Probing Social Identity Bias in Chinese LLMs with Gendered Pronouns and Social Groups

2025-10-08

AI Summary

This study systematically investigates social identity bias in Chinese large language models (LLMs), focusing on systematic in-group (“us”) versus out-group (“them”) preferences. To address the lack of language-sensitive evaluation tools for Chinese, we construct the first Chinese-specific bias assessment framework, integrating controlled prompting experiments with analysis of authentic human–AI conversational corpora—grounded in 240 socially salient Chinese demographic groups and gendered pronouns. Empirical evaluation across 10 mainstream Chinese LLMs reveals pervasive and statistically significant in-group favoritism and out-group negativity, which intensifies in realistic interactive settings. Our work provides the first cross-linguistic empirical validation of social identity bias in generative AI, uncovering its dynamic amplification mechanism in Chinese-language models. It establishes a reproducible methodology and empirically grounded foundation for bias detection, measurement, and mitigation in multilingual LLMs.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Cognitive Modeling & Cognitive Systems: Social Cognition And Interaction
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Large language models (LLMs) are increasingly deployed in user-facing applications, raising concerns about their potential to reflect and amplify social biases. We investigate social identity framing in Chinese LLMs using Mandarin-specific prompts across ten representative Chinese LLMs, evaluating responses to ingroup ("We") and outgroup ("They") framings, and extending the setting to 240 social groups salient in the Chinese context. To complement controlled experiments, we further analyze Chinese-language conversations from a corpus of real interactions between users and chatbots. Across models, we observe systematic ingroup-positive and outgroup-negative tendencies, which are not confined to synthetic prompts but also appear in naturalistic dialogue, indicating that bias dynamics might strengthen in real interactions. Our study provides a language-aware evaluation framework for Chinese LLMs, demonstrating that social identity biases documented in English generalize cross-linguistically and intensify in user-facing contexts.

Problem

• Investigating social identity bias in Chinese LLMs using gendered pronouns
• Evaluating ingroup-positive and outgroup-negative biases across 240 social groups
• Developing language-aware framework to assess bias generalization in real interactions

Innovation

• Evaluated bias with gendered pronouns and social groups
• Used Mandarin-specific prompts across ten Chinese LLMs
• Analyzed real chatbot conversations for bias dynamics