Artificial intelligence is now part of education, work and daily life. There is no turning back. We are all using it to search, summarize, translate, brainstorm, draft and check. The question is no longer whether AI should be used, but how to use it without weakening the human capabilities society needs.

This is why the debate about AI in teaching matters far beyond the classroom. Universities are where students are supposed to learn how to read carefully, think clearly, write coherently, speak precisely and exercise judgment. If AI helps them learn these things better, it is a valuable tool. If AI allows them to bypass the effort required to learn, it becomes a serious problem.

A recent Fortune article about Brown University illustrates the danger. Professor Roberto Serrano, who teaches mathematical economics, changed a midterm examination to a take-home format after a tragic campus shooting, as students were grieving and anxious. The result was shocking. Of 86 students, 40 scored a perfect 100. The class average was 96, far above the usual range of 65 to 80, even though the exam was harder than usual. Serrano later found evidence that many students had used AI. When he required an in-person final examination, the average fell to 48.

Teachers everywhere are seeing similar signs. Written assignments may look polished, structured and fluent. But that is not the same as understanding. A student may not have wrestled with the topic and missed the learning.

Cheating is the immediate problem, but dependency is the deeper one. If students use AI to avoid reading, writing, calculating, explaining and thinking, they are not only breaking academic rules; they are weakening their own capacity. They may become mere tool-users without understanding.

The same risk exists beyond campus. If companies use AI solely to cut workers, they may boost short-term efficiency while weakening the human knowledge and creativity embedded within their organizations. If we all rely on AI to interpret everything, we may appear more informed but be less capable of independent judgment.

This is the broader lesson from the history of technology. Powerful technologies are rarely dangerous because they are useless. They become dangerous precisely because they are useful, profitable and adopted at scale before societies fully understand their consequences.

Fossil fuels, industrial chemicals and digital platforms all brought enormous benefits. But each also caused harm when use became excessive, incentives became distorted, and governance lagged adoption.

The future belongs to those who can think with tools, speak with understanding and act with judgment. AI will change education and work, but it should strengthen human capability, not replace it

AI should be seen in this light. It can improve productivity, support research, assist decision-making, expand access to knowledge and help people do many tasks better. But if used indiscriminately, it can also encourage dependency, weaken genuine understanding and skills, displace work, concentrate power and consume large amounts of energy and water. The task is not to reject AI but to build guardrails that retain human judgment.

This is especially important in education. Detection software and plagiarism rules are not enough. The deeper task is to redesign learning and assessment so students must demonstrate understanding in ways AI cannot easily replace.

I teach an undergraduate course on risk, finance and sustainability. These subjects are especially exposed to superficial understanding. In the real world, there are few standard answers. They involve trade-offs, evidence, uncertainty, time horizons and institutional judgment.

In my most recent class, I have been experimenting with old-fashioned methods. First, I try to get students to talk more in class. This is not a casual discussion. I want them to explain the fundamentals in their own words. The point is succinctness. If students understand the fundamentals, they should be able to say them simply.

This is also good professional training. In working life, professionals must brief bosses, answer clients, question consultants, challenge models or speak in meetings. Clarity is a practical skill.

Second, I use multiple-choice questions that require more than memory. A challenging multiple-choice question asks students to identify the best answer among several plausible options. This tests comprehension. Students must read the question carefully, identify the issue and understand why one answer is stronger than another.

Third, I ask students to handwrite answers to narrative questions. These are not long essays. They are short responses that answer the question directly. The tests are open-book. Students may bring books and notes. But there are no electronics, and therefore no AI.

This combination is important. It is not about memorization. I am testing whether they can find, select and use knowledge under time pressure. Handwriting slows the process down. It forces students to decide what is worth saying. It reminds them that understanding is not the same as copying.

Fourth, I include a high-scoring group presentation. Students may use AI in preparing their work. Every group must follow a company and an industry for the duration of the course, so their knowledge accumulates over time. They cannot simply produce a generic presentation the night before.

Most importantly, they must present orally. They must explain the company, the industry, the risks, the financial implications and the sustainability challenges. They must show what they have learned. AI can help prepare slides, but it cannot stand in front of the class and demonstrate the students’ own grasp of the subject.

Another professor recently told me he is adding an oral component to final examinations. It is time-consuming for large classes but may become increasingly necessary. A few questions can quickly reveal whether a student understands the logic behind the topic.

These methods are pro-learning. Students should know how to question AI, check it, identify errors, improve prompts and apply judgment.

AI has become a normal professional tool. But tools are useful only when the user has competence. A calculator helps someone who understands mathematics. It does not create mathematical understanding by itself. AI is similar. It can extend learning, but it cannot replace the learner.

I do not pretend these methods are the final answer. Teachers everywhere are experimenting, and I am keen to learn from others who are finding effective ways to teach in the age of AI.

The central responsibility of teachers is to protect the learning process. We must ask students to speak, write, reason, question and defend their answers. We must make fundamentals visible again. We must reward clarity, comprehension and judgment.

The wider responsibility of society is similar. AI should not be allowed to hollow out human capability in the name of convenience or efficiency. It should help people think better, work better and solve harder problems. It should not disincentivize people from thinking.

The future belongs to those who can think with tools, speak with understanding and act with judgment. AI will change education and work, but it should strengthen human capability, not replace it.

The author is the chief development strategist at the Hong Kong University of Science and Technology’s Institute for the Environment.

The views do not necessarily reflect those of China Daily.