Researchers have created a tool called ‘xFakeSci’ that can tell the difference between real research articles and those made by AI chatbots like ChatGPT. In tests with 300 real and fake papers, xFakeSci correctly identified up to 94% of the fake ones. This success rate is almost twice as high as traditional data-mining methods.
Developed by researchers from the State University of New York, USA, and Hefei University of Technology, China, xFakeSci uses a new learning algorithm to spot AI-generated articles. The tool was tested using two datasets: one with nearly 4,000 real scientific articles from PubMed, a large database of research papers, and another with 300 fake articles created by ChatGPT.
To compare, the researchers used similar keywords for both sets of articles. They found that fake papers often lacked the variety of common word pairs (bigrams) like “climate change” or “clinical trials” that real papers have. The fake papers also connected these pairs differently in their content.
The tool showed accuracy scores between 80% and 94%, much better than common data-mining methods, which only reached between 38% and 52%. This highlights how xFakeSci could be a powerful tool to spot fake research, as AI writing lacks the depth and honesty of real scientific work.
Source: indiatimes
