Ethical AI Decision Making

AI systems that approve loans, screen resumes or rank applicants reduce a person to a single score built from weighted criteria — and when one of those criteria is a proxy that correlates with group membership, purely quantitative optimization can produce systematic unfairness without anyone writing a biased rule. This simulator renders two equally-qualified populations as grids of bars: height is each candidate's computed score, color is the resulting decision. Shift the accuracy and proxy-signal weights to watch approval rates between Group A and Group B diverge, move the threshold to see how strictness interacts with bias, and toggle bias correction to see a simple reweighing fix close the gap — while the disparate-impact ratio tracks whether the outcome would pass the common four-fifths fairness rule.