Medical journals are facing a growing influx of AI-generated research that is technically sound but scientifically trivial, according to a new report in MedPage Today. Experts describe the trend as a shift from obvious AI slop to a flood of "kind of meaningless" science. Elisabeth Bik, a science integrity consultant, said the field has reached a point where "we cannot distinguish fake from real."
Systematic reviews and studies built on public datasets are especially easy to generate with tools like ChatGPT, said Ivan Oransky of Retraction Watch. The number of submissions using public data sources such as CDC's WONDER and NHANES has skyrocketed since generative AI became widely available. In response, the publisher Frontiers began requiring new experimental validation for manuscripts based solely on public data, and its integrity team has rejected more than 12,000 such submissions in the past year.
Publishers are fighting back with screening tools, AI detection software, and human oversight. Springer Nature uses a combination of approaches, and NEJM Group is piloting AI detection tools. But detection remains an arms race. Bik noted that current tools are better at catching fraud in pre-AI research, and that her own methods are "based on the fraud of yesterday." Meanwhile, AI has also made it easier for authors to respond to fraud allegations with polished, AI-generated rebuttals.
Experts say the root cause is the publish-or-perish culture in academic medicine. As Oransky put it, if universities don't want the literature overwhelmed with garbage, they should stop incentivizing applicants to publish large amounts of low-quality work.