AI Tool 'xFakeSci' Outperforms in Identifying ChatGPT-Generated Research Papers

Researchers have developed 'xFakeSci', an AI tool capable of distinguishing original research articles from those created by AI chatbots like ChatGPT. The tool detected up to 94% of fake papers, surpassing traditional data-mining techniques. It works by analyzing the frequency and connections of bigrams in scientific texts.

Researchers have introduced 'xFakeSci', an AI tool designed to differentiate genuine research articles from those written by AI chatbots, including ChatGPT.

In a study involving 300 authentic and fabricated scientific papers, 'xFakeSci' successfully identified up to 94% of the fake articles, nearly doubling the success rate of common data-mining methods, according to researchers from the State University of New York and Hefei University of Technology.

The AI-based tool was developed using two distinct datasets: one comprising nearly 4,000 scientific articles from PubMed and another containing 300 fake articles generated by ChatGPT. The AI was trained on the real dataset and then tested for its efficacy on the fake articles, achieving impressive accuracy scores between 80% and 94%. The tool primarily analyzed the number of bigrams and their connections within the text, revealing that fake articles featured fewer and more broadly connected bigrams compared to genuine research.

(With inputs from agencies.)