Average accuracy can hide costly rare-defect failures: What remaining risk should change the requirements?
Average accuracy can conceal consequential rare-defect misses, so the business requirement must reflect impact and representative evidence.
The question
A manufacturing quality model was specified to maximize average defect-detection accuracy. Inspectors report that missing a rare defect can cause extended line stoppages, while false alarms are usually inexpensive. Testing contains few examples of that defect. What remaining risk should change the requirements?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Inspectors may need training to interpret the existing accuracy score.Training may improve interpretation, yet it cannot correct a metric that fails to represent rare, costly defect outcomes.
- Average accuracy can hide costly rare-defect failures. ✓The metric and sparse examples can obscure consequential errors, so requirements should reflect rare-defect impact and evidence limitations.
- Additional production data may increase the model’s average accuracy.More data could help, but volume alone does not ensure rare-defect representation or address the asymmetric operational consequences.
- False alarms may increase when inspectors emphasize rare defects.False alarms matter operationally, but the stated evidence makes missed rare defects the more consequential unresolved risk.
The trap
Look for asymmetric harms and sparse test examples; aggregate accuracy is insufficient when rare errors carry major costs. How to remember it
Average accuracy can conceal consequential rare-defect misses, so the business requirement must reflect impact and representative evidence.
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
- High-confidence routing errors can bypass the human: Which remaining risk should be addressed before →
- Confident extraction errors may persist on handwritten: Which residual risk remains? →
- Users may influence coworker evaluations with incomplete: Which concrete misuse risk remains? →
- 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.
