Machine-Human Boundary: A Systematic Survey of Machine-Generated Text Detection

The rapid development of Large Language Models (LLMs) has blurred the line between machine-generated text (MGT) and human-written text (HWT), raising ethical concerns. This paper reviews existing research on MGT detection. It introduces three types of detectors, manual, metric-based, and model-based, and analyzes their characteristics and representative cases. Three categories of benchmarks for evaluating detectors, namely dichotomy, demarcation point, and Mixtext, are also discussed. Current detectors have flaws, such as poor robustness and limited language support. Future research should prioritize the development of more robust and adaptable detectors and benchmarks, with an emphasis on leveraging textual correlation features to improve detection accuracy across diverse scenarios. This study summarizes the current state of MGT detection, providing a basis for future research in this field.