How is an author made? Assemblage, intra-action and agential cut in a case of co-authorship with ChatGPT

Abstract

Between 2023 and 2025, ChatGPT was listed as an author or co-author of articles published in indexed journals, even though high-impact journals and publication ethics organizations had banned this attribution in 2023. This article asks who, or what, is acting when a generative artificial intelligence system is called an author, and what authorship becomes after these systems. It starts from Foucault’s notion of the author function to indicate its limits in the face of writing with large language models, discusses the performance criterion of Turing’s imitation game, and articulates Hayles’s discussions of cognitive assemblages and the computationalization of the human with Barad’s notion of intra-action. Authorship attribution is approached as an agential cut, which separates what counts as subject and what counts as object. The analysis of Matusov et al. (2024), with ChatGPT-4 among its authors, describes three cuts: the invitation, which offers the system the position of co-author before any response; the conversation, in which researcher and system negotiate meanings and constitute each other; and the byline, which stabilizes authorship, while the paratexts concentrate accountability on the humans. The results indicate that anthropomorphizing effects do not precede attribution but are produced by these cuts, and that their reverse side is the computationalization of the researcher. It is concluded that authorship is neither only human nor only machinic, but intra-active, and that ChatGPT may not inaugurate, but rather make visible, an authorship that has always been an assemblage.