FMEA-AI: AI fairness impact assessment using failure mode and effects analysis

Recently, there has been a growing demand to address failures in the fairness of artificial intelligence (AI) systems. Current techniques for improving AI systems are focused on broad changes norms, procedures and algorithms used by companies that implement those However, some organizations may require detailed methods identify which user groups disproportionately impacted specific components their Failure mode effects analysis (FMEA) is popular safety engineering method proposed here as vehicle support conducting “AI impact assessments” organizations. An extension FMEA called “FMEA-AI” modification familiar tool engineers manufacturers can integrate moral sensitivity ethical considerations into company’s existing design process. Whereas current assessments focus helping regulators an aggregate risk level entire system, FMEA-AI helps multiple failure modes system. It also explicitly identifies considers objective definition proportional satisfaction claims calculating likelihood severity fairness-related failures. This help industry analysts adapt widely known incorporate considerations, promote overcome resistance change.