Probabilistic Causal Impact for Scalable Counterfactual Explanations

PCI bridges the gap between principled actual causality and scalable attribution methods like SHAP by treating explainability as a Monte Carlo estimation problem. It approximates Pearl's notions of necessity and sufficiency within probabilistic causal models to enable scalable causal attribution.