I feel like Morpheus offering you two pills, except both pills are just facts you’ve been avoiding.
Fact one: We’ve known for 40 years what actually works in education. One-on-one tutoring. Benjamin Bloom’s 1984 “two sigma” study showed students receiving individual instruction performed two standard deviations better than those in conventional classrooms — 50th percentile to 98th percentile.
Fact two: We’ve spent those 40 years pretending this doesn’t matter because individual tutoring doesn’t scale.
As a Stupidologist, I’m interested in why we’ve convinced ourselves that knowing the solution but not implementing it is somehow different from not knowing the solution at all. It’s a particularly sophisticated form of stupidity—acknowledging truth while living as if it’s false.
But here’s where it gets interesting: the constraint that made us stupid is gone.
Stanford’s 2024 study randomized nearly 1,000 students to receive tutoring from humans with or without AI assistance. Students whose tutors had AI support showed a 4 percentage-point gain in mastery. But the striking finding was this: the most significant benefits went to students with the weakest tutors. AI assistance improved their outcomes by 9 percentage points, essentially eliminating the gap between novice and expert human tutors.
A 2025 Harvard study compared students using sophisticated AI tutors against those in high-quality active-learning classrooms. The AI group learned more than twice as much material in the same time. Meta-analyses show intelligent tutoring systems outperform conventional instruction in 92% of controlled studies, with effect sizes large enough to move median students to the 75th percentile.
This isn’t hype. It’s peer-reviewed science involving thousands of students.
AI tutors work because they solve tutoring’s core constraints: they’re infinitely patient, available 24/7, can explain concepts dozens of different ways, and cost essentially nothing to scale.
A student stuck on derivatives at 11 PM can get immediate, personalized help. An adult changing careers can learn new skills without hiring expensive tutors. A child in rural Mississippi can access the same quality instruction as one in Manhattan.
The technology works. The question is whether we’ll use it to reduce educational inequality or deepen it.
One direction leads to AI tutoring as a luxury good. Wealthy districts and families get AI-enhanced education. Everyone else gets overcrowded classrooms and teacher shortages. The tutoring gap—already a major driver of inequality—becomes permanent and algorithmic.
The other path treats AI tutoring as public infrastructure, like libraries or public schools. Every student gets access. Teachers get training in how to deploy it effectively. We build quality control systems to catch AI hallucinations and verify accuracy. The technology becomes the foundation of genuine educational equity.
Right now, we’re choosing the first path—not deliberately, but by default.
AI tutoring is fragmenting into commercial products, each behind its own paywall, each gathering data on our children, each optimizing for profit rather than learning. Teachers get no training. Quality control is an afterthought. Privacy protections don’t exist. And we’re moving at breakneck speed with essentially zero public oversight.
This is how you screw up a revolution.
The difference between these paths isn’t technological. It’s political and economic. And the window to choose the better path is closing fast.
We made different choices with public libraries, public broadcasting, and public schools. We can make different choices here. But only if we hurry.
Universal access means it must be free for every student, regardless of where they live or their household income. I’ll keep pushing this point: this isn’t charity—it’s infrastructure, like roads or water systems.
The human-AI partnership? This works best if we agree that AI rides sidecar to human tutors—it doesn’t replace them. Teachers need training in how to use these tools, when to deploy them, and how to verify their outputs. AI handles routine explanation; humans provide wisdom, encouragement, and judgment.
Now, I wouldn’t be me if I didn’t scream about quality control. We can’t be so full of tech optimism that we ignore the fact that AI tutors hallucinate regularly—inventing sources, stating confident falsehoods, and making up facts. We need systematic verification, teacher oversight, and transparent error reporting.
To alleviate concerns about privacy protection, commercial AI tutors can’t be allowed to harvest detailed information about how millions of children think and learn. This data must be protected as rigorously as medical records.
The best AI tutoring systems learn from millions of interactions. They continuously improve. That improvement should benefit all students, not just paying customers.
None of this exists yet. There’s an ecosystem of special interest groups building the plane while flying it. Most of the building is being done by companies with profit motives that prioritize profit over the public good. This is why there needs to be public money and public oversight to create a public system of open learning.
And, as always, I see the important nuance. I’m equally an innovation hawk.
Families are seeking high-quality educational resources. They should find them.
If nothing else, AI can make intelligent tutoring economically viable at scale.
The pieces are in place for transformation. What’s missing is the political will to build public infrastructure rather than accept piecemeal solutions.
We need federal investment in AI tutoring research and deployment, modeled on the National Institutes of Health. We need state-level initiatives making AI tutoring (and human tutoring, BTW) freely available through public libraries, community organizations, and schools. We need teacher training programs that prepare educators to work alongside AI rather than compete with it. Colleges of education, please catch up.
We need privacy protections that treat children’s learning data as sacrosanct. That may be easier said than done. Just do it.
Most of all, we need to reject the assumption that education is a market where families compete for scarce resources. The genius of AI tutoring is that it’s not scarce. One system can serve millions simultaneously.
The only scarcity is the political imagination to treat it as a public good.
This piece was originally posted on Verbatim.