What do algorithms explain? The issue of the goals and capabilities of Explainable Artificial Intelligence (XAI)

Abstract

AbstractThe increasing ubiquity of machine learning (ML) motivates research on algorithms to “explain” models and their predictions—so-called Explainable Artificial Intelligence (XAI). Despite many publications and discussions, the goals and capabilities of such algorithms are far from being well understood. We argue that this is because of a problematic reasoning scheme in the literature: Such algorithms are said to complement machine learning models with desired capabilities, such as interpretability or explainability. These capabilities are in turn assumed to contribute to a goal, such as trust in a system. But most capabilities lack precise definitions and their relationship to such goals is far from obvious. The result is a reasoning scheme that obfuscates research results and leaves an important question unanswered: What can one expect from XAI algorithms? In this paper, we clarify the modest capabilities of these algorithms from a concrete perspective: that of their users. We show that current algorithms can only answer user questions that can be traced back to the question: “How can one represent an ML model as a simple function that uses interpreted attributes?”. Answering this core question can be trivial, difficult or even impossible, depending on the application. The result of the paper is the identification of two key challenges for XAI research: the approximation and the translation of ML models.