arXiv changes rules – AI studies among pre-prints

The impression is often deceptive. This is also the case with arXiv. It is an open platform for the exchange of research results from, among other fields, physics, mathematics, computer science, but also economics and more. The studies have not necessarily undergone a peer review process, as is customary for scientific investigations that appear in renowned journals and specialist publications, for example. This means that anyone could submit studies to arXiv until now. They are often so-called preprints. And this is precisely what is becoming the platform's downfall—it increasingly features AI-generated studies.

The area of computer science or informatics appears to be particularly problematic. A blog post states that arXiv has updated its moderation practices for review articles and position papers. Before they can be submitted, they must have been accepted by a journal or a conference, meaning they have successfully undergone a peer review process. The corresponding evidence must be submitted with the paper in the future, otherwise they cannot be published.

“This change is being implemented due to the unmanageable influx of review articles and position papers to arXiv CS.” And AI is largely to blame for this. Apparently, these are not only completely fabricated investigations but also studies that summarize the results and content of other studies. The blog post states submitted papers are sometimes even just bibliographies.

A lot of attention for AI studies

arXiv also writes that this is not a change in the terms of use. In fact, only studies that were not peer-reviewed were published in exceptional cases because moderators considered them exciting enough and the senders were trustworthy. However, since the platform's volunteer moderators do not have the capacity and expertise to review submissions professionally, arXiv is taking this step to rely on the expertise of others. Peer reviewers have concerns regarding data protection, ethics, security, and protection in newer technologies.

While submissions are increasing in all categories, there are particularly many in the field of computer science. This includes papers dealing with AI and large language models. Of course, attention has been particularly high here recently. For example, investigations by the media are also said to have been picked up, which subsequently failed peer reviews, meaning they did not meet scientific criteria.

arXiv is not the only one suffering. AI-generated studies are appearing everywhere in science. Entire so-called specialist magazines are appearing, which, of course, do not meet any scientific standards at all. A Swedish research team has already warned how harmful the influence of AI on science and thus society can be. Google Scholar and other databases are also affected, as are the actual peer reviewers of real specialist publications, who have to check more papers.

(emw)