Scientists from the University of Kansas have developed an AI detection tool with 99% accuracy, making it possible to identify scientific texts generated by AI, particularly ChatGPT. The detection tool was specially designed using a smaller dataset combined with human insight to recognize the difference between AI and human writing. It focuses on scientific writings found in peer-reviewed journals, making it more accurate than general-purpose detectors.
In a recent study published in Cell Reports Physical Science, Heather Desaire, a University of Kansas chemist who applies machine learning to biomedical studies, demonstrated the tool’s high efficacy. She hopes it can help defend scientific integrity by detecting when AI-generated text is presented as work from a human mind.
The reality is, with some guidance and effort, a high school student could do what we did. There are huge opportunities for people to get involved in AI, even if they don’t have a computer science degree, said Desaire. I would like to see people interested in AI realize that the barriers to developing real and useful products like ours aren’t that high. With a little knowledge and some creativity, a lot of people can contribute to this field.
Desaire said AI detection tools are urgently needed to prevent AI text generators like ChatGPT from polluting the literature with believable-sounding falsehoods, which could make their way into publications. To get accuracy, the team had to restrict themselves to a particular group of humans who write in a specific way. Existing AI detectors are useful for their intended purpose, but they are not as accurate as a tool built for a specific and narrow purpose, according to Desaire.
Desaire made her team’s AI-detecting code fully accessible to researchers who may be interested in building off it. ChatGPT is really such a radical advance, and it has been adopted so quickly by so many people; this seems like an inflection point in our reliance on AI, she added.