AI’s New Advances
At the end of last year, at someone’s request, I wrote a piece, Will AI Restructure the History of Science?. That happened to coincide with the release of gemini 3.0, and I thought it marked a watershed, the point at which AI moved from the level of a student to the level of a professional researcher. Less than half a year later, the AI tool landscape has once again shifted dramatically; gemini long ago ceased to be leading, and claude and chatgpt have reclaimed the lead in capability. More importantly, claude’s cowork and chatgpt’s codex have both undergone qualitative leaps forward too (perhaps spurred on by openclaw), and both have already broken out of the category of “programming tools,” becoming all-purpose work assistants capable of comprehensively controlling a computer and autonomously carrying out all kinds of tasks, including humanities research.
I have already begun using AI to help me write technology history. At first I used claude Opus 4.6 cowork to generate a draft, then I manually revised one chapter to show my orientation, and then I used chatgpt 5.5 codex to continue revising and expanding it (that is the step I am at now). Finally, I expand and revise it myself, and then I hand it back to AI for proofreading and polishing. I fed both AIs all my materials, including every blog post and published paper I have written, as well as my course PPTs, plus all the e-book resources I have collected. Anyway, I think the writing effect is very good; in many respects it writes better than I do. Of course, I still need to keep making substantial revisions, but in many cases the reason I revise is actually just to make the manuscript better fit my own train of thought, and not necessarily to make it better suited to readers’ tastes. I think that the 200,000 characters AI can write in a single day (depending on how much I pay; if I spend more, it can be even faster) are already at a level where they can be published directly, and they are better than most crude, shoddily made books on the market.
My use of AI is still only at the most elementary beginner level: I use only the official applications provided by mainstream large models, and I have not enabled full-control permissions. I have not used agent architectures like openclaw or hermes, nor have I used literature-management applications like Zotero or Obsidian, nor have I downloaded skill.md files developed by all kinds of netizens. My student is far ahead of me in this regard: he not only used a large number of AI tools to assist him in writing his dissertation, but also tried to summarize the workflow and form a skill template suitable for scholars.
Of course, I can still offer guidance on his dissertation, and there are indeed quite a few flaws and loopholes caused by reliance on AI. But I would not therefore draw a conclusion like “reduce AI use.” On the contrary, we believe this means that the student and AI still need further, conscious, reflective, and deeply integrated磨合. My attitude toward AI is actually quite similar to my attitude toward students: I should pay attention to his mistakes, but I need not condemn him outright; rather, I should communicate properly. What can be improved should be improved next time; what cannot be improved should be accommodated as best as possible. Make the best use of strengths and avoid weaknesses; bring myself into the collaboration or introduce more collaborators, so that after our complementarity the topics we complete together through cooperation can avoid these mistakes.
Refusing to Cooperate with AI Is Academic Corruption
Putting my hand on my heart, if a student can reach such a level—able to translate materials so comprehensively and meticulously, organize literature, and even write papers and monographs—then even setting aside the fact that his working speed is equivalent to thousands of times that of an ordinary student, even just from his mastery of existing academic resources, from the logic and organization in the way he answers questions, from the results of completing the tasks I assign, from his eagerness to promptly revise according to my suggestions, or from the benefits he can bring me—if a student at such a level came to apply for my graduate program, could I possibly refuse him? Could I refuse him? Would I dare refuse him?
If my research group were to turn away such an excellent member, or a whole group of them, and instead recruit more less capable “well-connected” people, who merely through blood ties (they are all carbon-based life forms, after all) obtain my attention and promotion, then what do you call that? That is典型的 academic corruption! Of course, the responsibility to teach and cultivate people is still there, and of course I still have to take in human students. But if one is thinking in terms of doing research and carrying out projects, then how can an upright advisor refuse to use AI?
As a researcher, one has the responsibility to grasp the most cutting-edge and the broadest relevant information as far as possible; this is an intrinsic requirement of academic research. For example, if an old scholar says, “I only read paper-based materials and never electronic materials,” is that acceptable? Fine, but then he must print out the necessary electronic materials and read them. If he says, “I only know how to look for materials in the library; I can’t search for materials on CNKI or Google Scholar,” is that acceptable? In a pinch, maybe, but then he must have students help search for and download the materials. But if this old scholar not only cannot look for electronic materials, but even looks down on and rejects students using electronic search tools, then one must say that this scholar is unqualified. He should have stepped down from the front lines of research long ago.
A few years ago, Zhai Tianlin was widely questioned for being exposed in a livestream as not knowing what “CNKI” was, and his plagiarism was dug up. Of course, his fatal mistake was plagiarism and theft, but “not knowing CNKI” was undoubtedly an obvious flaw; everyone knew that this should not be the case. Why? Isn’t CNKI just for searching Chinese journal articles? Can’t I just subscribe to the paper editions and read those? — When a scholar is doing ordinary reading and accumulation, relying on printed books and journals is of course fine, but once he needs to write a paper in a certain field, he needs to comprehensively search the relevant literature, and at that point he of course must use electronic search engines. Anyone who has used a search engine will not believe that anyone can do the same thing in a physical library.
The same logic applies to AI: AI is now replacing the position of search engines. Of course it still cannot replace Google Scholar, but it can already provide effective supplementation. And of course it goes far beyond that; it can play a role throughout the entire process of academic research, from conception to review and proofreading.
Traditional Learning Methods Also Depend on Technology
Many people resist AI or take pride in not using AI, and they are not without reason. Many studies have shown that relying on AI may improve creativity in the short term, but it also tends to produce homogenization of thought, and once AI is removed, creativity declines; if you rely on AI assistance when learning certain knowledge, you may feel that AI is useful while learning, but when tested again a month and a half later, you find that those who learned without relying on AI have retained it more firmly.
Since there is all this ironclad evidence, why still use AI? But if we look closely, what are those so-called non-AI control groups using? Courseware, databases, and ordinary search engines. The so-called “traditional learning methods” are not really all that “traditional” either; they are also tools that became popular only within the past twenty years.
And over the past twenty years, there have also been many comparative studies between these “traditional learning methods” and “more traditional learning methods.” For example, in 2013 and 2014 there were some famous studies (just ask AI at random; I won’t list them specifically). The research showed that taking notes by hand is more helpful for remembering knowledge concepts than typing on a keyboard, and using a computer in class lowers learning ability.
There has long been a special term for this phenomenon: “cognitive offloading.” It means that when we “outsource” certain links in cognitive activity to external tools, on the one hand we lighten the brain’s burden and make cognitive activity more efficient, while on the other hand we create mental laziness and harm memory and critical thinking.
The earliest discussion of this kind of phenomenon can be traced back to Plato and Zhuangzi. Plato cited a fable to point out that writing harms memory, meaning that people who rely on pen and paper become lazy about memorizing large amounts of information directly in their heads, thereby causing their memory capacity to decline. Zhuangzi, meanwhile, told the story of those “with mechanical hearts,” pointing out that relying on labor-saving machines does not merely make one lazy; it can even corrupt one’s morality.
But that is the essence of technology; it is the human destiny. The development of all technologies comes at a cost—they are “extensions of man,” and so they always expand human capacities while also making humans dependent. The memory of those who rely on pen and paper is irreversibly damaged; people who are used to farming lose the ability to hunt; those accustomed to industrial consumer goods lose the ability to make things by hand; those accustomed to navigation lose the ability to find their way……this is the cost of technology.
Many people insist on saying that “once the new technology is removed” you become incapable, but the question is: why would the new technology be removed in the first place? They will not disappear again—unless they are replaced by newer technologies. That is humanity’s fate.
So-called survival of the fittest means fitting the environment. Bats adapted to dark cave environments, and their eyesight degenerated; whales adapted to the marine environment, and their limbs degenerated……Evolution has no fixed direction, and any improvement in ability is not necessarily progress. The key lies in how one adapts to the environment and its changes. If the environment changes, and you then turn to mammals and say, “Remove those angiosperms; go back to the Jurassic and you won’t be able to adapt, you’re not as good as the dinosaurs,” so what? The times have changed; no matter how strong dinosaurs were, they were still going extinct.
And the human living environment is shaped mainly by technology. In the beginning, those who knew how to use stone tools were superior to hominids who were physically stronger. Nature would never ask, “If we remove stone tools and place humans and gorillas side by side, whose survival ability is stronger?” There is no such if. In fact, human beings used stone tools to hack through obstacles and thorns, and ultimately stood at the top of the food chain. After the Agricultural Revolution, nature would also never ask, “If we do not rely on grain and place a group of Sumerians and a group of hunting tribes in the wild jungle, which group survives longer?” In the industrial age, nature would also never ask, “If we remove precision machine tools and production lines and rely on workers to do everything by hand, whose manufacturing ability is stronger?” Nature is cruel; evolution is cruel. As long as you adapt well to the environment, no matter what degenerations you have, no matter how ugly and distorted you become, you can still survive.
Safeguarding Diversity
The dependence brought about by AI is not a problem, because we no longer need to consider an “environment without AI.” But the homogenization of thought caused by AI, or rather the loss of diversity and individuality, is an issue that needs to be taken seriously. When I wrote Thinking About AI earlier, the issue I cared about most, and have always cared about, was precisely diversity.
Of course, one must first recognize that the loss of diversity is also a kind of destiny inherent in technological development. We can see that writing technology accelerates the loss of linguistic diversity: on one small island in Vanuatu, 25,000 people possess 30 indigenous languages, and the spread of writing and education will obviously cause linguistic homogenization. The more classical an era is, the more pronounced the differences in cities and architecture among different cultures; the more modernized a metropolis becomes, the more its architectural styles tend to converge.
But the apparent reduction of diversity may, within an internal space, buy us richer diversity; civilization often develops in an “implosive” form. For example, while written language certainly reduces diversity at the level of spoken languages, it increases the diversity of literary works within a single language. Vanuatu has more than a hundred languages, but how many epics and bodies of knowledge preserved in those indigenous languages can compare with even a few libraries in English or Chinese? The cement forest of a modern city may look monotonously uniform, but the occupational forms and ways of life and entertainment it contains are probably a hundred times richer than what the ordinary inhabitants of a typical ancient city could access. Go looks more monotonous and dull than dou shou qi, but only when one goes deep inside does one discover its myriad changes.
Therefore, when discussing the loss of diversity, we need to proceed with caution: apparent homogenization may be in order to build a new “platform,” filtering out noise and providing common basic resources, on which we can more freely create new diversity.
It is hard to say whether a new technology is simply destroying diversity or absolutely increasing it; in fact, as with writing and architecture, many technologies exhibit both tendencies. The Internet and AI are probably no exception. How they will ultimately develop depends on our choices and efforts.
The diffusion of new technologies is almost an irresistible trend, difficult to reverse. But we can intervene actively in many ways to help promote diversity. For example, on the level at which diversity is disappearing, it is certainly hard to resist, but we can use translation to allow vanishing diversity to continue in another form. For instance, after writing became widespread, many traditional spoken languages and dialects came close to disappearing; yet through writing we can record these vanishing languages, preserving as much as possible their distinctive linguistic features and systems of knowledge. Of course, some ancient languages can also borrow writing technology to develop to a new stage. Admittedly, oral cultures preserved in written form can no longer continue many of their original qualities; this is a translation rather than a direct continuation, but these old forms of wisdom are also able to break out of their old boundaries, collide and resonate with more cultures around the world, and thereby stimulate richer diversity on a new platform.
Exploring New Environments Is the Scholar’s Mission
And in the new spaces opened up by new technologies, the implosion of diversity will not happen automatically; it requires people to explore and create it.
Although Plato criticized writing and believed that truth could not rely on books but required direct intuition with the mind’s eye, he also actively participated in writing practices and opened up the meaning of writing. Plato’s “expulsion” of the poets and his advocacy of the theory of Forms were all ways of developing a culture of writing. Plato’s dialogues, moreover, recorded Socrates’ words and deeds without resorting to writing, and became immortal classics.
Heidegger said that when facing technology one should adopt an attitude of “releasement” — a stance that neither blindly rejects nor mindlessly submits. In a sense, releasement means going with the flow, first taking technology as part of the “natural environment” — human beings cannot violate nature, but neither are they powerless before it. Floodwaters surge irresistibly, but one can channel them. Brave human beings can even ride the crest of the waves as “wave-riders.” Humans and environments are always in collisions of mutual adjustment and mutual transformation. There is always a need for some people to step into frontier fields filled with the unknown and explore the boundaries of the environment.
Mumford once satirized those romanticists who advocated escaping technology and returning to primitivism: “They recommend returning to the living conditions of the original pioneers, yet they do not possess the spiritual strength of those pioneers of the time.” For the primitives to which they wish to return were by no means content with the status quo either; they too struggled and took risks, bravely opening up their own frontiers. In ancient times, human beings crossed the Bering Strait on foot, drifted among Pacific islands in crude dugout canoes, and could even migrate from Southeast Asia to Madagascar in dugout canoes. Mumford asked: “If this is humanity’s attitude toward nature, then why is humanity so cowardly when faced with machine systems?”
In the face of the development of AI technology, we should also adopt the same attitude toward nature: acknowledge its existence, revere its power, study its laws, explore its boundaries, tame its wildness, and ultimately transform it into part of our ideal home.
Who should stand at the forefront of the era to explore and expand the boundaries of AI? Entrepreneurs and engineers are of course the pioneers of the technological age, but what about humanities scholars? Can humanities scholars calmly lie atop heaps of old papers, indifferent to the changing winds and clouds of the times?
Admittedly, the character of humanistic scholarship is conservative, and one of our missions is to safeguard the depths of human civilization, what is called inheriting the almost lost learning of the sages of old. But the reason it is worth “inheriting the almost lost learning of the sages of old” is also in order to “bring great peace to the world for all generations”! And what were those “sages of old” in the heaps of old papers themselves like? Confucius and Mencius traveled among the states, persuading feudal lords and criticizing contemporary ills; Socrates debated politicians and served as the gadfly of Athens; Plato made three journeys to Sicily to try to put the philosopher-king into practice; Aristotle instructed Alexander the Great; Hume served as a diplomat; Kant intervened in Enlightenment debates and discussed the future of law and political order; Hegel paid attention to the French Revolution and reflected on the order of the modern state; Marx need not be mentioned; Russell actively opposed war and nuclear weapons; Sartre took part in student आंदोलनों; Arendt关注 Nazi trials; Wittgenstein served in the army, worked as an elementary school teacher, and did manual labor as a porter; Heidegger also took part in pro-Nazi political movements (though a stain, it still counts as an act of active engagement with the world) … In short, modern humanities scholars comfortably hide in their studies and research those “sages and worthies of old,” but they forget that those sages and worthies were not confined to studies; most of them paid attention to current affairs, threw themselves into the secular world, and worried over the changes of the age and the fate of humanity.
If humanities scholars truly inherit the spirit of those sages and worthies of old, then they should make concern for and participation in the transformation of the times one of their missions. This is of course not to say that we should drift with the tide, but that we need to go one step further, to step onto the crest of the wave and walk at the very forefront of change. Humanities scholars should not wait until technological iterations have settled into dust, and most ordinary consumers have already become familiar and accustomed to them, before belatedly adopting new technologies. On the contrary, we should join the exploration while new technologies are still emerging and their direction of development is still unclear. Only then can we intervene in and guide technological development with the concepts and values we uphold, while also setting an example for the wider public in how to face these new technologies.
Specifically, working together with AI, exploring a path of mutual reinforcement between humans and AI, and while bringing into play AI’s efficiency and diversity, preserving human individuality and capacity for reflection — this is the scholar’s responsibility.
Anxiety Is Better Than Numbness
Many people are spreading something like an AI-anxiety syndrome: everyone is worried about being replaced by AI. But I think that in today’s world, if you are not anxious about AI, that is what is abnormal; you are far too numb. When such earthshaking shocks arrive and the environment we once knew has been completely transformed, how can you not feel tense? If you do not, then something is surely wrong with your cognition or your sensibility. Anxiety is a positive emotion: it shatters the illusion that you can simply remain content with the status quo and forces you to take action. Of course, excessive anxiety — flailing about like a headless fly — is certainly not good. But to say that you can stand apart from the AI tide, serene and unconcerned, is by no means anything to be proud of.
All kinds of work are facing the threat of AI, and knowledge workers face even greater pressure; the more one is a white-collar elite, the more easily one is replaced by AI. So why should scholars alone remain untouched? Is it because they think they have iron rice bowls and therefore can remain calm and unruffled? If so, this is what I mean by corruption. Relying on rigid authority and status, some scholars think the AI revolution has nothing to do with them, and even strike an air of transcendence, putting on a show of being above the world. This is a very ugly phenomenon.
Translated from the Chinese original with AI assistance. The original text is authoritative.
Leave a Reply