What Happens When Kids Use AI for Hard School Problems? A Teacher’s Observation
Last Tuesday, I flipped off the main classroom lights and was shoving my stack of graded quizzes into my tote when Mia knocked on the half-open door. She’s a quiet 7th grader who usually turns in all her work on time, but that week’s algebra homework had three multi-step word problems she’d left blank. She shifted her weight on the scratchy old carpet by the door, picking at the strap of her backpack with chipped blue nail polish, and said she needed help. I pulled my chair back out, asked her to walk me through what she’d tried so far. “I put the whole question into ChatGPT,” she said, “but the answer it gave me doesn’t match the back of the book. I don’t know what I did wrong.” I asked her to read the question out loud to me, then tell me what it was asking for. She stared at the page for 10 seconds, then said, “Wait. I didn’t actually read that far. I just copied it.”
I’ve been teaching middle school math for 12 years, so I’ve seen every trick in the book for getting through hard homework. Ten years ago, if a kid got stuck, they’d stare at the page for five minutes, complain that it was too hard, flip back three chapters to find a similar example, cross out two or three wrong starts, and then raise their hand to ask if their first step was even right. Now? Most kids who hit that wall don’t do any of that. They pull out their phone before they’ve even finished reading the question.
A month ago, I assigned a small group project where kids had to plan a 10-block neighborhood park, work within a $500,000 budget, and justify every line item. One group turned in a beautifully formatted, perfectly balanced plan in 22 minutes. Every number added up, every justification sounded professional. When I pulled the group aside to check in, I pointed to a line that said $11,500 for new playground mulch and asked one of the kids why they chose that number. He shrugged and said, “AI gave it to us. It seemed right.” None of the four kids in the group could tell me how they’d calculated how much mulch they’d need for the playground area they’d designed. They’d skipped the entire messy part of the problem.
That’s the change I’ve seen that doesn’t get talked about enough. It’s not that kids are cheating more, or that their grades are worse (though I do see a lot more high homework scores and low test scores these days). It’s that regular use of AI to solve hard problems has shifted how kids start any difficult task. The most important step of any hard problem—figuring out what the problem actually is, what you already know, and what you might try first—has become optional. That muscle, the one that lets you sit with being stuck for 10 minutes without quitting or running to a tool to give you the answer, doesn’t get used. It doesn’t grow.
I talked about this with a mom at parent-teacher conferences last month who has a 10th grader in high school chemistry. She said he gets 100% on every homework assignment, but fails every unit test, and no one could figure out why. He told her he uses AI to check his answers, but when she sat down next to him one night, she saw he plugs the whole problem into AI before he writes down a single number, then copies it over. He never practiced working through the problem on his own, so when the test comes and there’s no AI to ask, he doesn’t know where to start.
I’ve been testing small changes in my class this year, and I tell parents to try the same at home, that don’t require banning phones or AI entirely. The first one is the 10-minute pre-AI rule. Before any kid opens a tool to help with a hard problem, they have to write three things down on a piece of paper: what the question is asking for, one thing they know that could help, and at least one possible first step they could try. It doesn’t matter if the first step is wrong. The point is they do the work of unpacking the problem before they get an answer. I started this with Mia after that afternoon after class. She wrote down her three things, realized halfway through that the question was asking for total ticket sales, not the difference between adult and student sales like AI had assumed, and got the right answer on her own before she even thought to check AI.
Another small shift I tell parents to make: instead of asking your kid “did you finish your homework?” or “did you get all the answers right?” ask “what was the hardest problem you worked on today? Walk me through what you tried when you got stuck.” You don’t have to understand algebra or chemistry to ask that question. Even if your kid fumbles through the explanation, just having to talk through the process makes them engage with the problem in a way they wouldn’t if they just copied an AI answer. I’ve had parents tell me this has already changed how their kid approaches homework—they know they’re going to have to talk through the work, so they actually do the work first.
I don’t think AI is going anywhere, and I don’t think banning it is the answer. Some of my top students use it to work through concepts they don’t get after they’ve tried the problem on their own, and that’s a great use of the tool. What I do see is that small, consistent habits keep the problem-solving part of learning from disappearing. Some kids still build that tolerance for being stuck, even with AI around, if they’re expected to do the first step of the work themselves.
We’re all still figuring this out, honestly. Some weeks I see more progress, some weeks I catch half the class turning in AI-generated work that none of them understand. But the 10-minute rule and the three questions are working better than any big ban or lecture I’ve ever tried. It’s not a perfect fix, but it’s a start.
Please indicate: Thinking In Educating » What Happens When Kids Use AI for Hard School Problems