AI0-001 Implementing AI Solutions Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Report only the overall accuracy of the model across the entire patient population and confirm that it exceeds a pre-agreed target.

  Why it's wrong here

  Overall accuracy can hide large disparities because a group that is a small share of the population has little influence on the aggregate metric. A model can be highly accurate overall while systematically undertriaging a specific demographic group. The requirement is to detect group-specific undertriage, so a single population-level number is insufficient and may create false confidence. Subgroup metrics are needed to surface the disparity.

• ✗

  Remove all demographic features from the training data and assume the resulting model is fair.

  Why it's wrong here

  Removing demographic features does not guarantee fairness because other variables can act as proxies for group membership. For example, postal code or insurance type may correlate with demographics and preserve the disparity. This approach also makes it impossible to measure subgroup performance, since group labels are unavailable at evaluation time. The hospital would be unable to verify whether undertriage persists, so the requirement is not met.

• ✓

  Evaluate the model's performance separately for each demographic subgroup using metrics such as sensitivity and false negative rate, and remediate disparities before go-live.

  Why this is correct

  Subgroup evaluation with metrics like sensitivity and false negative rate directly detects systematic undertriage for a demographic group. Undertriage is a false negative in urgency classification, so a higher false negative rate for one group is the signal leadership is worried about. Remediating disparities before deployment, through retraining, reweighting, or threshold adjustment, addresses the requirement at implementation time rather than after harm occurs.

• ✗

  Require clinicians to override the AI recommendation whenever they disagree, and track the override rate as the fairness measure.

  Why it's wrong here

  Clinician overrides are a safety net, but the override rate is not a fairness metric. Overrides depend on clinician workload, trust, and awareness, and they may not be applied uniformly across groups. A high override rate could even mask a systematic model disparity rather than reveal it. The hospital needs direct subgroup performance measurement to detect undertriage, so relying on overrides alone does not satisfy the clinical leadership's requirement.