Detection of ChatGPT Fake Science with the xFakeSci Learning Algorithm

ChatGPT and generative AI tools are becoming the new reality. This work is motivated by the premise that ``ChatGPT content may exhibit a distinctive behavior that can be separated from scientific articles''. In this study, we demonstrate how we tested this premise in two phases and prove its validity. Subsequently, we introduce xFakeSci, a novel learning algorithm, that is capable of distinguishing ChatGPT-generated articles from publications produced by scientists. The algorithm is trained using network models driven from multiple types of data sources, such as ChatGPT-generated documents achieved by means of prompt-engineering, and PubMed articles. To mitigate over-fitting issues, we incorporate a calibration step that is built upon data-driven heuristics, including ratios. We evaluate the algorithm across multiple datasets covering publication periods and diseases (cancer, depression, and Alzheimer's). Further, we show how the algorithm is benchmarked against the state-of-the-art (SOTA) algorithms. While the xFakeSci algorithm achieve F1 score ranging from 80% - 94%, SOTA algorithms score F1 values between 38% - 52%. We attribute the noticeable difference to the introduction of calibration and a proximity distance heuristic, which we underscore this promising performance. Indeed, the prediction of fake science generated by ChatGPT presents a considerable challenge. Nonetheless, the introduction of xFakeSci algorithm is a significant step on the way to combating fake science.