Deep Model Decisions Resistant to Explanation
深度模型決策難以解釋
Verification & Validation
Risk Description
When a customer requests the reasons for an adverse decision, and the organization can supply only the model's output score without identifying specific grounds; due to unmitigated control gaps, after the customer files a complaint, the organization can neither explain the decision logic nor demonstrate that improper factors were not used, triggering compliance exposure and operational reputational costs.
Framework Mappings
EU AI ActArt.13、Art.86
ISO/IEC 42001Annex A.8.2
NIST AI RMFMEASURE 2.9
ISO/IEC 5338驗證與確效
MIT AI Risk RepositoryDomain 7
Risk Treatment & Implementation Guidance
Deploy explainability tooling for systems affecting individual rights so every adverse decision maps to concrete grounds; Keep decision audit records of input features, model version and output basis for reconstruction on appeal; Establish an appeal-response procedure that explains decisions in terms the individual can understand