Why China’s universities stopped fighting AI

(Edited and refined by Candice Chan, with the assistance of AI translation.)

Zhao Junming, a postgraduate student at Chongqing Technology and Business University, recently completed a master’s thesis on sports brand communication. In writing his 60,000-word dissertation, he used Google’s Gemini and Alibaba’s Tongyi Qianwen as AI assistants to develop the overall structure, while relying on Tongyi Qianwen to handle formatting.

Speaking to Lianhe Zaobao (LHZB), Zhao compared AI to a scholar with an encyclopaedic command of the literature.

“AI is like someone who has read a huge amount of literature,” he said. “Where a person might draw a few related ideas from one example, an AI model can generate many more. Its efficiency gains are even more apparent in tasks such as formatting.”

Zhao admitted that he barely used AI for his undergraduate dissertation two years earlier, but now, most of his fellow postgraduate students use AI to write their dissertations.

Zhao’s experience is far from unique. According to the results of a survey released in February last year by MyCOS, a higher education data and consulting organisation in China, nearly 60% of university faculty and students in China use generative AI either daily or multiple times a week. Among students, nearly 30% primarily use it for writing papers or assignments.

A report on university students’ mindset on using AI, jointly released last September by China Youth Daily, China Youth Campus Media and social networking platform Soul App, also found that 99.2% of university students surveyed had already used AI; about 65.9% said they would “instinctively turn to AI first” whenever they had a problem; and nearly half admitted they “cannot live without AI”.

The report found that AI tools had overtaken traditional search engines and social media as university students’ preferred source of information.

From outright bans to incorporating AI

As AI becomes an increasingly important tool for learning and research among university students, Chinese universities are also adjusting their approach to managing its use.

When ChatGPT sparked a global wave of interest in generative AI in 2023, Chinese universities initially adopted a cautious stance. During the 2024 graduation season, many institutions introduced regulations governing the use of AI.

Fuzhou University, Hubei University and Tianjin University of Science and Technology introduced AI checks for graduation theses, making the results part of the assessment. The Communication University of China required students who used AI to fully disclose the model, version, usage and content generation process.

Fudan University also specified six prohibited uses of AI in academic papers, including using AI tools to design research proposals, generate or alter original research data, and for language polishing and translation.

However, the explosive rise of generative AI has forced Chinese universities to rethink their stance. Over the past year, many have shifted from trying to curb its use to actively exploring how it can enhance teaching and research without compromising academic integrity.

In February last year, the Shenzhen University’s National Engineering Laboratory for Big Data System Computing Technology partnered with Tencent Cloud to launch a general AI course based on DeepSeek. The course covers fundamental knowledge, technical principles and practical applications, including AI-generated content.

Wuhan University, Beihang University and Shaanxi Normal University have also introduced related courses, incorporating AI literacy into their undergraduate talent development programmes.

Chen Xunwei, a senior engineer at Wuhan University who teaches courses including the programming language Python, told LHZB that AI education extends far beyond coding. It encompasses everything from foundational courses that build AI literacy to technical and application-oriented training, attracting students across the humanities, sciences and engineering.

“AI has become a productive force, and educators have a responsibility to expose students to the most advanced productive force available,” he said. “If you don’t use it, you will be overtaken by those who do. And that competitive advantage grows exponentially.”

Economist Ye Xiaoyang and Shen Meng, a director of Chanson & Co, both said the shift in universities’ attitude towards AI is inevitable.

Ye said universities initially adopted a cautious approach to protect academic integrity. But once a technology becomes ubiquitous, he argued, relying on outright bans is neither practical nor an effective way to govern its use.

Shen shared a similar view. “AI will play an increasingly important role in academic research, regardless of the wishes of any institution or individual,” he said. The more contentious question, he added, is not whether AI should be used, but whether it should remain an assistant or take the lead in the research process.

Experiments in AI-generated papers

While promoting the integration of AI into teaching, some universities have gone a step further, exploring AI in more pioneering ways.

According to information published on its official website, East China Normal University launched a large-scale social experiment entitled “AI as the First Author” in September last year. Using a quasi-field experimental approach, the initiative called for AI to be listed as the first author of academic papers, with humans as collaborators and gatekeepers.

In March this year, Nanjing University joined Tsinghua University, Fudan University, Zhejiang University and several other institutions in launching a competition titled “I Write History With AI”.

Chen Xunwei said the backdrop to these competitions was the dramatic leap in the capabilities of generative AI over the past six months.

When the universities published the findings of their respective experiments in March and June this year, they arrived at a similar conclusion: if an academic paper merely presents ideas and logical arguments, it is increasingly difficult to demonstrate genuine scholarly value. As a result, academic papers are gradually losing their traditional role as a measure of academic achievement.

Ye Xiaoyang expressed support for the experiments.

“This has reframed the debate. The question is no longer whether AI should be used, but whether institutions should acknowledge its role and develop better frameworks to govern it.’”

Wang Shengyu, a research assistant at the Centre for China Analysis under the Asia Society Policy Institute, expressed a similar view, noting that these experiments are consistent with the explorations and practices currently under way at the world’s leading research universities.

Guidelines issued in recent years by Harvard University, Cornell University and Northwestern University on the use of generative AI in research generally allow researchers to use tools such as ChatGPT in academic work, but attach conditions including the disclosure of AI involvement, human verification of AI-generated content, and researchers assuming ultimate responsibility for AI-generated material.

However, East China Normal University’s experiment went a step further by publishing several papers listing AI as the first author in the Journal of East China Normal University, which has considerable academic influence. The aim was to extend the discussion from principles governing AI use to the norms surrounding academic authorship, but the move sparked considerable controversy, and the relevant papers are no longer accessible on the journal’s website.

Critics argue that the essence of academic authorship lies in having someone who can be held accountable. AI may serve as an auxiliary tool, but it cannot replace human beings as the main academic agent or bearer of liability.

In a commentary, the Beijing Youth Daily argued that the experiment marks the beginning of an era of human-AI co-governance in academia and represents an attempt to break new ground. Its greatest value, the commentary said, lies in clarifying the boundary of responsibilities between humans and AI in the research process.

Chen believes this will be a common challenge that the academic community must face in the future.

“AI models are evolving too quickly. Most regulations inevitably lag behind technological developments to some extent, but such exploration is necessary.”

Shen Meng, however, said that universities exist primarily to cultivate students’ professional capabilities. If AI is allowed to undertake leading roles such as first authorship, faculty members and students will not receive sufficient training and their abilities will not improve.

“That runs counter to the mission of universities.”

‘AI checking AI’ distorts industry chain

While the role of AI in academic research remains under discussion, universities have widely introduced AI-generated content detection as part of the thesis review process, which leads to a new set of problems.

Chinese universities use AI detection systems to determine whether students’ writing overlaps linguistically or substantively with AI-generated text. However, AI detection platforms apply different standards and frequently misidentify original work as AI-generated. As a result, students are forced to spend additional time repeatedly revising their dissertations to reduce their AI detection scores, trapping them in a vicious cycle of “AI checking AI, and then using AI to evade AI”.

According to reports by the National Business Daily, the Economic Observer and other Chinese media outlets, some students have turned to online services that promise to “reduce AI detection rates” in order to meet the thresholds set by their universities. As a result, a distorted industry chain has emerged on platforms such as Taobao, Pinduoduo and RedNote (Xiaohongshu).

Ye Xiaoyang said it was understandable why universities had adopted AI to detect AI, but the outcome had become absurd.

“Students should be spending their time improving their research questions, evidence, arguments and expression. Instead, they are forced to learn how to satisfy the detector.”

He believes universities should distinguish between different forms of AI use.

“Using AI to correct grammar is not the same thing as using AI to fabricate references. A good system should not ask only one crude question, but rather ask more precise ones.”

From a technical perspective, Chen Xunwei said: “Using one probabilistic system to detect another probabilistic system simply doesn’t work.”

Wang Shengyu, meanwhile, argued that the current situation reflected bureaucratic inertia in university administration. If it is to be changed in the future, he said, the Ministry of Education will probably need to conduct systematic research and introduce policies to drive reform.

This article was first published in Lianhe Zaobao as “从严防死守到探索边界 中国大学AI攻防战转向？”.

Related: [Big read] China’s young workers pay the price of AI before reaping the gains | How Chinese talent powers US AI dominance