Abduction and Argumentation for Explainable Machine Learning: A Position Survey

Abstract

This paper presents Abduction and Argumentation as two principled forms for reasoning, and fleshes out the fundamental role that they can play within Machine Learning. It reviews the state-of-the-art work over the past few decades on the link of these two reasoning forms with machine learning work, and from this it elaborates on how the explanationgenerating role of Abduction and Argumentation makes them naturallyfitting mechanisms for the development of Explainable Machine Learning and AI systems. Abduction contributes towards this goal by facilitating learning through the transformation, preparation, and homogenization of data. Argumentation, as a conservative extension of classical deductive reasoning, offers a flexible prediction and coverage mechanism for learning -an associated target language for learned knowledge -that explicitly acknowledges the need to deal, in the context of learning, with uncertain, incomplete and inconsistent data that are incompatible with any classically-represented logical theory.

Argumentation: A calculus for Human-Centric AI

“…Premises of arguments directly provide an attributive element of an explanation, while the structure of the dialectic argumentative process can be used to form a contrastive part of the explanations, i.e., explain why some other inference or decision was not made. This link of argumentation to explanation and the general area of Explainable AI has recently attracted extensive attention by the computational argumentation community (Kakas and Michael, 2020;Čyras et al, 2021;Vassiliades et al, 2021). The challenge is how to turn argumentation into the language of explanation in a way that the explanations are provided at an appropriate cognitive level and are of high quality from the psychological and social point of view, e.g., they are naturally informative and non-intrusively persuasive (Miller, 2019).…”

Explainable and Ethical AI: A Perspective on Argumentation and Logic Programming

“…Some works exist in the most recent literature, that underline the possible synergies of integration of different logic approaches, especially when combined with statistical ML algorithms. For instance, [26] discuss abduction and argumentation as two principled forms for reasoning and fleshes out the fundamental role that they can play within ML surveying the main works in the area. More generally, abduction and argumentation have been combined in different ways in the literature, starting from Dung's foundational work [18] introducing the preferred extension semantics of abductive logic programs.…”

“…Argumentation is the spearhead of the proposed approach, yet -in order to tackle the AI requirements for ubiquitous intelligence -it should be strongly intertwined with logic programming and its extensions. In particular, our vision of symbolic intelligence leverages argumentation, abduction, inductive logic programming, and probabilistic logic programming, along the line of some recent research works-e.g., [22,26,31].…”

Chapter 11. Autodidactic and Coachable Neural Architectures

“…As in the previous section, we shall consider the case of training the neural module given an arbitrary fixed policy, but we shall, now, assume enhanced access, and, in particular, that the policy can be accessed through both an abduction and a deduction method. The abduction method receives an output and yields the set of inputs on which the policy would entail the output, based on whatever semantics the symbolic module adopts [17].…”