Glossary

Cortexa AI Glossary · Agents and connections

What does "human in the loop" mean?

From Cortexa Learn, by Cortexa Consulting. Last checked .

Your bank texts "Was this you?" before a big payment. That text has a name.


Was this you?

Your bank texts you before a large payment goes through. Was this you? Reply yes or no. A computer system did most of the work there. It spotted a payment that looked unusual and held it. But you get the last word, at the moment it counts. That's the idea behind a phrase you'll hear a lot with artificial intelligence (AI): human in the loop. It means a person reviews, approves, or can step in on what an AI system does, at the points where a mistake would matter.1

The loop

The loop is the cycle the system runs: it does something, checks the result, and goes again. Being in the loop means a person is part of that cycle, not watching from somewhere outside it. Their judgment feeds into what happens next. The opposite is a system that runs from start to finish while nobody looks, and a mistake only shows up once someone downstream trips over it. Both have their place. The trick is knowing which one you've got.1

Where a person can stand

There are three common places for that person.

  • Before it acts: the system suggests, and a person approves.
  • After it acts: a person reviews the work and fixes what's wrong.
  • Alongside it: the system runs on its own while a person watches, ready to stop it.

The third has its own name: human on the loop. The person supervises instead of approving each step. Most real setups mix them, with a tighter check at the riskier steps.4

Matching the stakes

Not everything needs a person. If a music app picks a song you don't like, you skip it. Nobody needs to approve a playlist. A letter about someone's medical results is different, and so is a loan decision. When a mistake would hurt someone or can't easily be undone, a person should be closer to the work. IBM's explanation makes the same point: human review matters most in high-risk and medium-risk uses, and in unusual cases the system wasn't trained for.12

A pause before acting

You'll see this more and more. As AI tools start doing tasks for you, the loop moves into everyday apps. A well-built assistant pauses and asks before it sends a message, pays for something, or deletes a file. That pause is the loop at work. Some people find it a little annoying at first. It costs you a few seconds, and it keeps the final say with you on the steps you can't take back.

A real check

A person only helps if they can really say no. That means having the time to look, the facts to judge, and the freedom to stop things. Say a reviewer is handed two hundred items an hour and clicks approve on each one. That isn't really checking. People also tend to trust a confident machine too much, especially when it's usually right. So a good loop makes the check easy to do properly. It shows the reviewer what changed and why, and it asks for a decision only when one is needed.3

Fixes that teach

There's one more benefit. When a reviewer corrects the system, those corrections can be collected and used to improve it, so the same mistake comes up less often. That's how many systems get better over time. The next time an app asks you to confirm something, notice it. Which steps in your own work would you want a person to see first?1

Works cited

  1. IBM, "What is human in the loop (HITL)?" (checked )
  2. NIST, "AI 600-1, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile" (July 2024) (checked )
  3. Parasuraman and Manzey, "Complacency and Bias in Human Use of Automation: An Attentional Integration," Human Factors 52(3) (2010) (checked )
  4. IBM Technology (Martin Keen), "What is Human In The Loop with AI? How HITL Shapes AI Systems" (video, 2026) (checked )