Peer Review System Strains Under Rising Paper Volume and AI

The scientific peer review system is buckling under unprecedented strain as the volume of research papers grows exponentially and AI tools make it easier than ever to generate submissions.

Researchers now collectively spend an estimated 15,000 years annually on peer review work—unpaid labor that would cost $1.5 billion if compensated in the United States alone. Papers indexed in major databases like Scopus and Web of Science are increasing at 5.6 percent per year, while the pool of volunteer reviewers struggles to keep pace.

Steven Mack, an editor for Human Immunology, now emails roughly 30 researchers to secure a single reviewer—up from just 5 to 10 emails five years ago for three reviewers. The shortage means editors increasingly assign reviews to people who lack qualifications or time to properly evaluate submissions.

Health economist Jason Semprini from Des Moines University experienced this firsthand when his HPV vaccine policy study was rejected based on a single reviewer who misunderstood the research question. "It's frustrating to see work held up because a reviewer didn't read it carefully," Semprini said.

Why it matters

Peer review has served as the legitimacy mechanism for scientific research for only about 50 years, but its current dysfunction threatens the credibility of published science. As AI makes paper generation easier and paper mills produce fabricated research, the system's inability to maintain quality control could undermine public trust in scientific findings—particularly concerning as interdisciplinary research requires more specialized knowledge to evaluate properly.

AI researchers pioneer alternatives

Computer scientists, facing two- to 10-fold increases in conference submissions since 2019, are experimenting with new models. Many AI researchers now publish findings on blogs rather than waiting for formal peer review.

The AI Alignment Forum uses community voting and commentary instead of traditional review. Oliver Habryka, CEO of Lightcone Infrastructure which runs the forum, argues this approach offers speed and transparency advantages. "By the time your thing passes peer review, there's a very substantial chance it's already out of date," he said.

Helen Qu, an AI researcher at the Flatiron Institute, now publishes exclusively on her blog after growing frustrated with conference submissions. However, philosopher David Thorstad from Vanderbilt University warns that abandoning peer review could sacrifice rigor and isolate work from mainstream academic discourse.

Shortcuts and reforms

A survey by Frontiers journals found over half of peer reviewers now incorporate AI into their work, though not always productively. Linguist Marijn van Putten from Leiden University spent two days tracking down a medieval Arabic source suggested by what turned out to be an AI-generated review—the reference didn't exist.

The NeurIPS AI conference is piloting custom AI tools designed to assist rather than replace reviewer judgment. Other reforms include preprints that allow research visibility during review, and efforts to redistribute review workload more equitably beyond male researchers in the United States and a handful of other countries.

Despite being "the bedrock of science" in Semprini's words, peer review at this moment "is not standing strong." The system's evolution—or potential collapse—will shape how scientific knowledge is validated and communicated in coming years.

These details were first reported by Ars Technica.