Scientific publishers are looking for an intelligent algorithm that recognizes the ChatGPT method

If you need an introduction to a scientific article, you can write it with ChatGPT’s chatbot and text generator. ChatGPT can ask its own research question based on a set of data… Very scientific article To write.

How can scientific journal publishers know which articles were written by humans and which were written by computers? This is a question that has been prevalent since the launch of ChatGPT last year.

Get it early this year nature in Sciences It states in its guidelines that ChatGPT may not be listed as a co-author by the authors. nature It was allowed from the beginning That researchers have used ChatGPT as a writing aid has always been noted. the magazine Sciences It had previously banned any use of ChatGPT, but this week it blocked it as well In the guidelines It states that ChatGPT, as long as it is referenced, can be used for any purpose.

Do it secretly

Jilli Zuidema, assistant professor of artificial intelligence at the University of Amsterdam, understands these rules: “In this way, the journals hope to prevent researchers from doing this secretly, and it can be better controlled.”

Zuidema has the idea that ChatGPT is already widely used in science to improve articles or write short pieces of text. “I haven’t done it myself yet, but it’s understandable that researchers do, because there’s pressure to publish, and you often spend a significant portion of your time writing down your findings.”

It becomes an issue when using ChatGPT not only for article optimization, but also to create entire articles from start to finish.

Crazy expressions

Despite the peer review process in which other researchers vet the article before it is published, sometimes articles written explicitly by ChatGPT come along. For example, there I found several articles Where suddenly in the text you find strange expressions such as “response renewal” or “As an AI language model, I…” clearly indicate that the text was generated by artificial intelligence.

To prevent articles written by ChatGPT from escaping attention, publishers are looking for an intelligent algorithm that recognizes ChatGPT style. nature Last week he wrote about a new model Which was developed by chemists and can identify with greater accuracy than current detectors whether ChatGPT has written a chemical article. This is still a fairly simple model, and Zuidema expects it to be less accurate in practice. “People will always try to find ways to fool detectors, for example by telling ChatGPT to use a certain typing style.”

According to Zuidema, in the long run it is better to require ChatGPT to build a watermark into texts that detectors can recognize, for example in the form of a pattern of synonyms in the text.