GPT detectors are biased against non-native English writers
2023/04/06 by Weixin Liang, Liang, Weixin, Mert Yüksekgönül +9 · 13 voices · 47 citations
Computer Science · Medicine · Psychology · Social Sciences · #Artificial Intelligence in Healthcare and Education #Artificial intelligence #Biology #Communication #Computer science #Conversation #Detector #English language #Ethics and Social Impacts of AI #Generative grammar #Linguistics #Mathematics education #Natural language processing #Psychology #Robustness (evolution) #Telecommunications #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2304.02819
published in arXiv (Cornell University) (Cornell University)
openalex publication_date 2023/04/06 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/01
Abstract
The rapid adoption of generative language models has brought about substantial advancements in digital communication, while simultaneously raising concerns regarding the potential misuse of AI-generated content. Although numerous detection methods have been proposed to differentiate between AI and human-generated content, the fairness and robustness of these detectors remain underexplored. In this study, we evaluate the performance of several widely-used GPT detectors using writing samples from native and non-native English writers. Our findings reveal that these detectors consistently misclassify non-native English writing samples as AI-generated, whereas native writing samples are accurately identified. Furthermore, we demonstrate that simple prompting strategies can not only mitigate this bias but also effectively bypass GPT detectors, suggesting that GPT detectors may unintentionally penalize writers with constrained linguistic expressions. Our results call for a broader conversation about the ethical implications of deploying ChatGPT content detectors and caution against their use in evaluative or educational settings, particularly when they may inadvertently penalize or exclude non-native English speakers from the global discourse. The published version of this study can be accessed at: www.cell.com/patterns/fulltext/S2666-3899(23)00130-7
Cited by
- Humanly: A Configurable and Traceable Environment for Human-AI Collaborative Writing
- The Widespread Adoption of Large Language Model-Assisted Writing Across Society
- Trusting AI to detect AI? A systematic evaluation of the reliability and robustness of current AIGC detection tools for student academic work
- Everyone is unique: Towards Behaviorally Heterogeneous Negotiation Dialogue Systems for Debt Collection
- From Pilots to Practices: A Scoping Review of GenAI-Enabled Personalization in Computer Science Education
- BAID: A Benchmark for Bias Assessment of AI Detectors
- Semantic Reconstruction of Adversarial Plagiarism: A Context-Aware Framework for Detecting and Restoring "Tortured Phrases" in Scientific Literature
- Identifying Bias in Machine-generated Text Detection
- Large Language Models and Forensic Linguistics: Navigating Opportunities and Threats in the Age of Generative AI
- Writing in Symbiosis: Mapping Human Creative Agency in the AI Era
- Does Scientific Writing Converge to U.S. English? Evidence from Generative AI-Assisted Publications
- How to Use Generative AI in Educational Research
- AI-Generated Text Detection in Low-Resource Languages: A Case Study on Urdu
- On the Detectability of LLM-Generated Text: What Exactly Is LLM-Generated Text?
- Position: LLM Watermarking Should Align Stakeholders' Incentives for Practical Adoption
- Can generative AI figure out figurative language? The influence of idioms on essay scoring by ChatGPT, Gemini, and Deepseek
- AdaDetectGPT: Adaptive Detection of LLM-Generated Text with Statistical Guarantees
- When AI Does the Work, What Is Learning For? Post-Instrumental Learning and the Risk of Capacity Dissolution
- What exam scores can and cannot prove about unauthorized AI assistance: Evidence from a highly public classroom episode
- Diversity Boosts AI-Generated Text Detection
- Trace Is In Sentences: Unbiased Lightweight ChatGPT-Generated Text Detector
- Gen AI in Proof-based Math Courses: A Pilot Study
- AI-Generated Content in Cross-Domain Applications: Research Trends, Challenges and Propositions
- Can LLMs effectively provide game-theoretic-based scenarios for cybersecurity?
- DACTYL: Diverse Adversarial Corpus of Texts Yielded from Large Language Models
- T-Detect: Tail-Aware Statistical Normalization for Robust Detection of Adversarial Machine-Generated Text
- AI writing detectors are ineffective, unreliable and harmful
- The AI 3 Model: Future Directions for Artificial Intelligence, Assessment Innovation, and Academic Integrity
- When Detection Fails: The Power of Fine-Tuned Models to Generate Human-Like Social Media Text
- ARB: A Matched Authorship-Rewriting Benchmark Dataset for AI-Text Detector Evaluation
- GradEscape: A Gradient-Based Evader Against AI-Generated Text Detectors
- Embedding Generative AI as a digital capability into a year-long skills program.
- Explainability-Based Token Replacement on LLM-Generated Text
- Trans-EnV: A Framework for Evaluating the Linguistic Robustness of LLMs Against English Varieties
- GPT Editors, Not Authors: The Stylistic Footprint of LLMs in Academic Preprints
- Generative artificial intelligence (AI) in higher education: a comprehensive review of challenges, opportunities, and implications
- Artificial Intelligence Bias on English Language Learners in Automatic Scoring
- From Trade-off to Synergy: A Versatile Symbiotic Watermarking Framework for Large Language Models
- From revolution to evolution: What generative AI really means for language learning
- Position: Academic Conferences are Potentially Facing Denominator Gaming Caused by Fully Automated Scientific Agents
- A Penny for Your Prompts: Experiments Detecting and Mitigating LLM Usage by Survey Respondents
- The Last Fingerprint: How Markdown Training Shapes LLM Prose
- AI Detectors Fail Diverse Student Populations: A Mathematical Framing of Structural Detection Limits
- Different Time, Different Language: Revisiting the Bias Against Non-Native Speakers in GPT Detectors
- On-Device Watermarking: A Socio-Technical Imperative For Authenticity In The Age of Generative AI
- Out of Style: RAG's Fragility to Linguistic Variation
- Detecting AI-Generated Text: Factors Influencing Detectability with Current Methods
Discussions
- GPT detectors are biased against non-native English writers [hn, 338 points, 274 comments]
- Many (all?) current ChatGPT detectors have not been adequately assessed for issues of algorithmic bias and therefore should not be used to accuse students of misconduct in their written work. This af [bsky, 64 points, 3 comments]
- arxiv.org/abs/2304.028... [bsky, 8 points, 1 comments]
- My understanding is that the detectors are not very good at this point (they're unreliable), and they're also biased against non-native speakers. So...use caution if you're going to use them. Paper o [bsky, 6 points, 1 comments]
- GPT detectors are biased against non-native English writers (2023) [hn, 2 points, 0 comments]
- In our Writing + AI workshops, I've mentioned that some research (arxiv.org/abs/2304.02819) has shown that AI detectors are biased against non-native English speakers. Reading more research today that [bsky, 1 points, 0 comments]
- Worse still, another study from Stanford showed that ZeroGPT and similar tools often flagged non-native English writing as AI simply because it lacked variation or complexity. arxiv.org/abs/2304.028 [bsky, 1 points, 0 comments]
- Hi Rachel, arxiv.org/abs/2304.02819 I found this after looking at a paper on the effectiveness of AI detectors, it gave glowing reports of Turnitin and Copyleaks (there were caveats, need to really re [bsky, 1 points, 1 comments]
- Aqui o estudo original GPT detectors are biased against non-native English writers Autors: Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, James Zou arxiv.org/abs/2304.02819 [bsky, 0 points, 0 comments]
- GPT detectors are biased against non-native English writers arxiv.org/abs/2304.02819 [bsky, 0 points, 0 comments]
- 🚨 Studie warnt! KI-Erkennungssysteme stufen Texte von Menschen, deren Muttersprache nicht Englisch ist, fälschlicherweise als von einer KI geschrieben ein. KI-Texte können nicht zuverlässig erkannt [bsky, 0 points, 0 comments]
- 🤔 GPT detectors are biased against non-native English writers https://arxiv.org/abs/2304.02819 [bsky, 0 points, 0 comments]
- arxiv.org/abs/2304.02819 www.researchgate.net/publication/... teaching.unl.edu/ai-exchange/... [bsky, 0 points, 0 comments]
Related