Advancing Machine-Generated Text Detection from an Easy to Hard Supervision Perspective

2025-11-02

AI Summary

Existing machine-generated text (MGT) detection methods treat labels as deterministic “gold standards,” overlooking label imprecision arising from the ambiguous human–machine boundary, inter-annotator inconsistency, and detector overintelligence. To address this, we propose a **knowledge-distillation-free, easy-to-hard supervised augmentation framework**: a lightweight, long-text-prior easy supervisor provides reliable initial supervision signals; its outputs are formally modeled as a performance lower bound of the main detector via structured feedback, and we theoretically prove its asymptotic convergence to the latent ground-truth labels. The framework integrates coupled optimization between the supervisor and detector. Extensive experiments demonstrate substantial improvements in robustness and generalization across challenging scenarios—including cross-large-language-model, cross-domain, mixed-text, and paraphrase-attack settings—validating its effectiveness.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Existing machine-generated text (MGT) detection methods implicitly assume labels as the "golden standard". However, we reveal boundary ambiguity in MGT detection, implying that traditional training paradigms are inexact. Moreover, limitations of human cognition and the superintelligence of detectors make inexact learning widespread and inevitable. To this end, we propose an easy-to-hard enhancement framework to provide reliable supervision under such inexact conditions. Distinct from knowledge distillation, our framework employs an easy supervisor targeting relatively simple longer-text detection tasks (despite weaker capabilities), to enhance the more challenging target detector. Firstly, longer texts targeted by supervisors theoretically alleviate the impact of inexact labels, laying the foundation for reliable supervision. Secondly, by structurally incorporating the detector into the supervisor, we theoretically model the supervisor as a lower performance bound for the detector. Thus, optimizing the supervisor indirectly optimizes the detector, ultimately approximating the underlying "golden" labels. Extensive experiments across diverse practical scenarios, including cross-LLM, cross-domain, mixed text, and paraphrase attacks, demonstrate the framework's significant detection effectiveness. The code is available at: https://github.com/tmlr-group/Easy2Hard.

Problem

• Addresses boundary ambiguity in machine-generated text detection
• Proposes easy-to-hard framework for reliable inexact supervision
• Enhances detection across cross-LLM and adversarial scenarios

Innovation

• Easy-to-hard framework enhances detectors via reliable supervision
• Longer texts mitigate inexact label impacts for supervision
• Structural integration models supervisor as detector's lower bound