Detecting AI-Generated Text: A Comparative Study of Machine Learning Algorithms

Li-jing Arthur Chang

As ChatGPT and other large-language-model (LLM) tools have made the generation of text via AI much easier than before, there is an increasing need to determine if humans indeed write the text we are reading. The study used six machine learning and deep learning algorithms to detect AI-created text. Using a balanced sample of AI-generated and human-written text, the results showed that deep learning algorithms outperformed their machine learning counterparts. A hybrid deep learning algorithm achieved the top accuracy rate of 0.974 (or 97.4%). Post hoc analysis showed that with a small fraction, such as 10%, of the full sample used in the present study, the hybrid algorithm achieved 0.928.

Full Text