The article discusses the complexities of explainable artificial intelligence (XAI), a rapidly developing field aimed at making AI decision-making processes transparent and interpretable. The authors introduce a new rule-based explainer, the Local Universal Explainer (LUX), which generates factual, counterfactual, and visual explanations. Unlike other algorithms, LUX focuses on identifying local high-impact concepts rather than generating data. When tested against other rule-based explainers, LUX was found to be simpler, more globally faithful, and more representative.
Publication date: 23 Oct 2023
Project Page: https://arxiv.org/abs/2310.14894v1
Paper: https://arxiv.org/pdf/2310.14894
