Definition
Human in the Loop (HITL) is a mechanism that keeps a person involved in an automated or AI workflow: the machine executes, while a human reviews, decides or corrects at key checkpoints.
How it works
Typically a model or program produces an output first, then a person confirms, edits or rejects it, and the confirmed result flows back into the process. In model training, humans scoring or ranking outputs are also the core of methods such as RLHF.
Usage
HITL appears in workflows with high stakes or accuracy requirements: content moderation, medical image reading, fallback decision-making in autonomous driving, and human approval before an agent takes critical actions.
Related and confusable terms
HITL stresses human intervention and is the counterpart of a fully autonomous agent: an agent completes tasks on its own toward a goal, while a HITL process waits for human sign-off at checkpoints. They are often combined—an agent handles most steps by itself, and key decisions are left to a person.