When Chinese students use AI to rewrite Wang Yangming: three levels of AI-enabled learning, and implementation points for universities and enterprises

This article was originally published in the expert argument column of “Technology Island (opens in a new tab).

This spring, I went south to Xiziwan twice and walked into the classroom of the Chinese Department of Sun Yat-sen University.

The person who invited me to serve as a tutor was Mr. Wu Mengqian, associate professor of the Chinese Department. The course he taught was called “Traditional Studies”, and he studied Wang Yangming, and the final assignment he asked students to hand in was not a report or a paper, but an interactive web page that people could click on with a mouse and actually play with it.

The names of those two workshops are very interesting: “When AI Becomes Wang Yangming: AI Command and Vibe Coding Implementation Workshop”. My task is to lead a group of Chinese students who have not written procedural programs to learn from scratch how to give instructions to AI and how to grow a vague idea into a running web page.

To be honest, I was a little curious when I received the invitation. The Chinese department and Vibe Coding sound like two different worlds.

Now that the semester is over, all four sets of works are online. Teacher Wu and the second-year master’s teaching assistant Liao Jieyu even used Vibe Coding to set up a results website (opens in a new tab) to fully present the works revised by students based on feedback. The feedback questionnaire showed that students’ overall satisfaction with this project was as high as 4.8 points (out of five points).

Well, this deserves a good chat. On the surface, it is a Chinese language course, but at its heart it is the same question that all universities and companies have been asking in recent years: Can AI make people learn better? If so, how?

When the technical threshold disappears, what is left is real

Let me talk about those four groups of works first.

“To Conscience. Planting the Heart” transforms the original written assignment “To the Conscience Diary” in the class into a healing plant cultivation game: you make awareness and practice once a day, and the plants in your heart will grow a little bigger. “Yangming Knowledge King” imitates a puzzle battle game. Players become Mr. Yangming and compete with thieves. The questions are not memorization test questions, but modern scenarios such as workplace integrity, AI ethics and online violence.

“The Smelting of the Heart: Wang Yangming’s Antidote to the Soul” captures Yangming’s metaphor of “quality is divided into two”, designs a life situation test, allows players to reflect on themselves, and finally produces an analysis report on the purity of gold. “Psychic Cultivation Mirror” combines MBTI-style psychological tests and uses important disciples of Yangming’s sect as personality representatives to help players see their own habits and recommend appropriate self-cultivation skills.

Did you find out? Among these four groups of works, the really interesting thing is the creativity of the students.

When it comes to writing web pages, AI takes care of it all. However, if the color is divided into two parts and translated into a gold purity report, Yangming’s disciples are used as personality prototypes, and the opponent who answers the question is a thief and the question is AI ethics, these AIs cannot do a single step. That is something that can only be developed if you have really read “The Record of Biography”, really thought about what Yangming is talking about, and really know where modern people are stuck.

In other words, when the technical threshold is wiped out by AI, what remains, exposed or seen is the depth of students’ understanding of the text.

This is the most interesting part of this course. Teachers and students in the Chinese Department of Sun Yat-sen University do not regard AI as an unattainable technical threshold. Instead, they use AI as a chisel to cut through the wall that was originally blocked by programmatic grammar, allowing humanistic literacy itself to become the main body of the work.

And there is a detail hidden here that many people will overlook, that is, the work must be able to be executed. If a report is written vaguely, the teacher may not be able to read it; if a web page is illogical, the user will get stuck in three seconds. When a student has to break down the abstract concept of “Conscience” into a process that the user can click on and the system can run, he is forced to face a question: What exactly is this concept?

Code is ruthless and forces you to think clearly.

When it comes to AI-empowered learning, I’ve seen too many occasions where it ends at the first level.

The first layer is the tool layer: AI helps you do what you would already do, just faster. Using AI to organize verbatim manuscripts, translate documents, and generate slides all belong to this layer. The efficiency gain is real, but it doesn’t change what you can make. AI at the tool level turns an average student into a faster average student. Most organizations get stuck in the first year after introducing AI.

The second layer is the method layer: AI allows you to do things that you could not do otherwise. The interactive web pages created by Chinese Department students are the method layer. Before the emergence of Vibe Coding, this kind of homework could not exist, not because the students were not creative, but because there was a year and a half of programming training in the middle. Once this wall falls, the form of homework can be redesigned, and once the form changes, the depth of learning will also change.

The keyword of the method layer is not faster, but redesigned. The question it asks is: What would this thing have looked like if that constraint were gone? This is where most university courses and in-house training should really invest. It’s a pity that many people have obtained the ability of the method layer, but only use it to speed up the tasks of the tool layer, like buying a car and only using it to drive circles in their own yard.

The third layer is the layer of subjectivity: AI forces you to face who you really are and what do you believe?

According to the feedback questionnaire from the course, many students admitted that before taking the course, they had doubts about whether AI should be used in humanities subjects and were worried about losing the subjectivity of learning. This concern is entirely reasonable. I even think that students who have such doubts are really thinking.

After implementation, their views changed. Huang Yuling, a senior student, said that if it can be used properly, AI “can actually become our good partner in translating classics.” Zhao Shiya, a junior, mentioned that while this kind of learning method provides a high sense of achievement, “it also allows you to check whether you have integrated what you have learned.” The one who impressed me the most was Zhang Ziyun, a junior. In the process of overcoming the team’s difficulties, she realized that “what she really needs to face is not the external environment, but her own inner emotions and persistence.”

You see, this sentence is no longer talking about AI. This is Yangmingology.

A course that allows students to experience the frustration of collaborating with AI first-hand what it means to have nothing outside their minds is already beyond the scores that can be scored on any evaluation scale. The role of AI here is not a teacher or an assistant, but a mirror. It reflects back to you the fuzziness of your thinking, the ambiguity of expression, and the hesitation of judgment, every inch.

Knowledge is the beginning of action, and action is the completion of knowledge. The real teaching material of this course is the process itself of students’ repeated running-in with AI.

What universities can do: Five replicable design pivots

I have organized the keys to the success of this course in the Chinese Department of Sun Yat-sen University into five fulcrums. They are subject-independent and can be used in sociology, law or medicine departments.

Pivot 1: Hand over something that can run, not a report that can be read. The work must be alive, interactive and breakable. Only executable output will honestly reflect the loopholes in thinking. Reports can be cloaked in rhetoric, but web pages cannot.

Fulcrum 2: Tie a non-negotiable professional standard. The four sets of works in this course at Zhongshan all have to take into account both the basis of original texts and the interactive transformation. If they only ask for fun, students will slip into playing games; if they only ask for quotations from classics, AI will not be able to help much. Real instructional design is to tie the parts that AI can help (form, implementation, interface) and the parts that AI cannot help (text, interpretation, judgment) into the same assignment, leaving students with nowhere to escape. Each discipline can find its own line: the law department is the basis of laws and judgments, the medicine department is the level of clinical evidence, and the business department is the source of financial figures. Draw that line clearly, and AI transforms from a ghostwriting tool into an amplifier.

Fulcrum Three: Publicly published and open for trial play. Teacher Wu specially held a results presentation meeting at the end of the semester, allowing students to present on stage, open trials, and conduct mutual evaluation in groups. This arrangement looks like a ritual, but is actually quality control. When you know that your classmates will play with your work and destroy it on the spot, the level of seriousness is completely different. Teacher Wu and the teaching assistant even set up the work as a public website. This step upgraded the homework into a work.

Fulcrum 4: Leave one round of correction. What is posted on the results website is the version that students have optimized based on teacher feedback to address deficiencies. The first version of AI’s output is often passable or mediocre. Real learning occurs at the moment when you get the feedback and then correct it again, because at that moment the student must judge: Which of the teacher’s words should be listened to, and which of them are my own persistence. This is the training of subjectivity.

Fulcrum 5: Teachers and teaching assistants themselves have to face consequences. Teacher Wu and Liao Jieyu used Vibe Coding to build the results website themselves. This detail is more important than anything else. For a teacher who has never personally wrestled with AI, it is difficult to judge whether the things handed in by students have really been thought about, or whether they were thrown to the AI ​​for random production. A teacher’s touch cannot deceive anyone.

What companies can do: Three cutting points and a common mistake

Changing the scene to an enterprise, these five pivots can be moved almost unchanged, but the nouns need to be slightly changed.

Let’s talk about the most common mistake first: starting with the import tool. Too many companies start their AI transformation this way: buy a batch of software licenses, hold a general lecture, send out a user manual, and then announce that the company has become AI-enabled. Checking after three months, only the IT department and a few fun-loving colleagues were using it. The problem is that this approach stops entirely at the tool level. You just gave everyone a faster pen, but you didn’t change the exam paper to write on.

The correct sequence is the other way around: first ask what the thing would look like if this constraint didn’t exist, and then go back and decide what tools to use. The following are the three easiest places to start and see results.

Cutting point one: Make internal knowledge a living thing rather than a dead document. Every company has a batch of documents that no one reads: newcomer manuals, SOPs, product knowledge base, and legal compliance regulations. The problem with them was never that they were poorly written, but that they were in the wrong form. No one is going to flip through a forty-page PDF when they need it. Students at Sun Yat-sen University have turned “The Record of Biography” into interactive tests, development games, and situational Q&A; using the same logic, your legal compliance regulations can be made into a situational judgment tool, and the newcomer’s manual can be made into an interactive guide that takes newcomers through a real process. Now, the cost of making these things has been so low that a patient senior employee and a weekend can make a prototype.

Cutting point two: Let the person who knows the business best make the gadgets he needs. The most painful gap in an enterprise is always that the person who understands the problem best does not know how to make the tool, and the person who knows how to make the tool does not understand the problem. The IT schedule has been postponed to next year, and the business units have no choice but to continue to use Excel to support themselves. The real subversion of Vibe Coding is to compress the threshold for making a gadget from a year of programming training to an afternoon of patience. Students in the Chinese Department can make web pages, and your business manager can certainly make quotation calculators, customer complaint classification boards, and scheduling conflict checklists. The premise is to clearly distinguish the boundaries: this type of tool is internal, auxiliary, and disposable. It involves customer data, cash flow, and external service systems. If it is time to follow the formal development process, you must follow the formal development process.

Cutting point three: Change the output of internal training from completing classes to delivering works. There is a piece of work between an internal training with a good reputation and a useful internal training. If you want to conduct internal AI training, my suggestion is not to teach tools, but to let everyone bring their own real and stuck work. Use the class time to do it. At the end of class, everyone hands over a running object and asks colleagues to click on it and play with it, and then leave a round of corrections. The only difference is that the student handed in a web page, while the employee handed in the solution to the process that had troubled him for three months.

As for that non-negotiable professional standard, it corresponds to your business bottom line in the enterprise: information security red line, legal compliance requirements, brand tone or digital source. Write it down clearly and make it an acceptance condition for every AI output. There is only one sentence in fulcrum five: If the supervisor himself has not used it, don’t expect the team to use it.

About the line: empowerment does not equal outsourcing

Finally, I have to bring the topic back to the student’s original doubts: Will the subjectivity of learning be lost when using AI?

My answer is: yes, and it’s very easy. There is only one dividing line, but it is extremely clear. AI is used to expand the radius of your expression, which is empowerment; AI is used to replace your judgment process, which is outsourcing.

When a student uses AI to make the Yangming theory he understands into a web page, his understanding does not decrease, he just finds a new outlet. Another student asked AI to help him summarize Yangming Studies and package it into a web page. The finished product may look similar, but nothing was left in his mind. It is difficult to distinguish the two from appearance, which is a common problem for all teachers and managers in the AI ​​era.

The only effective way is to shift the focus of evaluation from output to process and defense. After the work is completed, ask him to explain: Why did he choose this concept? Why this interactive form? What did the first version AI gave you look like, what did you change, and why did you change it? People who have really thought about it can answer these questions; people who outsource it cannot. Subjectivity is not preserved by refusing to use AI, that only postpones the problem. It is accumulated time and time again in every decision of “Should I accept AI’s suggestion?” The more you use it, the more you decide, and the stronger your subjectivity becomes. The premise is that you are really deciding, not just nodding.

Back to that sentence

On the home page of the results website, there is this sentence: “There are no external principles and no external things. Knowledge is the beginning of action, and action is the completion of knowledge. Today, AI is used as a tool to integrate classics into the web page, and knowledge and action are unified.”

I like the words “AI is the tool today” very much. Tools are tools and implements, not their masters or opponents.

Most of the discussion about AI in the past two years has been about whether it will replace me. But in that classroom in Xiziwan, I saw a completely different picture: a group of Chinese students who couldn’t write programs were re-translating a thinker from five hundred years ago to people of this era. The method they use is AI, but the translators are themselves.

4.8 points of satisfaction, not because the AI ​​is amazing, but because they were finally able to make what they had in mind, and there are actually people willing to play it. That feeling is easy to understand for anyone who has ever pressed execute late at night and watched their things run.

So, whether you are a university teacher, a business executive, or an office worker who is hesitating about whether to touch AI, I have only one question for you: If the barriers to programming, design, editing, or translation all disappear tomorrow, what exactly do you want to make?

Only by thinking clearly about this problem can AI come into play. If you don’t think clearly, no matter how powerful the tool is, it will only get you to a place you don’t want to go faster.