Skip to content

Glossary · AI and automation

Human-in-the-loopHuman review

Human-in-the-loop is the design in which a person confirms, corrects or rejects what an AI system proposes before it takes effect. The AI prepares the work; the decision still belongs to someone with a name.

Human-in-the-loop: the details

It doesn’t mean reviewing everything. The usual approach is a threshold: what the system resolves with high confidence and low risk goes through on its own, and anything doubtful or expensive goes to a review queue. The checkpoint goes where a mistake costs: a payment, an email to a customer, a change to a case file, a decision that affects a person.

For it to work, reviewing has to be faster than doing the task by hand. That means showing the proposal next to the evidence—the original document, the doubtful field highlighted—and recording what was accepted and what was corrected. Those corrections are also the best material for improving the system. For automated decisions that affect people, the GDPR recognizes the right to obtain human intervention.

Human-in-the-loop in a project

It’s the rule we apply in every automation: the AI proposes and a person or a written rule confirms. We design the review queue as one more screen of the product, with its permission, its keyboard shortcut and its log.

On this site

Where we apply it

All services

Glossary

Related terms

See the whole glossary

Get started

Does this come up in your project?

Tell us which system you have and what is still done by hand. We’ll tell you where we would start and whether it fits.

Before you write, see how we price a project or our solutions by department.

Fig. 07Morse keyH · E · L · L · O