AI failure is an incorrect, misleading, unsupported, or otherwise undesirable output or behavior. Because every AI system makes mistakes, responsible product work addresses them openly instead of treating engineers as solely accountable. section 12.3.2
Common failure patterns include:
- False positives identify something as relevant or true when it is not, such as forecasting a sales spike that never occurs. section 12.3.2
- False negatives fail to identify something important, such as a flash sale that unexpectedly depletes inventory. section 12.3.2
- Ambiguity or misinterpretation produces uncertain results when inputs or edge cases are unclear. section 12.3.2
- Biased or incomplete data can create systematically poor outcomes, such as understocking a growing region. section 12.3.2
Managing a forecast failure
A temporary historical anomaly triggers an oversized inventory order. Regular manual reviews can catch the false positive; planned campaigns should be entered so forecasts reflect real-world factors; uncertainty flags invite clarification. section 12.3.2
