## TL;DR

Explainable AI makes model decisions interpretable by humans. As models become more complex (deep NNs, LLMs), understanding WHY a model made a decision becomes critical for trust, debugging, and regulatory compliance (EU AI Act, GDPR). Methods: SHAP (feature importance), LIME (local explanations), attention visualization, saliency maps.

## Core Explanation

SHAP (SHapley Additive exPlanations): game-theoretic approach assigning each feature an importance score. LIME: locally approximate complex model with simple interpretable model. Saliency maps: highlight input pixels most influential for prediction. Limitations: explanations can be misleading, feature correlations complicate attribution. Post-hoc methods explain AFTER prediction; inherently interpretable models explain BY design.

## Further Reading

- [Interpretable Machine Learning (Christoph Molnar)](https://christophm.github.io/interpretable-ml-book/)