Towards a terminology for a fully contextualized XAI

Explainable Artificial Intelligence (XAI) has seen a surge in popularity in the past few years, thanks to new legislations that promote the “right to explanation”. Many popular methods have been developed recently to help understand black-box models, but it is not clear yet how an explanation is defined. Furthermore, the community agrees to say that many important terms do not have commonly accepted definitions. In this paper, we review the literature and show that there is a major issue concerning the definitions of terms such as explainability or interpretability. There is a lack of consensus that slows the development of this field. To address this problem, we propose a terminology that takes into account the context of an AI system, i.e., its users, purposes or design. This terminology is compatible with the majority of the definitions encountered in the literature so that it can be a foundation for future works.