Back to Glossary

Interpretability / Explainability (XAI)

Fundamentals

Understanding how and why a model makes decisions.


Explainable AI (XAI) makes model behavior understandable to humans.

  • Approaches: Intrinsically interpretable models (e.g., trees) and post-hoc methods (SHAP, LIME, Grad-CAM).
  • Goals: Trust, debugging, compliance, safety.
  • Limitations: Approximations can mislead; explanations must fit the target audience.