Service Quality Disparity Across Groups
服務品質跨群體差異
Operation & Monitoring
Risk Description
When users with a particular accent must repeat each interaction several times before being recognized correctly, taking several times longer than other users to complete the same task; due to unmitigated control gaps, because the system never refuses service, the complaint mechanism never captures the problem, triggering compliance exposure and operational reputational costs.
Framework Mappings
EU AI ActArt.10、Art.15
NIST AI 600-1Harmful Bias and Homogenization
ISO/IEC TR 24027
ISO/IEC 5338運作與監控
MIT AI Risk RepositoryDomain 1
Risk Treatment & Implementation Guidance
Establish cross-group performance benchmarks covering accent, language and demographic dimensions in validation and post-launch monitoring; Monitor service-equality metrics such as completion time and retry counts rather than refusal rates alone; Augment test data where minority-group samples are insufficient and continuously improve the model