AI is coming, can universities still be "too big to fail?

Professor Zheng Zhang

Writing |Zheng Zhang (Professor of the University of Hong Kong, Chief Scientist of the Hong Kong-Shanghai Innovation Research Institute)

●                  ●                   ●

Our previous two articles talked about the opening and closing of AI, and AI4S(AI to make scientific research a "scientific automaton", what are the side effects?)Let's talk again here.AI4Edu, which meansHow to Change Education in the Age of AIA problem that everyone should be more concerned about.

In November 2025, I will attend in Hong Kong.Digital Education Council（Digital Education Council）The Global Summit. The conference was hosted by the Hong Kong University of Science and Technology and was invited by the provost, deputy provost and dean of more than 100 universities from all over the world. The real theme of the meeting is only one: what to do when the AI comes.

One link is on-site voting. The question is: Is the university too big to fail?

At that time, I was still going through the formalities for joining HKU and thought to myself, okay, will I be laid off as soon as I return to school.

The appearance of "big but not falling" in such conferences is a necessary self-provocation by global universities, but it points to the future. It gives me the impression that if it weren'tAI suddenly came, now this set slowly repair.

The assembly line from general education to university and then to graduate education is a particularly large and complex ecology at present. Judging from the history of the development of human civilization, this educational assembly line is extremely young, but it has a history of more than 80 years. My mother is the first master's degree student in Fudan Foreign Studies Department. There are only two people in the department. I was very surprised to hear her say that the teachers in the department had no idea what to do with them. That was in the sixties of the last century.

However, this seemingly natural and stereotyped assembly line has many problems both at home and abroad. The most prominent are the following two.

First, those who can enter a good university are the lucky ones, and most of them are stopped.When a person is not trained in critical thinking and does not have a strong enough knowledge base-which is exactly what a good university can give-the person has no defense against misinformation and will become a conductor of misinformation.

To stop the spread of an infectious disease, each contact has to be passed on to less than one person on average, which isR is less than one. A study published by MIT in Science in 2018 counted more than ten years of data on Twitter. The time it takes for true news to reach 1,500 people is about six times that of false news. It's not how diligent the rumor mongers are, it's the crowd's.R is too high.

If everyone'sIf R does not fall, enlightenment and wisdom will not be completed.

Rumors run far ahead of the truth, and this is not a social network problem-I used to think it was. The essence of the "mob" is an educational issue, and the information society has only magnified it.

Second, the goal of this pipeline is to develop excellent experts. But "special" is also often "narrow". studyLiberal arts do not understand or even disdain to understand the most basic scientific knowledge, while science and engineering students are not aware of history and art, which is the norm. C.P. Snow said that the gap between the "two cultures" is running deeper and deeper under the goal of cultivating "experts.

Every major will develop its own thinking habits. It is a sharp weapon for this major, but it also shields other perspectives. I putThis kind of common sense that can't get out of the professional door is called "impassable general knowledge"。

For example: scientists pursue the only, universal truth, but in the world outside the study is often very axis of leverage; engineers seek the optimal solution in constraints, but human nature is not simplified into a string of constraints; investors talk about input and output, but not everything can be attributedROI(Return on Investment)Humanists emphasize narrative and freedom, but tend to underestimate the hard constraints of the system itself.

Each of these professional common sense has its own use, and the assembly line oriented to cultivate experts can only teach you one. The result is that the more successful, the narrower the vision and the deeper the prejudice.Successful people also hold enormous power over the distribution of influence and resources.

These two major problems are the result of the smooth operation of this assembly line.

In June this year, the Hong Kong Learning and Teaching Expo（Learning & Teaching Expo）In fact, in a three-person conversation hosted by the Digital Education Council, the question of "big but not falling" was thrown over again. My answer is: first remove the label "university" and see what it is.

It is a group of young people must spend four years. During these four years:

• This group of young people grew up together in the same place, at the same time, with each other's gaze and company.This is a "group" that overlaps in time and space ".

• It's a golden time for them to find the kind of person they want to be. It can be a teacher, a senior or a classmate that affects their life.

• At the end of the four years, what they hold in their hands is not a diploma, but a signal to the society they are about to enter: we are ready.

The University of Bologna, founded in 1088, is the oldest university in Europe. Source: University of Bologna

What makes everyone particularly anxious at the moment is: how can these young people be considered ready?

The anxiety of educational institutions is more specific: how to follow up the teaching and learning methods, and to put it bluntly, this job is not enough-AI ability is making full-channel self-study possible. Although there is no quantitative research, it is no longer a case that those outstanding young people who also have resources actively or passively jump out of this assembly line.

This question cannot be expected to be answered by the employer. The changes brought about by the AI are comprehensive. The company itself is adjusting, the position is changing, and the nature of the position is also changing. If the standard cannot be set, it will not give a clear signal.

Back to the university itself, the "group" that overlaps in time and space, and the people encountered in this group, are irreplaceable. Coupled with the huge ecological inertia, universities will continue to occupy this ecological niche, and they can indeed be "too big to fail".

There must be anxiety, but it is this question that must be looked at directly: what is "ready"?

For example, young people can use it freely.AI the tool and let it be its own extension-does this count?

This is a good but empty proposition. If it is implemented, it will bring a batch of answers with a short shelf life, one by one fast-moving targets.

For the first two years, everyone was busy teaching how to write prompt words for models, then how to pair code with Copilot, and thenVibe coding has now become how to keep an eye on a group of agents running by themselves.(Agent). In three years, there were three rounds, and each round was regarded as "the ability that students must master".

Bezos said in a conversation in 2012 that people always ask him what will change in the next ten years, but few people ask what will not change in the next ten years-and the latter question is more important because you can only put long-term investment on the same things.

Might as well refer to a case from nature.

Many people may not know that the synaptic connections in the baby's brain are "surplus" at first, and the density peaks in the first few years after birth, far exceeding the level of adults, and then in the process of crawling, touching and trying around, the useless ones are gradually trimmed off, and the useful ones are left behind and strengthened.

This magical design has only one point: the action that triggers the pruning at that stage. Since it is impossible to predict the future environment, simply do not design, let the environment to help choose.

In other words, it is not the answer itself that can be "anchored", but the ability to shape in an uncertain answer.

So, what are the constant abilities that education needs to cultivate?

• Ask good questions.

• Capture mistakes keenly and ask questions in a way that the other party can accept.

• Inspired by imperfect answers.

A good dialogue is never a one-way talk, but a jump between questions and answers. It makes no difference whether the opponent is human or AI.

On this educational assembly line, the space for asking questions is squeezed all the way.Asking questions in the classroom takes courage: worrying about asking "stupid questions", worrying about disrupting the rhythm, worrying about appearing to be out of step ...... So, not asking questions is the safest.

I try to set aside a few minutes to ask questions to my classmates in class. The effect can be imagined. Either I am completely cold, or I always ask questions from those activists. The more I ask, the more I am. During the four years of college, I think many students did not mention a single question in class.

Every encounter with knowledge does not include the opportunity to ask questions.

What about the Socratic teaching method?

Last year I used the big model to design a class about the big model itself,LLM4LLM, 14 topic conversations, 64 knowledge points, when designing I deliberately let it Socrates possessed: not directly give the answer, all the way to ask, forcing learners to push down.

The trouble is: tired. The somatosensory that has been questioned is being interrogated, not being explored. This teaching method may stimulate thinking, but it does not stimulate the ability to ask questions.

(If you want to know how I escaped "Socrates" to debug, simple: press all the way"I don't know. Tell me"。）

LLM4LLM is a half-project. But the harvest is also very real: as long as the knowledge points are sorted out, AI are fully capable of teaching. Sorting out knowledge points is the investment of most of the course. As for putting on the "Socrates coat", it is just a few words in the prompt.

This indicatesThe plasticity of AI in education is very strong, and there are many scenarios for application: changing homework, role-playing, and answering questions after class. There are many such products, which are speeding up the existing links without hurting the muscles and bones.

To reimagine how college education should be done, we must first look back: in these four years, what is the most expensive and most unrepeatable?

The greatest advantage of the university lies in the fact that one by one makes young peoplegroup of overlapping space-time: Campus, square, library, laboratory... abound.

No one wants to be stuck on their own lonely planet, especially young people.

How to make good use of this overlapping group of time and space is the true proposition of the university.

AI add in, make it possible to reallocate resources and makeAI do what they do best and leave the rest to humans.In these four years, a young man is looking for the kind of person he wants to be.——Not a model. No one wants to be carbon-based.Claude/GPT/DeepSeek 。

One of the reasons for Peking University's continuous flow of outstanding mathematical talents is that peers encourage each other and compete healthily. The picture shows a group photo of the 2007-level undergraduate alumni of the passive Peking University School of Mathematics at the 2026 International Congress of Mathematicians (from left: Zhao Zihui, Tang Yunqing, Wang Hong, Deng Yu, Sun Xin, Su Weijie). They are all excellent, including two Fields Medal winners and one of the highest statistics award winners. Source: School of Mathematical Sciences, Peking University

The transfer of knowledge is only part of it, and it is the most easily replaced part. Discussions, debates, fieldwork, experimenting with classmates, having a relationship, trying to start a career, doing something that has a high probability of failure-these are not.AI replaced.

College life is never and cannot be called a string of GPA (grade point average)The number is defined, but a trajectory that runs through school and life.

This trajectory can penetrate the framework of the current course: the distinction between cause and effect and correlation, sampling and significance in statistics, and data visualization. These skills are widely used in business, medicine, social science and other majors besides data science. Why should they be locked in their respective majors and not let students use them repeatedly in different scenarios?

The same ability can stand in several different scenes before it is learned.

This will be a very dynamic reform that is both unsettling and exciting.in absorbing reasonableWhile AI abilities, the most important thing is to cultivate and strengthen those abilities that remain unchanged.The three points I mentioned earlier-asking good questions, being keen to catch mistakes and asking them in a way that the other party can accept, and being inspired by imperfect answers-are the most important in my own opinion, and maybe there are others.

What remains unchanged is the beginning of the university. Just like the six goals set by the University of Hong Kong for undergraduates:

• The pursuit of academic and professional excellence, critical intellectual inquiry, and lifelong learning

• Dealing with unfamiliar situations and ill-defined problems

• Critical self-reflection, a deeper understanding of others, and adherence to personal and professional ethics

• Cross-cultural communication and global citizenship

• Communication and Collaboration

• Leadership, and initiatives for the betterment of the human condition