Unmeasured subgroup false negatives may delay or deny: Which remaining risk most directly concerns
Aggregate efficiency and accuracy do not resolve excluded-subgroup harms, especially when review targets only uncertain cases.
The question
A housing-application system reduces processing time and improves aggregate prediction accuracy. However, evaluation excludes applicants with irregular income, and staff review only cases flagged as uncertain. Which remaining risk most directly concerns affected-person outcomes?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Processing-time improvements may be overstated by inconsistent workload measurement.Measurement validity matters, but this does not address the stated population exclusion or the consequences for applicants.
- Aggregate accuracy may decline after expanding the evaluation population.A possible metric change is not itself the central harm; the decisive issue is whether excluded people receive inaccurate housing outcomes.
- Unmeasured subgroup false negatives may delay or deny housing access. ✓Excluded applicants may experience inaccurate adverse outcomes, while aggregate gains and selective review do not establish fairness for that subgroup.
- Staff may spend more time reviewing uncertain applications than planned.Review workload is a legitimate operational concern, but it does not explain harm to excluded applicants receiving unexamined decisions.
The trap
When aggregate metrics improve but subgroup evidence is missing, prioritize the plausible subgroup harm over operational gains. How to remember it
Aggregate efficiency and accuracy do not resolve excluded-subgroup harms, especially when review targets only uncertain cases.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Development questions
- Low-confidence itinerary answers may still be inaccurate: Before accepting residual risk, which demonstrated →
- High-confidence routing errors can bypass the human: Which remaining risk should be addressed before →
- Average accuracy can hide costly rare-defect failures: What remaining risk should change the requirements? →
- All 426 Understanding How to Govern AI Development questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Development ·
Every answer, right and wrong, comes with its own explanation.
