The Emergence and Resolution of Bias Issues in Large Language Models
DOI:
https://doi.org/10.54097/6gmdyh60Keywords:
Bias of Large Language Models, Output Issue, Development of LLM.Abstract
The issue of bias in the output results of large language models has sparked extensive discussions and research in academia and industry. This article systematically analyzes the manifestations and formation mechanisms of bias in big language models, and points out that current governance methods still have limitations in data transparency and causal reasoning mechanisms, and explores future development directions. Research has found that bias is mainly divided into two aspects: "insufficient interpretability" and "narrow generalization ability". Their formation is related to data-driven bias inheritance, black box characteristics and optimization bias of model architecture, and bias reinforcement brought about by feedback loops. Although some current governance methods, such as adding interpretability tools, multimodal fusion, and data synthesis and reinforcement, have to some extent alleviated bias issues, effective governance has not yet been carried out from the root causes and processes of bias. Based on the limitations of the governance methods mentioned above, research proposes that in the future, efforts can be made to build causal frameworks, promote scenario based dynamic reasoning, establish trustworthy data systems, and establish interdisciplinary collaboration mechanisms to address both the symptoms and root causes of bias, providing theoretical support for the development of more fair, reliable, and responsible new language models.
Downloads
References
[1] Bai, Q., Kuang, Q., Xia, X., Yan, S., & Zhang, S. (2025). Prejudice mitigation: A study on the recognition efficacy and cognitive adaptation interventions of large language models. Journal of Education and Media Studies, 3, 17–27. https://www.cnki.net
[2] Dai, G., & Huang, B. (2025). A discussion on the cognitive characteristics and developmental transformation paths of large language models in translation. Chinese Translators Journal, 4, 122–129. https://www.cnki.net
[3] Huang, H., Li, S., Lan, T., Qiu, Y., Liu, Z., Yao, J., Zeng, L., Shan, Y., Shi, X., & Guo, Y. (2024). A survey on the safety of large language models: Classification, evaluation, attribution, mitigation and prospect. CAAI Transactions on Intelligent Systems, 20(1), 2–32. https://link.cnki.net/urlid/23.1538.TP.20241212.1608.002
[4] Li, Z., & Yao, F. (2025). Challenges and responses of AI-generated content: Exploration and reflection from the perspective of sustainable development of audiovisual content platforms. Radio & Television Research, 9, 118–124.
[5] Meng, L., Wang, X., & Han, C. (2025). Global AI development trends and future prospects. Contemporary China and World, 3, 25–32.
[6] Raisch, S., & Fomina, K. (2024). Combining human and artificial intelligence: Hybrid problem-solving in organizations. Academy of Management Review, 50(2). https://doi.org/10.5465/amr.2021.0421 DOI: https://doi.org/10.5465/amr.2021.0421
[7] Taani, O., & Alabidi, S. (2024). ChatGPT in education: Benefits and challenges of ChatGPT for mathematics and science teaching practices. International Journal of Mathematical Education in Science and Technology. Advance online publication. https://doi.org/10.1080/0020739X.2024.2357341 DOI: https://doi.org/10.1080/0020739X.2024.2357341
[8] Wang, Y., & Li, M. (2025). Artificial intelligence application and corporate sustainable development: A test based on innovation capability and absorptive capacity. Journal of Environmental Economics, 10(3), 47–72.
[9] Xu, Y. (2024). “The AI alignment problem” posed by DeepSeek: A philosophical challenge—From the perspective of the relationship between “pre-judgment” and “prejudice”. Journal of Philosophy and Technology, 2(7), 108–118. https://www.cnki.net
[10] Zhou, W., Zhu, X., Han, Q., Li, L., Chen, X., Wen, S., & Xiang, Y. (2025). [Article title not provided]. IEEE-CAA Journal of Automatica Sinica, 12(1), 1–26. https://doi.org/10.1109/JAS.2024.124983 DOI: https://doi.org/10.1109/JAS.2024.124983
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Education and Educational Research

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.









