The peer review system in science, essential for validating research, faces an unprecedented crisis of overload, intensified by the rise of artificial intelligence. With a growing volume of articles and few volunteer reviewers, the effectiveness of this pillar of academia is in question, generating concerns about the quality of scientific information disseminated. On August 10, 2026, researchers expressed the urgent need to redesign existing processes to ensure the credibility of knowledge.
Overload in the academic evaluation system
Researchers in different areas face increasing difficulties with peer review, the process by which experts evaluate the validity and relevance of studies before their publication. This crucial step, generally anonymous and voluntary, has been compromised by the exponential increase in the number of scientific articles. It is estimated that the number of publications indexed in databases such as Scopus and Web of Science grows at a rate of 5.6% per year.
This expansion places a heavy burden on the scientific community, which collectively dedicates 15,000 years of volunteer work to the review annually. If this effort were remunerated, the costs in the United States alone would reach 1.5 billion dollars. Difficulty finding qualified and available reviewers is a growing problem.
Haseeb Irfanullah, a member of the editorial board at Learned Publishing, reported his constant difficulty in finding reviewers. Steven Mack, editor of Human Immunology, had to contact about thirty researchers to get just one reviewer for an article, in contrast to five to ten contacts needed five years ago. Scarcity leads editors to assign reviews to people who are less qualified or have limited time. Sebastian Lourido, a microbiologist at the Whitehead Institute, described the system as “extremely time-consuming and painful”, impacting the careers of researchers who depend on publications for jobs and funding.
A more recent foundation than you might think
Although it is a central pillar in current research, peer review only became a universal feature of academic publishing about 50 years ago. Before the 19th century, scientific societies already reviewed articles, but with varied internal workings and a distinct research philosophy, as explained by science historian Aileen Fyfe, from the University of St Andrews.
In the past, science was done by “independent gentlemen”, not university professors, and publication was not linked to career advancement. Working for a scientific journal was a pleasurable activity, which motivated the donation of time as a reviewer.
Learn more: Crisis in reviewing scientific articles can be solved with new artificial intelligence agents
The universalization of peer review occurred largely by chance, as a strategy to solve an image problem and legitimize science. In the 1970s, with the United States economy in crisis, the National Science Foundation (NSF) found itself under public pressure to justify its spending. The use of external evaluators to analyze funding requests, an unusual practice at the time, gave legitimacy to decisions, preventing politicians from assuming direct control over the allocation of resources. “It was really a marketing strategy by scientists to convince other interested parties to embrace peer review as a fundamental pillar of science,” said Melinda Baldwin, a science historian at the University of Maryland.
The United Kingdom experienced a similar situation in the late 1980s and early 1990s, when scientific controversies, such as those related to HIV or cold fusion, generated great public interest. Scientific organizations responded by asserting that legitimate science was recognizable by passing peer review.
The influence of artificial intelligence and new alternatives
The cadence of review requests is driven not only by the increase in studies, but also by the proliferation of journals and “special issues” for which researchers create content. Many journals now publish more manuscripts because they are exclusively online, eliminating paper limitations. Furthermore, interdisciplinary research requires more time and knowledge for review, and artificial intelligence (AI) contributes to the overhead, making it easier to write and submit articles. The existence of “article mills” that publish fake work also exacerbates the problem. Irfanullah noted, “Sometimes I think we need to start talking about the decline of the publishing sector. It’s extremely harmful!”
Researchers in the field of AI feel this explosion of content intensely. The number of papers submitted to major AI conferences has increased two- to tenfold since 2019, resulting in reviews from reviewers who “simply don’t understand the field well enough to evaluate,” according to Haewon Jeong, a computer scientist at the University of California, Santa Barbara.
More on this story: Japanese philosophy shapes Dwayne Johnson’s gym project in Hawaii
In response, the field of AI has experimented with informal alternatives, such as blogs. Because it is a new and rapidly evolving field, many researchers do not depend on formal publications for career advancement. Helen Qu, an AI researcher at the Flatiron Institute, chose to publish her research exclusively on her personal blog, hating the routine of submissions.
The AI Alignment Forum, for example, is a collaborative blog where members vote for or against content and post comments, replacing traditional peer review. Oliver Habryka, CEO of Lightcone Infrastructure, the company responsible for the forum, recognizes that there is no guarantee of reflection in the votes, but highlights that the posts receive much more reactions than the few peer reviews of a journal.
On the same topic: Teenager kills mother and brother after using AI to create fantasies about the crime
- Advantages of this system include:
- Speed:Crucial in a rapidly evolving field; Traditional articles may become outdated before publication.
- Focus on attention:The voting system directs attention to quality research and counterarguments, avoiding reliance on social media algorithms.
- Accessibility:The conversational style of many blogs makes complex concepts easier to understand, as pointed out by Jeong.
However, there are disadvantages. Vanderbilt University philosopher David Thorstad warned that accessibility can compromise rigor, and charismatic voices can stand out regardless of the quality of the work. Furthermore, moving away from peer review can isolate work, making it difficult to cite in journals that require the process. Despite this, Habryka notes that currently, “less than 50% of the things you actually want to cite in a [machine learning] paper end up being properly peer-reviewed.”
Pathways to Peer Review Reform
Faced with the crisis, many researchers are looking for shortcuts and reforms. A consolidated tactic is preprints, formal publications posted on sites like arxiv.org so that research becomes visible while awaiting the traditional process. Preprints are now seen as a legitimate basis for applications for grants and jobs in some fields.
- Other ideas focus on changes to the review process:
- Workload redistribution:Assign reviews more equitably rather than disproportionately to researchers from certain countries.
- Remuneration for reviewers:Some advocate payment, with evidence showing that $100 to $300 per review speeds up the process without compromising quality, creating a sense of deadline.
- Separation of the review and submission process:A community of researchers reviews articles before formal submission, allowing authors to contact multiple journals without repeated rounds of peer review.
Linguist Marijn van Putten from Leiden University had a frustrating experience when receiving an apparently AI-generated review, with “crazy references” and typical features of automated text, resulting in a “huge waste of time”. Chris ChoGlueck, a philosopher at New Mexico Tech, warned about breaching confidentiality when sending unpublished manuscripts to chatbots that can retain data. The NeurIPS AI conference will test a custom AI tool to assist reviewers without replacing their judgment.
Cameron Neylon, an advocate of academic communication, questions whether all scientific articles really need peer review, suggesting that studies reach a minimum level of interest before being submitted. It is difficult to imagine radical change as the system is rooted in career progression and public trust. However, if publishing continues to grow at its current rate, the system will fail. Neylon suggests that research areas encourage quality over quantity.
More on this story: NASA’s Webb Telescope discovers new giant planet with innovative method
Peer review is resilient, but its form is changing. It is likely to remain in some form, but perhaps less rigorous and more fragmented. Mack states that the average number of reviewers in immunology has decreased to two, and some economics journals now pay reviewers, while preprints are routine in many areas. The system, therefore, seeks a new balance for its survival in the era of artificial intelligence.