Detecting AI-Generated Text: Factors Influencing Detectability with Current Methods

• DOI: 10.1613/jair.1.16665
• Publication Date: 2025-04-09T02:45:42Z
• Authors: Kathleen C. Fraser; Hillary Dawkins; Svetlana Kiritchenko

Large language models (LLMs) have advanced to a point that even humans difficulty discerning whether text was generated by another human, or computer. However, knowing produced human artificial intelligence (AI) is important determining its trustworthiness, and has applications in many domains including detecting fraud academic dishonesty, as well combating the spread of misinformation political propaganda. The task AI-generated (AIGT) detection therefore both very challenging, highly critical. In this survey, we summarize stateof-the art approaches AIGT detection, watermarking, statistical stylistic analysis, machine learning classification. We also provide information about existing datasets for task. Synthesizing research findings, aim insight into salient factors combine determine how “detectable” under different scenarios, make practical recommendations future work towards significant technical societal challenge.