For thirty years I taught high-school mathematics in Canada. For one year before that, behind the Iron Curtain, I taught it for about fifteen US dollars a month. The two systems had almost nothing in common. But the communist one got one strange thing right.

In the Polish People's Republic, almost everything was censored. Mathematics was not. As I write in my book: "Geometry proofs, logical reasoning, deep algebra, physics — the regime left it all intact because it could not figure out a way to weaponize it for the party." For decades, people were taught, for free, how to think. Logic. Analytical reasoning. Proofs.

The regime never saw the trap it set for itself. Teach a whole population to follow an argument, and one day they stop believing arguments built out of slogans.

What the new AI tutors actually do

I think about that every time another AI tutor rolls into a classroom. Khan Academy's Khanmigo has scaled to hundreds of districts and [hundreds of thousands of students](https://is4.ai/blog/our-blog-1/top-10-ai-tutoring-systems-2026-learning-outcomes-208) in barely a year. And here is the detail that matters: it is built to refuse. Ask it for the answer and it will not give it. It asks a question back. It hints. It makes you do the work.

The early evidence says that refusal is the whole game. A 2025 Harvard study found students learning with an AI tutor absorbed more than [twice as much in less time](https://www.engageli.com/blog/ai-in-education-statistics) than peers in a traditional active-learning class. In a randomized trial, students whose tutoring was supported by Google's LearnLM were about [5.5 percentage points more likely](https://www.future-ed.org/research-notes-two-emerging-strategies-for-using-ai-in-tutoring/) to solve brand-new problems on their own afterward — 66 percent versus 61 — than students working with human tutors alone.

The tools that help all share one trait: they make the student reason. The tools that hurt share the opposite one: they hand over the answer and let the reasoning rot. It is no accident that around [70 percent of teachers](https://www.solidaitech.com/2026/06/ai-in-education-complete-guide.html) now worry AI is weakening students' critical thinking. Both things are true at the same time. The design decides which one you get.

The rigor we threw away

Here is what worries me, with thirty-plus years of chalk on my hands. North American schools quietly stripped the rigor out of math long before AI showed up. Logic was removed from the curriculum. Proofs in geometry were removed entirely. Students were asked to think deeply less and less often.

Now drop a tool that produces a flawless answer in two seconds on top of a curriculum that already stopped asking kids to prove anything. You do not get smarter students. You get faster surrender. The communist classroom, for all its poverty, forced the struggle that builds a mind. An answer-vending AI removes that struggle. A Socratic one restores it. Same technology, opposite result — and the difference is entirely in how it is built and how it is used.

Seven ways to make an AI tutor build thinking, not replace it

  • Choose tutors that withhold the answer. If a tool hands over solutions instead of hints, it is a cheat sheet, not a tutor.
  • Make the student attempt first. Ten minutes of honest, messy struggle before the app ever opens.
  • Ask for the why, not the what. Have the student explain the steps back to you with the screen closed.
  • Use AI to check reasoning, not to produce it. "Where did my logic break?" is worth more than "what is the answer?"
  • Keep the proofs alive. Demand at least one "justify it" or "show your work" on every assignment, with or without AI.
  • Separate admin from instruction. Let AI draft the rubric and the email; never let it do the thinking that was the lesson.
  • Watch the clock honestly. Homework that drops from an hour to ten minutes is not efficiency. It is outsourcing.

What we actually decide

The regime that raised me tried to control what people thought, and accidentally taught them how to think. We face the opposite danger now: the most powerful learning tool ever built, handed to kids inside a system that already forgot why struggle matters. The technology will not decide how this turns out. We will.

Bogdan Marzewski, Author of 'From Communist Paradise to Freedom'

Bogdan Marzewski, Author of 'From Communist Paradise to Freedom'
Previous Posts
June 3, 2026 · AI & Learning

Mastering the Thinking Pattern: Why AI Needs a Teacher's Wisdom

AI is now in 83% of classrooms — but the teachers seeing real results aren't the ones handing it the keys. They're the ones who know what AI can't replace: the discipline of deep thinking. A retired…

Read more →
May 20, 2026 · AI & Learning

Resilience in Learning: From 1989 Scarcity to 2026 AI Abundance

In 1989 Poland, the communist regime accidentally taught an entire generation how to think — and that skill brought down the wall. In 2026, AI is doing something similar. Here is what a retired math…

Read more →