AI has not yet taken over scientific research

Artificial intelligence can search through information at lightning speed and improve texts. Yet four years after the launch of ChatGPT, it is becoming clear that AI is not yet going to replace scientists, writes Hans Van Eyghen.

It has now been four years since ChatGPT went online. Since then, a great deal has been written about the dangers and possibilities of using AI. AI has also repeatedly been the subject of academic gatherings, such as the recent opening of the academic year at Tilburg University.

Many questions surrounding the use of AI are ethical or societal in nature, such as ‘Is the environmental pollution caused by AI acceptable?’ and ‘Will AI make jobs redundant?’.

Other questions are more intellectual or epistemic in nature and thus touch on the very heart of what researchers do. From the outset, attention was drawn to the intellectual capabilities of large language models such as ChatGPT. These systems enable you to search through enormous amounts of information very quickly. They can also summarise or rehash information in understandable language.

It was soon claimed that AI systems could therefore take over scientific research. There were reports of academic papers generated by AI being reviewed by scientific journals. AI was expected to fundamentally change the scientific landscape.

Now that four years have passed, the overall impact of AI on research seems to have been relatively modest. Although AI-generated papers have been accepted by academic journals, their number has remained limited. At least one paper was accepted by AI Magazine. Furthermore, at least one paper was accepted at a machine learning conference.

Furthermore, two scientists attempted to submit AI-generated papers to predatory journals and were successful in 92% of cases. There is also evidence that passages in papers are often generated by AI systems. According to one study, 57% of the papers examined show signs of AI use.

The idea that content produced by AI systems could match human scientific research therefore still seems like science fiction. This does not mean that AI has not changed scientific research. Whereas research used to generally begin with mulling over an idea internally, followed by a search on Google Scholar, many researchers now start a conversation with an AI bot and ask targeted questions to find information.

Having academic texts proofread by an AI system or asking for suggestions on phrasing is also not uncommon. When a larger part of the research is outsourced to AI, things often go wrong. It is often quickly apparent when a text has been generated entirely or largely by AI. The text reads less smoothly, contains errors and fails to reach the standard of human work. The fact that it is mainly predatory journals that tend to accept AI papers is telling in this regard.

Using AI systems for research therefore offers opportunities, but does not replace the role of the scientist. It is better not to outsource creative thinking and reasoning to AI (at least not yet). For other research tasks, such as looking things up and polishing texts, the use of AI is not problematic.

_{Advertenties.}