Can AI Be Explainable?

This chapter provides an introduction to Explainable AI (XAI), its purposes, key concepts, and terminology, and its importance to Responsible AI. It discusses the aim of explanations to improve understanding and to calibrate trust and reliance, leading to improved system use, and how this contributes to improved transparency in Responsible AI. It describes global and local explanations and how information in explanations can be presented in various ways. The current state of the art in XAI techniques, such as LIME, SHAP, Counterfactual explanations, Grad-CAM, and their limitations, are presented. Multiple perspectives on explanations from computer science, HCI, and social science that question the scope of explanations and how explanations are generated are discussed. As a consequence, opportunities and challenges for the application of XAI in Responsible AI, such as XAI in fairness and bias, the development process, and technological advances, are identified. Reflections are offered on why explanations are not commonly employed in Responsible AI so far, in order to set future research directions and encourage the removal of barriers standing in the way to its successful adoption in practice.