Shulou(Shulou.com)11/24 Report--

CTOnews.com, June 8 (Xinhua)-- earlier this year, Som Biswas, a radiologist at the Tennessee Health Science Center, attracted attention when he published an article in the journal Radiology assisted by ChatGPT, an artificial intelligence chat robot, entitled "ChatGPT and the Future of Medical Writing." He said he used and edited the text generated by ChatGPT to raise awareness of the usefulness of the technology. He also revealed that he then used ChatGPT to publish 16 journal articles in four months. Some journal editors also report that they have received a large number of articles written in ChatGPT.

In response, Heather Desaire, a chemistry professor at the University of Kansas, and her team developed a new AI detection tool that can efficiently and accurately distinguish whether scientific texts are generated by humans or ChatGPT. Their findings are published in the journal Cell report physical Science.

Professor Desaire said she and her team first analyzed 64 "opinions" articles in the journal Science, which are summary articles that review and evaluate current research. They then analyzed 128 articles generated by ChatGPT on the same research topic. By comparing the two, they identified 20 features that can help determine the identity of the author of a scientific text.

They found that there were significant differences between human scientists and ChatGPT in paragraph complexity, sentence length, punctuation and vocabulary use. For example, human scientists are more likely to use parentheses, dashes, question marks, semicolons and uppercase letters, while ChatGPT is less common. Human scientists also prefer to use "ambiguous language", such as "however", "although", "but" and so on. In addition, human scientists write with both short and long sentences, while ChatGPT is more average.

Based on these 20 characteristics, they used a ready-made machine learning algorithm XGBoost to train their AI detection tool. They tested the performance of their AI detection tool on 180 articles and found that they were very good at judging whether a scientific article was written by a human or ChatGPT. " This method has more than 99 per cent accuracy, "Professor Desaire said, adding that it is much better than existing tools, which are trained on a wider range of text types rather than scientific texts.

Professor Desaire says the AI testing tool can help journal editors deal with a large number of articles written using ChatGPT, giving them priority to which articles are worthy of review. She added that the tool can be adapted to different areas, such as to detect student plagiarism, as long as it is trained in the right language. "you can adapt it to any field you want, as long as you figure out which features are useful."

CTOnews.com notes, however, that not everyone agrees that this AI detection tool is very useful. Dr Vitomir Kovanovi Qing, of the Centre for change and complexity Learning (C3L) at the University of South Australia, said the comparison made by Professor Desaire and her team was unrealistic because they only compared 100% AI-generated text to 100% human-generated text, without taking into account the collaboration between humans and AI. When scientists use ChatGPT, there is often a degree of human-computer cooperation, such as editing text generated by AI, he said. This is also necessary because ChatGPT can sometimes make mistakes and even generate fictional references. But because the researchers compared only two extreme situations, their success rate was improved.

Dr. Lingqiao Liu of the Institute of Machine Learning at the University of Adelaide also believes that in the real world, the accuracy of this AI testing tool may be lower, leading to more misclassifications. Dr Liu, an expert on developing algorithms to detect images generated by AI, said: "in terms of methodology, this is fine, but there are risks in using it."

On the other hand, Dr Liu points out that it is also possible to instruct ChatGPT to write in a certain way, allowing 100 per cent of the text written by AI to pass the test. In fact, some commentators even talk about an "arms race", referring to the competition between those who try to make machines more like humans and those who try to expose those who use the technology for malicious purposes.

Dr Kovanovi Qing called it a "meaningless competition" because of the strong momentum and potential positive impact of the technology. AI testing "misses the point, and I think we'd better focus on how to use AI effectively," he said. He also opposes the use of anti-plagiarism software to assess whether college students use AI writing, saying it puts unnecessary pressure on students.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.