The Academic Assembly Line

1What is wrong with research becoming a job?#

Let me be clear at the outset: this essay is not an exercise in nostalgia. Professionalization gave science stable funding, systematic training and a community that hands itself on from generation to generation, and most of what modern science has achieved rests on that system. There is nothing wrong with scientists drawing salaries, and nothing shameful in chasing fame and fortune. In the real history of science, the great figures mostly moved in wide circles and fought for recognition within their community; they were less lonely fighters than ambitious, sociable operators who could not bear obscurity and knew how to work every side.¹ Nor is there any going back from Big Science. Science is now bound up with capital and politics in a complicated “complex”, and only by facing that reality squarely can we hope to keep the spirit of free inquiry alive in the marketplace of reputation and reward.²

The trouble lies elsewhere. Learning for pleasure is perfectly natural and deeply human: small children delight in learning to walk and to talk. Learning for marks, grade-point averages, certificates and jobs is the alienated form.³ The same holds for research. If you study a problem because it is interesting, with fame and fortune as incidental rewards, you are doing science; if you study a problem only because it will bring the next paper, the next grant, the next evaluation, inquiry has become a way of earning a living. People of the first kind also weigh gains and losses, and people of the second kind are not necessarily less clever. The difference is whether the research itself is still the point.

I have lived inside this system myself. When I joined the faculty of Tsinghua University in 2017, I signed a tenure-track (pre-tenure) contract: “renewed every three years, up-or-out at the end – quite a lot of pressure”.⁴ In 2024 I gave up my post and became an independent scholar.⁵ What follows is observation, but it also draws a little on experience.

2When did being a scientist become a job?#

“Scientist” is a young word in English. At the 1833 meeting of the British Association for the Advancement of Science, someone proposed coining it on the analogy of “artist” – and that someone was in fact William Whewell himself. The following year he recorded the episode in a review of Mrs Somerville’s new book.⁶ Before then, nature was studied mostly by aristocrats, clergymen, physicians and amateurs of every kind, who lived on family fortunes, church livings or other professions; research was something they pursued in their leisure (see the age of the amateur). Afterwards, the German-style research university joined research to teaching; the seminar, the laboratory and the doctorate became the standard route for training researchers, and science gradually turned into a profession with a way in, a ladder and a salary (see papers and the professionalization of the scientist).

On 7 November 1917 Max Weber addressed students in Munich on “Science as a Vocation”; the lecture was published in 1919.⁷ The German word Beruf means both a “calling” and an “occupation”, yet Weber began with the cold realities of the second. Whether a Privatdozent (an unsalaried lecturer), let alone an assistant, will ever rise to a full professorship, or even to the directorship of an institute, is, he said, “simply a gamble”; “I know of hardly any career on earth in which chance plays such a role.” And so: “academic life is a wild gamble.”⁸

Weber also saw scholarship being industrialized. German universities, he said, were developing in the American direction, and the great institutes of medicine and natural science had become “state-capitalist enterprises”. As in every capitalist enterprise, the worker had been separated from the means of production: the assistant depended on equipment provided by the state, and on the institute’s director, much as a factory employee depends on the boss; his position was often as precarious as that of any quasi-proletarian. Weber granted that the technical advantages of such operations, “at once capitalist and bureaucratized”, were beyond doubt.⁹ He granted, too, that science had entered a stage of specialization never known before, and would stay there for all time to come; today a truly definitive, first-rate achievement is always a specialist’s achievement.⁹

Yet in the same lecture Weber insisted on something else: passion. Anyone who cannot put on blinkers, so to speak, and lose himself in the idea that the fate of his soul depends on whether he gets this one conjecture right at this point in this manuscript had better keep away from scholarship. Without this strange intoxication, laughed at by every outsider, without this passion, a person has no calling for scholarship and should do something else, “for nothing is worth anything to a human being, as a human being, that he cannot do with passion”.⁹ He took issue with a view then common among the young: that science had become a sum in arithmetic, to be manufactured by the cool intellect alone, in laboratories or statistical card indexes, as in a factory. Inspiration cannot be forced, and it has nothing to do with cold calculation.⁹

More than a century ago, then, Weber laid this contradiction on the table: professionalization and specialization are our fate, but passion and inspiration are the soul of science. The history since has been the story of the first pair’s triumphant march.

3How was research broken down into an assembly line?#

Behind professionalization lies the logic of industrial society as a whole. Technology has always tended towards specialization, and since the Industrial Revolution the division into specialities and the assembly line have become the underlying logic of the entire technical system. The principle of the assembly line is simple. A crew of all-round craftsmen who work because the work interests them is no match for a crew of dull, rigid workers each drilled in a single motion. The assembly line itself shuts out creativity: if you are told to give a screw one and a half turns, you give it one and a half turns; if you are clever enough to give it an extra half-turn on your own initiative, you only cause trouble.³ The heart of industrial production is copying, and it requires people, too, to be copyable and interchangeable like parts. Posts are interchangeable: when one person is laid off, another can step in. Schools, in turn, became places for turning out “copies” in bulk.¹⁰

In the twentieth century science was drawn into the same logic. In the 1930s Ernest Lawrence built cyclotrons at Berkeley and pioneered a new way of doing research: “Big Science”. The cyclotron was complex in every dimension. Theorists and experimenters had to join forces to make any headway; it was so expensive that the old university system could not possibly support it; and its daily upkeep required a great deal of tedious labour. Most of that work demanded neither lofty scientific ideals nor rich creativity – it was little more than mechanical drudgery – and the people now jokingly known in China as “research labourers” (literally “research migrant workers”, hired hands for routine lab and data work) had already appeared.² During the Second World War, Oak Ridge used Lawrence’s device to separate uranium isotopes, and the Tennessee Eastman company recruited some ten thousand young local women to watch the meters. Kenneth Nichols, who was in charge of the project, recalled comparing the production figures of the various units and pointing out to Lawrence that the young “hillbilly” girl operators were outproducing his scientists. Lawrence said his men were experimenting with ways to improve operations, so Nichols challenged him to a production race between his PhDs and the local girls. Lawrence agreed – and lost. The girls won, Nichols wrote, because they were “trained like soldiers ‘to do or not to do—not to reason why’”, whereas “the scientists could not refrain from time-consuming investigation of the cause of even minor fluctuations of the dials”.¹¹ The assembly line has no need of curiosity; on the assembly line, curiosity is even a nuisance.

Big Science succeeded, but at a price. Lawrence himself came to reflect on the culture of his laboratory. Putting the cyclotron’s performance first had bred sloppiness and carelessness; the laboratory missed many discoveries that its cyclotrons could easily have made, and fell behind European colleagues with far humbler equipment.² Too much attention to technical detail and practical results got in the way of the vision and insight that research needs.

So the assembly line entered the laboratory, and it stayed. Paula Stephan, an economist who has studied the economics of science for many years, observes that in the United States university labs are “overwhelmingly staffed by graduate students and postdocs”: they are flexible, they are temporary, and they are cheap. This pyramid system (“some would call it a scheme”, she adds) only works if research funding keeps growing.¹² A research group is like a small factory. The professor is both factory manager and salesman, forever applying for grants and publishing papers to keep the team fed; the graduate students and postdocs are the skilled hands, each completing one module after another along a finely divided production line. This division of labour is extremely efficient, and it also puts many people, from their first day in the trade, in a position from which they can see nothing but their own step in the process (see the triumph and cost of specialization).

4Why does more knowledge make people narrower?#

Specialization has a deeper source as well: knowledge itself is accumulating at speed. At the opening of Little Science, Big Science (1963), Derek de Solla Price wrote that “using any reasonable definition of a scientist, we can say that 80 to 90 percent of all the scientists that have ever lived are alive now”; by his estimate, science doubles roughly every fifteen years.¹³ The more knowledge there is, the heavier the load a person must shoulder before reaching the frontier.

The economist Benjamin Jones calls this “the burden of knowledge”. Analysing more than two million US patents from 1975 to 1999, he found that the average size of inventing teams rose from 1.73 to 2.33 people, an increase of 35%; that among inventors who first appeared after 1985, aged between 25 and 35 at the time, the age at first invention rose by about 0.66 years per decade; and that solo inventors found it ever harder to move to another field, with specialization rising by about 6% per decade.¹⁴ He titled the paper “The Burden of Knowledge and the ‘Death of the Renaissance Man’”. As the frontier recedes, each person can only train for longer and longer in a smaller and smaller patch, and then team up with more and more others.

The “Renaissance man” is the “all-sided man” (l’uomo universale) described by Jacob Burckhardt, who saw him as a phenomenon peculiar to Renaissance Italy.¹⁵ Scholars in the age of print could still dream of surveying every book. In the sixteenth century the Swiss scholar Conrad Gessner compiled his Bibliotheca universalis, which set out to list every work in Latin, Greek and Hebrew published in the century since the invention of printing. After that the number of books grew exponentially, and no one could ever again compile a bibliography of everything. After the Enlightenment, scholarship entered its age of specialization: science became a “learning of separate branches”, scholars explored ever narrower fields, and it grew ever harder to bring all human knowledge together into one.¹⁶ Neighbouring fields might as well be separated by mountains; like the builders at the foot of Babel, we no longer understand one another’s speech (see why interdisciplinarity matters).

Will the trend go on for ever? Jones’s argument rests on a hidden premise: that the burden of knowledge can only be carried by people, in their own heads. This is exactly the premise AI changes. It has long been impossible for any person to survey all the world’s books; AI really can. Digitized books can not only be gathered in one place but also read through and synthesized, and in breadth of reference and cross-disciplinary synthesis AI already outdoes the very best scholars.¹⁶ If machines can share the load, the “death of the Renaissance man” need not be our fate (see why a crisis is also a turning point).

5Where does the pressure to publish come from?#

The origins of the phrase “publish or perish” are unclear. Some trace it to the early twentieth century, when a geographer used it to urge his learned society to publish more; by the 1920s and 1930s scholars were using it in discussions of promotion at American universities; and in 1942 the sociologist Logan Wilson used it in his book The Academic Man, which was long taken to be its earliest source.¹⁷ At first it even carried a note of encouragement; only later did it become the sword hanging over every young scholar’s head.

Where the pressure comes from is not hard to see. Papers are counted, and so they have become the currency of an academic career: promotions, grants, bonuses and talent “hats” (honorific titles) are all converted at a rate set by the number of papers and the rank of the journals (see metrics and paper-counting). In 1981 William Broad, a reporter for Science, wrote an article called “The Publishing Game” about a notion current in some circles, the “least publishable unit”: a euphemism for slicing data thin, four short papers where one long one would do.¹⁸

The most extreme case is the “hyperprolific” author. In 2018 John Ioannidis, Richard Klavans and Kevin Boyack searched the Scopus database for authors who, in at least one calendar year between 2000 and 2016, had published more than 72 full papers – one every five days on average. They found more than 9,000. Of these, 86% were in physics, where large collaborations list every member as an author, and so were excluded; once indistinguishable names and erroneous records had also been removed, 265 remained. The number of such authors grew about twentyfold between 2001 and 2014, while the total number of authors grew only about 2.5-fold. The researchers sent a questionnaire to the 81 hyperprolific authors of 2016. Twenty-seven completed it, and 19 of them admitted that for more than a quarter of their papers they had failed to meet at least one of the internationally accepted criteria for authorship. There is evidence, the authors wrote, that the rising number of authors per paper owes less to the real needs of team science than to the pressure to “publish or perish”; they also stressed that they had no evidence that these authors had done anything improper.¹⁹

In China, the pressure to publish takes forms of its own. Research grants and projects impose publication requirements on those who take part, and sometimes a paper counts as an output only if it carries particular names; students must publish in order to graduate and to find jobs; and many journals discriminate by status, so that authors from elite universities or with senior titles find it easier to get published, while students with lesser qualifications or from weaker institutions find it harder.²⁰ My own position is that there is no need to bow too low to such realities, nor to reject them too fiercely: provided integrity is preserved and principles upheld, a small compromise does no great harm.²⁰ But when compromise becomes the norm, publication turns from communication into assessment, and from assessment into survival. Salami-slicing, honorary authorship, padding, paper mills – all are products of the same incentives (see the dam bursts: paper mills and waves of retractions).

6Why do academic careers look like a Ponzi scheme?#

The pressure to publish is so intense because there is a pyramid standing behind it. In 2014 Richard Larson of MIT and his colleagues borrowed the epidemiologists’ “basic reproductive number” to estimate how many PhDs a professor trains, on average, over a whole career. For engineering in the United States the figure is 7.8. In a steady state only one of these can take over the professor’s post, which means that only about 12.8% of PhDs can find a place in academia; for half of all PhDs to stay in academia, the field would have to grow by 14% a year. For biomedical engineering the figure is 13.6, for environmental engineering as high as 19.0, and for MIT as a whole about 10.²¹ The authors admit that this is a “back-of-the-envelope” estimate, and it leaves out PhDs who do not want to become professors. A follow-up study the next year estimated that in the United States there is roughly one tenure-track position for every 6.3 biomedical PhDs.²²

Other data point the same way. According to a 2011 feature in Nature, between 1998 and 2008 the annual output of science doctorates in OECD countries rose by nearly 40%, to around 34,000; in the United States, 55% of biology PhDs in 1973 secured tenure-track posts within six years of their doctorate, whereas by 2006 only 15% held tenured posts six years after graduating.²³ A 2012 report from the National Institutes of Health found that the proportion of biomedical PhDs going into tenured or tenure-track faculty posts had fallen from about 34% in 1993 to about 26%.²⁴ And data from the American Association of University Professors show that in autumn 2023 about 68% of faculty at US colleges and universities were off the tenure track, against about 47% in 1987.²⁵

In 2010 The Economist ran an article headlined “The disposable academic”. “The fiercest critics compare research doctorates to Ponzi or pyramid schemes,” it reported; universities, for their part, “have discovered that PhD students are cheap, highly motivated and disposable labour”.²⁶ The essence of a Ponzi scheme is that it can only be kept going by an endless stream of newcomers: those who got in early are paid out of what latecomers keep putting in. The academic pyramid is, of course, not a fraud; nobody sets out to deceive anyone. But its structure really is like this: once expansion stops, the people at the base discover that the places they were promised do not exist.

China’s pyramid is now being squeezed by two forces at once. On one side, more and more doctoral students are being admitted: 55,000 doctoral students graduated in China in 2016, and 112,900 in 2025; in 2025, 201,600 new doctoral students were admitted and 753,600 were enrolled.²⁷ On the other side, births have plunged. The number of births in 2016, as revised after the seventh national census, was 18.83 million; in 2025 it was only 7.92 million, about 42% of the earlier figure.²⁸ Eighteen years from now, far fewer people will be starting university. Universities have introduced a “pre-tenure/tenure” system, popularly known as “up-or-out”: the usual arrangement is two three-year contracts, and anyone not promoted when they run out may have to leave. But there is still no official national figure for the share who pass the review.²⁹ The scattered data one can find describe only one stage at individual institutions. In Wuhan University’s first round of reviews in 2018, 48 people formally applied and 6 were promoted to associate professorships on the permanent staff; those who did not pass still had a second three-year contract.³⁰

I have argued that even without the AI revolution, the Ponzi structure of humanities education was close to collapse. If the best future open to the students you train is to become teachers themselves, you need an ever-growing number of faculty slots to house your academic children and grandchildren. In the past, population growth and the expansion of universities supplied plenty of posts; now the population is shrinking and university enrolment has peaked, so the number of posts can only fall. If humanities scholarship keeps insisting on the ivory tower’s own standards of reproduction, it is bound to become unsustainable.³¹ Science and engineering may not be quite so extreme – industry can still absorb large numbers of PhDs – but the logic of the pyramid is the same.

More than a century ago, Weber had already put the question to every young person hoping for an academic career. He began with a cruel reality of his day: if the candidate was a Jew, one naturally told him, “abandon all hope”. But anyone else, too, had to be asked in all conscience: do you really believe you can bear to watch, year after year, one mediocrity after another climb past you, without growing embittered and corrupted inside? Everyone, he said, replies: of course, I live only for my calling. Yet in his experience very few had endured it without inner harm.⁸

7Why do people who don’t love knowledge go into research?#

Together, the pyramid and the assembly line have changed the people who go into research. Our present model of teaching is alienated because technological society needs vast numbers of “cogs”, and to become a cog one has to master the corresponding specialist skills; so many people throw themselves into study not for pleasure but because they have to earn a living. The aim of their learning is not to pursue excellence, change the world or enjoy themselves, but to clear the threshold for becoming a cog. And because everyone is competing to become a higher-grade cog, learning keeps sliding into “involution” – a zero-sum race in which everyone must run harder just to stay in place – and, for fairness’ sake, can only be measured by objective standards such as marks and diplomas.³ Research, too, has become one such way of earning a living. Many people do a PhD, take a postdoc and apply for faculty jobs not because some problem keeps them awake at night, but because this is a respectable, stable job. They do not necessarily lack a love of knowledge, but they do not necessarily have one either; the assembly line never needed love anyway, only compliance.

None of this is a matter of anyone’s moral failings. With the university reforms of the eighteenth and nineteenth centuries and the institutionalization of primary and secondary schooling, learning and research both became occupations – jobs. Work in the modern mechanized industrial system is, as Marx said, alienated: in his labour the worker does not affirm himself but denies himself, and feels at ease only outside his work. He can neither enjoy the pleasure of creating while he works nor gain a sense of achievement when the work is done; all he gets is money, and all he can hope for is to buy some happiness with it after hours. Marx made the point long ago: labour is alienated not because the pay is too low, but precisely because pay is all there is.³² The alienation of research works the same way. Higher salaries, bigger bonuses, more talent hats for researchers – none of it changes one thing: if every reward of research can be converted into money and metrics, research itself is no longer the goal.

There are two misunderstandings to avoid here. One is high-minded purism: treating fame and fortune as the enemies of science. Chasing reputation and reward, wanting to win – these have always been part of the scientific spirit. Priority disputes run through the whole history of science, and scientists are still combative big boys (see contests and the race for knowledge). The question is not whether there is a prize, but whether the prize is all that is left. The other is elitism: the belief that only a handful of geniuses are fit to do science. Most scientific work has always been small-scale and unsung – improving an instrument a little, adding a little precision, laying another brick within an existing framework;¹ such work can be every bit as absorbing, so long as the person doing it still cares about it.

This is also where involution comes from. When everything is measured in money and metrics, there is no floor any more. One person says he will do it for fifty thousand, another says forty thousand will do; between fifty thousand and fifty cents there is no qualitative line anywhere, and the inevitable result is a race to the bottom.³³ Involution in academia means researchers competing over who has more papers, who has fewer weekends, who can put up with more. What it wears away is precisely the passion Weber spoke of.

8Can science turn from a career back into a labour of love?#

AI is changing the premises of all this. Any skill that large numbers of people can be drilled into step by step, AI will soon acquire too, and often better: drawing, literature reviews, programming. Every kind of learning that is force-fed, dull and organized as a step-by-step training routine can more or less be declared obsolete. This is not just a problem for vocational schools; it has reached the top levels of research. Tools like AlphaFold have long since shaken up molecular biology, and some “research labourers” will face unemployment as well.³ More fundamentally, the industrial age demanded that people be copyable because industrial products depended on copying. Information, by contrast, can be copied almost for free: build one AI, and it can be copied for everyone to use. Once copying is no longer society’s main task, education and research have a chance to stop aiming at the production of replaceable consumables and to return to forming whole human beings.¹⁰

The end of the assembly line need not mean unemployment; it can also mean liberation. The telegraph made the running messenger obsolete, yet the marathon, which descends from a messenger’s run, became a sport, and the ability to run long distances still has its uses. Sport has long since ceased to serve production, and machines long ago surpassed human strength; yet in some places sport is organized entirely through clubs, and the people who train there do it mainly for enjoyment and in pursuit of excellence. Ideally, the arrival of AI would turn many professions back into “liberal arts”. When a craft is no longer a way of making a living, there will still be people willing to put effort into learning it, as every sport and game shows; and once people no longer have to spend most of their lives on the skills that earn them a living, they may be able to devote more of their energy to learning for its own sake.³

Weber said something in that lecture that is often overlooked. An amateur’s inspiration can carry exactly the same scientific weight as an expert’s, or more; we owe many of our very best problems and insights precisely to amateurs. The amateur differs from the expert – as Helmholtz said of Robert Mayer – only in lacking a secure working method, and is therefore usually unable to check, gauge or carry through the implications of his inspiration.⁹ Today AI can, to some extent, fill exactly this gap. It can keep a layperson company while they read through the literature, check a derivation, get the code running, and turn an inspiration into something that can be tested. In the comment threads of the Erdős problems website, a researcher who works for a customer-service company by day can collaborate with Terence Tao; a 23-year-old with no professional training in mathematics used AI to crack open a problem that had stood unsolved for sixty years³⁴³⁵ (see amateurs and ten thousand agents).

But we must also face another possibility. Once the skills people live by become obsolete, society as a whole may fall into chaos or drift; many people may want to learn nothing at all and do nothing but scroll through short videos. Even as the professions become “liberal arts” again, they may also be “de-massified”, turning back into “elite arts” practised by a very small few.³ This need not be so frightening: no ordinary person was ever under any obligation to advance science. That everyone can be a scientist does not mean that everyone has to be one; but for those with a genuine passion for knowledge, the door has never been open so wide.

Universities may not be the protagonists of this transformation. The resistance to university reform is too great; in many respects universities have not even adapted to the information age yet. Every great upheaval is a good moment for innovation, and every one is also bound to see a very high rate of failure. I would rather pin my hopes on an environment of companies and civil-society groups that will rekindle the light of inquiry outside the ivory tower; what traditional universities can do is stay out of the way as far as possible, not tie their students’ hands, and make knowledge as open and widely shared as they can.³¹ A curation platform, a new journal, a game engine for research: each is a small step in that direction (see what we set out to build). Replace distribution according to labour with distribution according to delight, make asking questions, checking results, explaining clearly and appreciating good work into tasks that are worth doing and that earn recognition,³⁶ and research may yet turn from a rigid profession back into a lively labour of love (see can everyone be a scientist?).