Authorship after AI? Responsibility moves first in a stepwise disclosure experiment

Abstract

Cataloging, rights clearance and archival provenance depend on identifying an author, which generative artificial intelligence (AI) unsettles. This study asks whether lay publics treat authorship, responsibility and creative production as one judgment or three when AI involvement is disclosed. In an online survey of Japanese adults (N = 250, quota sampled by age and sex), participants chose who was a fictional book's author, who bore responsibility for its content and who produced its creative content, after each of three successive revelations of AI involvement (within-subjects; McNemar, Cochran's Q, Holm corrected). Attribution to the named author fell on all three questions, but unevenly. Responsibility fell furthest, and most participants relocated it to the editor and development team, while authorship and creative production fell less and ceased to differ from each other. The questions already differed before disclosure, which reshaped that difference rather than creating it. The robust structure is bipartite, not tripartite. Disclosure effects are upper bounds: an editor was visible from the first stage and information was only ever added. The byline and the AI system shared a name, so the convergence may reflect shared reference. The sample is Japanese and the study was exploratory. Where earlier work measured each attribution separately, this study tests within participants whether the three differ from one another at each stage. Readers withdraw accountability from a named author while leaving naming and credit intact, a distinction the single authorship field does not record.