A Conceptual Framework of AI Transparency from Sociotechnical Perspective

Transparency serves as a cornerstone principle in artificial intelligence (AI) ethics and governance, playing a crucial role in upholding ethical standards and ensuring responsible AI deployment. Despite its critical importance, the concept of AI transparency remains fragmented in literature, highlighting the necessity for a unified and comprehensive understanding. This article addresses this imperative by firstly conducting a systematic literature review about existing varied definitions of transparency to deepen our understanding of AI transparency. Then, the three key aspects of AI transparency, that is: 1) transparent to whom; 2) transparent of what; and 3) how to be transparent, are examined. Building upon this groundwork, we propose a novel sociotechnical framework that uniquely integrates both intrinsic and extrinsic dimensions of AI transparency while accounting for the roles of internal and external stakeholders, resulting in a three-layered AI transparency framework encompassing intrinsic, internal, and external transparency. This comprehensive framework not only deepens our understanding of AI transparency but also provides a structured roadmap for navigating the complex sociotechnical landscape of AI systems. Our main contribution lies in developing, this novel conceptual framework that enriches both the theory and practice of AI transparency and provides guidelines for designing and deploying transparent AI systems in the future.