Foundations of AI Explainability

Explainability is a family of distinct demands (transparency, post hoc explanation, fidelity, intelligibility to an audience) that arise where a decision problem cannot be fully formalized; every explainability claim must say which demand it meets and for whom.