When AI Writes, Who Is the Author? Authorship as Governance in Generative AI

Abstract

Debates over generative AI often ask whether texts produced with large language models are “really” authored by the human user. These debates typically assume that authorship resides in the production of linguistic tokens. This paper challenges that assumption by reconceptualizing AI-assisted writing as a structured, iterative process rather than a discrete act of generation. I introduce the concept of Corrigible Judgment Execution (CJE) to describe how human agents orient communicative purposes, solicit probabilistic proposals from generative systems, evaluate outputs against normative and strategic criteria, impose constraints, and determine convergence. Drawing on the concept of entropy from information theory, I argue that generative models operate within high-uncertainty possibility spaces, producing text that is semantically plausible but purposeless until embedded within governed processes of constraint and ratification. Meaningful communication emerges not from the elimination of stochasticity but from its disciplined reduction. Authorship, therefore, cannot coherently be located in token production. It consists in the sustained authority to govern the reduction of communicative entropy and to ratify convergence under conditions of accountability. By relocating authorship from inscription to governance, this account clarifies responsibility in human–AI co-production and situates AI-assisted writing within broader sociological debates about agency, expertise, and automation.

Ähnliche Arbeiten