Textbook of AI

Abstract. Argues that post-hoc explanations of black-box models are fundamentally unreliable for high-stakes decisions and that practitioners should instead design intrinsically interpretable models whose decision logic can be verified directly.