New paper proves no AI feature ranking is reliable when data is correlated

Researchers from an anonymous team (Caraker, Arnold, Rhoads) have released a landmark paper proving that no feature attribution ranking can simultaneously satisfy three desirable properties—faithfulness, stability, and completeness—when features are collinear. For collinear pairs, ranking effectively reduces to a coin flip. The proof is quantitative: the attribution ratio diverges as 1/(1-rho^2) for gradient boosting, is infinite for Lasso, and converges for random forests. The authors characterize the entire design space: only two families of methods exist—faithful-complete methods that are unstable (rankings flip up to 50% of the time) and ensemble methods like DASH that are stable and report ties for symmetric features.

To mitigate the problem, the team introduces DASH (Diversified Aggregation of SHAP), a Pareto-optimal ensemble method that achieves the Cramér-Rao variance bound with a tight ensemble size formula. In a survey of 77 public datasets, 68% exhibited attribution instability. The framework includes practical diagnostics—a Z-test workflow and single-model screening tool—and has direct consequences for fairness auditing: SHAP-based proxy discrimination audits are provably unreliable under collinearity. The entire impossibility theorem, design space theorem, and diagnostics are mechanically verified in Lean 4 (305 theorems from 16 axioms, 0 sorry)—the first formally verified impossibility in explainable AI.