Explainable AI is a field focused on making machine learning systems transparent and understandable to humans, especially when they are used in high stakes decisions such as medicine, finance or autonomous driving. Modern AI models often function as black boxes: they can achieve very high predictive performance, but their internal reasoning is opaque. Explainable AI tries to bridge this gap by providing human interpretable explanations for model behavior.

A central distinction is between intrinsic and post hoc explainability. Intrinsically interpretable models, such as simple decision trees or linear models, are structured so that their predictions can be directly understood. Post hoc methods, in contrast, take a complex model such as a deep neural network and generate explanations after training. These explanations may focus on individual predictions, such as indicating which input features were most influential, or on global behavior, such as summarizing how the model behaves across many cases.

Common techniques include feature importance scores, saliency maps for images, counterfactual examples that show how small changes in input alter a decision, and surrogate models that approximate a complex model with a simpler one in a local region of the input space. Choosing the right explanation depends on the audience and task: domain experts, regulators and lay users each need different forms and levels of detail.

Explainable AI also raises philosophical and practical questions. Explanations must be faithful to the underlying model, yet simple enough for humans to grasp. They can influence trust, reveal biases and guide model improvement, but they can also mislead if poorly designed or optimized only for appearance.