The Dream of Adaptive Learning Has a Human Problem
What Neal Stephenson's Primer got right about AI education — and what we keep getting wrong.
When I worked adjacent to EdTech, it felt like everyone was selling the same dream: a learning experience tailored to each individual student, delivered at scale, without requiring a human teacher for every interaction.
Khan Academy has Khanmigo. Duolingo replaced human contractors with AI-generated content, and its CEO declared that AI is “a better teacher than humans,” predicting that schools would survive mainly as childcare1. Google has embedded Gemini chatbots into the Chromebooks distributed to public school children, sometimes without parental knowledge2.
In 2024, Arizona State University President Michael Crow cited Neal Stephenson’s The Diamond Age as the inspiration for the university’s “Realm 5”: infinitely scalable education delivered through massively distributed, personalized, adaptive learning3.
That sent me back to the book to understand the vision. And as generative AI becomes more deeply embedded in our world, the vision is worth examining.
The Dream Already Failed
So far, the track record is grim.
In 2019, E. Tammy Kim investigated Silicon Valley’s “personalized learning” reforms in Rhode Island schools. Tech-industry-funded programs had promised to revolutionize education through adaptive software and one-to-one Chromebook initiatives. What Kim found instead was overworked teachers embracing the reforms simply to secure basic resources, while critics warned that the model accelerated privatization and turned public schools into “big-data siphons”4.
By 2024, IBM’s $100 million Watson education project — once heralded as the future of AI tutoring — had become a cautionary tale; after five years, IBM conceded that its dazzling supercomputer made a lousy teacher5. That same year, the neuroscientist Jared Cooney Horvath argued in Harvard Business Review that more than forty years of research suggests generative AI could actually harm learning. Human understanding, he wrote, depends on empathy, the slow accumulation of knowledge, and sustained attention — exactly the things a tool that instantly supplies the right answer tends to short-circuit6.
Then came ChatGPT. By 2025, The Atlantic was reporting that students were using AI to generate essays and finish homework, sidestepping the intellectual struggle that once gave schoolwork its meaning7. A New York City public school senior, Ashanty Rosario, wrote a first-person account of watching her classmates outsource their thinking to AI, killing the “frantic productivity” that had once bonded them together8. By 2026, a viral AI agent called “Einstein” could complete a student’s coursework end to end — watching lectures, writing papers, taking quizzes — by logging into their Canvas account9.
The personalized learning dream had become a personalized cheating machine.
A Novel That Got There First
Stephenson’s Primer is nothing like these tools.
In The Diamond Age, the central technology is an interactive nanotech textbook called the Young Lady’s Illustrated Primer. The book finds its way to Nell, a girl growing up in poverty in a future Shanghai. Although it was never meant for her, it teaches her language, logic, social graces, self-defense, and computation — all calibrated to her developmental stage and cultural context.
It is the ultimate personalized learning tool. And it works. Nell transforms from a neglected child into a resourceful, capable young woman. The Primer is, by any measure, the most effective educational technology ever devised.
On its surface, the novel looks like a triumph of the AI education dream. The technology delivers on every promise EdTech and AI companies are still making.
But we should not lose sight of why the Primer works.
The Human in the Machine
The Primer was built by the nanotech engineer John Hackworth. Every story it tells, every challenge it sets, every moment of scaffolded difficulty reflects a deliberate choice about what this particular child needs at this particular moment. As Gideon Dishon argues in a recent analysis of the novel, the Primer functions simultaneously as a personalized tutor and a vehicle for enculturation — transmitting specific cultural values and ways of being in the world10. Its cultural responsiveness is a design choice, not merely a feature spit out by an algorithm.
Cultural responsiveness, as a concept in education, was articulated by Gloria Ladson-Billings in 1995 — the same year Stephenson published the novel. Her theory of culturally relevant pedagogy held that effective teaching must account for the specific cultural context of each learner: students learn best when their backgrounds are treated as assets rather than obstacles, when the curriculum speaks to their lived experience, when the teacher understands who the student is beyond their test scores11.
The Primer also works, given the current landscape of AI in education, because it makes Nell struggle. It does not hand her answers; it puts her in situations where she has to find them. It presents challenges calibrated to push her just past her current ability, then offers support when she is genuinely stuck. The personalization is not in the content delivery. It is in the timing of the struggle.
But the deepest reason the Primer works is also the easiest to miss: there is a person behind it. The book’s voice — the warmth that answers Nell, follows her, stays with her for years — is not generated. It is performed by Miranda, a human ractor hired to voice the Primer, who over thousands of hours comes to love the child she has never met. N. Katherine Hayles has argued that the Primer’s significance lies precisely in what it reveals about the boundary between human understanding and technological mediation12: the technology works when it is in dialogue with a human relationship, and falters when it operates alone. The thousands of identical Primers handed to the orphaned “mouse army” run on synthetic voices instead — and those girls, with the same hardware and the same lessons, do not become Nell.
The Quiet Argument
The Diamond Age does not reject the Primer. Nell benefits enormously from it. Her life is materially better because the technology exists.
But the novel never lets us forget what that technology rests on. Strip out Miranda and you do not get a lesser Primer; you get the mouse army’s Primer — the hardware and little else. Stephenson tucks his thesis into that contrast: the machine is a multiplier of human attention, not a replacement for it.
That is the part the current rush keeps getting backwards. The companies promising the Primer at scale are selling the one thing the novel insists cannot be scaled — Miranda herself — and stripping out, in the name of efficiency, the struggle and the relationship that did the actual teaching. The dream of adaptive learning was never really held back by a technical problem. Stephenson saw, thirty years ago, that it was a human one. We are only now finding out he was right.
Footnotes
“Duolingo CEO says AI is a better teacher than humans—but schools will still exist ‘because you still need childcare.’” Fortune, May 20, 2025. https://fortune.com/2025/05/20/duolingo-ai-teacher-schools-childcare/
Winter, Jessica. “What Will It Take to Get A.I. Out of Schools?” The New Yorker, April 23, 2026. https://www.newyorker.com/culture/progress-report/what-will-it-take-to-get-ai-out-of-schools
Crow, Michael M. “Augmented Intelligence: Reimagining How the World Learns.” ASU+GSV Summit, 2024. https://asugsvsummit.com/video/augmented-intelligence-reimagining-how-the-world-learns
Kim, E. Tammy. “The Messy Reality of Personalized Learning.” The New Yorker, July 10, 2019. https://www.newyorker.com/news/dispatch/the-messy-reality-of-personalized-learning
Berdik, Chris. “AI Can’t Replace Teaching, but It Can Make It Better.” WIRED, July 10, 2024. https://www.wired.com/story/what-aspects-of-teaching-should-remain-human/
Horvath, Jared Cooney. “The Limits of GenAI Educators.” Harvard Business Review, July 16, 2024. https://hbr.org/2024/07/the-limits-of-genai-educators
Shroff, Lila. “The AI Takeover of Education Is Just Getting Started.” The Atlantic, August 12, 2025. https://www.theatlantic.com/technology/archive/2025/08/ai-takeover-education-chatgpt/683840/
Rosario, Ashanty. “I’m a High Schooler. AI Is Demolishing My Education.” The Atlantic, September 3, 2025. https://www.theatlantic.com/technology/archive/2025/09/high-school-student-ai-education/684088/
Shroff, Lila. “Is Schoolwork Optional Now?” The Atlantic, April 10, 2026. https://www.theatlantic.com/technology/archive/2026/04/ai-agents-school-education/686754/
Dishon, Gideon. “AI based personalized learning in ‘The Diamond Age’: Artificial subversiveness and human feeling machines.” Educational Philosophy and Theory, vol. 57, no. 10, 2025, pp. 907–919. DOI: 10.1080/00131857.2025.2494589
Ladson-Billings, Gloria. “Toward a Theory of Culturally Relevant Pedagogy.” American Educational Research Journal, vol. 32, no. 3, 1995, pp. 465–491. DOI: 10.3102/00028312032003465
Hayles, N. Katherine. “Is Utopia Obsolete? Imploding Boundaries in Neal Stephenson’s The Diamond Age.” In World Weavers: Globalization, Science Fiction, and the Cybernetic Revolution, Hong Kong University Press, 2005, pp. 95–110. DOI: 10.1515/9789882203129-008
Finn Agler is the blogging pen name of R.G. Luchmun, a cultural anthropologist writing science fiction about first contact, time travel, and the places where cultures collide. Follow along for more intersections of anthropology and speculative fiction.