It's Not What AI Takes. It's What Potentially Never Forms.

The risk is not what AI takes from an adult mind, but what it may prevent a child's developing mind from ever building.

The AI-and-cognition debate is usually about what adults might lose. The sharper question is what a developing mind might never build.

An adult arm immobilized in a cast loses strength. The muscle weakens, the coordination dulls, the grip fades. Remove the cast, begin rehabilitation, and most of it comes back, so the capacity was built once, and the body remembers how to find it again.

Now immobilize a limb through the years when strength and coordination and the neural pathways governing movement are first supposed to form. That is no longer a story about strength lost. It is a story about strength that never arrives. There is no built capacity waiting to be recovered, because the capacity was never built. Rehabilitation has nothing to rehabilitate.

That difference, specifically between a capacity weakened and a capacity that never formed, is the one the AI-and-cognition debate keeps stepping over. I think it is the whole question.

The Worry We Already Have

Most of the worry about AI and the mind is a worry about subtraction. We talk about losing focus, losing memory, losing the patience to think a problem through when a finished answer sits one prompt away. The neuroscience behind that worry is sound. The brain essentially runs on use, with connections strengthening when they are exercised and weakening when they are not, and a faculty you stop exercising will quietly decondition.

What the Research Actually Measured

This year’s evidence points the same direction. An MIT Media Lab team wired up participants writing essays with an LLM, with a search engine, and with nothing at all, and watched brain connectivity scale down with the amount of outside help: strongest in the unaided group, weakest in the LLM group, whose members later struggled to quote work they had finished minutes before.

The researchers called the result cognitive debt. A separate study from Microsoft and Carnegie Mellon, surveying hundreds of knowledge workers, found that the more someone trusted the AI, the less critical thinking they reported bringing to the task, with automation quietly removing the routine reps through which judgment stays sharp. A review of seventeen studies found the same shape: AI lifts performance on lower-order work and erodes it on the higher-order reasoning that gets offloaded.

Unfortunately, both those findings measured adults (knowledge workers, university students) and fundamentally people whose judgment formed before the tool arrived. They are studies of deconditioning. They tell us what happens when a built capacity goes unused. They do not reach the question underneath.

The Question Underneath

Judgment is not a fact you store. It is a capacity you build, and you build it the only way capacities get built: by bearing the load. You learn to synthesize by struggling to synthesize. You learn to write by failing to write and trying again. You learn to reason through uncertainty by sitting inside the uncertainty long enough that something in you reorganizes around it. The friction is not the obstacle to the learning. The friction is the learning.

Which raises a question none of the current studies are built to answer. What happens when the tool that can do the work shows up before the capacity to do it has formed? When a student hands off synthesis before ever learning to synthesize, expression before finding a voice, the weighing of a hard call before having weighed a single one?

Developmental neuroscience says those two situations are not the same in kind. In the experiments that helped earn them a Nobel Prize, David Hubel and Torsten Wiesel showed that a kitten deprived of normal input to one eye during a brief early window never develops normal vision in that eye, because the cortex wires around the deprivation and does not unwire later. Deprive an adult cat the same way and it recovers. The difference is not severity. It is timing. During a critical period, experience does not refine a circuit that already exists; it determines whether the circuit forms at all. The human version of that early visual deprivation has a name, amblyopia, and a treatment window that closes. Miss it, and the door does not reopen.

What We Can and Can’t Claim

We know these windows exist for vision, for hearing, for language. Whether the higher-order capacities like synthesis, judgment, an authored voice, even have anything like the same windows is not settled. That uncertainty is the honest center of this, so let me put it plainly.

That experience shapes developing brain structure: established. That some capacities have sensitive periods, after which they are hard or impossible to build: established for the senses, for motor control, for language. That cognitive offloading reduces engagement and, sustained, can dull a skill.

What is not established by any long-run data, is that generative AI causes lasting underdevelopment of judgment in the humans now growing up with it. The studies run months, not years. The samples are small. Most are correlational and self-reported, and the strongest of them say, in their own words, that they are preliminary. Anyone who tells you the damage is proven is letting the fear outrun the data.

But anyone who tells you the risk is imaginary is making the opposite error, and the larger one. The mechanism is real. The early signal is strongly suggestive, especially in at least one large sample, the youngest users leaned on these tools the most and scored the lowest on the thinking the tools displaced. And the thing at stake here, whether a capacity forms at all, is exactly the kind of thing that, once missed, may not come back. There is no rehab plan for a capacity that never formed. That asymmetry is why the formation question deserves more weight than the deconditioning one, even though it is the one we can prove the least about today.

Scaffold or Cast

Which is where the tool stops being the question. AI can do two opposite things to a forming mind, and which one it does has almost nothing to do with whether it was used.

It can act as a scaffold. The word comes from how we already understand learning: a temporary support that lets someone work just past their current reach while they build the underlying capacity, designed to come down as competence rises. Used that way, AI lets a learner stretch beyond their level while still doing the construction themselves (the wrestling, the drafting, the deciding), so the capacity forms under the support and the support eventually falls away.

Or it can act as a cast. A cast takes the load off entirely. That is precisely what you want after an injury, and precisely what you do not want anywhere near a capacity still trying to form. The output looks finished. The grade comes back fine. And the developmental weight that the capacity needed to bear in order to form never lands at all.

What Chess Taught Me

I watched this run as a controlled experiment for fifteen years, without meaning to. My three sons played competitive chess, and the engines that can solve any position were available to all of them. The rule in our house was that you found the move yourself first, by sitting with the position, reasoning it through, and committing to a line. Only then checked it against the engine. Most kids ran it the other way, leading with the engine and absorbing its answer. They built the capacity in that order, and I believe it helped them excel on the national and internal chess level. The capacity formed in the children who did the work before they verified it, and it formed because they did the work first. Same tool. Opposite order. The order was everything.

The distinction will never show up in whether a student used AI. It shows up in what the student was still required to do.

The Real Question

So the practical question is not whether to allow these tools. The “horse is out of the barn” so to speak. That argument is over. It is whether we can shape their use, and our children’s use, to keep the load on the human at the exact point a capacity is still forming. Get that right, and AI is the finest scaffold ever built. Get it wrong, and we will have cast a generation’s judgment before it ever bore its own weight and unlike a muscle, that is not the kind of thing that comes back.