Is AI Dependency Changing How Students Learn Difficult Subjects? What I’ve Seen As A Middle School Math Teacher
Last Tuesday, I wiped crumbs of leftover fun-size Snickers off my desk after school and looked across at Mia, one of my 8th grade algebra students, staring at a quadratic equation problem that had stumped her all period. Before she’d even picked up her pencil to write a single number, her hand drifted to her phone in her hoodie pocket. I didn’t even have to ask what she was reaching for; she pulled up ChatGPT before I could open my mouth to remind her of our class rule. “I just know I’m gonna mess it up anyway,” she mumbled, when I raised an eyebrow. “Why waste the time if I can just see how it’s done right the first time?”
I’ve been teaching middle school math for 12 years. Long enough to remember when kids snuck answers off the back of the textbook, when they copied the kid sitting next to them who always finished early, when they’d pull up answer keys on Google five minutes before class started. Every parent that comes in for a conference these days asks the same question: Is AI dependency ruining how my kid approaches hard stuff? Are we watching them lose the ability to work through hard problems on their own?
For the first year after ChatGPT blew up, I panicked about it too. I made kids do all homework on paper in class, I installed blockers, I lectured about how relying on AI would catch up to them. But this year, after watching hundreds of kids work through everything from pre-algebra to 8th grade geometry, I’ve started to think we’re overstating the big, seismic change AI brought. The dependency isn’t new. It’s just easier now.
Last month, I had Jake, a 7th grader who spends his free time programming video games, come to me with an extra credit problem he’d been stuck on for three days. He’d plugged it into AI the first night, got an answer back that didn’t make sense. He spent the whole lunch period in my room, picking apart AI’s steps, crossing out what didn’t work, asking why the model picked one formula over another. He ended up solving it a different way than AI did, and turned it in with notes on where the AI went wrong. That’s the same kid who would have spent three nights arguing with me about a hard problem 10 years ago, and who would have used a Google answer key to check his work before coming in anyway. AI didn’t change his approach to hard stuff; he just used it as a starting point, same as he would any other tool.
Mia’s case is the one that made me rethink the whole conversation. Mia’s had test anxiety since 4th grade, she cries before every quiz because she’s convinced she’s “bad at math.” Ten years ago, when a kid like that got stuck, she’d sit at her kitchen table staring at the problem for 20 minutes, then call a friend to get the answer, or throw her backpack down and tell her mom she couldn’t do it. The end result is the same: she avoids the struggle. AI just lets her get to that end result in 30 seconds instead of 20 minutes. The problem isn’t AI. The problem is that she’s scared of getting things wrong, and that’s been true long before any of us heard of generative AI.
What’s changed is that we have to be more intentional about building that struggle into the work, instead of hoping kids will do it on their own. I don’t ban AI in my class anymore. What I do enforce is a simple three-step rule: write down the first three steps you think you need to take to solve the problem, even if every single one is wrong, before you pull up any tool to check. That’s it. It takes 30 extra seconds, it’s easy for kids to do, and it forces them to actually engage with the problem before they outsource the work. I told Mia’s mom this rule when we had our conference last month; she’d been fighting with Mia every night about hiding her phone to use AI, and had banned it entirely with no luck. She started asking Mia to show her three steps before letting her check her work with AI, and she said the nightly fights dropped off almost immediately. It doesn’t matter if you as a parent don’t understand algebra, by the way: you don’t have to check if the steps are right, you just have to ask to see them.
Another thing I do, and parents can do this too, is swap the question. Instead of asking “did you use AI to do this?” ask “can you walk me through this one step right here?” If they can’t explain it in their own words, that’s a sign they haven’t actually done the work, no matter if they got the answer from AI or from the kid down the street. That’s concrete, it doesn’t require you to be a math whiz, and it cuts through all the yelling about cheating.
I still have days where 30% of the homework I get is clearly straight from AI, no work shown, no explanation. That’s frustrating. But that’s not new. Ten years ago, 30% of the homework I got was copied from the kid who finished first. Twenty years ago, it was copied from the back of the book. The tool changed, but the behavior didn’t. Kids who want to learn how to work through hard stuff use AI as a tool. Kids who want to avoid hard stuff use AI to skip the work, same as they’ve always done with every other tool we’ve given them.
I don’t think we’ve found the perfect balance yet. There are still days I wonder if we’re letting too much slide, if the kids who avoid the struggle now will hit a wall in high school that they can’t AI their way out of. But so far, the biggest change I’ve seen isn’t that kids are more dependent than they used to be. It’s that we’re finally talking about how uncomfortable most kids are with struggle, even when that struggle is the whole point of learning a hard subject. Right now, the three-step rule is working for me, and for most of the parents I’ve shared it with. We’ll see how it goes from here.
Please indicate: Thinking In Educating » Is AI Dependency Changing How Students Learn Difficult Subjects