How Does AI Perceives the World? Moral Differences and Value Alignment in Generative Large Models
DOI:
https://doi.org/10.63313/SSH.9094Keywords:
LLMs, Moral Differences, Value Alignment, Technology for GoodAbstract
The widespread adoption of generative large language models (LLMs) has sparked prominent ethical controversies, with inter-model moral disparities and their knock-on effects becoming a key bottleneck to the development of technology. Taking ERNIE Bot 4.0, iFlytek Spark V4.0 and ChatGPT-4 as research objects, this paper conducts interactive experiments and finds that these models vary notably across five core moral foundations (harm/care, fairness/reciprocity, in-group/loyalty, authority/respect, and purity/sanctity) and four dimensions: moral emotions and attitudes, moral knowledge and cognition, moral sensitivity, and moral decision-making. Such divergences stem from deep-seated issues including ambiguous ethical guidance, inherent moral biases, and flawed value cultivation. To address these problems, this paper concludes that LLMs need to achieve value alignment in three aspects: moral framework, value prioritization, and context-specific evaluation criteria. It can be realized through three major approaches—technological empowerment, multi-stakeholder collaboration, and dynamic iteration—so as to drive the sustainable development of LLMs.
References
[1] Lu Yu, Yu Jinglei,Chen Penghe, et al. Educational Applications and Prospects of Generative Artificial Intelligence: Taking ChatGPT as an Example[J]. Chinese Journal of Distance Education,2023,43(4): 24-31+51.
[2] Zhang Zhi'an, Li Xinying, He Hanfu. From Public Communication to Intelligent Communication: Paradigm Changes in Journalistic Practice and Practical Teaching[J]. Journalism University, 2024(5): 1-14+116.
[3] Wang Xiaoli, Li Weixin. Challenges of Generative Artificial Intelligence to Human Subjectivity and Countermeasures [J]. Journal of Jiangsu University (Social Science Edition), 2025, 27 (4): 91-101+124.
[4] He Yuhua, Li Xia. New Challenges and Countermeasures for Disinformation Governance of Generative Artificial Intelligence: From the Perspective of Agile Governance[J]. Governance Studies, 2024, 40(4): 142-156+160.
[5] Xiao Hongjun, Zhang Lili. Theoretical Deconstruction and Governance Innovation of Ethical Anomie of Large Models[J]. Research on Financial and Economic Issues, 2024(5): 15-32.
[6] Qi Chenhang. Risk Characteristics and Governance Paths of Disinformation of Generative Artificial Intelligence[J]. Information Studies: Theory & Application, 2024, 47(3): 112-120.
[7] Han Xuzhi. Logical Renewal and Path Optimization of Generative AI Governance: From the Perspective of Human-Machine Relationship[J]. Administrative Law Review, 2023(6): 30-42.
[8] Luo Guojie. Ethics[M]. People's Publishing House, 1989: 7.
[9] Li Ping. Fundamentals of Ethics[M]. Capital University of Economics and Business Press, 2004: 20-21.
[10] Tang Kailin. Ethics[M]. Higher Education Press, 2001: 38.
[11] Qu Peiheng. A Brief Discussion on the Definition of Morality[J]. Morality and Civilization, 1987(3): 27-29+44.
[12] Silvia T,Marta G,Natalia J.How do Personality,Team Processes and Task Characteristics Relate to Job Satisfaction and Software Quality? [J].Information and Software.2009,51(3),627-639.
[13] Felicitas K,Kees V.O,Martin P.Is there an Ethics of Algorithms? [J].Ethics Inf Technol,2011,(13):251-260.
[14] Yotam L,Shlomo M.Professional Ethics of Software Engineers:an Ethical Framework[J].Science&Engineering Ethics,2016,(22):417-434.
[15] Graham J,Nosek B A,Haidt J,Iyer R,Koleva S,DittoPH. Mapping the Moral Domain.Journal of Personality and Social Psychology,2011, 101(2):366-385.
[16] Forbes M, Hwang J D, Shwartz V,Sap M,Choi Y.Social Chemistry 101: Learning to Reason about Social and Moral Norms[EB/OL].In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing,EMNLP’20,2020,653-670. extension://ngbkcglbmlglgldjfcnhaijeecaccgfi/https://aclanthology.org/2020.emnlp-main.pdf
[17] Li Hong, Sun Shu. On Moral Values and Their Evaluation Methods[J]. Psychological Exploration, 1992(4): 28-31.
[18] Zeng Xiong. Current Situation, Problems and Normative Approaches of Value Alignment of Large-scale AI Models[J]. E-Government, 2025(2): 34-44.
[19] Zheng Huangjie. From ChatGPT to DeepSeek: Normative Approaches to Agent Value Alignment[J/OL]. Studies in Science of Science, [2025-06-17].
[20] Luo Yichen, Zhou Xinheng. Types and Standardized Paths of Employment Gender Discrimination in Algorithmic Automated Decision-making[J]. Human Resources Development of China, 2025, 42(1): 92-103.
[21] Council of the European Union.Proposal for a Regulation of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act) and Amending Certain Union Legislative Acts-General Approach.Brussels,Belgium. 2022.11.25[EB/OL]. https://data.consilium.europa.eu/doc/document/ST-14954-2022-INIT/en/pdf
[22] Hu Zhengrong, Yan Jiaqi. A Comparative Study on Value Alignment of Generative Artificial Intelligence——An Experiment Based on Comments Generated from the Top Ten International News Events (2012—2023)[J]. Journal of Journalism University, 2024(3):1-17+117.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 by author(s) and Erytis Publishing Limited.
This work is licensed under a Creative Commons Attribution 4.0 International License.
