Peer review is overwhelmed—can it survive in the AI era?
By Tom Beckwith ·
The problem with science—and indeed, with any complex system built on shared trust—is that its growth rate has become functionally ungovernable.
The sheer volume of work is already a geopolitical crisis
The problem with science—and indeed, with any complex system built on shared trust—is that its growth rate has become functionally ungovernable. We are witnessing an exponential surge in intellectual output that existing infrastructure simply cannot absorb. Consider the numbers: the number of papers indexed in Scopus and Web of Science is increasing at 5.6 percent per year, while researchers globally are devoting a collective 15,000 years of work to peer review every single year. This isn't merely an academic bottleneck; it’s a systemic failure of vetting capacity.
The current model—the painstaking process that requires a constellation of voluntary effort—is fraying under the strain. As arstechnica.com reported, reviewers are struggling immensely; one immunogeneticist noted, "I struggle like anything to get peer reviewers." The sheer velocity is staggering: the number of papers submitted to top AI conferences has increased two- to 10-fold since 2019. When Oliver Habryka stated that “by the time your thing passes peer review, there’s a very substantial chance it’s already out of date,” he wasn't making an observation; he was delivering a warning about temporal irrelevance.
The calculator analogy is dangerously insufficient
The reaction from established publishers—the cautious pronouncements emanating from Elsevier and others—is to treat AI as merely an "assistive tool." Alina Helsloot stated that while AI can improve efficiency, it cannot replace expert judgment. This sentiment echoes across the board: Professor Jim Jansen noted that a fully AI-driven review is unacceptable; Dr Francesco Mangano concluded that AI cannot replace experience or passion.
This cautious insistence on human oversight, however, is dangerously insufficient. It treats this crisis like a diplomatic disagreement over protocol when it is, in fact, an infrastructure collapse. To suggest that better matching of reviewers to papers—or adding safeguards and ethical oversight—will solve the problem ignores the fundamental mechanism at play. This pattern of exponential productive capacity outpacing vetting mechanisms has historical precedent: the Industrial Revolution. That shift was not managed by simply issuing new guidelines for factory workers; it required a complete, structural overhaul of how goods were produced, distributed, and regulated. The sheer scale demands a paradigm shift, not just better process management.
Trust is now measured in computational cycles
The core issue facing academia—and any knowledge-based society—is that the mechanism designed to preserve credibility has been overwhelmed by its own success. As laneblog.stanford.edu detailed, systematic misconduct and organized "paper mills" have led to a surge of fraudulent publications that are outpacing legitimate outputs. This is not simply about poor writing; it is about fraud at scale.
The current system demands voluntary dedication—a commodity far scarcer than the data itself. We must acknowledge that the existing structure cannot handle this velocity. The only viable path forward requires treating peer review less like a scholarly courtesy and more like a critical utility service, one capable of scaling its capacity through mandatory, standardized processes. The time for incremental fixes is over; we need institutional mechanisms to manage the sheer weight of intellectual output, or the entire edifice of trust will crumble into irrelevance.