Stockertim Dictionary · Artificial intelligence
Human on the loop
Also called HOTL
In short
Human on the loop is an arrangement where people supervise an AI system as it runs and intervene when needed, rather than approving every action first.
- Intermediate
- Concept
- Checked
In detail
Human on the loop sits between full manual approval and full automation. The system acts by itself, and people watch dashboards, alerts and samples, ready to step in, correct a result or halt the process. It suits tasks with many small, reversible decisions, where approving each one would be slow and would wear reviewers down.
It works through monitoring and limits. The system logs what it does, flags unusual or low-confidence cases, and stays inside boundaries such as a value cap or a list of allowed actions. Supervisors review flagged items, check random samples and look at trends. A clear stop switch lets them pause the system at once if something drifts.
The arrangement matters because it shifts where the risk sits. Errors can happen before anyone sees them, so the approach is safer for actions that can be undone, such as tagging documents or ordering stock, than for irreversible ones such as making a payment or rejecting a job applicant. Good design keeps the most serious actions with a human in the loop.
At work, teams choosing this pattern write down what the system may do alone, what triggers an alert, how often samples are checked and who can stop it.
Worked example
An AI tool tags incoming invoices at Kite & Co Accountants, and staff supervise it.
- Invoices tagged in a week: 2,000
- Flagged for low confidence at 3%: 2,000 x 3% = 60 invoices reviewed
- Random sample of the rest at 2%: 1,940 x 2% = 38.8, so about 39 invoices checked
- Total human checks: 60 + 39 = 99
Staff check about 99 invoices instead of 2,000, while the flags and the sample keep watch on the rest.
In practice
Operations and finance teams use human on the loop for high-volume, reversible tasks: alerts for unusual cases, regular sample checks, a weekly trend review and a kill switch. Irreversible or high-impact actions stay behind an approval step.
Watch out
- On the loop is not the same as in the loop: the system acts first and people intervene afterwards.
- Monitoring nobody reads is not supervision; alerts need an owner and a response time.
Learn it properly
Official sources
- AI Risk Management Framework · National Institute of Standards and Technology
- NIST AI RMF Playbook · National Institute of Standards and Technology
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.