Artificial intelligence (AI), and especially its so-called "generative" aspect found in many chatbots (GPT, Gemini, or Claude, for example), continues to disrupt the scientific world. Now, one of the pillars of research quality control is being shaken: peer review . This system relies on the evaluation of an article submitted to a journal by several experts in the field. Their reports are used by the publisher to decide whether or not to publish the manuscript and to request revisions from the authors.

“But this system is suffering ,” notes Thomas Lemberger, head of Open Science at EMBO Press, a publisher specializing in life sciences. The problems are well-known: slowness (several months, even more than a year), difficulty in finding experts, volunteers who are overworked, and, unfortunately, the process is not infallible. In 2024, Nature and Science each retracted three articles, according to the database of the specialized media outlet Retraction Watch .

AI is exacerbating the situation with automatically generated reports, which are no longer produced by peers. The company Pangram (which markets, among other things, software for detecting AI-generated text) estimates that 21% of the 70,000 reports submitted for a major machine learning conference in 2024 (the International Conference on Learning Representations) were "entirely written by AI ." And half of them contained "AI interventions ." Incidentally, a few hundred of the 19,000 papers were themselves entirely automatically written.

‍