JMIR Publications

ABSTRACT

Generative artificial intelligence (AI) is rapidly changing the conditions under which scientific manuscripts are produced, revised, and disseminated. Current debates in scholarly publishing have appropriately emphasized authorship, disclosure, transparency, and the unreliability of fabricated or unverifiable content. However, a more subtle risk is emerging in health sciences publishing: AI can increase the expressive fluency of manuscripts while weakening the density of human, clinical, and epistemic judgment behind scholarly claims. This article names this phenomenon expression-rich but judgment-thin scholarship and situates it within the broader problem of research waste. In this framing, AI-induced research waste is not limited to false citations, fabricated text, or undisclosed tool use. It also includes manuscripts that are linguistically polished, formally compliant, and citation-rich, yet contribute little to clinical understanding, practice reasoning, or trustworthy knowledge construction. Health sciences publishing is especially vulnerable because its claims often imply consequences for patients, clinicians, institutions, and policy. A manuscript can be syntactically coherent while remaining clinically thin: it may summarize evidence without weighing applicability, state implications without context, or produce recommendations without defensible professional judgment. We argue that the next stage of responsible AI governance should move beyond tool disclosure toward clinical judgment traceability. Editors, reviewers, and authors should be able to identify the human reasoning behind problem framing, evidence interpretation, contextual limits, and claims for practice. The aim is not to reject AI-assisted scholarship, but to preserve the human infrastructure that makes health sciences knowledge accountable, situated, and clinically meaningful.