What I Teach My 4th Graders to Prepare Them for an AI-Driven World (It’s Not Just Coding)
Last month, during our 4th grade personal narrative writing workshop, I leaned over Javi’s desk and paused. The room smelled like dried glue sticks and the cherry lollipop someone had dropped under the reading rug three days prior, and rain streaked the cinder block windows so the playground oak looked blurry at the edges. Javi’s story about making tamales with his abuela over Christmas break was three paragraphs long, perfectly spelled, no run-on sentences, and it didn’t sound like him at all.
I pulled up the chair next to his desk and nodded at the screen. “Remember last week when you told me she lets you dump the whole bag of masa even when it makes the pot bubble over on the stove?” I said. “Where’d that part go?”
He shrugged and scuffed the toe of his sneaker against the leg of the desk. “I typed ‘story about making tamales with grandma for 4th grade’ into ChatGPT,” he said. “This doesn’t have any mistakes. It’s better, right?”
I didn’t take his Chromebook, or mark the assignment zero, or shoot off a panicked email to his mom. That month, I’d already gotten six nearly identical “what I did over summer break” essays from other kids, all generated the same way. Banning the tools would just mean kids used them at home on their phones and lied about it, the same way my generation hid SparkNotes from our English teachers. The question wasn’t how to keep AI out of my classroom. It was what I needed to teach these 9- and 10-year-olds that AI couldn’t do for them.
That afternoon, I pushed our writing deadline back a week, and told everyone to close their laptops. We spent 10 minutes doing what I called “bad detail brain dumps,” on scrap paper, no erasing allowed. A bad detail is the messy, specific thing no one else would know to put in a story. I wrote my own example on the whiteboard: when I make chocolate chip cookies with my little sister, we always eat so much raw dough we don’t have room for the baked ones, and my mom yells but she eats a spoonful too. No generic “we had fun” lines allowed.
Javi ended up adding three paragraphs of bad details. His abuela turns her old Vicente Fernández radio up so loud the neighbors bang on the wall, and she sings off key the whole time. Masa gets stuck under your fingernails and doesn’t wash off for two days, no matter how many times you scrub. His baby cousin dropped a finished tamale in the dog’s water bowl, and the dog ate it anyway, corn husk and all. When he read that last line aloud for share circle, the whole room laughed so loud the teacher next door poked her head in to see what was going on. He still used AI to fix his comma splices before he turned the final draft in. I didn’t mind.
That same kind of practice showed up a few months later, when we worked on our class vegetable garden, four raised beds tucked between the school parking lot and the playground fence. I split kids into groups and asked them to calculate how many lettuce seedlings we could buy with our $30 PTA budget. Half the groups typed the full problem into a math solver AI, and came back with the same answer: 72 seedlings, spaced six inches apart in four 4×8 beds.
When we trundled outside with tape measures and a pack of popsicles to mark the spots, they figured out the problem fast. One of the beds has a rusted sprinkler head dead center that takes up almost a square foot of space. The edge closest to the fence gets almost no sun after 1 p.m., so nothing will grow there. The AI didn’t know any of that. It only had the generic numbers we gave it. The group ended up planting 58 seedlings, and three of those died in the first cold snap anyway. We got 11 pounds of lettuce out of the deal, enough for two salad bars in the cafeteria.
After that, I started requiring a “first guess sticky note” for every project that uses a digital tool. Before anyone types a question into a search bar or AI chatbot, they have to write two things on a 3×3 note: what they think the answer will be, and one specific detail about our situation that a generic tool won’t know. It takes two minutes, no fancy training required. I’ve suggested the same trick to parents at conferences, for everything from calculating how much paint to buy for a bedroom to planning a weekend road trip. You don’t need a special curriculum to do it. You just pause before handing the whole problem over.
Last semester, we started an AI mistake jar on the shelf by the pencil sharpener. Any time someone finds a wrong, made-up, or just plain boring claim from an AI tool, they write it on a slip of paper and drop it in. Once a month, we read through them over Goldfish crackers. Mia found one that said all bees die after they sting someone, which she knew was wrong because her grandpa keeps honeybees and only honeybees die after stinging. One boy asked for a list of Dr. Seuss books and got a fake title about a penguin that doesn’t exist. Another asked when our town public pool opens for the summer, and got an answer that said it’s open year-round, which everyone knows is ridiculous—it closes every September and doesn’t open again until June.
I get at least two emails a week from parents asking what they should be doing to get their kids ready for a world where AI is everywhere. Most of them ask if they should sign their kid up for extra coding classes, or ban chatbots entirely from homework time, or buy some expensive online course that promises “AI literacy” for elementary kids.
I tell them coding is great if their kid likes messing with computers, but most of these kids won’t grow up to build AI systems. They’ll grow up using them, the same way most of us use cars without being mechanics. The small, daily habits matter more than any extra class. When your kid gets a generic answer from an AI tool for homework, don’t yell about cheating. Just ask, “What’s one part of this that only you could add?” When you’re making dinner and pull up a recipe, ask what the recipe doesn’t know about your oven that runs 25 degrees hot, or your kid’s nut allergy, or the extra chocolate chips you always add to banana bread. When they’re writing a book report, ask if the AI’s take matches what they actually thought when they read the book. If it says the ending was “satisfying” and they actually threw the book across their bed when the dog ran away, that’s the part that matters.
I saw that play out in homework club a few weeks ago, with a 3rd grader who’d let AI write her entire report on Because of Winn-Dixie. I sat down next to her instead of closing the tab, and we deleted the generic lines one by one, replaced them with the parts she actually remembered. It took 10 minutes. She got a B+, and she told me later it was the first book report she didn’t hate turning in.
I don’t have all the answers. I don’t know what jobs these kids will be applying for in 15 years, or what AI tools will be able to do by then, or what new problems will come up that none of us are thinking about right now. I still catch kids trying to turn in fully generated essays when they’re tired or stuck, and we still talk through it, one bad detail at a time. Javi still uses AI to fix his grammar, and he still writes all the messy, weird parts first on scrap paper. Our lettuce had aphids this spring, and half the jar’s mistake slips are just kids writing “it said my homework would be easy and it wasn’t.”
Last week, he brought me a new story he wrote, about teaching his abuela how to use ChatGPT to write grocery lists. She uses it, he told me, but she always adds extra chiles to the list anyway, no matter what the recipe says. That feels like enough for now.
Please indicate: Thinking In Educating » What I Teach My 4th Graders to Prepare Them for an AI-Driven World (It’s Not Just Coding)