Question 2
Domain 6: Human-AI Interaction and OversightWhat human-in-the-loop workflow would best meet these requirements?
Correct answer: B
Explanation
This workflow fits human-in-the-loop because it keeps people in the decision path through "rule-based escalation" when risk is high or model confidence is low, and it gives reviewers "context provision" by showing the agent analysis. It also supports oversight with "human override tracking" to learn from disagreements and "audit logging" to document human involvement.
Why each option is right or wrong
A. Have humans review all decisions (no AI autonomy).
B. Implement intelligent routing with: rule-based escalation (amount > $50K or confidence < 85%), context provision (show agent analysis to reviewers), human override tracking (learn from disagreements), and audit logging (document human involvement).
The requirement is met by a controlled review workflow that triggers human intervention when the transaction exceeds $50,000 or model confidence falls below 85%, which are explicit escalation thresholds rather than ad hoc review. The inclusion of reviewer context, override capture, and audit logging satisfies the core governance elements of human-in-the-loop operation: decision support, documented human participation, and traceable disagreement handling for later model improvement.
C. Let AI make all decisions but allow humans to override after the fact.
D. Randomly route 10% of decisions to humans.