As AI systems increasingly impact society, the EU AI Act (AIA) is the first serious attempt to contain its less desired effects. Among others the act proposes audit as a mechanism and compliance products as tools for organizations to demonstrate compliance. In this paper, a framework for managing AI compliance, APPRAISE, is proposed. The framework is built upon the rationale that driving a balance between generating shareholder value through innovation in AI systems and managing compliance through organizational processes will eventually result in value that is responsible. By adhering to AIA compliance products, the framework operationalizes and hence safeguards compliance. Furthermore, a two-phase experiment with a limited scope is presented. The experiment aims to measure the extent to which companies coordinate technical elements of AI systems to ultimately comply with the AIA. In the first phase a survey is conducted and in the second phase the survey results are validated with a couple of respondents to generate additional in-depth insights and root causes.

翻译：随着人工智能系统对社会的影响日益加深，欧盟《人工智能法案》（AIA）成为首个旨在遏制其不良影响的严肃立法尝试。该法案将审计作为合规机制，并将合规产品作为组织证明合规性的工具。本文提出了一个管理人工智能合规性的框架——APPRAISE。该框架基于以下理念：在通过人工智能系统创新创造股东价值与通过组织流程管理合规性之间取得平衡，最终将产生负责任的价值观。通过遵循AIA合规产品，该框架将合规性付诸实践并加以保障。此外，本文还呈现了一个范围有限的两阶段实验。该实验旨在衡量企业在多大程度上协调人工智能系统的技术要素以最终符合AIA要求。第一阶段进行了问卷调查，第二阶段则通过与部分受访者的验证调查结果，获取了更深层次的洞见与根本原因。