What happened
Researchers developed a unified mathematical framework that uses Shapley values — a tool from game theory — to both measure unfairness in AI systems and identify which specific features (like age or marital status) are causing it. This matters because it lets companies and regulators actually see which input variables are driving discriminatory outcomes, rather than just knowing an AI system is unfair without understanding why.
Why it matters
For the first time, you can use the same mathematical tool to diagnose unfairness and explain it in concrete terms — which features are responsible for gender discrimination in hiring or lending decisions — making it harder for companies to claim they don't know why their AI discriminates.