CAS Architecture Enables Causal Attribution for Explainable AI

The Causal Attribution Score (CAS) uses an interventional coalition game and causal Shapley contributions to provide local and global causal explanations. In simulations, coalition-aware CAS achieved a mean Local CAS MAE of 0.107, significantly outperforming standard normalization methods.