AI Ethics Case Library: Fairness Metrics and Thresholds (2D)
Top-down dot histograms of two groups sorted by a threshold classifier, with live selection rate, true-positive rate, false-positive rate and precision for each group, ROC curves and the trade-off between fairness criteria. Same shared model and numerical checks as the 3D scale.
Sort a crowd of 400 simulated applicants with a score-threshold classifier and watch three group-fairness criteria (equal selection, equal opportunity, equal precision) pull against each other. Move each group's threshold, base rate and measurement bias; a Blender balance scale tilts with the selection gap while live rates, ROC curves and numerical checks show why they cannot all be zero. Educational model, not guidance for real decisions.
Top-down dot histograms of two groups sorted by a threshold classifier, with live selection rate, true-positive rate, false-positive rate and precision for each group, ROC curves and the trade-off between fairness criteria. Same shared model and numerical checks as the 3D scale.
2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install