What is human-in-the-loop AI?
Human-in-the-loop means a person reviews, approves or corrects an AI system's output at defined points before it takes effect, rather than letting it act unchecked.
In more detail
The design question is where to put the person. Reviewing every output defeats the point, and reviewing none removes accountability. A common pattern is confidence-based routing: the system proceeds when checks pass and sends exceptions to a reviewer with the evidence attached. Another is approval by risk: reading and drafting are automatic, while sending money, contacting customers and changing master data need a click. Review also produces training data, since each correction can feed back into rules or test sets. Done badly, reviewers rubber-stamp because the queue is long, so track the override rate. If reviewers never change anything, loosen the gate. If they change a lot, fix the system.
Example
An AP team lets the system post invoices that match their purchase order exactly. Mismatches land in a queue showing the invoice, the PO and the line that differs, and the approver picks accept, adjust or reject.