An analogical argumentation framework for argumentative local explainers

Abstract

A local explanation method (LE) in explainable artificial intelligence (XAI) is basically a two-step procedure: first construct a naively explainable model approximating the black-box model in need of explanations; then extract an explanation from the approximate model and return it. The extracted explanation aims to be just analogous to the target explanation, suggesting that users should rely on analogical arguments to transfer certain properties observable on the former to the latter. In this article, assuming a hypothetical expert user whose knowledge satisfies certain conjectures, we reconstruct the structures ‘ reason  therefore  conclusion ’ of these analogical arguments and study conditions for ensuring the truth of the reason ; conditions for ensuring that the conclusion follows necessarily from the reason ; as well as counter-arguments the user has to consider. It is argued that the presented findings shed light on the internal reasoning of an expert user at the end of User–LE dialogue. On this basis, the article then pursues a computational argumentation-based approach to elevate existing system-centred LEs to user-centred XAI. Technically the paper develops an argumentation framework for the so-called Argumentative Local Explainers ( A r g L E ) which extend local explanation methods with the internal reasoning of a hypothetical expert user . Formally, we show that A r g L E can compute the expert user’s beliefs about the target explanation. It is envisioned that the integration of A r g L E and existing models of argumentation dialogues helps to create conversational agents that can guide non-expert users to receive system-centred XAI explanations in a way that expert users do.