Explainable AI. Explain for whom. What. And why

Highly complex AI systems are often so-called black boxes. Their results are often difficult to understand. However, comprehensibility is crucial in medical diagnostics or quality assurance. Explanations of AI decisions must therefore be understandable, especially for the respective target group: developers receive information on how to improve AI systems and consumers can find out which factors influenced their credit decision. The white paper addresses the topic of ‘Comprehensible AI’ by looking at the question of who, what, how and why something should be explained. It begins by defining key terms in the subject area, presenting the main objectives of explainable AI and highlighting XAI methods and recent trends. Various personae are used to illustrate how individual differences in terms of prior knowledge, objectives and explanations present themselves.