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