What Children Understand About AI That Adults Tend to Forget

4 days ago
2 min read

We often hear that AI is too complex for children. That you first need to understand mathematics, statistics, neural networks.

My experience running workshops at Codeacademy123 tells me the opposite.

Children arrive without preconceptions

When I introduce an AI tool to an 8-year-old, they don't wonder whether the machine "really thinks." They explore. They test. They observe what works and what doesn't.

Adults, on the other hand, often arrive with a verdict already formed: either AI is magical and all-powerful, or it's "just a statistical parrot" repeating what it has seen.

Both positions miss something essential.

What "compressing well" actually means

A language model doesn't memorize. It compresses. And compressing well means learning abstractions — patterns, structures, analogies across very different domains.

This is exactly what children do when they learn.

When a 6-year-old understands that adding and counting objects are the same thing, they are making a compression. They are generalizing. They are building an abstraction that will make learning multiplication easier later on.

LLMs do something structurally similar — at a scale and across data we can barely imagine.

Four levels of novelty

In my workshops, I help children distinguish what AI does well from what it does less well. We work with four levels:

Repeating or rephrasing what has already been said — AI excels at this
Recombining existing ideas in new ways — AI is already very useful here
Surfacing latent intuitions — paths that were present in the data but never explicitly formulated — this is where things get genuinely interesting
Major conceptual breakthroughs — a truly original idea with no precedent — this is far more debatable, and children quickly understand why

This framework helps children avoid two traps: blind fascination and outright rejection.

What children do naturally

Children ask questions without fearing they'll look ignorant. They rephrase. They push back. They say "but why isn't it working?" without giving up.

That is exactly the right way to interact with an LLM.

The real skill to develop — for a child as much as for an adult — is not knowing whether AI "thinks." It's learning to ask good questions, explore hypotheses, and surface ideas you wouldn't have found on your own.

What I try to pass on

At Codeacademy123, the goal is not to train AI users. It's to develop people who can dialogue with these tools — using them to amplify their own thinking, not to replace it.

Children grasp this very quickly. Often better than we do.

Because they haven't yet learned to be afraid of not knowing.