AI Glossary
Human-in-the-loop
human in the loop, HITL
Human-in-the-loop is a way of working in which a person approves, corrects, or blocks the model's action at chosen points before the result moves on. It pairs the speed of AI with oversight on higher-risk decisions.
- A person approves or corrects the result at a key point in the process.
- It is used where a model's error is costly or hard to undo.
- It requires clear thresholds: what passes automatically and what goes to sign-off.
In the human-in-the-loop approach, the model does not act fully on its own. At set moments its proposal goes to a person, who approves, corrects, or rejects it. This applies above all to decisions with real consequences: sending a letter to a customer, approving a payment, or making a change in a production system.
The point of this arrangement is to control risk without giving up pace. The automation handles routine cases, and a person steps in where judgment, accountability, or knowledge beyond the model's data is needed. The condition is to set the approval thresholds clearly; otherwise oversight becomes either a fiction or a bottleneck.
Related terms
In guides
Related articles