China's Grand AI-Education Experiment and Why Your Kid Is Already in One

The Experiment Nobody Opted Into

MIT Technology Review reported on it. Hacker News lit up about it. Most US parents missed it entirely.

A Chinese ed-tech company called Squirrel AI built an algorithm that curates every lesson for every student based on real-time performance data. China deployed it at scale, across millions of kids, before Western education researchers had finished writing the first think-piece about whether anyone should. The speed of that rollout made headlines. But the speed is not actually the uncomfortable part.

The uncomfortable part is this: the same feedback loop is already running on your kid’s device.

Every Adaptive App Is the Same Experiment

Here is how the loop works, regardless of which platform your kid uses.

The app observes your kid’s responses. It builds a model of where they struggle. It serves content calibrated to that model. It adjusts based on how they perform. The data flows to the company. The profile gets richer. Your kid’s learning path gets narrower and more tailored.

This is not a flaw in the design. This is the design. And in narrow terms, it works. The test-score gains documented in cases like Squirrel AI are real. Adaptive platforms are genuinely good at finding weak points and drilling them.

The question MIT Tech Review raised, and that most ed-tech marketing carefully avoids, is what the loop is actually building over time.

Agency vs. Dependency: The Only Question That Matters

There is a real difference between a kid who improved their score because the algorithm found their weak points, and a kid who learned how to find their own weak points.

One of those kids, when they hit a problem they have never seen before, knows how to orient themselves. They can frame the problem, figure out what they do not know, and decide how to approach it. The algorithm is a tool they pick up and put down.

The other kid is good at following the next step the algorithm feeds them. They are fast, accurate, and genuinely optimized for the format the platform rewards. But without the feed, they are waiting.

That gap is not obvious from a grade report. It compounds quietly for years. And almost no adaptive platform has a financial incentive to tell you which type of learner your kid is becoming, because engagement metrics look identical for both.

What the Algorithm Cannot Teach

Personalized learning platforms are built around a specific definition of progress: performance on measurable tasks. That is a reasonable thing to optimize for. It is also a dangerously narrow slice of what it means to be capable.

The skills that actually compound over twenty years are the ones that resist easy measurement. Framing a problem from scratch. Reasoning under uncertainty, when there is no right answer loaded into a database. Deciding what to learn next and why, without a system nudging you toward the next module. Communicating an idea to someone who does not already agree with you. Knowing when a tool is helping you think and when it is thinking for you.

These are not soft skills or bonus content. They are the operating system everything else runs on. And they are almost completely absent from the feedback loop that powers most adaptive learning tools, including the ones currently sitting on your kid’s home screen.

What a Different Kind of School Looks Like

Globeskool is an online school for kids aged 8 to 16. Core subjects, yes. But the curriculum is built around a different question than most adaptive platforms ask.

Not: what content does this kid need next?

But: is this kid learning to direct their own learning, or just getting better at following directions?

The practical difference shows up in how work is structured. Real projects over passive consumption. Problems that require judgment, not recall. Technology as a tool the kid controls, not a system the kid is fed through. Critical thinking, problem solving, communication, creativity, and working with technology treated as core curriculum rather than afterthoughts.

The goal is a kid who, at 16, can sit down with an unfamiliar problem and know how to think about it. Not because an algorithm served them the right preparatory content, but because they have spent years practicing the act of thinking itself.

The Real Opt-Out

China did not ask permission before running its experiment. The US version of the experiment is already running, just with a friendlier interface and less public visibility about what is being built and for whom.

You cannot fully opt out of a world where algorithms shape learning environments. But you can choose what your kid is learning to do inside that world. The difference between a kid who uses the algorithm and a kid who is used by it is a real difference, and it is not determined by the algorithm.

If you want to see where your 8 to 16 year old actually stands right now, the Globeskool assessment takes about 5 minutes. It does not tell you what content your kid needs next. It starts with the right question.

Take the free assessment