Automated Explanation Selection for Scientific Discovery
Markus Iser
TL;DR
Addressing the need for trustworthy AI in safety-critical domains, the paper proposes a cycle that blends machine learning with automated reasoning to generate and select explanations. The method encodes learned models into a formal language and uses deductive reasoning (e.g., SAT/MaxSAT/SMT) to deduce explanations, which are then evaluated and used to form testable hypotheses. A taxonomy of explanation-selection properties from sociology and cognitive science is introduced, with abductive (AXp) and contrastive (CXp) explanations and a multi-objective optimization framing; formal guarantees simplify evaluation and enable verifiable explanations. The work argues that this approach improves reliability, provides principled hypothesis generation, and offers a productive pathway for future research in XAI-driven scientific discovery.
Abstract
Automated reasoning is a key technology in the young but rapidly growing field of Explainable Artificial Intelligence (XAI). Explanability helps build trust in artificial intelligence systems beyond their mere predictive accuracy and robustness. In this paper, we propose a cycle of scientific discovery that combines machine learning with automated reasoning for the generation and the selection of explanations. We present a taxonomy of explanation selection problems that draws on insights from sociology and cognitive science. These selection criteria subsume existing notions and extend them with new properties.