Causality problems in machine learning systems

The topic of causality represents a major issue for the contemporary machine learning (ML) research agenda. Although the technical literature treats it as a single problem, causality is a family of distinct, albeit related, issues. This chapter presents a taxonomy that distinguishes among three different causality problems in ML systems: the problem of causal opacity, the problem of causal interpretability, and the problem of causal reliability. Based on this distinction, it provides an overview of the various methodologies and techniques adopted to address these issues in the fields of explainable artificial intelligence and causal ML.