Kiss AI for a more efficient and effective peer review in science

It is expected that in 2025, around three million items will be indexed in Scopus and the Web of Science. If each one undergoes the peer examination of two experts, and 2 million additional articles undergo an examination by peers, but are rejected at around 10 million exams per peers which will be carried out this year – an amazing number which should increase as the biomedical company, and the number of reviews with peers will increase.

According to an editorial of the journal Intensive care medicineIn the coming years, artificial intelligence (AI) should be part of the future of peer exam.

The peer review in biomedical journals has been essentially unchanged for many decades. Although the compensation of peers examiners would likely help to receive opinions in a timely time, it is probably not possible on a large scale. In addition, peer exam has well -known limitations. “”

Howard Bauchner, MD, pediatric professor at the Chobanian & Avedisian Medicine School of the University of Boston

“We believe that the peer exam should include a form of initial revision of the AI, helping the publishers in the decisions on which the articles send for an external review by peers,” adds Bauchner, who is also the former editor of the Journal of the American Medical Association.

Bauchner describes the limits of peer exam and defines the different types: double blind, simple and open review. He describes one of the greatest trials ever conducted by comparing the double blind review to a review in a single stroke. “When the examiners were aware of the identity of the authors (in a single blind), they gave a more favorable note from countries with higher English skills and higher income. Since.

Bauchner also underlines that several independent groups which already offer an AI review of articles, largely as a service for the authors before the submission of articles, have already experienced good results. He quotes a particular study, where the authors found that the comments of GPT-4 see again as more useful than the comments of certain examination peers.

In addition, he considers that the AI ​​will be good to assess whether an article follows the appropriate report guidelines, which is often noted by the authors as requested by the journals, but without proof that the Peers examiners really verify the adhesion to these directives. In addition, Bauchner believes that AI can be able to detect fraudulent research more effectively than examination peers.

“While he continues to improve,” he said. “It is time to adopt a different approach, an approach that is probably more effective and more effective by AI.”