Justice Metrics
Quantitative measures (e.g., disparate impact, equal opportunity) used to assess fairness and nondiscrimination in AI decision‐making.
Definition
Statistical metrics that quantify fairness across demographic or protected classes. Disparate impact measures outcome rates, equal opportunity measures equal true-positive rates, and calibration assesses predicted risk accuracy. Governance frameworks mandate selecting appropriate justice metrics per use case, setting acceptable thresholds, and regularly reporting them to oversight bodies to ensure nondiscrimination throughout the AI lifecycle.
Real-World Example
A predictive-policing model’s outputs are evaluated monthly: the department calculates disparate-impact ratios for stops by race and finds the ratio exceeds the 0.8 threshold. They pause automated patrol recommendations, recalibrate the model with fairness constraints, and verify post-deployment that stop rates meet justice-metric targets.
.svg)
