Definition
Explainability is the ability to give a person a useful account of why an AI system produced a particular output or action. "Useful" depends on the audience and the decision. A model engineer debugging a failure, an applicant challenging a denial, and an auditor testing compliance need different kinds of explanation.
Explanations may come from an inherently understandable model, analysis of internal behavior, or a post-hoc method applied after the prediction. They should be tested for faithfulness, stability, and relevance. A fluent rationale generated by the same model can be persuasive while bearing little relationship to the mechanism that produced the answer.
Distinguish it from nearby terms
Interpretability focuses on understanding a model's representations, mechanisms, or decision process. Explainability focuses on communicating an account that serves a human purpose. Transparency discloses facts about the system and its development; disclosure alone does not explain a particular result.
Check your understanding
A lending system lists "recent credit history" as the reason for a denial, but changing that feature does not affect the result. The statement may be readable, yet it is not a faithful explanation of this decision.