Explainability addresses the 'black box' problem where complex models like deep neural networks produce accurate results through opaque processes that users cannot interpret. This creates risk when decisions affect individuals' rights, safety, or significant interests and regulators require justification. A bank using AI for loan decisions must explain why applications were denied to comply with fair lending laws and build customer trust. Techniques for explainability include model-agnostic methods like LIME and SHAP, inherently interpretable models, feature importance analysis, and decision pathway visualization, balanced against the accuracy-interpretability tradeoff.