The experiment ran for twenty years. The results are in. The digital classroom failed, and it is being dismantled by the people who built it. Board resolutions are taking it apart; New York City has barred individual devices in preschool through second grade and set recommended screen-time limits in the grades above it, and Canon-McMillan in Pennsylvania is pulling Chromebooks out of kindergarten through second grade and buying printed textbooks and workbooks. Burke County, North Carolina, voted in 2025 to make pen and paper the primary mode of learning in its classrooms. Sweden ran the same vote at national scale in 2023, announcing an end to the rule that had made devices mandatory in its preschools and putting 685 million kronor into printed books, with half a billion a year announced behind it.
I taught high school science for twelve years. I watched the devices arrive with their promises attached. The pitch was always the purchase price, and the promises that rode with it: an easier job for teachers, and students prepared for the future. The costs came later, in currencies the pitch never mentioned.
Maine was the first state to put a laptop in every middle schooler's hands, Apple iBooks for about 17,000 seventh graders in the fall of 2002, under a four-year contract worth 37.2 million dollars that bundled warranty and repair with the machines, because breakage was predictable. Dive in, the pitch said, and Maine was the first of dozens to jump. At Piscataquis Community High School, about half of the laptops needed at least a minor repair within seventeen months; forty were sent back to Apple for screen repairs, and the school ran out of its ten loaner machines in the first year. An Associated Press reporter sitting in a Maine classroom counted three of sixteen students without a computer that day: two students whose machines were broken and shipped back to Apple, and a third student who had lost his computer privileges. Multiply that arithmetic by every fleet bought since, and add the subscriptions and the replacement cycles. Districts are doing that math now, with pandemic money gone and pandemic fleets worn out, and Chapel Hill-Carrboro, North Carolina, found $ 435,000 to redirect to print curricula by canceling technology contracts.
Three of sixteen is not small when the entire lesson is designed as a digital activity, and three students have no machine: away for repair, left at home, or uncharged. A non-digital option has to exist for those students, which doubles the work; the teacher scaffolds the same lesson across two media for the same class. Questions of equity aside, and that is a big aside, the arrangement asks teachers for more work in the name of making teaching easier.
Nearpod is a typical case study. It was an excellent tool for teachers. It let them build lessons that looked dynamic, full of beautiful interactives, and it spread fast. I converted about a third of my lessons to Nearpod because I liked it so much. After a few months, students hated it. Any time a teacher said “open your Chromebooks and follow the link to Nearpod,” students would groan and whine before they even saw the lesson. Students crave human problems. That is something a digital classroom cannot provide.
A classroom device is a door that opens both ways, and on the other side of it is every distraction the entertainment industry has built. UNESCO estimated that about 40% of the world's education systems would restrict phones in schools by the end of 2024. Thirty-five American states had limits in place by the 2025-26 school year. Sweden collects phones at the door now, by national rule.
Teachers named the problem before the policies moved. In Pew's 2024 survey, 72 percent of high school teachers called phone distraction a major problem in their classrooms. What changed when the phones went away showed up first in the places scores do not measure. Phones in Focus, which has gathered responses from more than 68,000 educators, reports louder lunchrooms, and the noise is conversation: students talking face to face, learning to socialize again in one teacher's phrase, with fewer phone discipline problems and more engagement in class. The same survey carries a warning. Teachers estimate that one in three students uses a school-issued laptop for texting or social media during class, so part of the distraction simply moved to the very machine the school handed out.
Sweden acted after its fourth-grade reading score in PIRLS slid from 555 to 544. No assessment cycle has run since its reinvestment in printed books, so Sweden has no recovery to report yet. Where phone bans have been measured, the gains are real and small: English schools saw a modest improvement after bans, concentrated in their lowest-achieving students, and a recent English study found no attainment difference at all. Nothing in the record shows reading levels back where they stood before the screens arrived. The decline is documented; a recovery is not. The reading research points in the same direction: across dozens of studies, students comprehend informational text, the kind schools assign, somewhat better in print than on a screen, and they judge their own understanding less accurately after reading on a screen. That is the whole of the measured evidence there is, and the case for paper does not depend on it.
Then COVID put every student in the country on a managed device full-time at home, and the last cost came with it. Students learned the systems by living inside them. The adults charged with holding them accountable learned the systems as a work tool, well enough to assign and collect. The gap between those two fluencies is where accountability went. In September 2025 Google added a homework help button to Chrome without announcing it. It appeared inside course sites, Canvas among them, and opened Google Lens on whatever question was on the screen, even during live tests. Berkeley's notice to its faculty stated plainly that no administrator control existed to turn the feature off. Google paused the button only after the Washington Post called, and the same capability remains one right-click away. A Los Angeles English teacher watched his struggling students start acing tests and could not find the method. A student showed him. The adult learned how the cheating worked from the student who did it, and the tool that made it possible was shipped by the company that manages the school's devices.
My own district learned of it the same way. Most answers were evaluated electronically, so teachers often never saw what their students had submitted, and no alarm rang anywhere in the system. A student had to tell a teacher that students had found a feature that unlocked the correct answers, even on state exams.
None of this is the fault of the teachers who leaned on the machines. A teacher carrying five sections, an overpacked schedule, and planning periods that vanished years ago will hand work to whatever tool offers to take it, and blaming her for taking the offer misunderstands the job she was handed. The arrangement failed. The people inside it were doing the work in front of them. What the technology push owes us now is an honest accounting: keep what it taught, abandon what it proved it cannot do.
The digital classroom was an arrangement: the screen as the default medium, the conduit every lesson had to pass through. Sweden's budget, New York City's rules, and the phone laws now in force across most states are dismantling that arrangement. Paper is resuming the default position it held for a century. The device stays in the building, one teaching tool among others, pulled out where it earns its place and left in the cart where it does not. Technology is always best as a support for learning, not its main driver.
A written page holds the student's thinking where a teacher can see it: the first explanation, the correction, the place where the evidence is pushed and the idea moves. Handwriting is slow, and the slowness does the work: nobody can transcribe a lecture by hand word for word, so the student has to decide what the words mean as she writes them down. Paper asks nothing of the network. It does not update overnight, it does not ship new features in September, and its failures stay small: a torn page, a lost notebook, one student's work at a time.
A hands-on investigation recorded on paper stays alive in a way a simulation does not. The mold grows where it wants, and a radish sprout leans toward the window. A measurement refuses to repeat itself, and the journal page holds the whole untidy record, crossings-out included. A student's notebook shows a mind changing as evidence arrives: a claim written large on Monday, crossed out by Friday, and rewritten smaller where the evidence ran out. A simulation arrives sterilized: its data behaves, its variables hold still, and the student never meets the resistance that makes the work real. In science, wrong answers are most of the content, and a student who has only ever run clean simulations has been protected from the science itself.
Computers have a role in the classroom. Text-to-speech reads print aloud for students whose disabilities keep them from reading it themselves. A tablet a low-vision student can adjust beats a large-print book she cannot. Adaptive software that teaches at the student's actual level has the strongest trial evidence in the field, as shown by the Mindspark evaluation in India. Those are tools doing work paper cannot do, and a written lesson can call on every one of them. The ordinary tools earn their place beside them. A spreadsheet in a science classroom holds the class data, performs the arithmetic, and leaves the interpretation to the student. A word processor makes revision cheap, and revision is where writing is learned. The computer belongs in the lesson, directed to the task that needs it.
The arrangement breaks down at the thinking itself. A tool that carries the labor leaves the thinking: the spreadsheet computes the mean, and the student still has to say what it means. A tool that carries the thinking leaves the student with nothing. A generation of tools now answers the question on the screen before the student has finished reading it. The distinction is one I have fully made in The Tool Can't Think for You. Learning is learning what you do not know, techniques and processes that are unfamiliar. The process is not instantaneous; learning takes time and effort, and the minutes spent not knowing yet are where persistence is built. If the technology circumvents that time and that effort, the learning may not be happening. Those minutes are productive struggle, and the argument for protecting them is one I have made before, in Championing Uncertainty.
Artificial intelligence is where the offloading stops being partial. A chatbot produces the finished product, the paragraph, the solution, the explanation, in the student's voice if asked. The policy world is drawing its line at the same place. New York City has paused student-facing generative AI from its youngest grades through eighth grade for this school year, nearly 600,000 students, and a bill in Albany would prohibit most AI in classrooms before ninth grade statewide. The exceptions are the support tools: assistive technology in the city policy, diagnostics, and interventions for students with disabilities in the Albany bill. AI that reads text aloud to a student who cannot decode it is a support. AI that writes the essay is a substitution, and a student whose essay was written for him never gets his minutes of not knowing at all.
Most students’ work on the machines was ungraded practice: scenarios and investigations where a wrong turn cost nothing. The assessment at the end caught the misunderstandings and corrected them. That is a property of the design, and paper holds it when the material is built for it. The practice has to be the science itself, though: a student who works a simulation somebody else scripted performs science rather than doing it, and the performance is the science pantomime. Where the practice is a real investigation, the work teaches the skills of the Science and Engineering Practices (SEP) themselves. Then the assessment challenges them to apply those skills in a scenario they have never seen, which is where a performance expectation under NGSS, the IB frameworks, or any honest standard is met. A memorized answer fails as soon as the scenario changes. Reasoning builds.
A student who can name what she is doing as she does it can carry the process into the next problem. A student who only follows the steps has nothing to carry. The Syzygy framework builds that recognition on four questions an investigator asks:
What am I observing? The Crosscutting Concept (CCC), the lens the student chooses for the situation in front of her.
How will I investigate it? The Science and Engineering Practice (SEP), also hers to choose.
How does it fit what I am learning? The Disciplinary Core Idea (DCI) is supplied by the design, the phenomenon, and the evidence package.
How can I show what I understand? The Performance Expectation (PE) is provisional while the investigation runs and formal at the assessment.
Two of the questions belong to the student, and two to the design, and the split is the method: if the teacher assigns all four, the student performs the motions of inquiry without doing any. Asked on every investigation, the questions become a habit, and the student starts to see the pattern of her own process. The framework document lays out the four questions in full, and it is free on this site.
The Syzygy labs are print-first by that design and were, before it was a position anyone had to defend: paper journals, organizers on paper, recording done by hand at the bench where the work happens, the standard held back until the class has something of its own to reason about. The device comes out when the lesson calls for it. The rest of the time, the page is the platform, and the page is enough.
Citations
