Governance

modes of human involvement

Human-in/on/over-the-loop models varying by degree of intervention.

Definition

A taxonomy of human oversight models defining varying degrees of human involvement in AI decision-making: human-in-the-loop (humans directly make or approve decisions), human-on-the-loop (humans supervise automated decisions and can intervene), and human-over-the-loop (humans set policy, audit and adjust automated systems); selection is risk-informed based on harm severity and reversibility.