Chapter 3 — From Teaching Machines to ChatGPT: A Short History

Last updated: July 5, 2026

Every generation has met a technology it was told would transform education, and the history is worth knowing because the present rhymes with it so exactly. In 1913 Thomas Edison predicted that books would soon be obsolete in the public schools and the whole system changed within ten years — confident, specific, and wrong in every particular. That prediction inaugurated a pattern repeated for over a century: a transformative technology arrives, its champions declare the classroom obsolete, schools invest, the revolution quietly fails to occur, and the next technology arrives to declarations that this time is different.

The first half of the twentieth century ran the experiment three times — film, radio, television. The radio "schools of the air" reached millions of students before fading; Benjamin Darrow, founder of the Ohio School of the Air, wrote in 1932 that the dominant aim of education by radio was "to bring the world to the classroom" — a line that could close a 2026 keynote. Instructional television reran the arc at higher altitude: the Midwest Program on Airborne Television Instruction flew aircraft in circles over Indiana, broadcasting to six states until the economics collapsed. The broadcast era's lesson, and Larry Cuban's in Teachers and Machines, is that transmission is not teaching: content could be delivered cheaply and uniformly, but no broadcast could see the learner or respond to confusion.

Then came the machines that responded. In the 1920s Sidney Pressey built a machine to administer and score tests, and promised an "industrial revolution" in education; it shipped in 1930 at fifteen dollars apiece — more than half a year's per-pupil spending — and ceased production with most of the run unsold. In the 1950s B. F. Skinner's teaching machines promised individualised, self-paced mastery, in language — "personalized learning," liberation of the teacher — that today's adaptive-learning decks reproduce almost verbatim. Each arrived as revolution and settled, at best, as a modest tool. Pressey's lesson was that being right about pedagogy is not sufficient; the economics and institutions of schooling decide adoption.

The pattern repeated through computer-assisted instruction and PLATO, through the intelligent tutoring systems of Chapter 4, and most vividly in the MOOC "year" of 2012, when Sebastian Thrun predicted that within fifty years only ten institutions would deliver higher education worldwide, then conceded eighteen months later that "we have a lousy product." The definitive study of 5.63 million learners found completion settling at 3.13 percent in 2017-18, down from nearly 6 percent. The edtech graveyard is instructive: inBloom, a hundred-million-dollar data venture that launched with nine states and was gone within a year, igniting more than 400 student-privacy bills; Knewton, whose founder called his platform "like a robot tutor in the sky" before its assets sold for under seventeen million dollars; AltSchool, which burned a reported forty million a year; the Summit backlash, in which nearly a hundred Brooklyn students walked out rather than be personalized at; and the 2U/edX bankruptcy. Bold theories of transformation have a poor survival rate.

What survived did so quietly, by doing one thing reliably rather than by remaking pedagogy. The overhead projector arrived with no manifesto and replaced nobody. The calculator stuck only once the SAT made peace with it in 1994, when 87 percent of students brought one, rising to 95 percent by 1997. The learning-management system, unloved and unhyped, was installed at 99 percent of US colleges by 2014. None was sold as a revolution, and none has been declared a failure. Cuban's verdict in Oversold and Underused: even Silicon Valley's classrooms treated abundant computers as souped-up typewriters — access had been solved, integration had not. Durable adoption tends to be narrow, boring and useful.

Hype-cycle scholarship — the Gartner curve fits education uncannily well — predicts this shape well enough that the burden of proof sits with anyone claiming AI is the exception. The one honest reason it might be is the argument of Chapter 1: unlike its predecessors, generative AI produces the work itself rather than merely transmitting or drilling content, and it reached majority student use through the phone with no institutional decision at all — the gatekeeping that killed Pressey's machine simply did not apply. That could genuinely break the pattern, or merely accelerate it into a faster, sharper cycle.

Either way, the history hands the reader a discipline rather than a prediction: discount the transformation narrative, ask what specific and durable job the tool actually does, and demand the evidence past the enthusiastic first year — which, as the next chapter shows, is precisely where most effects shrink.

Sources. [Primary empirical] Skinner, B. F. "The Science of Learning and the Art of Teaching." Harvard Educational Review 24(2): 86–97, 1954. (Record: https://www.semanticscholar.org/paper/1cbd547821eeadd8805f95335ecca59c05703e34) · [Primary empirical] Cuban, L. Teachers and Machines: The Classroom Use of Technology Since 1920. Teachers College Press, 1986. https://books.google.com/books/about/TeachersandMachines.html?id=uQeEn1vEUSQC · [News] Pappano, L. "The Year of the MOOC." The New York Times, November 4, 2012. https://www.nytimes.com/2012/11/04/education/edlife/massive-open-online-courses-are-multiplying-at-a-rapid-pace.html