On November 12, the world’s first system-level vertical domain ethical large language model, WENDAO, was officially launched at SEU. This marks a significant shift in
China’s approach to AI ethics governance, moving from “passive compliance” to “active co-governance.” The model is designed to serve as an ethical thinking partner and
decision support system for users, helping research institutions, technology companies, ethics review bodies, policymakers, and the general public navigate complex
technological and societal issues by gaining insight into diverse values. Currently, the model is available for free to users.
Currently, general-purpose large models arewidelycriticizedfor “hallucinations” and a lack of transparency in ethical reasoning. WENDAO was developed to address these
concerns. Dr. Miao Yujie, a PhD student from the School of Information Science and Engineering, SEU, explained that “hallucinations” refer to instances where AI models
provide responses to user queries without factual grounding. “These answers lack sources and could very well be fabricated.”
As a specialized ethical large model, WENDAO has created a functional matrix that covers the entire process of ethical governance, focusing on five core areas: For ethical
risk assessment and auditing, the system identifies potential risks in algorithms, such as fairness, privacy, and security issues, allowing for the proactive prevention of ethical
risks; for ethical dilemmas and decision-making simulation, the model assists users in multi-perspective reasoning and decision optimization by constructing realistic moral
scenarios; for ethical alignment assistance in design, the model operationalizes values into executable, quantifiable engineering standards and algorithmic norms, offering a
full-chain solution for AI products from concept design to compliance implementation; for dynamic knowledge base and case studies, the model gathers global ethical
standards and case studies, supporting intelligent search and situational learning to enhance the efficiency of acquiring professional knowledge; for ethical frontier
exploration and paradigm innovation, through deep literature mining and theoretical analysis, WENDAO assists scholars in understanding the development of ethics and
contributes to the evolution of academic research paradigms.
The theoretical framework of WENDAO integrates both Chinese and international ethical frameworks alongside China’s current laws and regulations, forming an analysis
system rooted in the Chinese context and characteristics. It enables the systematic identification and assessment of various risks in AI applications and offers actionable
governance pathways within the Chinese context. When providing suggestions, the model clearly marks the reasoning basis and the sources of ethical principles.It also
highlights uncertainties in its conclusions, reflecting the “human-in-the-loop” collaboration concept and the explainable AI governance approach. The model can be used as
an intelligent assistant for government policy-making, regulatory ethics review, corporate compliance development, and public awareness.
The model was developed under the leadership of Prof. Wang Jue, Director of the AI Ethics Laboratory of the School of Humanities, SEU in collaboration with the SEU
National Mobile Communications Research Laboratory, National Key Laboratory of Millimeter Waves, the Ministry of Education’s Frontier Science Center for Mobile
Information Communication and Security, and the Jiangsu Provincial Ethics Development Think Tank. This interdisciplinary collaboration fully reflects SEU’s distinctive feature
of “interdisciplinary integration across the humanities, sciences, and engineering,” highlighting the university’s innovation in the development of new liberal arts and providing
a reference for global AI ethics governance with Chinese characteristics.
In the future,WENDAO will continue to optimize its human-computer interaction mechanisms, focusing on systematization and phased development. Feedback channels will
be opened to both experts and users, creating a positive feedback loop for model iteration and expanding its application across research, industry, and education.
Source: SEU News Network, Modern Express
Translated by: Melody Zhang
Reviewed by: Gao Min
Edited by: Li Xinchang
