Hu Yilin on AI Writing, Authorial Responsibility, and the Academic Paper System
Introduction
After science fiction writer Hao Jingfang spoke about AI’s participation in novel writing, a dispute broke out over whether “AI writing should be labeled” and whether “AI as the chief writer still counts as creation.” What such arguments really touch is not how much AI a particular writer used, but how authors, readers, and the academic community should rethink creation, responsibility, and evaluation once AI has entered the formal writing process.
Hu Yilin’s position is very clear: he not only supports AI-assisted writing, but also AI-led writing. Human beings can in the future completely become second authors, editors, question setters, or planners. The issue is not who typed out the text word for word, but who posed the questions, organized the materials, judged the quality, and takes responsibility for the final work.
He has already discussed AI writing and educational institutions several times on his blog. This article mainly presents further views from that interview.
Whether to Label AI Should Not Become a Moral Tribunal
When it comes to AI creation, the most common demand is “it must be labeled.” Hu Yilin does not object to disclosure, but he opposes treating disclosure as a universal moral obligation.
His principle is: whether to label it first depends on the rules of the publishing platform, journal, or press. If the platform requires labeling, then label it; if there is no requirement, the author may decide for themselves.
“I tend to be frank and honest; one should not lie,” he said. “If you clearly used AI and insist you didn’t, that of course is wrong. But I also don’t think one must proactively disclose it.”
In his view, using AI is not a stain; it does not need to be reported like a confession. The real issues are honesty and contract. If a publisher markets a book as “ten years of painstaking work” and “purely hand-crafted,” when in fact much of it was written by AI, then the problem is not AI itself, but false advertising.
In other words, the boundary should not be drawn between “used AI” and “didn’t use AI,” but between “lied” and “misled readers.”
He even believes that as AI becomes the default writing environment, in the future it may be “purely human creation” that will need to be specially labeled. Just as writers today do not specially declare that they used search engines, electronic dictionaries, databases, and typesetting software, AI too will gradually shift from a special tool to infrastructure.
Readers Buy the Work, Not the Author’s Sweat
In the Hao Jingfang incident, one typical reader emotion was: I bought a human writer’s painstaking effort, not a machine-generated product. Hu Yilin offered a sharp analogy in response.
“You buy grain so you can eat your fill and eat well, not so you can obtain the crystallized sweat of the farmer. If it doesn’t taste good, you say it doesn’t taste good—what does that have to do with whether the farmer sweated or not? The farmer got on a tractor and isn’t tired anymore; does that mean you stop eating? The same logic applies to spiritual food. Whether something is good to read or not is one thing; whether it contains painstaking effort or not is another; there’s no need to drag these things together arbitrarily.”
This line of reasoning shifts evaluation of creation from “labor ethics” back to “work quality.” Readers of course may criticize an AI-involved novel as unreadable, clichéd, false, or aesthetically impoverished; but if the work itself stands, then to deny its value merely because the author sweated less and used a more efficient tool is, in his view, to mistake suffering for the source of value.
That said, he does not deny that literary and intellectual products involve “consumption of personality.” When readers read a certain writer, they are often also encountering a certain experience, a certain eye, and a certain spirit. But Hu Yilin believes that what really matters in this kind of personality consumption is not how much physical labor the author expended, but their life experience, taste, and judgment.
“When it comes to consuming personality, what generally should matter is a person’s experience and taste, not their sweat,” he said. “We can reasonably believe that someone with insight and taste, working with AI, can produce works of greater taste. In an era when everyone can use AI, this kind of individual taste will instead become even more important.”
Once everyone can call on models, the model itself no longer constitutes a scarce resource. What is scarce is what questions one asks, what materials one chooses, what expressions one preserves, and what mediocrity one refuses.
AI Can Lead, but Humans Must Be Responsible
If AI is the chief writer, can a human still sign as “author”? Hu Yilin’s answer is yes.
He believes that as long as a human plays substantive roles in the work—such as second author, editor, planner, or question setter—they can still sign as “written by so-and-so.” If the work is closer to material compilation, selection, or anthology, then it can be signed as “compiled by” or simply “edited by.”
The core of authorship is no longer “personally writing every sentence,” but “bearing judgment and responsibility for the final work.”
“Responsibility, of course, is borne by people; that’s one of the rare irreplaceable things about human beings,” he said.
This also means that the author cannot push mistakes onto AI. AI can become a powerful external organ for writers, researchers, and editors, but it cannot become a black hole for responsibility. Those who sign their names enjoy the reputation, income, and influence brought by the work, and must also bear responsibility for factual errors, hallucinated citations, plagiarism-like formulations, and aesthetic failure.
“Written by AI” is not a reason for exemption. Just as traditional authors cannot push errors onto secretaries, research assistants, or editors, signatory authors in the AI era cannot use the internal production process as a shield.
To Evaluate an Article, One Should Look at the Article Itself
In academia, AI writing is often discussed within the framework of “ghostwriting.” Hu Yilin believes that this framework confuses evaluation of text with evaluation of behavior.
“Reviewing an AI paper and reviewing a human paper should follow the same rules: judge the matter itself, not the person. As for ‘ghostwriting,’ that does not belong to evaluating the text in the first place, but to evaluating a person’s behavior,” he said. “For example, if Zhang San stole my gold, then of course he is at fault, but the purity of the gold itself has nothing to do with whether it was stolen or not. Since AI writing does not involve theft, it is irrelevant to evaluation. To evaluate an article, one should start from the article itself.”
This is not to say that academic misconduct does not exist. Fabricating data, inventing citations, concealing experiments, lack of verifiability, lack of reproducibility—these are still serious problems. But whether a paper was drafted by AI does not automatically constitute a negative proof of its quality.
In his view, universities’ detection of so-called AIGC rates is essentially institutional laziness. It tries to replace genuine academic judgment with technical indicators, yet it can neither reliably identify the extent of AI participation nor answer whether the text makes an original contribution, whether the evidence is solid, or whether the argument holds. For related views, see his earlier articles on AI writing and the university paper system.
The Traditional Paper Format Has Reached the Time for Reconstruction
Hu Yilin further believes that AI will not only change the way writing is done, but also the form in which research results are published.
He mentioned the paper The Last Human-Written Paper, coauthored by 37 authors from Stanford, CMU, Michigan, and other institutions. In his view, the importance of this paper does not lie in discussing “whether AI can be used in papers,” but in more fundamentally questioning whether static papers are still suitable for research communication in the age of agents.
Hu Yilin believes that the traditional format of the research paper was originally designed for the age of print journals. In the age of AI and open-source collaboration, research results should be published more like GitHub projects: including the entire process log, version history, data, code, failed paths, model calls, and peer discussions. How AI participated need not be explained in a separate statement, because it will be naturally recorded in the research process.
This will also change the focus of peer review. Reviewers should not only look at a beautifully structured PDF with polished rhetoric, but should check whether the research process is traceable, whether the data are real, whether the conclusions are reproducible, whether failed experiments have been hidden, and whether the code and materials can withstand continuation by others.
He said that peer review should have emphasized reproducibility in the first place. In reality, even top journals do not always prevent “obviously fake” data from entering the publication system. This shows that the problem is not AI, but that the existing evaluation mechanism itself is already in need of reform.
AI Does Not Necessarily Exacerbate Academic Inequality
Some worry that if papers become open research repositories, academic evaluation will become more engineering-like and more platform-based. Those who understand GitHub, agent workflows, and log management will have an advantage over traditional text writers.
Hu Yilin does not deny that a new system will bring new skill requirements, but he believes one cannot imagine traditional academia as a world that is equal, pure, and low-threshold.
“Does traditional academia not require management?” he asked rhetorically. Scholars control resources through academic status, possess large numbers of students and assistants; a few people have expensive laboratories and databases; academic conferences are often also spaces for cultivating connections; and academic leaders can even influence or control the direction of trends.
By comparison, although AI access also requires money and experience, the barriers are not especially high. Compared with laboratories, databases, student labor, international conference networks, and academic cliques, AI tools are in a certain sense more widely accessible.
AI will not automatically bring fairness, but it may weaken some of the old monopolies on resources. In the past, young researchers lacked assistants, editors, and the ability to organize materials; today, at least, they can obtain basic collaborative capacity through AI. Real competition will still return to problem awareness, judgment, taste, and responsibility.
Conclusion: From Personally Creating to Responsibly Creating
Hu Yilin’s view of AI writing can ultimately be summed up as a redistribution of values.
He does not regard AI as an external threat to literature, scholarship, or education; rather, he sees it as a change in the very environment of knowledge production. The old problems do not disappear because of this: false advertising is still false advertising, forged data is still forged data, bad work is still bad work, and academic gatekeeping may still exist in new forms.
But AI forces people to stop using the crude standard of “whether it was done by one’s own hand” to evaluate a work. Human value has not disappeared; it has shifted from handcraft-style text production to the parts that are harder to replace: asking questions, organizing materials, judging quality, shaping taste, and bearing responsibility.
Having AI as the main writer does not mean the death of the author. On the contrary, it may force the author, for the first time, to answer seriously: if it is not because you typed out every word by hand, then what exactly did you contribute to this work?
Hu Yilin’s answer is: the contribution is not in sweat, but in vision; not in keyboard labor, but in responsibility.
Translated from the Chinese original with AI assistance. The original text is authoritative.
Leave a Reply