Risks and guardrails12 of 13
Doing the hard part yourself
Struggling doesn't mean that kids aren't learning; rather, struggle literally is them learning, and handing a task to a machine prevents them from having a learning experience.
How to explain it, by age
Same idea every time. Pick the version that fits for the child's age, and level up when it requires more complexity.
Ages 4-6
Long before chatbots, the idea is just that the hard part is the good part. "It is meant to be tricky. That feeling you get in your body when it is hard can feel uncomfortable, but it means your mind is getting smarter." Then hold yourself back and let her keep going.
Try this
Next time she is stuck on a puzzle or a zipper, count silently to twenty before helping. Most of the time she gets there. The twenty seconds are harder for you than for her.
Listen for
"I did it myself." That is the feeling you want them to have, and at this age it needs no explanation at all.
If the child already has more access than most: If she already asks an AI agent for answers, start asking her to guess first, every time, before anyone checks.
Ages 7-9
Now connect it to the machine. "If you ask it for the answer, you get the answer. But your brain does not get any stronger, because the part where you learn was the part you skipped. It is like watching someone else do basketball practice." Kids this age understand practice completely.
Try this
Ask it to explain something she already knows well, then have her grade it like a teacher. Being the expert for once flips the whole relationship, and she will find mistakes.
Listen for
Her distinguishing "it told me" from "I figured it out". Once she can hear the difference, she starts noticing which one she has done.
If the child already has more access than most: Agree on a rule she helps write: attempt first, ask second, and be able to explain the answer to somebody afterwards.
Ages 10-12
Name the trap. Researchers call it cognitive surrender: falling back on the machine at the very first sign of difficulty, which is, of course, the moment the learning was about to happen. It feels efficient. It produces the illusion of learning, where you recognize the answer and could not have produced it.
Try this
Have her do a piece of homework with AI, then close everything and redo it cold a few days later. This is more convincing done once than argued about ten times.
Listen for
Her catching herself reaching for it out of reflex, with no need in sight.
If the child already has more access than most: Introduce the better question: is it doing the hard part, or making a harder part possible? Using it to attempt something beyond her alone is a different act from using it to avoid something within reach.
Ages 13+
This is now documented at the top of the education system. MIT spent five months studying its own students and concluded that learning requires productive struggle, that offloading it produces an illusion of learning, and that early signals point to weakened memory, eroded confidence and undermined mastery. Worth noting who is saying it – an institution that helped build this technology and is not remotely against it.
Try this
Have him read the MIT report's principles himself, particularly "augmentation not automation", and then write his own AI policy for his own work. If he writes the rule himself, he's more likely to follow it.
Listen for
Him drawing the line somewhere specific and defensible, rather than either refusing the tools or using them for everything.
If the child already has more access than most: The same report found AI increasing isolation by displacing study groups and time with teachers. Ask whether he has noticed that in his own week, and what the impact was.
Why this one matters most day to day
Most of this curriculum is about systems your child does not control – who owns the feed, where the training data came from, what the company earns. This one is about a choice they make at the kitchen table on a Tuesday night, and has the most immediate consequences.
It is also the concept with the shortest distance between understanding and behavior. A child who grasps that the struggle is the learning has a reason to sit with a hard problem.
Watching is not doing
Computing education researchers have a name for observing without doing: pseudo-apprenticeship.
An apprenticeship works because the apprentice does the work with somebody watching. Take away the doing and leave only the watching, and what remains looks like learning from the outside but develops nothing. A child who follows along as an LLM solves their problem has had a demonstration, not instruction. Demonstrations are pleasant and they feel productive, which is why this is hard to spot in your own kid.
The giveaway is that your kid can’t do the next problem.
Not an argument against the tools
It’s worth being careful here, because the obvious version of this lecture backfires.
Telling a teenager not to use AI for schoolwork is both unenforceable and sometimes incorrect. These tools are useful for having a concept explained in a different way, for checking finished work, proofreading drafts. It’s not about whether to use it but what you are using it for.
Is AI doing the hard part, or making a harder part possible? That question works at every age, on tools that do not exist yet.
The source is the interesting part
The strongest version of this argument currently comes from MIT: five months of work, students and faculty across every school, published in August 2026.
It is no longer the only version. Computing education researchers studying university beginners found that people who produced correct work with AI help overestimated how much they had understood. Different field, different method, same finding.
That matters, because MIT builds this technology. The report is enthusiastic about what AI makes possible, and it still concludes that a student who lets it do the thinking has been robbed of the knowledge they sought to achieve. That is a much harder argument to wave away than the same point made by somebody who dislikes computers.
Misconceptions to address
That finishing faster means learning faster. Speed and learning come apart completely once a machine can produce the answer, and the feeling of having understood something arrives either way.
That this is about cheating. Cheating is a school rules problem. This is about the kid cheating themselves out of the thinking, which nobody catches and no policy fixes.
That AI is useless for schoolwork. It is good for some things: explaining a concept a different way, checking finished work, arguing back. The question is always whether it is doing the hard part or making the hard part possible.
Adults do exactly the same thing at work, then find it surprising in children.
Running this with a class
Set the same problem twice, a week apart. The first time, let students use whatever they want. The second time, no tools, no notes. Discuss what the first attempt felt like at the time.