From Silicon Valley to the World Economic Forum, boosters insist that artificial intelligence will transform education—closing learning gaps and personalizing instruction for every child.
We’re betting it won’t.
Two years ago, we argued here in “Transformative for the motivated and mere meh for the unmotivated” that AI would supercharge motivated learners while doing little for the unmotivated. We proposed a bet on whether Khan Academy’s AI-powered tutor would lift outcomes broadly.
Those data still aren’t out, so here’s our new wager: If AI really delivers for typical students—not just the self-starters—then by 2030, the schools, districts, or countries that use the most AI should show the biggest test-score gains.
Our bet: There will be no relationship.
If anything, AI might widen gaps—slightly increasing the number of motivated kids racing ahead while the unmotivated stand still, not much moving the average.
We expect NAEP scores to fluctuate for all sorts of reasons—pandemic recovery, curriculum reforms, demography, funding. But if AI is the breakthrough its boosters claim, the pattern should be unmistakable: The more AI in classrooms, the larger the gains. We don’t think that pattern will appear.
Early randomized trials offer little reason for optimism. One review examined 2,500 studies of AI and learning. Only 51 met even minimal standards. Just eight were in K–12, nearly all with small samples and short-term outcomes like unit exams or teacher-designed rubrics.
No scholar has yet published a yearlong study of hundreds of students comparing heavy AI tutoring, AI-assisted teaching, and business-as-usual instruction with state test results.
A few results are intriguing but narrow. Jill Barshay wrote how AI temporarily helped Turkish high schoolers learn more science—until they forgot it faster. Stanford’s co-pilot study found AI nudges made human tutors to be 4 percent more effective. In another paper, those Stanford scholars evaluated something we created: our AI “talk meter,” which improved human tutor behavior by prompting the educator to talk less and getting the kids to talk more.
These were modest steps, not revolutions.
To be clear, some of our friends who build AI education tools are seeing legitimate, measurable gains—real learning improvements worth celebrating. Yet those impacts remain small, far from the sweeping transformation boosters predict.
David Labaree once wrote that schools are “remarkably resistant to change.” That remains true. Over decades, districts have implemented waves of “innovations”—ed tech, credit recovery, teacher evaluation systems, high-dosage tutoring, advisory programs—without mastering the details needed for impact.
AI is likely to follow that pattern. Some teachers are genuinely thriving with AI lesson planning and grading help—saving time. We applaud them. But there’s no evidence yet that this enthusiasm will translate into large-scale classroom transformation—or measurable learning gains.
Here’s what we see.
The good news: Some teachers are bringing back blue books to defeat AI-generated essays. Still, most students are using AI as a substitute for effort. The “who cares if nobody writes anymore” argument misses the point. Writing remains a key proxy for learning to make a reasoned argument. When students outsource that process, they lose more than grammar practice—they lose thinking practice.
Motivated teachers do get some genuine value from AI: better coaching scripts, faster feedback, lighter grading loads. But the scale of benefit is small. It’s unlikely to move national scores.
AI confers enormous advantages on kids who already care. The curious, the ambitious, the self-starters—they’re thriving. They’re using AI to learn coding, foreign languages, physics tricks, or even bass fishing at warp speed. Most kids, though, use AI like they use every other piece of ed tech: “How can I exert the least effort without getting in trouble?” That’s the “10 percent problem.” Roughly one in ten students use AI (or Khan Academy, or similar tools) as intended. The rest don’t.
AI may double that fraction—from 10 to maybe 20 percent. That’s progress. But it’s still far from the sweeping transformation boosters envision.
The missed opportunity
The most underused AI move remains simple: “Explain it to me like I’m in third grade.”
That command could unlock understanding for many struggling students. It’s not dumbing down; it’s scaffolding comprehension. But too often, students instead use AI to complete work they don’t understand—then hand it in for a grade. Teachers, aware of this, are reverting to handwritten work. It’s a modern replay of Ted Sizer’s Horace’s Compromise: both sides quietly lowering expectations to survive the system.
In many classrooms, kids now spend long stretches on assignments that have near-zero learning value, such as texts that are too hard to comprehend and problem sets that are too complex to solve independently. AI might at least prune those, helping students reach the same modest level of understanding more efficiently.
Imagine an unmotivated teenager told to clean his room. It takes 20 minutes to make the bed (poorly) and pick up his clothes (a few still miss the hamper). Now imagine a technology that achieves that same “sort of clean” in ten minutes. That’s not transformation. It’s efficiency.
But if AI can reclaim time otherwise lost to low-yield work, we could spend that time-dividend wisely. Earmark it for what teens need most outside school:
- Real sports and fitness they actually enjoy.
- Part-time jobs and volunteering.
- Pleasure reading and musical instruments.
- Unstructured adventures with friends.
AI may not lift NAEP scores or shrink achievement gaps. But if it frees hours for teens to build skills, health, and human connection, that’s the kind of learning worth betting on.
Robert Pondiscio once captured a hard truth about Success Academy charter schools, which he and we admire immensely. Robert wrote: “We can either attempt to serve all disadvantaged children equally, or we can do all in our power to ensure that receptive and motivated students can reap the full benefit of their talents and ambitions.” He was describing Success Academy’s model, but the same logic fits AI in education.
Early AI tools seem to, like other edtech, mostly help the receptive and motivated—the kids already leaning in—but that’s still real progress. If technology deepens engagement for those students, and even pulls a few more into that “motivated” camp, it’s a gain worth celebrating, not dismissing.
Mike Goldstein and Sean Geraghty founded the Center For Teen Flourishing.
