Glossary
Human in the loop
Human in the loop means a person reviews or approves an AI system's action before it takes effect, rather than only auditing it afterwards.
Before, not after
The defining feature is timing. A human in the loop approves before the action happens; a human on the loop reviews afterwards. Both are legitimate, and they suit different risks — anything irreversible or outward-facing wants the first.
Where the gate belongs
At the edge: on the thing that leaves — a message, a payment, a published post — rather than on every internal step. A gate per step trains people to click through without reading, which is worse than no gate because it manufactures the appearance of oversight.
The failure mode
Approval fatigue. If the queue is too long or the items too trivial, the reviewer stops reviewing. The fix is fewer, higher-stakes gates, not a better interface.
Related terms
FAQ
What does human in the loop mean?
A person reviews or approves an AI system's action before it takes effect. It contrasts with human on the loop, where a person reviews afterwards. Anything irreversible or outward-facing generally wants approval before rather than after.
Where should the approval gate sit?
At the edge — on the thing that leaves, such as a message, payment or published post — rather than on every internal step. A gate per step produces approval fatigue, where the reviewer clicks through without reading, which manufactures the appearance of oversight without the substance.