"It's hard for me to imagine how my university can live without AI." In the past three years, a silent but dramatic change has been taking place on university campuses - AI is being used by college students to write homework. And when we heard such voices from college students more than once, we thought: It’s time to talk to college teachers and college students about “writing homework with AI”.

We want to know how students get lazy? How does a teacher “defeat magic with magic”? In the end, how can students make good use of AI tools?

This "new cat-and-mouse game" in university classrooms: students use AI to "ghostwrite" and teachers try their best to "catch monsters" is now deeply present in every corner of higher education, whether it is paper writing in literature, history and philosophy, code debugging in the computer department, or even news writing in the journalism department, AI is everywhere.

Behind the teacher-student game, a deep question is emerging - when higher education is fully penetrated by AI, will teaching and learning be redefined?

The all-around teaching assistant, or the culprit who cheats on homework?

"The 7,000-word final report was written in three or four days."

This is the real-life experience of a science and engineering interviewee. With the assistance of AI, tasks that once required weeks of hard work can now be produced as quickly as products on the production line. For college students who are in exam week and final season, AI has become the ultimate "savior". Some students jokingly call their AI memberships "paying for technology."

In the eyes of front-line teachers, this change was both sudden and profound.

In the journalism department of a comprehensive university, teacher Yu Wenzhou (pseudonym), who teaches the "News Writing Class", only belatedly discovered in January 2025 that as early as 2024, first-year students were using AI to generate homework. In the spring semester in March, he saw with his own eyes in a news writing class that students directly used AI to generate news reports - not even an interview was needed, and a neat press release was produced. Of course, most news facts are made up by AI.

In a CCTV news report, college teachers warned students. Source: Internet

Cheng Yunlai (pseudonym), a teacher at the Foreign Languages ​​School of a certain 985 university, observed that students used AI answers to design PPTs, summarize literature, and polish language. In 2025, the graduation thesis and graduate theses of senior students "have greatly improved their language quality, and there are almost no grammatical errors in the entire text." But she also pointed out that articles written by AI often have a "translation tone" and are vague in content. "There are no detailed examples, and the speech draft is very empty. It talks about general principles and cannot reflect people's real experiences."

Teacher Peng Jian from the Communication University of China summarized the typical characteristics of AI homework: Although the writing is neat, there are deviations in conceptual understanding and confusing logical segmentation. What is even more fatal is that AI will "tell lies seriously." He came across a paper on investment and financing in the film industry. The article described conclusively that a certain company completed a financing with a specific structure in a certain month of a certain year, but "people who understand the industry will know at a glance that it was compiled by AI."

Teacher Wang Lingzi from the School of Journalism and Communication of Zhejiang University of Communication can even tell which big model the students are using - the content generated by Wenxinyiyan looks like liberal arts students, while Tongyi Qianwen is more like science students.

A survey of more than 3,000 undergraduates in 13 universities across the country showed that not only do college students generally use generative artificial intelligence to assist learning, but more than half of them use it frequently, with the total proportion of "frequently used" and "always used" exceeding 50%.

Teacher-student AI attack and defense battle

Faced with the menacing AI, teachers have to pick up the "demon mirror" and try to find traces of AI in students' homework. But this game is far more complicated than imagined.

During the interview, we found that some college teachers rely on various AI detection tools-GPTZero, Copyleaks, Turnitin, etc. However, the accuracy of these tools varies and sometimes gives very different results, making technical testing only a reference rather than a basis for conclusive conclusions.

Therefore, "human eye recognition" becomes the real line of defense. Experienced professors can smell the smell of AI in assignments with “smooth language, impersonality, and watertight logic.” Teacher Peng Jian relies on this kind of academic experience: "Although the articles generated by AI are logically rigorous, they often forcefully juxtapose things that are not in the same dimension - for example, in the four-point analysis, three points are divided by function and one point is divided by effect, which are not in the same dimension at all."

Teacher Cheng Yunlai mainly relies on personal experience and occasionally uses the iWrite platform for testing in writing classes. "If AI is found to be used improperly, a warning will be given in class, which will seriously affect the final grade," Cheng Yunlai said.

But in the three private universities, the way of identifying AI takes on a different look.

Wu Yuan (pseudonym) works as a software engineering teacher at a private university. He feels this more directly.

Wu Yuan said frankly that more than 80% of the homework can be completely completed by AI. Whether it is multiple choice questions, short answer questions, writing code, or doing programs, students do not even check, "paste it and hand in the homework."

Wu Yuan mainly relies on two points to identify AI jobs: one is the "AI flavor" - the machine feeling that hits the face; the other is the "repetition rate" - when the central idea of ​​multiple jobs is almost just a different wording, and it can even be seen that it comes from the same AI model, the answer is self-evident.

Even more thorny is the practical dilemma of performance assessment. Wu Yuan revealed that according to school regulations, student scores need to show a normal distribution. "If you all get very low marks, when you write your score analysis report at the end of the semester, the Academic Affairs Office will ask you why it is not a normal distribution." What's even more ironic is, "The things written by AI, such as programs, are indeed correct. Although you know that it is AI, you feel that it is a bit perfunctory and you are not happy, but you can't give low marks. At least its score points are correct."

Teacher Zhao Bin from Fudan University gave "Option B". As one of the first teachers in China to introduce ChatGPT into teaching, he never sets restrictions on students' use of AI and even encourages students to try boldly. He invented an assessment method called "filling the holes you dig yourself" - at the end of the semester, he collected the assignments students completed with AI assistance during the semester and fed them to the AI, allowing the AI ​​to generate exclusive test questions based on each person's assignments.

This means that if a student simply copies the content of AI without truly understanding it, then in the final exam he will face an examiner who is “fed” by him and who best understands his blind spots in knowledge. Zhao Bin said that this not only supervises students, but also makes them understand that AI is just a mirror, and what they really have to face is themselves.

Screenshot of Teacher Zhao Bin’s course group. The group chat is called “Asking Good Questions to AI”

Students are also "evolving".

Faced with the teacher's siege, they quickly mastered the "anti-reconnaissance" skills:Adjust the tone of the AI's output to make it more like "an undergraduate writing a paper at two in the morning, with a hint of despair and messiness"; first write the outline or draft yourself, and then let the AI ​​fill it in to ensure a "human touch"; even deliberately introduce typos, grammatical errors or verbal slurs, pretending to be "written by one's own hands".

In the United States, there are also counter-complaints from students, questioning a professor who opposed the use of AI by students and using AI to write teaching materials and grading standards, so the student asked the school to refund his tuition. During the interview, Wu Yuan also mentioned that he had received a system document from the school leader, which read "very AI-smelling". In this regard, Wu Yuan joked: "The boy who slays dragons eventually becomes a dragon."

This kind of game imposes huge emotional costs on teachers.

Teacher Yu Wenzhou said frankly: "I thought that after teaching news writing for so many years, I was afraid that the children would not be able to write. In a large class of 120 people, I taught one by one and corrected each word. The result? The children used a bunch of AI Generated things come to deceive you... Sometimes I feel like a fool." He mentioned Li Shande in Ma Boyong's novel "The Lychees of Chang'an" - the honest and kind-hearted man who worked hard for a lychee that he didn't need at all. "The emotional impact was really huge."

Faced with this situation, colleges and universities began to issue various regulations. At the end of 2024, Fudan University issued the "Regulations on the Use of AI Tools in Undergraduate Thesis (Design) (Trial)", proposing "six prohibitions", prohibiting the use of AI tools in six aspects, including research design and data analysis, original data collection, creation of result pictures and important illustrations, thesis writing, defense and inspection, and confidential content.

Tsinghua University has also issued the "Guiding Principles for the Application of Artificial Intelligence in Education", which strictly prohibits the direct copying or simple paraphrase of AI-generated texts and codes and then submits them as academic achievements. The "Student Guide to Generative Artificial Intelligence" jointly issued by East China Normal University and Beijing Normal University requires students to mark relevant content in red when using AI to complete assignments, and the directly generated content must not exceed 20% of the full text.

But the introduction of regulations does not mean the problem is solved. Deng Jianguo, a professor at the School of Journalism at Fudan University, pointed out that students also have an involution mentality when using AI - if they declare in advance that AI is used in their assignments, they may worry that the teacher will lower their scores; while those students who do not mark but do use AI polishing may receive higher scores.

"In the context of fierce competition for GPA, this involution mentality may make it difficult for students to regulate how to use AI." During the interview, some teachers and students told us that the AI ​​usage regulations issued by the school are more regarded by them as a statement by the school. Some students think, "If I use AI without knowing it, what can the school and teachers do to me?"

Liu Yun (pseudonym), a doctoral student in the liberal arts field at Peking University, provides another perspective. As a senior doctoral student, she has completed her courses and mainly uses AI to assist in writing her doctoral thesis. "I use AI to help me process some English documents. I used to read one article in an hour, but now I put it in AI to translate and refine it, and I can read three articles in an hour." Her tutor strongly advocates the use of AI. "I feel that since this thing has come out, you should be able to use it."

Listen to the echoes of thinking in the noise of AI

As the game of "catching monsters" and "counter-reconnaissance" intensifies, a deeper question begins to emerge: Where is the end point of this game? If AI has entered the education ecosystem unstoppably, how can we coexist with it?

Liu Yun told an "advanced use" case. Recently, in the process of writing her graduation thesis, she felt that writing clear, logical, and summarizing subtitles was a difficult problem in thesis writing, and AI was very good at this.

During the actual operation, she did not simply let AI polish it. Instead, she first found a doctoral thesis in her major that she felt particularly comfortable reading, fed it to AI, and let AI analyze the language style of that paper. Then she took the text she wrote and asked the AI ​​to polish it according to the analyzed style. Liu Yun found that the logical chain after AI processing was indeed clearer.

"I have compared the generated results of the models and summarized a set of experiences on which jobs various AIs are suitable for. If the answer given by AI is particularly strange, try another model, or I can directly choose not to use it." In her view, AI is just a tool, and the right to filter and judge is always in her own hands.

The educational community’s understanding and discussion of AI are also gradually deepening.

Xue Lan, dean of Schwarzman College at Tsinghua University, believes that after artificial intelligence enters campuses, it is possible to promote the formation of a personalized teaching model and make up for the "shortcomings" of the current education model. For example, building a personalized learning center and forming a personal learning hub can provide application scenarios for students to understand knowledge.

In July 2024, Ding Kuiling, President of Shanghai Jiao Tong University, proposed the concept of "AI + HI" (artificial intelligence + human intelligence) at the World Artificial Intelligence Conference, focusing on the value of "what makes people human", helping professional education to free itself from complicated general knowledge and grasp people's core values ​​and competitiveness.

In the AI ​​usage specifications released by Jiaotong University, AI applications are further divided into four types: prohibited use, limited use, encouraged use, and open use, and promote the "AI+ education and teaching" graded classification reform. This refined management approach is replacing the “one size fits all” ban thinking.


In March 2025, Shanghai Jiao Tong University released the "Specifications"

In January 2026, Fudan University officially launched the AI ​​3A education co-creation platform and simultaneously released the "Generative Artificial Intelligence Education and Teaching Application Guidelines".

This over-10,000-word guideline breaks down AI application scenarios into specific links such as classroom teaching, academic evaluation, and independent learning support, and provides action suggestions, ethical warnings, and tool recommendations for each link. It is clarified: AI can be used for auxiliary links such as language polishing and structural suggestions, but it cannot replace core academic labor such as research conception, theory construction, and data analysis.

The model recommended by Fudan teacher Zhao Bin is the "AIGC Collaboration 1-2-3 Rule": 10 minutes of independent thinking, 20 minutes of human-machine collaboration, and 30 minutes of group co-creation. This model hopes to reasonably allocate time and energy, allowing AI to undertake mechanical and repetitive tasks, including data collection, preliminary data analysis, etc., and use the saved time for more creative and thinking tasks.

This ability to “ask good questions” is the core quality that education in the AI ​​era needs to cultivate. As Yang Zongkai, President of Wuhan University of Technology, said: “When it comes to the biggest change, I think we need to cultivate two types of talents in the future, one is talents who can ask questions, and the other is talents who can solve problems.”

As early as March 2023, Teacher Zhao Bin published an article in The Paper: Asking questions will only require higher, not lower, students’ knowledge reserves, ability to think about problems, and critical thinking skills. However, in reality, many people use AI with the opposite logic - students let AI write code so that they don’t have to learn code; they let AI write copy so that they don’t have to think about copy; they let AI do data analysis so that they don’t have to understand statistics.

The consequences of this are: students will never be able to develop their core judgment, and people will always stay below the productivity threshold. The more developed AI is, the easier it is to be replaced.

And those who can really get dividends from AI use AI like this:They have mastered the underlying professional logic and used AI to complete repetitive, low-value execution work for themselves, thereby using the time saved to hone higher-level judgment, decision-making and system design capabilities, making their core competitiveness stronger and stronger. Teacher Zhao Bin pointed out.

Looking back at education itself, is education the transfer of knowledge or the training of thinking? Is it being reshaped by artificial intelligence? Professor Mutlu Kukurova, director of the Learning Analytics and Artificial Intelligence team at University College London’s Knowledge Lab, believes: “AI tools alone are unlikely to transform education.”

After the outbreak of generative AI, Mutlu Kukurova will pay more attention to students' potential to ask and solve problems when assigning homework. He mentioned that the transformation of the education system is a socio-technical ecosystem issue, and today, the essence of education has not changed.

If the nature of education has not changed, what about the boundaries between teaching and learning? Perhaps it lies in "thinking" itself.

Teacher Peng Jian from Communication University of China pointed out that many people regard thinking as labor, but in fact it is a right. "If he completely relies on AI and doesn't think for himself, will he still be a complete person with the ability to think and make judgments?"

The wave of AI sweeping through universities is already unstoppable. No one can shut out AI, but we must continue to try to figure out the boundary between humans and AI. Of course, college students must take the initiative to understand and use AI. The problem is never the tool itself, but how to use it. What we object to is letting AI take over the process of thinking, creating and growing for ourselves.

In the future, education may no longer compete with who remembers things better, but with competition: who can ask good questions; who can inject unique and warm thoughts into the sameness generated by AI; who can use AI to break through the productivity threshold and spend time improving their own cognition and judgment.

When everyone is using AI, we must still hear the echo of our own thinking amidst the noise of AI.