Glossary
Human-in-the-Loop
An AI system design where humans review or approve specific model outputs before they take effect — the standard pattern for any AI decision with regulatory or safety consequences.
Human-in-the-loop (HITL) is a pattern where the AI proposes and a human disposes. The model produces a recommendation, draft, or action; a qualified human reviews it; the action only takes effect after approval. The human's review is logged, timestamped, and tied to a user identity.
HITL is not a sign that the AI is not good enough. It is the design choice that makes the system defensible. For a clinician documenting a visit, an AI-drafted note is reviewed and signed before it enters the chart. For a paralegal generating a contract clause, an attorney reviews and approves before it goes to the client. The AI is a productivity multiplier; the human's judgment and license remain the source of authority.
The practical design decisions are where to put the review boundary, what the human sees (the proposed action plus the reasoning, or just the action), and how friction-free the approval path is. Heavy friction kills adoption. No friction defeats the purpose. The right answer is usually a streamlined diff-style review with one-click approval, plus a logged audit trail.
For regulated industries, HITL is often the only credible answer to questions like "what if the AI gets it wrong?" The answer is "a human reviewed it before it took effect, and that human is on the record."
Related terms
Agentic Workflows
Multi-step AI workflows where the model selects tools, makes decisions, and chains actions — rather than producing a single response from a single prompt.
Audit Logging for AI
Recording every model invocation — the requesting user, the input, the retrieved context, the model's output, and the tools it called — in a tamper-evident log.