Rethinking Transparency in Artificial Intelligence: From Definitions to Functions

Abstract

Transparency has become one of the defining principles of contemporary artificial intelligence governance, appearing throughout legislation, standards, regulatory frameworks, and ethical guidelines. Yet despite this widespread consensus, AI transparency remains conceptually unstable. This paper argues that the instability does not arise because transparency lacks definition, but because artificial intelligence inherited multiple transparency traditions that evolved independently across law, science, engineering, public administration, archival practice, information security, philosophy, and related disciplines. Through a historical and comparative reconstruction, the paper shows that each tradition evolved to solve a different informational problem and consequently developed distinct objects, audiences, mechanisms, and criteria of success. Many contemporary debates over AI transparency therefore reflect the convergence of legitimate but historically distinct transparency functions rather than conceptual disagreement. Building on this reconstruction, the paper argues that transparency requirements should be specified according to their intended function rather than treated as a single universal obligation. It examines the implications of this perspective for AI development, governance, and standards, introducing transparency profiles as a conceptual framework for matching transparency functions to different application domains. Rather than offering another universal definition, the paper provides a historical and philosophical foundation for understanding what AI governance inherited and why that inheritance matters.