When ChatGPT and other generative artificial intelligence can produce scientific articles that look real — especially to someone outside that field of research — what’s the perfect strategy to determine which of them are fake?
Ahmed Abdeen Hamed, a visiting research fellow at Binghamton University, State University of Latest York, has created a machine-learning algorithm he calls xFakeSci that may detect as much as 94% of bogus papers — nearly twice as successfully as more common data-mining techniques.
“My primary research is biomedical informatics, but because I work with medical publications, clinical trials, online resources and mining social media, I’m all the time concerned concerning the authenticity of the knowledge someone is propagating,” said Hamed, who is an element of George J. Klir Professor of Systems Science Luis M. Rocha’s Complex Adaptive Systems and Computational Intelligence Lab. “Biomedical articles specifically were hit badly throughout the global pandemic because some people were publicizing false research.”
In a brand new paper published within the journal Scientific Reports, Hamed and collaborator Xindong Wu, a professor at Hefei University of Technology in China, created 50 fake articles for every of three popular medical topics — Alzheimer’s, cancer and depression — and compared them to the identical variety of real articles on the identical topics.
Hamed said when he asked ChatGPT for the AI-generated papers, “I attempted to make use of very same keywords that I used to extract the literature from the [National Institutes of Health’s] PubMed database, so we might have a typical basis of comparison. My intuition was that there should be a pattern exhibited within the fake world versus the actual world, but I had no idea what this pattern was.”
After some experimentation, he programmed xFakeSci to research two major features about how the papers were written. One is the numbers of bigrams, that are two words that steadily appear together equivalent to “climate change,” “clinical trials” or “biomedical literature.” The second is how those bigrams are linked to other words and ideas within the text.
“The primary striking thing was that the variety of bigrams were only a few within the fake world, but in the true world, the bigrams were rather more wealthy,” Hamed said. “Also, within the fake world, despite the proven fact that were only a few bigrams, they were so connected to all the pieces else.”
Hamed and Wu theorize that the writing styles are different because human researchers do not have the identical goals as AIs prompted to provide a chunk on a given topic.
“Because ChatGPT remains to be limited in its knowledge, it tries to persuade you by utilizing essentially the most significant words,” Hamed said. “It shouldn’t be the job of a scientist to make a convincing argument to you. An actual research paper reports truthfully about what happened during an experiment and the strategy used. ChatGPT is about depth on a single point, while real science is about breadth.”
To further develop xFakeSci, Hamed plans to expand the range of topics to see if the telltale word patterns hold for other research areas, going beyond medicine to incorporate engineering, other scientific topics and the humanities. He also foresees AIs becoming increasingly sophisticated, so determining what’s and is not real will get increasingly difficult.
“We’re all the time going to be playing catchup if we do not design something comprehensive,” he said. “We’ve got a number of work ahead of us to search for a general pattern or universal algorithm that doesn’t depend upon which version of generative AI is used.”
Because though their algorithm catches 94% of AI-generated papers, he added, meaning six out of 100 fakes are still getting through: “We have to be humble about what we have completed. We have done something very necessary by raising awareness.”
