Human-in-the-loop workflows place a person at defined decision points rather than allowing a system to act autonomously. The human may review every output, review flagged items only, or approve actions before execution.
The design question is where review is placed and whether it is genuine, since review that is technically required but practically perfunctory provides no real control.
Alternative Names:
HITL, Human Oversight
Why it Matters?
For legal work, human review is not a design preference but the mechanism by which professional responsibility is discharged. The failure mode worth guarding against is review theater, where volume and interface design make meaningful examination impractical and approval becomes reflexive. Systems that surface uncertainty and cite sources make review real; systems that present confident undifferentiated output make it nominal.
Frequently Confused with
Related terms
Frequently asked questions
Where should the human sit in the workflow?
What makes human review meaningful rather than nominal?
