Decoder. plain-English AI glossary

HITL

● Core“human in the loop”

Also called Human in the loop

A person checks or approves the AI’s work before it counts — the safety catch on high-stakes steps.

Think of it like

A spellchecker that suggests, but you still have to hit accept.

Example

The AI drafts the refund; a human clicks approve before any money actually moves.

How it actually works

The real design question is where the human sits: on every action (slow, safe), on random samples (a spot check), or only on low-confidence cases where the model itself asks for help. Most teams end up with the third.

For product teams

A human approval gate on risky actions — your control point for compliance and trust.

For engineers

An approval checkpoint that blocks an action until a human confirms (all / sampled / low-confidence).

Evals — The judgment that becomes your evals.
Agents — The opposite pole from a fully autonomous agent.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome