At the very beginning of this book, we tried to clarify the difference between the term interpretability and the term explainability. In that context, we said that Interpretability is the possibility of understanding the mechanics of a machine learning model but this might not be enough to answer “Why” questions that are questions about the causes of a specific event. We also provided in Table 9.1 of Chap. 1 (don’t worry to look at it now, we will start again from this table in the following) a set of operational criteria based on question to distinguish between Interpretability as a lighter form of Explainability. As we saw, Explainability is able to answers questions about what happens in case of new data, “What if I do x, does it affect the probability of y” and counterfactual cases to know what would have changed if some features (or values) would not have occurred. Explainability is a theory that deals also with unobserved facts toward a global theory, while Interpretability is limited to make sense of what is already present and evident. The question is: Why are you getting back to this point in this chapter about making science with ML? The answer, long story short, is that Explainability is exactly what we need to climb “the ladder of causation” (we will talk about it in a while). We will use XAI in the domain of “knowledge discovery” with a specific focus on scientific knowledge. To recall what we already discussed: