But the fight over A.I. in schools isn’t just a piece of the broader opposition. The logic of how A.I. infrastructure is being built remains somewhat intuitive and recognizably American — big companies with cash to burn buying up plots of land to develop for their own purposes. If a protester stands up at a Utah town hall to demand, about a 62-square-mile, $100 billion project proposed for Box Elder County, “who asked for this?,” the answer isn’t difficult to come up with. It’s the developer trying to build the data center and the hyperscalers desperate to make use of it.
With schools, it’s a lot less obvious how exactly we got here in the first place, with A.I. planted so quickly in classrooms and curriculums that parents and advocates are left to try to weed it out. Or at least institute some common-sense guidelines: more public input, more transparency about what products are being used, more restrictions on data collection and student privacy and above all, perhaps, more rigor regarding whether or to what degree the tools being used are actually helpful to student achievement. This isn’t just resistance from the familiar coalition, in other words. It’s parents shocked at how quickly some very new technology has so hastily remade school for their kids. These are 5- and 6-year-old brains we’re now running our Amira experiment on. Most Americans first heard of L.L.M.s less than four years ago, after all.
And it isn’t just the smartphone backlash that makes this rush to embrace A.I. so confounding. It was just four or five years ago that Covid-era school closures had produced widespread anxiety — among parents, among policymakers — about the inadequacy of remote learning and the costs to students of conducting school through a small, personalized screen. Much of that alarm looks to me, in retrospect, excessive, given the way that what was called “pandemic learning loss” did not significantly accelerate long-running declines in test scores that both predated school closures and have outlasted them. But the impression lingers enough that parents object when local officials propose that, in the wake of massive snowstorms, schools might go remote for a single day. Working from a laptop in class is not the same thing as remote-only school, of course, but the lessons we thought we learned five years ago — about the downsides of screen dependence and the importance of embedding learning in a social environment — would certainly seem to counsel some caution.
Three or four years ago, the arrival of sophisticated A.I. chatbots produced a wave of panic about the catastrophic impacts on learning in high school and especially college: an epidemic of cheating and plagiarism; a workaround for deep reading and critical thinking; and the onset of what has been called “cognitive atrophy,” the result of turning away from education as a kind of intellectual training and toward an approach to schoolwork as a matter of mechanical output. (You generate a term paper the way you buy a Snickers at the vending machine.) Much of the alarm was about how kids were essentially cheating themselves out of an actual education by using A.I. tools; the basic worry was that A.I. would find its way into the classroom, and into schoolwork, almost regardless of the precautions taken by teachers and administrators. But in many cases, now, it is the schools themselves that have incorporated A.I. tools into their curriculums, and while not all such tools are the same, to date there is pretty limited evidence that they might help.
So how did it happen all over again, and so fast, with yet another new technology? One partial explanation comes from genuinely hopeful evangelism, the possibility that new tools might allow us to do better. Another comes from a kind of related desperation about how hard it has been to end the stagnation of American educational achievement. (Would-be reformers have worried that American students were falling behind for many decades, and the past dozen years have traced some striking test-score declines.) To their boosters, the tools promise to unburden teachers and deliver something between a smaller class size and a genuinely personalized tutorlike school experience — a favorite fantasy for tech-minded reformers. And while many parents are likely to feel a reflexive queasiness about turning kindergarten over to A.I. tools, it’s not entirely clear how representative those voices are. There is also a fear of being left behind — among parents, schools and districts — and a sense that, if the world really is changing this fast, maybe kids really would benefit from early exposure.
