Human oversight in AI refers to the practice of keeping humans informed, involved, and in control of AI system decisions and actions, particularly in consequential contexts.
AI systems can produce errors, reflect biases, and behave unexpectedly. Human oversight means that people — not just automated systems — review, validate, and take responsibility for consequential decisions that AI systems inform or make.
The appropriate level of oversight depends on the risk level of the application. Routine low-stakes tasks may require less oversight than decisions affecting people’s rights, safety, or livelihoods.