Interpretability

Interpretability describes how readily a person can understand the relationships and decision logic inside an AI model. A small decision tree or linear model may be interpretable by design, while a large neural network usually requires specialized analysis. Interpretability tools can examine feature importance, internal representations, sensitivity to changes, influential training examples, or local explanations for individual outputs. These methods answer different questions and can be unstable or misleading if treated as proof of causation. The required level also depends on the audience and risk: a developer debugging a model, a regulator reviewing compliance, and a customer contesting a decision need different kinds of explanation.