The Policy Fiction of Human Authorship in AI-Mediated Science

Abstract

The growing integration of generative artificial intelligence into scientific work has destabilized traditional assumptions about authorship, accountability, and intellectual responsibility. While existing academic and publishing norms continue to treat human authorship as the central unit of responsibility, contemporary knowledge production is increasingly organized through distributed “human - AI” interaction. This article addresses the resulting gap between formal authorship rules and the actual epistemic structure of AI-mediated science. To analyze this problem, the article develops the Epistemic-Legal Authorship Decoupling Model (ELAD). The model distinguishes between formal authorship claims and substantive epistemic authorship by decomposing scientific work into weighted cognitive components and evaluating human origination, human epistemic control, and formal responsibility across them. The model demonstrates that authorship can remain formally intact while substantive human contribution and control erode significantly. In such cases, authorship is less an accurate description of epistemic agency than an institutional fiction necessary for governance and liability. The article concludes that the key problem in AI-mediated science is not AI use as such, but the decoupling of formal responsibility from substantive epistemic contribution, which calls for a rethinking of authorship norms in research and scholarly publishing.