The AI-Education Death Spiral: What Actually Dies When Kids Let AI Do Their Thinking

The Grade That Should Worry Every Parent

An economics professor at Brown University gave a take-home midterm this semester. His class averaged a 96. His historical average had sat in the high 60s for years.

He suspected most of his students had used AI. He was probably right.

The reflex reaction is familiar: blame the kids, roll out detection software, ban laptops, require handwritten responses. Repeat next semester when the workarounds evolve. This cycle has a name in some corners of education discourse. The AI-education death spiral. And the name is accurate, because each loop leaves students a little further from the thing schools were supposed to build in the first place.

The Wrong Problem Is Getting All the Attention

Here is the part that gets buried in the cheating conversation: the students who outsourced that midterm did not just skip an assignment. They skipped the friction. And friction, the kind that comes from sitting with a problem you cannot immediately solve, is how judgment gets built.

If your kid has never had to wrestle a math problem to the ground without help, they have no baseline for what “hard but solvable” feels like versus “I am genuinely stuck and need support.” If they have never had to find a flaw in their own argument before submitting it, they cannot tell when an AI response sounds plausible but is factually wrong.

That inability to evaluate the output of a tool is not a moral failing. It is a skill gap. And it is one that AI detectors, handwriting bans, and honor codes cannot fix, because none of those things teach the skill. They only punish its absence after the fact.

The Skill Nobody Is Naming

There is a meta-layer sitting above both “how to use AI” and “don’t cheat.” It sounds like a few simple questions:

• Can I describe what I am trying to figure out before I open the tool?
• Is working through this myself building something in me, or just producing an output?
• Would I recognize a wrong answer if the model gave me one?

These are judgment questions. And judgment is exactly what gets bypassed when a student goes directly from “here is the prompt” to “here is the AI response” without a stop in between.

The students at Brown did not fail because they used AI. They failed because no one had ever helped them understand when using it is fine and when sitting in the discomfort is the entire point of the exercise. Those are completely different situations, and knowing which is which is a learnable skill. Schools are just not teaching it.

What the Fix Does Not Look Like

It does not look like more surveillance. Proctoring software and AI detectors are already an arms race that education is losing, and they address none of the underlying problem. A student who passes a proctored exam without cheating still may not know how to evaluate sources, spot a logical gap, or articulate a problem clearly enough to get useful help from any tool, human or AI.

It also does not look like “no AI, ever.” That position is both unenforceable and, more importantly, wrong. AI is a real tool. Kids who grow up knowing how to direct it, question it, and use it in service of their own thinking will be in a meaningfully different position than kids who either avoid it entirely or let it think for them entirely. The goal is neither extreme.

What It Actually Looks Like

It looks like a kid who can close the laptop, make a real attempt, and then return to the tool with specific questions instead of a blank request for a finished product.

It looks like a kid who reads the AI output with the same skepticism they would apply to a source they found online, because that is exactly what it is.

It looks like a kid working on a real project with actual stakes, where the output matters to someone beyond a grade, and the kid knows that if the reasoning is wrong, the project fails.

That last part is important. Passive consumption of AI-generated content feels smooth and low-effort because nothing is real enough to break. Real projects break. Real problems push back. That friction is not a bug in learning. It is the mechanism.

Why This Is a Design Problem, Not a Character Problem

Schools built their assessments around exactly the tasks that AI handles best: recall, summarization, template-based writing, pattern-matching problems with clean solutions. Then the tools arrived that could do all of those things faster and with better surface-level polish than most teenagers. The system was not designed to survive that.

The response has mostly been to try to preserve the old design through enforcement. A more useful response is to build something different: learning that asks kids to direct the tool rather than be directed by it, to produce judgment not just output, and to understand that some struggle is load-bearing. You cannot remove it without weakening the structure.

That is the argument Globeskool is built around. Core subjects plus the skills schools consistently skip, including the meta-skill of knowing when your own thinking is irreplaceable. Not because AI is dangerous. Because you should be in charge of the tool, not the other way around.

If your child is between 8 and 16 and you are thinking about what their education is actually building, the free assessment takes about 5 minutes.

Take the free assessment