The battle with the ``thesis factory'' where AI mass-produces fake research is intensifying

With the advent of AI that can generate natural sentences at the same level as humans, such as ChatGPT, and AI that can generate highly accurate photos and illustrations simply by entering text, fake science There have been reports of cases where papers are written by AI. Companies that mass-produce and sell fake scientific papers are called `` paper mills '', and it is a problem that greatly shakes the reliability of the review process of academic journals and the performance evaluation of researchers.

AI intensifies fight against 'paper mills' that churn out fake research
https://doi.org/10.1038/d41586-023-01780-w

``The ability of paper mills to generate plausible data has really skyrocketed with AI,'' said Jennifer Byrne, a molecular biologist and publication integrity researcher at the University of Sydney. Also, Jana Christopher, an image data integrity analyst at a German academic publisher, said, ``I have seen fake microscope images generated by AI. Proving it remains a challenge,' he said.

In May 2023, the UNITED2ACT Summit was held in the United Kingdom, hosted by the non-profit Committee on Publication Ethics (COPE) and the International Association for Scientific, Technical and Medical Publishing (STM). At this summit, researchers, research analysts, representatives of academic publishers and editors gathered and various discussions were held. .

One of the measures discussed at the Summit is the mandatory submission of raw experimental data. In addition to adding it to the paper, it is a method of receiving the raw data by an academic publisher, confirming that the data is genuine, and adding an electronic watermark. However, Christopher said it is important for publishers to unify requirements for raw data submission, as requirements for raw data submission vary widely between publishers. Sabina Aram, director of publication ethics and integrity at a British scholarly publisher, agreed with Christopher, but noted that introducing such uniform requirements would take time.

STM is consolidating resources on similar tools while continuing to develop software to detect papers created by Paper Mill. The increase in paper mill papers has increased the demand for technologies that detect fake papers at the time of submission and identify whether already published papers are fake.

Mr. Aram's publisher also uses tools to detect fake papers. According to Mr. Aram, about half of the `` ethical cases '' that are considered fraudulent in the paper are paper mill papers created by AI. The number of ethical cases has increased more than tenfold from 2019 to 2022, and the number of ethical cases that occurred at the time of article creation in 2023 has already exceeded the number of occurrences in 2022.

Deborah Kahn, a research consultant at COPE, said, ``There is no doubt that AI will be used to assist researchers in writing papers, but there is still a difference between legitimate papers drawn with AI and completely fabricated papers. It will be necessary to distinguish between the