The article, titled “AI can help scientists publish less”, reflects on a tension that is becoming increasingly urgent. AI tools are rapidly lowering the cost of producing scientific papers, from devising research plans and writing code to drafting manuscripts and polishing prose. But the work of reading, refereeing, judging and absorbing new results remains slow, human and scarce.

If this asymmetry is left unchecked, Bertone argues, AI could intensify an already overloaded publication culture, producing more articles than scientific communities can seriously read, review and absorb.

Rather than treating AI only as a threat to the integrity of the scientific record, the Comment asks whether it can also become part of the remedy. Used well, AI could lift routine labour from scientists, strengthen the early stages of scientific evaluation, and make contributions such as code, datasets, benchmarks and reproducibility packages more visible. Under the right incentives, these changes could help science move away from publication volume as a proxy for progress.

Quality over quantity

Bertone introduces the idea of negative epistemic value: a paper does not have to be wrong to subtract from the collective production of knowledge. If it adds too little understanding compared with the time and attention it demands from editors, referees, readers and colleagues, it can slow down the collective production of knowledge.

The Comment also connects the AI debate to ongoing reforms in research evaluation, including moves away from journal impact factors and toward recognition of broader scientific contributions. Bertone argues that such reforms will require not only new criteria, but also new practical tools and infrastructure. AI, he suggests, may help make these reforms operational by giving scientists back some of the time needed for reflection, judgment and genuinely difficult work.

The scientific community should aim for more than defending itself against a flood of AI-assisted papers. Used wisely, AI offers an opportunity to correct distortions in the publication system, reduce the pressure to publish continuously, and improve both science and the lives of scientists.

Publication

AI can help scientists publish less. Gianfranco Bertone. Nature Astronomy (2026)