The European Union has proposed the Artificial Intelligence Act which introduces a proportional risk-based approach to AI regulation including detailed requirements for transparency and explainability. Many of these requirements may be addressed in practice by the field of explainable AI (XAI), however, there are fundamental differences between XAI and the Act regarding what transparency and explainability are. These basic definitions should be aligned to assure that regulation continually translates into appropriate technical practices. To facilitate this alignment, we first give an overview of how XAI and European regulation view basic definitions of transparency with a particular focus on the AI Act and the related General Data Protection Regulation (GDPR). We then present a comparison of XAI and regulatory approaches to identify the main points that would improve alignment between the fields: clarification of the scope of transparency, the legal status of XAI, oversight issues in conformity assessments, and dataset-related transparency.

翻译：欧盟提出了《人工智能法案》，该法案引入了基于风险的比例化监管方法，包括对透明度和可解释性的详细要求。这些要求在实践中可能由可解释人工智能（XAI）领域解决，然而，XAI与该法案在透明度和可解释性的定义上存在根本差异。这些基本定义应保持一致，以确保监管能够持续转化为适当的技术实践。为促进这种一致性，我们首先概述了XAI和欧洲法规如何看待透明度的基本定义，特别关注《人工智能法案》及相关的《通用数据保护条例》（GDPR）。随后，我们比较了XAI与监管方法，以确定能够改善这两个领域之间一致性的主要问题：明确透明度范围、XAI的法律地位、合规评估中的监督问题以及与数据集相关的透明度。