In a 2023 TED Talk watched by millions of people, the American educator and entrepreneur Sal Khan declared that AI was about to deliver “probably the biggest positive transformation that education has ever seen.” The founder and CEO of Khan Academy was touting the company’s new educational chatbot, Khanmigo, claiming it promised to be an “amazing personal tutor” to “every student on the planet.” By 2024, OpenAI CEO Sam Altman was chiming in that AI was on the verge of delivering, for students, “virtual tutors who can provide personalized instruction in any subject, in any language, and at whatever pace they need.”

But by this spring, Khan had admitted that the release of Khanmigo was “a non-event” for many kids. Although access exploded, from reaching 40,000 students in 2023 to nearly 1 million this year, actual uptake—whether students use it—has stagnated.

A tool designed to respond to questions and ask follow-ups can’t help a student who doesn’t engage or know what to ask. Khanmigo, like so many other ed-tech tools, has floundered because it hasn’t solved the challenge at the center of education: How do you motivate students to experience the discomfort of learning something new? An AI tutor may be able to deliver math problems that are perfectly calibrated to a student’s level. But it can’t make the student actually do the problems.

“Learning is hard work,” Kristen DiCerbo, Khan Academy’s chief learning officer, told us. “It’s cognitively effortful and not experienced as fun. How do we get kids to want to do that?” AI is a powerful tool, she added, but it can’t be expected “to bridge that motivation gap.” Although AI tutors have sometimes proven valuable in low-resource schools in developing countries, a recent Stanford review of all of the available research into the use of AI in K–12 schools found that educational benefits for students generally were limited.

Read: Is schoolwork optional now?

Only about one in three students is highly engaged in school, according to U.S. census data—a share that has remained stable over the past decade. These students, who also tend to come from wealthier homes with two educated parents, may well be motivated to seek extra guidance from a bot. But a motivated minority will not produce a revolution.

Even among the driven few, only a fraction of kids use ed-tech tools such as Khanmigo enough to see any gains. Laurence Holt, the author of The Science of Tutoring, calls this the 5 percent problem: About 5 percent of students use education technology as intended, thus reaping the learning benefits. That means that instead of democratizing access to affordable tutors, these tools could very well widen inequality by supercharging students who are already motivated to get ahead.

Personalized 24-hour systems and adaptive algorithms held such promise, but apparently no amount of animation or gamification will convince a student to care about learning if they don’t already. Khan has lately hedged any talk of a digital transformation. “I think our biggest lever is really investing in the human systems,” he said in a Chalkbeat interview in April.

Essentially, these ed-tech experiments have driven home what educators have long intuited: Learning is a largely social and relational enterprise, and bots have yet to replicate the value of a human touch. Teachers are still our best source of motivation for students, not only because strong ones know how to push kids to learn new things, but also because education works best when it happens in a group.

Ron Ferguson, the director of Harvard’s Achievement Gap Initiative, has found that successful teachers motivate students by pressing them “to think rigorously and persist in the face of difficulty,” creating moments of fruitful collaboration along the way. Students have a deeper understanding of thorny concepts when they discuss and debate them together, and they feel inspired to care more about mastering quadratic equations when they see their peers are trying to do the same.

This is not surprising. People who hate exercise are more likely to push themselves and hold themselves accountable in group fitness classes than they are on their own. The problem is that too many teachers are failing to motivate students, and the peer effect can go both ways, depressing student achievement in places where ambition isn’t valued. Many students come to class with different backgrounds, interests, and learning needs, and are greeted with a curriculum that can feel rigid, boring, and far removed from the world around them. Strong teachers who adeptly exploit group dynamics may be essential to academic excellence, but this approach is woefully hard to scale.

The solution is not to presume that more easily scalable digital tools will magically solve these problems, but to improve the performance of teachers in the classroom. This starts with the hiring process. In public high schools, where student disengagement is highest, teachers are typically selected on the basis of which degrees and credentials they’ve earned, and of how familiar they are with the subject matter. But many of these teachers come from graduate programs that prioritize theory over practice, and knowing various pedagogical approaches does not necessarily translate into teaching well. Navigating the fog of a student’s confusion is a powerful skill that most educators need help developing. Most good teachers aren’t born but made, with plenty of coaching and feedback.

Read: The human skill that eludes AI

Ferguson led a study in which researchers surveyed hundreds of thousands of sixth to ninth graders in classrooms across the country about what made their teachers effective. He found that the teachers who inspired students to work hard and aspire to go to college shared a number of key qualities. They were caring, with a knack for making students feel like their success genuinely mattered to them, and they were captivating, capable of sparking and sustaining the interest of students, even those who arrived in class apathetic and incurious.

These qualities needn’t be innate. Studies by Roland Fryer Jr. of Harvard University show that the observations and notes of more experienced peers and mentors can help teachers improve in the classroom. Teachers can learn to create a sense of community in their classroom, call on students regardless of whose hand is raised, and hold everyone to realistically high standards.

“We are social beings,” Mary Burns, a former teacher and current educational technology practitioner and researcher, told us. “We want to learn with and from other people.” Burns points to the learning loss during the coronavirus pandemic as evidence of what happens when we underestimate the value of learning communally. When students were isolated at home, without peers and often beyond the reach of teachers, “we saw a psychic break,” she said.

Training AI in the skills of the best teachers would seem to be far easier to scale than finding and training more teachers. But until we figure out how a bot might motivate young people to learn and do hard things, even the most advanced AI won’t serve most students. Many kids “do not care about the things we are teaching them,” Justin Reich, the director of the MIT Teaching Systems Lab, told us. Instead, he said, they care about impressing their teachers and cooperating and competing with their peers. “They care about the people.”