Keeping people in control.
THE DEFINITION
A person remaining responsible for and able to review a system's output or decisions.
WHY IT MATTERS
AI cannot be accountable, notice that it is wrong, understand consequences, or exercise judgement about an exception.
WHAT MEANINGFUL OVERSIGHT REQUIRES
A person who reads the output Who has the knowledge to evaluate it Who has authority to reject it Who has time to do so properly
THAT LAST POINT
Oversight that is nominal is not oversight.
A person approving a hundred outputs an hour provides no protection.
WHAT UNDERMINES IT
Automation bias Volume Familiarity, as verification decays
WHERE IT IS ESSENTIAL
Decisions affecting people, anything published, financial transactions, safety-related matters.
WHAT TO BUILD IN
Enough time to review, sampling where full review is impractical, and periodic checks that review is happening.
RELATED TERMS
Automation bias, accountability, explainability.
