86. An AI inspection model has high overall accuracy. What would you ask about missed defects?
HardQuality & improvement
Technical
Show model answer
I'd examine defect prevalence, error definitions, severity and performance across the relevant conditions and products. High accuracy can coexist with consequential misses, especially when acceptable items dominate the data or evaluation does not reflect current production, lighting, variation or failure modes.
I'd require specialist evaluation and monitoring with clear use boundaries. A model score is not release authority or safety proof; uncertain and changed conditions need accountable appropriate handling.
What interviewers listen for
- Examine consequential false negatives
- Validate against actual operating conditions
- Preserve qualified release and monitoring authority