AI-generated deluge strains academic peer review as submission volumes surge

Journal editors and peer reviewers are facing a significant surge in submissions of high-quality, AI-generated research papers, creating a "deluge" that strains the peer-review system. Generative AI tools are enabling the mass production of papers using public datasets, resulting in content that is difficult to distinguish from human-authored work due to improved coherence and reduced hallucinations. Several journals, including Security Dialogue and Accountability in Research, report substantial increases in submissions, with some seeing rises of up to 100 per cent and 60 per cent respectively, alongside difficulties in recruiting reviewers.

Experts warn that the current academic incentive structure, which rewards publication volume, exacerbates the problem, potentially undermining the integrity and efficiency of scientific research. The influx threatens to overwhelm academic publishing, grant assessment, and research evaluation processes, with some journals reporting significant increases in submissions.

Peter Degen, a postdoctoral researcher at the University of Zurich Center for Reproducible Science and Research Synthesis, highlighted the scale of the issue after discovering that a 2017 paper he had worked on was being cited hundreds of times by AI-generated studies. These papers utilised public datasets from the Institute for Health Metrics and Evaluation at the University of Washington to produce endless variations of statistical predictions. Degen noted that while earlier AI papers contained obvious errors, the new wave of content is far more difficult to filter out, placing immense pressure on an already limited peer-review capacity.

Matt Spick, a lecturer at the University of Surrey and associate editor at Scientific Reports, observed a similar trend involving the US National Health and Nutrition Examination Survey. He reported an explosion of papers following a similar formula, often finding misleading correlations that lacked scientific merit. Spick tested an AI tool called Prism, developed by OpenAI, which produced a coherent paper on eggplant ripening in under 30 minutes. He warned that the technology is reaching a point where it is nearly impossible to distinguish between human and AI authorship, risking a system that publishes data irrespective of whether it constitutes new knowledge.

The strain on the system is evident in editorial workflows. Marit Moe-Pryce, managing editor of Security Dialogue, stated that submissions had increased by 100 per cent over the previous year. She described a "gray mass" of articles that are coherent and well-structured, making it difficult to identify fraudulent content. Moe-Pryce noted that the workload has become unmanageable, with fewer reviewers responding to requests due to fatigue and the sheer volume of material.

David Resnik, an associate editor at Accountability in Research, reported a 60 per cent surge in submissions, including papers that ironically mined data from Retraction Watch to discuss academic fraud. Resnik highlighted the difficulty in finding reviewers, sometimes sending out 20 requests to secure two responses. He also noted a survey by Frontiers indicating that more than half of researchers have used AI assistance in peer review, raising concerns about the reliability of the evaluation process itself.

The problem is compounded by an academic culture that prioritises publication metrics. An analysis in Quantitative Science Studies attributed the exponential growth in papers to commercial and professional incentives rather than rapid scientific progress. Vincent Larivière, editor-in-chief of Quantitative Science Studies, argued that the conflation of productivity with publication counts has distorted science, causing research to gravitate toward small, tractable problems. He called for a reform of what matters in science, suggesting that the current system incentivises the mass production of low-value papers.

As AI capabilities advance, the scientific community faces a critical juncture. The STM Integrity Hub, an initiative launched to combat paper mills, is now engaged in an "arms race" with AI. Joris van Rossum, the project’s program director, suggested that publishers may need to shift towards requiring researchers to demonstrate the authenticity of their work, rather than relying solely on detection methods. Meanwhile, experts like Reese Richardson from Northwestern University argue that without changing how prestige and resources are awarded, the incentive to mass-produce papers will persist, further straining the scientific enterprise.