Published in Science Daily of China (2025-02-21, p. 4, Culture). This should be the longest article I have ever published in a newspaper; they gave me a full page. My original text was basically not changed much, except that the section headings were altered a bit and the prose was lightly polished. What I am posting here is my original draft; anyone who wants to reprint it can go find the version posted by Science Net.
Welcoming the AI Era
In the courses on history of science and philosophy of technology that I had been teaching at Tsinghua, I had already run into the shock of AI. Back then I gave ChatGPT the final exam questions I had set; at the time it was basically able to reach a moderately above-average level. Now I have tried the more advanced 4o and the newly released deepseek, and both can easily reach excellent, surpassing all of the students on many questions, and even doing better than my own answers.
Of course, at times AI’s answers still have many unsatisfactory aspects—for instance, they are too formulaic, too rigidly organized, fond of beating around the bush, affected and artificial, pretending to know what they do not know, and producing fake citations, and so on—but these faults are in fact even more serious in many students. Besides, AI is easier to train: some students never change despite repeated instruction, letting things go in one ear and out the other, whereas AI, with a little correction, can immediately produce answers that satisfy me much more.
So if the students our courses produce are not even as good as AI, is there still any need to study this course?
In the AI era, the educational goals and training methods of every discipline need to be rethought and redefined—which things no longer need to be learned, which things should be taught in new ways, and which things are still worth insisting on through traditional teaching methods?
Everyone can notice AI’s impact, but many people remain unmoved; with each technological improvement they like to find some evidence that AI still falls short of human beings, and then they feel reassured, as though AI were no serious threat. I see this ostrich-like attitude in many philosophers and humanities scholars.
Of course there are also people who think AI challenges human education and work, but much of the analysis they offer is, in many cases, not even as good as AI’s. I have seen some papers that use endless empty clichés to comment on the impact of AI technology on education, intermittently inserting such apparently impressive terms as “embodiment,” and in the end conclude that AI cannot replace face-to-face teaching by teachers. But if you replace AI with “the internet,” “electronic communication,” or even “printing” throughout the whole article, it still works.
If these philosophers and humanities scholars who can do nothing but speak boring nonsense can be eliminated by AI, that would be a good thing too.
More Than a Tool
Of course there are also many people who can take AI’s impact seriously, but many of them understand AI only in terms of a “tool”—AI provides new, more efficient teaching methods and productivity tools. This is certainly true, but it is not necessarily the most crucial issue. For example, we can of course regard the steam engine as a donkey that is more efficient at turning a mill, and the steam locomotive as a donkey that can carry heavier loads and run faster… But in fact the steam engine was far from merely replacing animal power; more importantly, it brought about systemic changes across society—the Industrial Revolution brought by the steam engine created entirely new ways of life, relations of production, and even political and economic systems. These revolutionary effects are very hard to understand as just “a stronger donkey.”
In the teaching of the humanities, we must be even less confined to a utilitarian perspective; we need to pay more attention to how AI shakes the meaning or purpose of education itself.
For example, after information technology appeared, did it become more conducive to training students to have greater memorization and retention? It seems so at first glance: in the past, people could only pore over heavy encyclopedias, which were expensive and inconvenient; searching for entries and copying notes by hand was not very efficient. Now we can search and consult encyclopedic knowledge online anytime and anywhere, free of charge and rapidly, and marking up key points on electronic devices is more convenient. So students in the information age clearly have better conditions for exercising their memory. But the more crucial question is: since encyclopedic knowledge is accessible anytime, anywhere, why should the new generation of students bother to memorize it by rote?
Many new technologies in history have been like this. For example, bronze and iron tools are more conducive to hunting, but once people mastered bronze and iron tools, they had already entered the agricultural age, and all kinds of hunting techniques became less important. Gutenberg’s printing press was similar: at first people thought the press, as a medium of dissemination, would make the spread of the Bible more convenient, but in the end people discovered that the printing press actually promoted secularization.
The development of AI is like this too: what AI improves as a tool may not be what will ultimately be promoted throughout the entire AI era. How things will specifically develop in the future is still something on which it is hard to give a clear judgment now.
Technology Returning to Life
My personal judgment, or rather my hope, is that on the one hand AI, as a productivity tool, can greatly enhance productivity; on the other hand, in the new AI era, productivity will become less important for human civilization.
If the future is not like this—if people still regard productivity as the most important value, or even the only value—then this world will become extremely dangerous.
If the final meaning of human effort and construction is to improve productivity, then it is easy for us to draw such a conclusion: those who contribute more to improving the productivity of society as a whole are more valuable; if a thing can produce more productivity, then that thing is more worth doing. But what will happen when human beings discover that they are far inferior to AI in improving productivity? What if human beings find that many things only create trouble for AI, and that the only time they can play a role in bringing productivity into full play is when they are used as consumables, as “batteries,” for AI? Then what? What meaning, then, remains in human effort, or even in existence itself?
There are indeed people who believe that human beings should go extinct, and that silicon-based life should carry on human civilization. But if AI really awakens to self-consciousness, would it also be trapped by narrow instrumentalism and define the meaning of its existence as being a productivity tool? I believe that a higher intelligence could never regard its own value as that of an ox or a horse. Intelligent life—whether human or AI—should liberate itself from the fate of oxen and horses and pursue richer purposes.
Productivity is of course important, and the significance of science and technology in improving productivity is of course undeniable. But we must always remain clear-headed: production is important because it enriches human life. We must not turn it around and say instead that the purpose of human life in the world is to produce.
The logic of productivity is linear and expansive: it seeks more and more, the more the better. But life is diverse and bounded: it seeks what is just right, stopping at the highest good. From the perspective of a factory, producing bread is of course the more the better; but from the perspective of a person, even the person who loves bread most could not possibly eat dozens of loaves a day. From the perspective of a factory, specialization and standardization are the winning strategy, and a bread assembly line cannot be used to produce wine; but from the perspective of a human being, even the greatest bread lover always needs a variety of foods.
The Wisdom of Life
How to adjust the steps and rhythm of an assembly line in order to produce more bread—that is one kind of wisdom. But how to arrange the pace and rhythm of life so as to enjoy life better—that is an even greater kind of wisdom. Foolish people, when they see something delicious, can only wolf it down and gorge themselves; as a result, they can neither savor the wonderful details carefully nor maintain their health, thereby increasing future suffering. How to be relaxed yet restrained, how to enjoy yet not become addicted, how to expand the space for free choice yet not fall into a bewildering confusion—these are all forms of the wisdom of life.
One of the great meanings of learning knowledge is that it helps us improve our taste and broaden our horizons. For example, a materially poor person may have eaten nothing better than pancakes; if that person’s knowledge is equally impoverished, and he imagines the emperor’s life as nothing more than “carrying a gold shoulder pole and eating pancakes one by one and throwing each away after one bite,” then even if he suddenly becomes rich, he will still find it hard to discover refined pleasures. This is also why gambling is prevalent in many impoverished areas, and why many formerly poor lottery winners, nouveau riche, or people who have profited from urban redevelopment demolitions are more likely to squandary their wealth in drunken revelry—because their spiritual world has never been enriched, and thus they do not know how to seek out and savor more kinds of pleasure, nor do they know how to restrain and regulate their desires in a moderate way.
One great meaning of enlightenment education is to teach everyone the capacity to bear and enjoy freedom. We need learning, not in order to memorize many facts by rote so as to cope with exams or show off our learning, but because the more we know, the greater the space of choice becomes; we need to delve deeply into certain fields of specialization, because the rich pleasures of many activities can only be appreciated through deep study and practice—for example, go is nothing more than black and white stones, and to an outsider it seems dull and tedious, but in the eyes of an expert it changes in myriad ways and has endless interest; we need philosophy and history in order to enhance our capacity for introspection, understand where we came from and where we are going, and think about our boundaries and possibilities; we need mathematics and science in order to cultivate rationality and a commitment to truth, so that when choosing a way of life we can calculate gains and losses and identify traps, while many fields of technology are games deeper and more enchanting than go, and studying them in depth also brings intrinsic joy; we need literature and art in order to enhance our aesthetic taste and imagination; we need political science and sociology in order to participate better in public life; we also need physical exercise in order to improve our quality of life…
Learning to Become Human
We find that once we put aside utilitarian goals such as “training instrumental people” or “producing human resources,” then when human beings are freed from the urgency of making a living and focus on enjoying life, education and learning still have many meanings—perhaps we can say that education and learning return to their most fundamental purpose: “learning to become human,” that is, cultivating a healthy and free personality.
The teaching tradition of ancient Greece originated in the gymnasium, and the original meaning of “school” is “leisure”; the Greeks studied during the leisure time of physical exercise. Plato’s Academy and Aristotle’s Lyceum were both established on the sites of former gymnasiums.
The Greek free citizen’s enthusiasm for physical exercise was not in order to sell labor power more effectively, but as part of the pursuit of freedom. On the one hand, they pursued bodily fitness; on the other hand, they also enjoyed the pleasures of competitive play in athletic events such as the Olympic Games.
It is easier for us to understand the meaning of exercising the body: even if I were like a Greek free citizen, or if in the future I could, thanks to technological support, be spared the need to run around for a living, I would still need to exercise my body. Even if cars long ago surpassed the human body in speed, we still run. Maintaining bodily health is not so that one can serve as someone else’s tool, but so that one can live better oneself.
The Greeks believed that the human soul, like the body, also needs continuous “exercise” in order to avoid decay. The health of the soul is likewise not for working for others or contributing things to others, but for a better life. Therefore, when machines surpass the human mind in many respects, we still have reason to exercise our own minds.
The education of Greek free citizens had, at its deepest level, the purpose of bodily fitness or the health of the soul. And a well-rounded person possesses qualities such as “wisdom, courage, justice, and moderation.”
Ancient China was similar in this respect as well. When Confucius discussed “becoming a complete person through learning,” he said: “If one had the knowledge of Zang Wuzhong, the freedom from desire of Menggongcho, the courage of Bian Zhuangzi, and the skill of Ran Qiu, and refined them with ritual and music, one could also be considered a complete person. … Seeing profit, one should think of righteousness; seeing danger, one should be ready to give one’s life; after long association, one should not forget the words spoken in ordinary times. One could also be considered a complete person.” In short, this likewise includes wisdom (knowledge), bravery (courage), justice (righteousness), and moderation (freedom from desire), while additionally emphasizing talent (skill), civility (ritual), and aesthetic sensibility (music).
Granted, the sages of antiquity may have enjoyed the sustenance provided by slaves or by laborers at the bottom of society, and thus were able to abstain from production, unburdened by utilitarian thinking, pursuing higher tastes. But I believe that with the progress of technology, ultimately everyone will be liberated from forced labor and be able to think more freely about life and meaning.
Because human beings are weak, they are unique
Some say, “When human beings think, God laughs.” So if AI were to acquire even higher intelligence than human beings, would human philosophical reflection also become laughable in its face? The answer to this question depends on how we understand the meaning of “philosophy.”
“Philosophy” comes from Greek and is a combination of “love” and “wisdom.” The earliest philosophers used this term to distinguish themselves from the many “sophists” of the time who taught the art of debate, or what one might call “rhetoric.” What philosophers pursue is not those verbal tricks that help you sound dazzling in a speech, aggressive in a debate, or pretentiously bookish when dropping citations; what philosophers pursue is wisdom itself, not for the purpose of persuading others, but to satisfy their own desire to know.
The Delphic maxim says, “Human, know yourself,” and Socrates said, “The unexamined life is not worth living.” The ultimate object of philosophy is always each person oneself. We can neither substitute self-examination with examining others, nor substitute self-examination with being examined by others. Philosophical reflection cannot even be replaced by other human beings, let alone by AI.
Of course, human beings need to communicate with one another and draw inspiration and contrast from others’ words and deeds. In this sense, AI will certainly also become a good teacher and friend for human philosophical reflection, and we can indeed gain inspiration from AI. But these inspirations cannot replace communication between person and person, because other people’s lives and situations are always closer to one’s own and more likely to resonate.
And in this respect, AI can never replace human beings, not because it is too weak, but because it is too strong—transcending humanity’s inherent finitude and therefore never able to empathize with human beings.
The key point is that human beings have bodies. This is not to say that putting AI into a bionic robot counts as giving AI a body. For the human body not only provides the capacity for movement and operation; more importantly, it provides a kind of boundary or constraint. A person cannot detach themselves from their body, and their body is only ever “one”; its sensory capacities are limited, its reaction speed is limited, and it is inevitably subject to aging, sickness, and death. Mechanical limbs, by contrast, are not a constraint for AI, because AI is essentially a mass of data, and that data can be copied onto any digital medium. The same set of data can be implanted into this robot as well as that robot dog; it can have identical backups on countless computers in China and the United States at the same time; it can also be run simultaneously, modified into countless different versions, and even merged again at any time.
“To be, or not to be” — for human beings, “that is the question.” The question of life and death is philosophy’s eternal first question. But for AI, even if it too may have its own ontological questions, its situation is certainly very different from that of human beings.
Because I can only ever be “one,” I need to think carefully: should I dine out tonight or go home to eat? Tomorrow should I drink with Zhang San or go on a date with Li Si? Next year should I look for a job or take the graduate entrance exam? In the future should I return to my hometown and lie flat, or strike out overseas? In the end, should I live my life in peace and stability, or with thunderous momentum and grandeur? … Human beings make choices and decisions through “wisdom” and “taste,” fundamentally because of their own finitude. If I could be copied infinitely, could talk with Zhang San and Li Si at the same time, could be here and there simultaneously, would I still need to care so much about how to compare and choose? Or rather, could the concept of “I” still be formed?
Of course, in terms of storage capacity, transmission speed, and computing power, AI is still limited, so it is still possible to form some kind of concept of “self.” But this mode of life and human life are ultimately quite different, and human beings and AI find it difficult to empathize with one another on the most fundamental philosophical questions.
Against AI hallucinations
You may say: although AI, in essence, cannot empathize with human beings’ questions of life and death, it can imitate human beings and provide speeches and theories better than those of philosophers; if human beings want to learn philosophy, it should be enough to talk with AI.
And I think that perhaps in many fields AI can indeed become a good teacher, but at least in philosophy and other humanistic fields, AI still has a long way to go.
One flaw common to all major AIs at present is the so-called “hallucination” phenomenon: that is, AI may solemnly reel off a whole lot of statements that seem highly reliable, but in fact are just made-up nonsense.
This is because the current training methods for AI determine that it will always give answers that are more popular with ordinary people. Trainers, annotators, and ordinary users are all participating in the taming of AI, reinforcing the iterative paths that can produce better answers.
So today’s AI is rather like the “sophists” of ancient Greece: they were good at teaching debate skills, and their rhetoric was judged by how well it could move more listeners, rather than by the aim of pursuing truth. Thus, answers that are riddled with errors in the eyes of professional scholars, so long as they can bluff more ordinary people, will not receive correct feedback in version iterations and therefore cannot be gradually corrected.
Of course, in some specialized fields, especially in science and engineering, AI is more easily subjected to specialized training. For example, when ChatGPT first came out, it spoke with seeming authority on literature and art, yet was often mistaken when doing math and logic problems; some AIs could even not tell which was larger, 9.8 or 9.11. But after subsequent specialized reinforcement training, AI’s performance in these fields of mathematics, science, programming, and other science-and-engineering areas has already improved significantly.
On the one hand, this is because many scientists and engineers have participated more actively in AI development; on the other hand, it also benefits from the fact that these disciplines are relatively highly formalized—right is right and wrong is wrong, making it easy to form strict standards of judgment and provide rapid and accurate feedback for AI training.
The humanities are much more difficult. First of all, scholars in the humanities, especially philosophers, often have no public standard by which to judge things. Philosophers do not curry favor with the masses; on the contrary, they always have to stand apart and voice distinctive and sharp criticism. The greatest social responsibility of humanistic intellectuals is criticism: they dare to say things ordinary people do not like to hear. Many classical philosophers were even devoted to constructing complex systems of thought; taken out piecemeal, their words are often obscure or puzzling, and only by deeply studying the whole process of their thinking can one appreciate the profundity and brilliance of their thought.
And AI finds it difficult to develop the kind of personal independence and unity of thought that philosophers have, and it is also difficult for us to train it into such a state.
Of course we must take this trend seriously. We cannot, because training in this area is troublesome and difficult, simply stop responding and allow AI hallucinations to run rampant. First of all, the entire industry should encourage scholars in all fields, especially scholars in the humanities and social sciences, to participate actively in the work of training and correcting AI. This requires us to bridge the divide between the humanities and the sciences and eliminate prejudice.
The hierarchy of academic contempt is especially severe in China. Programmers from engineering backgrounds often look down on the humanities, thinking that the humanities are nothing more than being adept at empty talk, stock phrases, and repetitive platitudes (and this is not entirely unfair); while scholars in the humanities also often do not value the frontiers of technology, and poor scholars only know how to pad papers and make grandiose claims, while many excellent scholars also only know how to bury themselves in dusty old archives and seldom take the initiative to pay attention to emerging technologies. The urgent task at hand is still to promote dialogue between the humanities, the arts, and the technological fields, so as to jointly foster the development of technology and correct possible dangers.
In short, in the AI era that is now approaching, philosophers on the one hand must return to their original aspiration, transcend utilitarian attitudes, and, like the philosophers of ancient Greece, take “love of wisdom” as a way of life for exercising the soul. On the other hand, they must also keep an eye on the frontier, set aside divides and prejudices, engage in close exchange with the tech industry, and participate together in the process of technological evolution.
Translated from the Chinese original with AI assistance. The original text is authoritative.
Leave a Reply