AI-Generated Studies Overwhelm Medical Journals, Experts Warn
Experts say artificial intelligence now enables researchers to produce studies with no obvious errors but little scientific value, making it harder for journals to detect AI involvement. Systematic reviews and analyses of public datasets are particularly easy to generate, leading to a surge in submissions. One publisher, Frontiers, now requires additional validation for manuscripts based solely on public data or computational methods.
The surge in AI-assisted submissions is concentrated in studies built on publicly available datasets, such as CDC's WONDER and NHANES. Frontiers, an open-access publisher, rejected more than 12,000 such submissions in a single year, including over 3,000 NHANES-based manuscripts. The publisher now requires original experimental validation for any paper relying solely on computational analysis of public data.
Detection tools remain imperfect. Programs like Pangram, Imagetwin, and Proofig can flag AI-written text or duplicated images, but experts note these systems are calibrated to catch older forms of fraud. As AI improves, distinguishing machine-generated research from genuine work becomes increasingly difficult for both automated screeners and human reviewers, leaving journals to rely on evolving policies and manual oversight.
The influx of low-value AI-generated studies could strain peer review systems, delaying publication of meaningful research and eroding trust in medical literature. Clinicians who depend on journals for evidence may find it harder to identify reliable findings, potentially influencing patient care decisions. Researchers competing for publication slots could face increased pressure from high-volume AI submissions. If detection tools lag behind AI capabilities, the integrity of the scientific record may be compromised, affecting systematic reviews and clinical guidelines that rely on published evidence.