Attribution-Scores and Causal Counterfactuals as Explanations in Artificial Intelligence

Two main explanation approaches exist for AI systems: Attribution-scores (which identify which input features most influenced a decision) and counterfactuals (which show what would need to change for a different outcome), each offering different insights into how AI makes decisions.

Counterfactuals require careful logical reasoning: The paper emphasizes that counterfactual explanations must be logically sound to be trustworthy—simply saying "if X were different, Y would happen" isn't sufficient without rigorous reasoning backing up that claim.

Explanations bridge the gap between AI capability and human understanding: The paper highlights that explainable AI is essential precisely because newer AI developments have become harder to interpret, making explanation methods critical for responsible deployment.

Attribution-scores and counterfactuals serve complementary purposes: Attribution-scores answer "what mattered in this decision?" while counterfactuals answer "what would change the outcome?"—organizations may need both approaches for comprehensive AI transparency.

Logical foundations matter for AI accountability: The paper stresses that explanation methods grounded in formal logic and causality are more reliable for high-stakes decisions than ad-hoc approaches, supporting better governance of AI systems.