What it means
Human-in-the-loop describes a system design in which a person reviews, approves, corrects or completes AI outputs at defined points before they take effect. The human is a required step in the process, not an optional observer.
Real-world example
In a benefits office, AI drafts a summary of each application and flags missing documents. A caseworker must review the summary, check the flags against the original documents and make the decision. The system cannot send any notice to the applicant without the caseworker's approval.
Why it matters for AI adoption
Human review only protects people if reviewers have the time, information, training and authority to disagree with the AI. Otherwise review can become a formality, and people may accept AI suggestions too readily (automation bias). Designing meaningful review is a UX and organisational challenge, not just a checkbox.
Often confused with
Human-in-the-loop means a person is involved in individual decisions. Human oversight is broader: the ability to monitor a system as a whole, understand its limits, intervene and stop it. A system can have oversight without a human reviewing every case.
Questions to ask
- At which points must a person review AI outputs, and what information do they see?
- Do reviewers have enough time and authority to overrule the AI, and how often do they?