Identical errors, different responsibilities: epistemic duty in AI-assisted scientific writing

Abstract

Generative artificial intelligence has introduced a peculiar asymmetry into scholarly responsibility. Two authors may publish the same false claim generated by the same system, although one accepts it without scrutiny while the other conducts a serious but unsuccessful review. Conversely, an author may publish a correct claim without undertaking any meaningful verification. These cases show that factual correctness, procedural diligence, and ethical responsibility cannot function as interchangeable properties. This article develops a multidimensional account of responsibility for AI-assisted scientific writing. We argue that scientific reliability depends on the factual accuracy, evidential support, and appropriate qualification of published claims, whereas ethical responsibility also depends on the quality of the author’s epistemic conduct, including the competence, proportionality, and independence of the verification process. Disclosure of AI use and nominal human review therefore do not establish responsible authorship on their own. We distinguish between authorship responsibility, epistemic responsibility, and corrective responsibility to explain why an honest error need not constitute misconduct while still creating an obligation to correct the scholarly record. Responsible AI-assisted authorship does not require infallibility, but it requires proportionate diligence, epistemic humility, and continued accountability for published content.