What do we do about AI and homework?

Why the obvious rules both fail

Forbidding it does not work, and not only because it is unenforceable. The tools are useful, your child will use them for the rest of their life, and if there’s a rule that treats them as contraband, kids will learn nothing except to hide the tab.

Permitting it freely fails in a different way, because nobody gets caught. Homework gets done faster, the work looks fine, everyone is pleased, and the learning simply does not happen. There is no moment where this becomes visible, which is exactly what makes it something to pay attention to.

So draw the line around the task – don’t make the rules about the tool.

The question that actually works

Is it doing the hard part, or making a harder part possible?

There are reasonable uses of AI, which are often better than working alone. Having a concept explained a fourth way is valuable when the textbook and the teacher have both failed. AI can also be great for checking finished work and asking why something is wrong, proofreading drafts, or getting unstuck at eleven at night when the alternative is giving up.

But what’s not reasonable is using AI to produce the thing they were meant to produce: writing the essay, solving the problem set, summarizing the book instead of reading it.

Asking whether this technology is doing the hard part or supporting a child in getting through the hard part is also applicable to tools that do not exist yet.

Say why, or the rule will not hold

Whatever you land on, the reason needs to be communicated clearly.

A white paper for university computing departments makes this point better than anything else written this year. It argues that making a rule saying you can’t use AI to assist your work reads as a prohibition for its own sake. But the same rule restated as “this task exists to teach you something you will need later” connects the restriction of AI to a cognitive benefit. Their argument is that people who understand a limit engage with it honestly, and people who do not simply route around it.

Which is worth taking seriously, because a rule your child treats as arbitrary turns into a detection problem, and you will struggle to maintain it. Secret use is worse than negotiated use in every respect: you cannot see it, you cannot shape it, and you find out about it too late.

So say the reason out loud, every time, even when it feels labored. Do this not because your child will agree, but because a reason can be argued with whereas a decree cannot, and you would much rather have the argument.

What the evidence says

This is now documented at the top of the education system. MIT spent five months studying how its own students use AI and reported that learning requires productive struggle, and that offloading it leaves an illusion of learning in its place where students recognize the answer but aren’t able to produce it.

The report names the trap precisely. Cognitive surrender is falling back on the machine at the very first sign of difficulty, which is the moment the learning was about to happen. Students told the committee the pull was strongest when a deadline was close, which is worth knowing, because it means this is a problem of pressure rather than character.

Two findings from a survey of 1,204 children aged 9 to 17 sharpen this. Kids who find it hard to stay focused on schoolwork lean on AI the most, which means the tool is most attractive exactly where the struggle would have done the most good. And schools are teaching rules while judgment goes untaught: three quarters of kids say their school has said what they may and may not use AI for, but only just over half have been taught how to use it safely. The gap between those two numbers is the space this page is trying to fill.

The early signals the MIT report records are worth stating plainly – diminished critical thinking, weakened memory, eroded confidence, undermined mastery. There was also a social cost that surprised the committee, which is that leaning on a chatbot displaces the study group, the classmate and the teacher.

Notice who is saying this. MIT builds this technology and the report is openly enthusiastic about what it makes possible. It still concludes that a student who lets it do the thinking has been robbed of the learning journey.

What this looks like in daily life

Attempt first. Whatever the rule ends up being, the first pass is theirs. The struggle was the homework all along.

Then the explain-it test, which does most of the work on its own: they have to be able to talk through the finished piece without looking at it. A kid who used AI well can do this. A kid who let it do the work cannot, and both of you find out in about thirty seconds, without an argument about what happened.

Ask the school what its policy actually is, subject by subject. Most schools have something and many have not told parents. Your kid is likely getting different rules in different classrooms, which is confusing for them and useful for you to know.

Do the homework alongside them sometimes, on your own laptop, on your own work. Nothing teaches this faster than a child watching an adult decide not to take the shortcut.

What to say

When they say everyone uses it:

“Probably. What did you use it for?”

That question is not a trap and should not sound like one. Half the time the answer is completely reasonable and you can say so, which is what buys you the other half.

When you suspect the essay is not theirs:

“Talk me through it. Not to catch you out, I just want to hear it.”

When they say AI explained it better than the teacher:

“Good, that’s a real use. Now do the next one without it and see if it stuck.”

When they are up against a deadline and reaching for AI:

“You’re not stuck, you’re late. Those are different problems.”

When this doesn’t apply

Assistive use is a different category entirely. For a dyslexic child, a nonspeaking child, one with a working memory difficulty, or one writing in an additional language, these tools can remove a barrier that was never the point of the exercise. The hard part there is the thinking, and the tool is clearing the obstacle in front of it. That is the best possible use.

Older students working at the edge of what they can do alone are also a different case. Using AI to attempt something past that edge, and understanding what comes back, is the thing the MIT report calls augmentation, and it is worth encouraging.