Rethinking Transparency in Artificial Intelligence: Historical Functions and Functional Adequacy

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Transparency has become a defining principle of artificial intelligence governance, yet it remains conceptually unstable. Existing scholarship has shown that AI transparency is multidimensional, audience-dependent, and broader than explainability or disclosure. This paper addresses a prior question that those classifications leave open: why do legitimate transparency requirements differ, and why do they conflict when applied to the same AI system? Through a historically informed comparative reconstruction, the paper argues that artificial intelligence inherited distinct transparency traditions from law, science, engineering, public administration, archival practice, information security, and philosophy. Each developed in response to a different informational problem and therefore carries its own objects, audiences, mechanisms, limits, and criteria of success. Their convergence within AI governance explains why transparency cannot be represented as a single movement from opacity toward visibility. The paper distinguishes transparency functions from the mechanisms used to support them and introduces functional adequacy as an evaluative criterion: a transparency arrangement succeeds only when it enables the intended audience to perform the epistemic or institutional action the requirement was meant to support. Transparency profiles apply this account to particular systems by specifying required functions, audiences, purposes, mechanisms, limits, and evidence of success. The contribution is not another inventory of transparency types, but an explanation of their historical plurality and a framework for determining whether particular transparency arrangements perform the functions claimed for them.