Specialists from the U.S. have developed an algorithm that can distinguish scientific articles and other publications written by real people from the texts of ChatGPT and other neural networks with 99% probability. The first results of testing the algorithm were published in an article published in the journal Cell Reports Physical Science.
"We tried hard to create an accessible method so that with little guidance, even high school students could build an AI detector for different types of writing," says first author Heather Desaire, a professor at the University of Kansas. "There is a need to address AI writing, and people don't need a computer science degree to contribute to this field."
Professor Desaire and her colleagues produced a series of papers published in the journal Science and used it as training material to create a scientific version of ChatGPT that can generate such texts.
An interesting fact drew their attention: people prefer to use more complex paragraph structures than the neural network does, and also use adverbs and certain other words very often, namely "but", "however", "although".
On the other hand, the artificial intelligence system likes to use the words "researchers" and "others", which real scientists rarely do. In addition, in the case of humans, it is more likely that the sentence length and structure are variable.
According to the researchers, the algorithm could also be adapted to look for traces of creativity in ChatGPT and other artificial intelligence systems in other forms of written text. In the future, this will allow to quickly identify the cases when schoolchildren and students try to present the creativity of neural networks as their own.