Do kids still need to learn math when AI can do it for them?

Do kids still need to learn math if AI can do it? Yes, and the reason is not "because school says so." The people building the tools, and the labor economists watching what they change, keep landing in the same place: when the answer becomes free, the thinking behind the answer becomes more valuable, not less.

When the answer is free, the reasoning is the point

David Deming, the Harvard economist and dean of Harvard College, said this plainly to his own students in 2026: ["You can ask AI about anything, and it'll give you the answer, but the skills you acquire in the process of developing expertise are more valuable than ever."][1] He models it by choosing to work problems himself rather than outsource them, not out of nostalgia, but because the doing is where the reasoning gets built.

The point is not that AI is bad. Outputs are cheap now and judgment is not. A child who can produce a plausible answer to a fraction problem, and check whether it makes sense, holds something the chat window does not: the reasoning that lets you tell a right answer from a confident-sounding wrong one.

The people building these systems tend to land in the same place. Andrej Karpathy, a founding member of OpenAI and formerly the head of AI at Tesla, was asked what a child should actually learn for an AI future. His answer was mostly math, with some physics and computer science, and not because the other subjects do not matter. Math, in his view, is the best way to build the thinking itself: working with ideas rather than memorizing facts a machine can recite faster than any of us. It is a striking thing to hear from someone whose whole job is teaching computers to do the reciting.

Powerful tools raise the value of expertise, not lower it

MIT labor economist David Autor made [the same argument from the labor side in 2024][2]: "Rather than making expertise unnecessary, tools often make it more valuable by extending its efficacy and scope. And the more powerful the tool, the higher the stakes." To use freely available electrical expertise on a home project, he still needed his own foundation: enough procedural knowledge to handle high voltage safely, enough judgment to tell when the job had gone off script.

Employers are voting the same way. The WEF Future of Jobs Report 2025 found that nearly 70 percent of employers globally now name analytical thinking as an essential core skill, the single most-cited one on the list. The demand is not shrinking as the tools get better. It is growing.

The math a child builds now is the foundation for the analytical work later. A language model cannot hold the number sense in your head for you.

Why fluency still matters when a calculator is in every pocket

There is another worry sitting underneath all of this: maybe the old curriculum is the wrong curriculum, and drilling procedures is beside the point in a world with calculators in every pocket.

Jo Boaler, a Stanford professor of education, argues that the single most valuable math skill a person can develop is stepping back and judging whether a calculation is reasonable. That is precisely the skill that checks a machine's answer. It is also the skill you cannot build without doing enough math yourself to have a feel for what right looks like.

Bill McCallum, the lead writer of the Common Core math standards, made the calm version of the fluency point recently: procedural fluency and fact recall are not busywork. They free working memory for solving harder problems and learning concepts. A fifth-grader who has to think hard about six times seven has no room left for the reasoning the actual problem requires. A fifth-grader who knows it cold has that room.

Picture a fourth-grader named Ava, checking whether the AI's answer of 3/8 to a mixed-number problem could possibly be right. A solid feel for how fractions behave, that adding two pieces bigger than a half should land above one, that 3/8 is smaller than a half, catches the mistake in a moment. Without that feel, the plausible-looking answer wins.

The strongest skeptic, honestly answered

The most serious version of "why learn math if AI can do it" comes from Conrad Wolfram, who argues that schools should teach computational thinking over hand calculation. Some of what a fifth-grader does with a pencil today, a computer will do faster and better every year.

But Wolfram is explicit about what does not change. Using computers, he writes, is not cheating and does not make everything easier. The human job becomes structuring problems and formulating the right questions. That is more math reasoning, not less. Whether a child practices it on paper or on a screen, the underlying number sense still has to be built inside the learner's head. A tool that produces an answer does not install the judgment to evaluate it.

The goal was never a human calculator

Learning math in the age of AI is not about competing with the machine on arithmetic. It never was. The goal is a child who can tell when an answer is wrong.

The real question is not whether kids still need to learn math. It is which specific foundations this child has solid and which are still wobbling. A report card will not answer that; it shows a grade, not the exact skills two rungs down that are giving way. This is where Level Up starts. The assessment traces down through the prerequisites to find the skills a child has not yet mastered, K-8, and the daily practice fills those gaps in the order that builds real understanding. The point was never to out-calculate a computer. The point is that when the computer answers, the child can tell.

[1]: https://www.thecrimson.com/article/2026/4/14/deming-ai-literacy-ec50/ [2]: https://www.noemamag.com/how-ai-could-help-rebuild-the-middle-class/