In 1913 Thomas Edison told a New York newspaper: "Books will soon be obsolete in the public schools. Scholars will be instructed through the eye." Motion pictures, he believed, would teach every subject. Radio, television, computers and the internet each arrived with similar promises. Schools did change, but books survived, and the history of educational technology is a useful guide to judging the newest arrival: artificial intelligence.

A century of promises

In the 1920s the psychologist Sidney Pressey built a machine that quizzed students and moved on only after a correct answer. In the 1950s B. F. Skinner promoted teaching machines that broke lessons into small steps with immediate feedback. Television courses, computer labs and interactive whiteboards followed. Around 2012 massive open online courses, or MOOCs, offered free university classes to anyone online, and some predicted the end of the campus. Completion rates for MOOCs turned out to be very low, often in single digits, and many of those who finished already held a degree. The One Laptop per Child project set out to give inexpensive laptops to millions of children in poorer countries, and a large randomized evaluation in Peru found no improvement in math or reading scores, though children did become more skilled with computers.

Why technology often disappoints

A long-running argument in the field holds that the medium matters less than the teaching method it carries. A film, a tablet or a lecture can each deliver good or poor instruction. Devices that simply put content in front of students rarely help much, because motivation, attention and practice are the hard part. Technology works best when it supports methods known to work: retrieval practice, spaced review, quick feedback and practice at the right level. Well-designed adaptive software and computer tutors for math have shown solid gains in some studies, precisely because they do these things for many students at once.

Technology can also harm. Studies link heavy recreational screen use to distraction, and many countries and school systems restricted phones in class during the early 2020s. The COVID pandemic was an unplanned test. When schools moved online in 2020 and 2021, students in many countries lost substantial learning, with the largest losses among poorer students, who were less likely to have quiet space, devices and fast internet. Screens did not replace a teacher in the room.

AI as tutor

Chatbots built on large language models, which spread rapidly after late 2022, can explain ideas, generate practice questions and respond to a student's mistakes in plain language at any hour. That looks like an answer to Bloom's two sigma problem: a patient personal tutor for every child. Some early trials of carefully designed AI tutoring have found promising gains. For students who have no one at home able to help with homework, the potential for equal opportunity is real.

AI as a shortcut

The same tools can write an essay or solve a problem set in seconds. If a student hands off the thinking, the learning does not happen, since the effort of retrieval and reasoning is what builds understanding. A randomized study in Turkish high schools illustrates the risk. Students given an unrestricted chatbot did better on practice problems but worse on a later exam taken without it, compared with students who had no chatbot. A version designed to give hints rather than answers largely avoided the harm. AI also states wrong facts with confidence, so students need the skill to check it. Tools that claim to detect AI writing are unreliable, and false accusations hurt honest students.

What stays human

Some parts of education depend on people. Teachers notice the child who is struggling at home, build a class where students argue respectfully, and model what it looks like to care about the truth. If schools exist partly to form citizens and to help children flourish, those purposes need human relationships. A tool can do much of the explaining, but the purposes of education still have to be chosen by people, and that choice returns us to the question that opened this course: what is school for?

Worked example: an AI policy for a high school

A high school must decide what to do about AI. A total ban is hard to enforce and leaves students unprepared for tools they will use at work. Unlimited use risks hollowing out practice. A policy built on the evidence might do several things. It allows AI for explanations, feedback on drafts and generating practice quizzes, all of which support retrieval and feedback. It moves key assessments into class, written by hand or spoken aloud, so teachers can see what students actually know. It asks students to note how they used AI on each assignment. It teaches students to check claims from AI against reliable sources. And it gives every student access, so the tool narrows gaps instead of widening them. The policy treats AI the way Edison's films should have been treated: as a powerful medium whose value depends on how it is used.