Deutsch - English - Español - Français - Italiano
Two loud resets. Same stopping point on both sides: how success is measured. Second in a four-part series on what the machine is doing to school, and on who will decide what happens to the human.
In Texas, a school without teachers still measures itself by yesterday’s tests
Alpha School, founded in Austin in 2014, sells itself in one phrase: “2 Hour Learning”. Two hours of academics a day, delivered end to end by adaptive software and AI tutors at each pupil’s own pace. The rest of the day goes to practical-skills workshops: public speaking, entrepreneurship, teamwork. No teachers, just “guides” who motivate and supervise without teaching. Every pupil gets thirty minutes a week one-on-one with theirs. Tuition runs from $10,000 to $75,000 a year depending on the campus, $40,000 at the Austin flagship. Nearly a dozen new campuses are set to open in the autumn of 2026, on top of some fifteen already running. The claims keep climbing: the brand pledges “2X in 2 hours”, the network reports gains more than twice the national average, and co-founder MacKenzie Price talks of pupils learning “ten times faster” than in an ordinary school.
The theoretical ground is old and serious. Benjamin Bloom’s “2 sigma problem”, laid out in 1984, found that the average pupil under one-to-one mastery tutoring reaches the level of the top 2 % of an ordinary class. One-to-one tutoring was unaffordable at system scale. The machine, the promise runs, would make it affordable.
Now watch what Alpha does with that promise. The school that has done away with teachers, classes and timetables proves itself with the very tools of the school it claims to leave behind: the same curricula, and above all the same standardised tests. Every result Alpha reports is expressed in percentiles of national assessments. Process replaced; goals and quality control untouched. A private operator selling to parents has to prove itself against a shared yardstick, or its results measure against nothing. It can’t change the yardstick, so it runs faster along the one already in place.
A few warnings on the record. Alpha’s results come from internal data, never independently checked, on an affluent, self-selected population. The Pennsylvania Department of Education, reviewing a charter application tied to the network, called the model “untested” and lacking evidence it meets state academic standards. And Alpha’s “AI”, by the founders’ own admission, is mostly adaptive software of the IXL or Khan Academy sort, older than the large language models. The showroom runs ahead of the workshop, as readers of this series know by now.
The USA’s mainstream path is grafting AI onto the old machine
Alpha is the exception. The landscape has picked the other road by a wide margin. According to the Center for Democracy and Technology survey published in October 2025, 85 % of USAmerican teachers and 86 % of pupils used AI during the past school year, for planning lessons, marking, differentiation — for the existing tasks of the existing school. Fifty-five per cent of teachers say the tool freed up direct time with pupils. The graft fits the shape of the teacher’s job for the same reason the factory humanoid fits the shape of the worker: nobody wants to rebuild the building. The institutional debate does exist, and it settles into a formula. A report from the Center on Reinventing Public Education urges schools to be “ambidextrous”, improving what exists with one hand and building the new with the other, a formula that spares them the trouble of choosing.
The pupil can’t leave the workshop: delegation destroys the product
Here the industrial analogy that runs through this series hits its limit, and the limit is more instructive than the analogy. On the factory floor, the product doesn’t care who makes it. The machine welds, the weld holds, and the worker walks out. In school, the “product” is a transformation of the pupil, and the pupil can’t walk out. Only the teacher can. And if the machine does the work in the pupil’s place, the product never gets made.
One large randomised trial has documented this. About a thousand Turkish high-school students, in mathematics, split into three groups: no AI; raw access to GPT-4; and a GPT-4 tutor wrapped in pedagogical guardrails. Published in PNAS, the journal of the US National Academy of Sciences, under the title “generative AI without guardrails can harm learning”, the study reports: the raw-access group did 48 % better on the assisted exercises, then 17 % worse than the control group on the exam taken without help. The guardrailed tutor, though, lifted assisted performance without hurting the exam. Delegation is what automation does at work. In learning, it destroys the product. That is the structural difference between the factory and the school, shown experimentally. A word of method: the negative effect is the most fragile of the three results, close to the threshold of statistical detection; this is one trial, not a law. A counterpoint exists, and it fits the same mechanism. A randomised trial at Harvard finds students learn about twice as much, in less time, with a purpose-built AI tutor than in an active-learning class. A World Bank pilot in Nigeria, after six weeks of teacher-supervised sessions, measures a 0.31 standard-deviation gain on the final assessment and 0.23 on English alone, the programme’s main target. (The standard deviation is educational research’s usual unit for effect sizes; a gain of that order moves the median pupil from roughly the 50th to the 62nd percentile, which the authors put at one and a half to two years of ordinary schooling.) Everything turns on the guardrails from the PNAS title: on who controls the relationship between the pupil and the machine.
In Hangzhou, the state overhauls goals and measurement, but the exam holds
The Chinese mirror flips the Texan case, item by item. Where Alpha swaps the process and keeps the goals and the checks, the augmented Chinese school keeps the process (classroom, teacher and workbook stay in place) and changes the other two. The state redefines the goals from above. The government’s January 2025 blueprint elevates AI-driven educational reform to national strategy, aiming at an education “among the best in the world” by 2035, in a country where the number of schools has dropped from over a million in the 1970s to 470,000 today. Measurement changes shape. At the Xiang Lake Future School near Hangzhou (an operator with more than thirty schools and 60,000 pupils), the pupil’s pen tracks every stroke and its pressure, and the instruments run from the canteen tray to the heart-rate wristband. The third episode will follow this migration from product to process, which reaches well beyond China.
Yet at the end of this thoroughly remeasured school, the gaokao stands. Intact. Sole judge of who goes where. The Chinese state has shown it can redefine what school is meant to produce and how the producing is observed. It has not touched the exam that decides everything. The most radical instrumentation in the world leads to the most classical exam in the world. Two readings are open: real redefinition of purposes, or fine-tuning of the drill for the same exam. Both probably run in parallel; the available sources, all warmly disposed to the system, don’t allow a verdict.
Zhu Yongxin, or how a reset gets absorbed: 2019, 2023, 2025
One man lets us measure, on paper, the exact space of the Chinese reset: Zhu Yongxin, central figure in Chinese pedagogy and deputy secretary-general of the Chinese People’s Political Consultative Conference, a voice from inside the apparatus. In 2019, his book Future School: Redefining Education proposed a structural demolition nobody in the West dares to name. Schools would be replaced by “learning centres” without fixed classrooms or unified textbook, without timetables or age cohorts. Teachers would become companions of learning, some going freelance. Verification would centre on a “credit bank” recording learning across a lifetime. And the shock formula: the diploma will no longer count. Even in that most radical version, one detail stayed put. The ministry keeps “the setting of national standards and the organisation of evaluation”. Schools can dissolve, so long as the state keeps the yardstick.
In 2023, right after ChatGPT, Zhu republished his programme in the field’s leading academic journal, in full, but locked down. The article opens with China’s shortage of graphics processors and closes on national technological self-sufficiency, revised data-security laws, and a list of foreign AI products to authorise on the education market. One sentence has vanished. “The diploma will no longer count” isn’t there any more. The gaokao isn’t attacked, isn’t defended. The question has evaporated. The same year, in a doctrinal text, the derivation of ends starts from Marx and lands at the “builders and successors of socialism”. In 2025, in a dialogue published by the same journal with Turing Award laureate John Hopcroft, the learning-centres thesis is nowhere. Zhu defends the irreplaceability of teachers and the perfecting of what already exists, textbooks and teacher training included. It is the USAmerican guest who speaks against entrance examinations. Zhu himself only proposes to abolish the humanities-sciences split within the gaokao, and never touches the exam itself.
The radical structure of 2019 has not been disowned. It has been wrapped in the language of national security, stripped of its most cutting proposition, and quietly moved off screen. Zhu’s arc is that of a brilliant thinker who ends up tucking his most original ideas under the carpet of the exam.
Moving the yardstick takes a sovereign, and even the sovereign backs off at the exam
Alpha changed the process and kept the yardstick because a private operator has no choice: the common yardstick is its only proof. China changed the goals and the measurement because only a state can impose the metric. Moving the yardstick is sovereign work, not entrepreneurial. Yet the sovereign itself, having instrumented the school down to the canteen tray and let the dissolution of the school form get into print, has not touched the gaokao. Its boldest thinker has pulled from his programme the one sentence aimed at it. On both sides of the Pacific, the reset stops in the same place: the exam that turns a school career into a life.
The yardstick is where school meets society, the point where scarce places get handed out. Pedagogy can be rebuilt without anyone’s permission. Changing how success is measured, though, means opening up the question of what society promises the measured. Neither Alpha nor China will go there. While they hold back, the measuring instrument is already mutating, quietly: pen stroke by pen stroke, oral by oral, fabrication note by fabrication note. That is what the next episode is about.
All our publications are free and Copyleft. They may be republished, in whole or in part, under following conditions:
→ keeping the title and content
(text, images, editor’s or translator’s notes, boxes)
→ crediting the author(s) and/or translator(s)
→ inserting a clickable link to the source.
Any monetization or commercial use of these publications is prohibited and will be subject to legal action.