Framing what can be explained - an operational taxonomy for explainability needs

Abstract Modern software systems are becoming increasingly complex to operate and to understand. In some cases, where usability and transparency cannot address these problems on their own, providing explanations to the user can be an efficient solution. As a result, explainability has gained much traction as a non-functional requirement of software systems. Understanding what system requires which explanations is necessary to facilitate requirements engineering for explainable systems. To understand how different kinds of explanations relate to different kinds of software, an explainability taxonomy that applies to a variety of different software types is needed. In this paper, we report on the development and evaluation of a taxonomy for explainability needs. In an online survey, we asked 84 participants to state their questions and confusions concerning their most recently used software systems. We identified 315 explainability needs from the survey answers, from which we derived the taxonomy. We then evaluated the taxonomy in two focus groups with six stakeholders each. Drawing from the insights that we gained through this research, we present the three major contributions of this work: 1) an operational taxonomy for explainability needs in everyday software systems, 2) an overview of how the need for explanations differs between different types of software systems, and 3) a detailed evaluation of the taxonomy applied within a practical scenario.