Leading academics from top Chinese universities are raising alarms about the hidden dangers of artificial intelligence in education. As the new semester begins, presidents and chancellors from institutions including Peking University and Fudan University are cautioning students against over-reliance on AI tools, warning that such dependency can lead to "cognitive offloading" and "illusory mastery." This creates a troubling scenario where students merely appear to understand subject matter without genuinely processing knowledge.

In August, the Massachusetts Institute of Technology released a report evaluating AI usage on its campus, concluding that many applications of the technology are depriving students of valuable learning opportunities. While these warnings are necessary, they raise an important question: if classrooms, assignments, and assessments continue operating unchanged, can we truly prevent intellectual outsourcing? Universities cannot simultaneously embrace AI while shifting all responsibility for genuine learning onto students.

A recent survey conducted by a media outlet in collaboration with researchers from Minzu University of China examined 1,953 undergraduate students across more than 400 universities nationwide. The findings are striking — 99.18% of respondents use AI tools, over 60% express confusion about the boundaries of acceptable use, and fewer than 30% report that their institutions offer systematic AI-related courses. This data raises an urgent question: students are already using AI, so why hasn't pedagogical policy caught up?

The immediate impact of artificial intelligence is that a polished assignment no longer proves that a student has truly mastered the material. Essays may demonstrate structural completeness, and presentations can impress with visual aids, yet whether students have read original sources, verified evidence, or engaged in genuine reasoning remains nearly impossible to determine from the final product alone. This exposes a longstanding weakness in university education — many assignments have always been exercises in assembling materials and meeting formatting requirements. AI simply accelerates this existing process.

Universities must resist importing corporate efficiency standards directly into the classroom. In the workplace, eliminating redundant labor increases productivity; in learning, however, certain labor that appears unnecessary is precisely what builds capability. When students struggle through difficult texts and revise arguments repeatedly, they are not merely producing a final deliverable — they are developing skills through the process itself. Removing that training eliminates what might be considered the essence of education itself.

Reform must therefore begin at the level of each course and each assignment. Clear guidelines should specify which components may involve AI assistance, what must be completed independently, and what disclosures are required when submitting work. The MIT report cited earlier explicitly requires courses to provide clear AI usage policies. These rules may vary by discipline, but students should never have to guess, nor should boundaries be drawn after assignments are already submitted.

Consider how the same tool serves different purposes across disciplines. A literature course training close reading skills might require students to engage with original texts and develop interpretations first, then use AI to explore alternative perspectives. A course on translation and revision might directly use AI-generated translations as discussion material. The same instrument deployed in different pedagogical contexts can yield entirely different outcomes. Vague exhortations to "encourage" or "prohibit" AI usage cannot substitute for thoughtful instructional design.

Assessment must also shift from evaluating finished products to verifying actual competence. After submitting a paper, students should be asked to explain why a key quotation supports their conclusion. After programming a solution, they should face modified conditions and be asked to adapt their code while explaining their reasoning. After presenting research findings, they should be questioned about sample selection and which evidence contradicted their initial hypotheses. These follow-up inquiries enable instructors to see precisely how deeply students understand their work.

Simply reformulating questions as open-ended prompts will not solve the problem, however. AI can generate position papers with clear arguments and complete reasoning chains. Draft histories and revision records offer clues but cannot automatically prove that thinking occurred. A combination of in-class writing, live demonstrations, oral discussions, and extracurricular projects should form the assessment framework — testing both the ability to work independently and the capacity to solve problems using available tools.

Faculty accountability matters equally. If instructors use AI to generate course materials, design assignments, and grade work without verifying content quality or providing targeted feedback, then demanding independent thinking from students lacks credibility. Tools may assist educators, but responsibility for course quality and evaluation outcomes must remain with the instructor.

Reform cannot become an additional burden for teaching staff. When a single instructor faces over one hundred students, conducting individual viva examinations and tracking every draft version requires institutional support. Without teaching assistants and adjusted workload calculations, these requirements quickly devolve into another round of bureaucratic paperwork. What universities genuinely need to invest in is faculty time spent understanding and mentoring students.

There is also cause to resist turning AI usage rates into a new form of administrative achievement. The number of "AI+" courses offered or platform queries answered does not directly demonstrate improvement in student judgment. Mandating AI use in every course may even crowd out essential independent reading and foundational training that should be preserved.

Universities must teach students to use AI while also helping them build the disciplinary foundation necessary to evaluate it. Without reading literature, how can one identify fabricated citations? Without performing derivations, how can one spot gaps in an answer? Critical thinking grows through engagement with concrete knowledge and repeated practice — it cannot be instilled through slogans about "maintaining independent thought."

The greatest danger after AI enters academia is what might be called pedagogical stagnation: students submit assignments, instructors complete grading, institutions accumulate impressive metrics, yet students fail to develop corresponding abilities. A university degree ultimately certifies the training one has received and the judgments one can make. Universities cannot outsource that responsibility.