JMIR Publications

ABSTRACT

Current debate over artificial intelligence (AI) in scientific research and publishing centers on plagiarism, hallucination, disclosure, and authorship policy. This paper argues that these concerns address symptoms of a deeper structural shift in how scientific claims are produced. Drawing on social epistemology, distributed and extended cognition, automation research, responsibility-gap theory, and metascience, the paper develops a graded, multidimensional account of the relationship between researchers and the AI systems they increasingly rely on, rather than treating human and system-generated contributions to scientific cognition as a simple binary. It distinguishes integrated epistemic extension, in which a researcher retains comprehension, evaluative capacity, control, and answerability over delegated work, from derivative epistemic substitution, in which these capacities fall below the threshold required for genuine answerability, even though formal authorship persists. This produces what the paper terms responsibility asymmetry: a mismatch between the answerability formally assigned to an author and the epistemic competence that author retains over the content of their claims. The paper specifies the conditions under which this asymmetry arises, distinguishes it from mere reliance on AI tools, and situates the institutional incentives that encourage it within existing metascientific accounts of perverse publication pressures. It evaluates existing governance instruments, including International Committee of Medical Journal Editors (ICMJE) authorship criteria, Committee on Publication Ethics (COPE) guidance, and contributorship frameworks such as CRediT and MeRIT, against this account, and proposes a risk-graduated governance model, comprising an epistemic participation profile, tiered cognitive provenance reporting, an enforcement-specified standard of reasoning auditability, and an equity-sensitive treatment of AI use, designed to be administratively implementable rather than aspirational.