot studies with companies, researchers, and students.

Organizational culture matters as much as policies: The research demonstrates that moral exercises can help build a responsible AI culture within organizations, suggesting that fostering the right values and dispositions among AI actors may be as important as implementing formal oversight rules and procedures.

Practical application shows both promise and limitations: While pilot studies across three different contexts (industry, academia, education) showed potential for moral exercises to improve human oversight capabilities, the paper acknowledges limitations, indicating this approach works best as part of a broader strategy rather than as a standalone solution.

Human judgment requires intentional development: The findings suggest that effective human oversight of AI systems isn't automatic—it requires deliberate cultivation of ethical reasoning and virtues through structured practices, not just hiring knowledgeable people.