Human in the loop
A design where a person reviews, approves or corrects an AI system's work at defined points. In practice it can mean a person with real judgment, or a person who has become the loop.
"Human in the loop" means that a person takes part in an AI system's process: reviewing outputs, approving actions, handling exceptions or correcting errors. It is often presented as the safety net that makes automation acceptable.
The phrase hides two very different situations. A human staying in the loop applies judgment at moments that matter: approving a consequential action, resolving an ambiguous case, setting policy. A human becoming the loop is consumed by supplying context, checking every output and repairing mistakes, so the person is effectively doing the work through the machine.
Telling them apart
Measure supervision load: how much human attention a dependable outcome requires, and why the human is involved. Mandatory judgment, such as a regulatory approval, is a very different investment than preventable repair of a recurring error.
A well-designed loop also needs somewhere for uncertain cases to go, which is the role of abstention, and it should be reported next to automation rate, not hidden by it.
Read more in The hidden metric in AI automation is supervision.
Related terms
Supervision load
The human attention an AI system still consumes — context supply, review, approval, correction and exception handling — measured per dependable, completed unit of work.
Abstention
A deliberate option for an AI decision system to decline to decide and return uncertain or high-consequence cases to a person, rather than forcing every input into yes or no.
Automation rate
The share of work an AI system completes without a person doing it. A useful headline number that says nothing on its own about the human attention left behind.
Used in these essays
The hidden metric in AI automation is supervision
As AI moves from assisting work to leading it, hours saved stop telling the whole story. The scarce resource shifts to human supervision: approvals, exceptions, context and judgment.
The next AI interface may never be seen
Agents are turning software capabilities into an interface of their own. The next enterprise design problem is deciding what should be callable, by whom, and under which boundaries.