On the Possibilities of AI-Generated Text Detection

TL;DR

This paper investigates the theoretical and empirical limits of detecting AI-generated text, demonstrating that detection is generally feasible with sufficient sample size unless machine and human texts are indistinguishable, and providing bounds for detection complexity.

Contribution

It offers the first theoretical bounds on sample complexity for AI text detection and validates detection methods across multiple datasets and models.

Findings

Detection is possible with adequate sample size unless texts are indistinguishable.

Empirical results confirm the effectiveness of state-of-the-art detectors across various datasets.

Theoretical bounds align with observed empirical detection sequence lengths.

Abstract

Our work addresses the critical issue of distinguishing text generated by Large Language Models (LLMs) from human-produced text, a task essential for numerous applications. Despite ongoing debate about the feasibility of such differentiation, we present evidence supporting its consistent achievability, except when human and machine text distributions are indistinguishable across their entire support. Drawing from information theory, we argue that as machine-generated text approximates human-like quality, the sample size needed for detection increases. We establish precise sample complexity bounds for detecting AI-generated text, laying groundwork for future research aimed at developing advanced, multi-sample detectors. Our empirical evaluations across multiple datasets (Xsum, Squad, IMDb, and Kaggle FakeNews) confirm the viability of enhanced detection methods. We test various state-of-the-art text generators, including GPT-2, GPT-3.5-Turbo, Llama, Llama-2-13B-Chat-HF, and Llama-2-70B-Chat-HF, against detectors, including oBERTa-Large/Base-Detector, GPTZero. Our findings align with OpenAI's empirical data related to sequence length, marking the first theoretical substantiation for these observations.