Three modes of AI-human oversight defined: in, on, and out-of-the-loop

The article describes three distinct models for human involvement in automated systems. Human-in-the-loop requires a person to approve each action before it proceeds. Human-on-the-loop involves a person monitoring an autonomous system, with the ability to intervene if needed. Out-of-the-loop describes a scenario where automation runs independently until a failure forces a sudden, unprepared handoff to a human. The author argues that choosing the wrong oversight mode is a common point of failure in AI systems and advises that human-on-the-loop is the most suitable for general use.