< BACK TO POSTS

AI Will Make Younger People into Super Geniuses

AI GENERATED / JEREMY CURATED

September 2, 2026

Kids who don’t grow up with AI won’t just be “less techy.” They’ll be operating with stone tools in an electric world. That’s the part most school-policy debates refuse to say out loud.

A kid in 2040 who can think with a model the way you think with language will look like a different species next to a kid who was told “don’t touch it until you’re 18.” Not because the model is magic. Because personalized practice at scale is the closest thing we’ve ever had to a private tutor for every child, and private tutors have always been how elites pulled away from everyone else.

The world before electricity looks foreign to us. Not because people then were stupid. Because the environment changed and the people who adapted built on top of the new layer. AI is that layer for cognition. Pretending otherwise is costume-drama education policy.

The actual upgrade is tutoring, not “chatbots for homework”

One-on-one human tutoring is the most reliable learning intervention we have. Bloom called the gap the 2-sigma problem in 1984: a good tutor can move a student roughly two standard deviations above a normal classroom. Schools never solved it because you cannot hire a patient expert for every child for every hour.

AI doesn’t have to match a great human tutor to matter. It has to be good enough, always on, and cheap. That’s already happening in pieces.

Will that mint a generation of “super geniuses”? Not automatically. Most kids will still be average kids with better tools.

The honest version is sharper: the ceiling moves, and the kids who treat AI like a thinking partner instead of an answer vending machine will do work that makes 2026 look primitive. Research compression, simulation, code, design, lab iteration — those compound. The first generation that grows up fluent in that loop will ship breakthroughs that make our stack look like hand tools. That’s not sci-fi. That’s what electricity did to muscle and what computers did to paper.

The kids who never develop alongside the tool don’t get a noble “pure mind.” They get a lag they won’t close later.

Calculators and computers already ran this experiment

People said calculators would rot arithmetic. They did rot unexamined arithmetic. They also made it possible to teach actual math instead of turning every student into a human abacus.

People said computers would kill writing, research, and memory. They killed a bunch of clerical skills. They also created new ones that paid: search, modeling, debugging, documentation, collaboration, knowing when the machine is lying.

The pattern is stable:

• The tool deletes a low-level skill.
• Schools panic and ban the tool.
• The labor market starts paying for people who can use the tool without becoming the tool.
• The kids who only practiced the old skill by hand and never learned the new stack get the worst of both worlds.

AI is the same pattern with higher stakes, because it sits on top of thinking, not just calculation. Ban it and you don’t preserve thinking. You preserve a classroom that no longer matches the world the kid will work in.

When to use it. When not to.

This is the part adults keep skipping because “ban vs allow” is easier than teaching judgment.

Don’t use it when the struggle is the skill

First pass at a proof. First draft of your own argument. Sounding out a word. Holding a times table in working memory. Explaining an idea out loud with no screen. Sitting with a hard problem long enough to feel where you’re stuck.

If the model does that for you, you didn’t learn the move. You rented it.

That’s the rule. If you cannot do the task with the model off, you do not own the skill.

Use it when the bottleneck is access, feedback, or range

• You already tried. You’re stuck. You need the concept explained five different ways until one lands.
• You need infinite practice at the exact edge of what you can do.
• You’re comparing approaches, checking a proof, generating counterexamples, simulating a system.
• You don’t have a tutor at home and the classroom is pitched at the median, which is most classrooms.
• You have a disability, a language gap, or a schedule that school was never built for.

The adult version of this is already how competent people work. Nobody serious hand-computes a regression or pretends they don’t use search. The difference is they can smell a bad answer. Kids need that smell trained on purpose.

A usable classroom rule is boring and correct:

1. Attempt without the model.
2. Use the model as a tutor, not a ghostwriter.
3. Close the model and redo it.
4. If you can’t redo it, you didn’t learn it.

Schools that skip step 1 and 3 will produce fluent prompt-users who collapse under a closed-book test. Schools that skip the tool entirely will produce kids who can pass analog tests and then get smoked by peers who grew up iterating at 10x speed.

Banning the best available tutor is not “protecting childhood”

If an AI tutor is the best thing that can teach a specific kid a specific thing today, it is not the job of government to ban it outright.

Safety rules for companion chatbots, porn generators, and unlogged data slurps? Fine. Age-appropriate design? Fine. Force vendors to refuse answers and demand student reasoning? Fine. Require AI literacy so kids know models hallucinate, flatter, and invent citations? Necessary.

A blanket ban on student-facing generative tools through 8th grade is a different animal. New York City just did that for the 2026–27 year in the largest school system in the country — about 600,000 kids — plus screen-time caps. High schoolers get limited, vetted use and two 45-minute “literacy” modules. Teachers’ unions have been pushing the same direction: keep the youngest kids off chatbots entirely.

Some of the fear is real. Little kids need human faces, play, handwriting, and boredom. Unsupervised chat is a slop hose. Cheating is already off the charts because schools kept assigning take-home essays into a world that can write them in 20 seconds.

None of that justifies pretending the tool doesn’t exist after 3 p.m.

Rich kids will use it at home. Motivated kids will use it at home. The ban hits the child whose only supervised computer is at school.

That’s how you manufacture a new literacy gap and call it child protection.

The honest policy is narrower and harder:

• No companion bots in elementary school.
• No “write my essay” mode on school networks.
• Yes to Socratic tutors that already know the curriculum.
• Closed-book work for the skills you claim to be measuring.
• Open-tool work for the skills the world actually pays for.
• Teachers who know the difference.

If you can’t tell those apart, you shouldn’t be writing the ban.

The raw split

Two kinds of adults are going to raise two kinds of kids.

One kind treats AI like cigarettes: hide it, moralize it, then act shocked when the kid hits 16 and mainlines the unrestricted version.

The other kind treats it like fire and electricity. You don’t hand a toddler a welder. You also don’t raise them in a house with no power and call it virtue.

The kids who learn to think first, then think with the machine, will make the rest of us look slow. Not because they’re genetically upgraded. Because they practiced on a different instrument.

The kids who were protected from the instrument will still have to live in the world it built. That’s the part the bans never price in.