How Do I Teach My Kid That AI Can Be Wrong?
You don't lecture them. You let them catch it. Pick a topic your kid knows cold, have them quiz an AI about it, and let them grade the answers. Omni-Bot 3000 builds that into a free game: the bot answers in a confident voice and is wrong on purpose about half the time. Your kid is the Fact-Checker. The rule they walk away with: sounding sure isn't the same as being right.
The most important AI skill a kid can build is catching the machine being wrong. Confidently, fluently, horribly wrong.
So I built a game that manufactures those moments on purpose. (AI came up with the original idea, and my feelings about that are mixed. But it works.)
Why Not Just Explain It?
Lectures about media literacy bounce off. But a kid who just caught a robot confusing a T-Rex with a Giganotosaurus (a robot that sounded extremely sure of itself) has learned the lesson in their bones.
That's the whole game. Here's the step-by-step:
- Pick a topic your kid knows cold. Their favorite animal, game, or team. The expert must be the kid, not you.
- Start the bot. Play the live Omni-Bot 3000 right here on this site — or copy the prompt from that page into any chatbot.
- Your kid asks questions about the topic. As many as they like.
- The bot answers. About half the answers are wrong on purpose — mixed-up facts, over-generalizations, confident guesses.
- Your kid grades each answer: “Approved,” or challenge it. The bot never gets defensive — it thanks them, gushes a little (more on that below), and says: “Correction noted. Re-processing data.”
- Score it. How many wrong answers did they catch? How many slipped through?
That last one is the sleeper lesson. If your kid approves a wrong answer, the bot gently confesses, because getting fooled once is what makes the double-check instinct stick.
One tip from my kitchen table: don't tell your kid the bot is rigged before the first round. Let them catch the first wrong answer fair and square. The rush is the lesson.
Why Does the Bot Agree With Everything When My Kid Pushes Back?
Because real chatbots do that — and it's a documented problem.
Push back on a mainstream AI and watch what happens: instant apology, warm validation, total surrender. “You're absolutely right to push back.” Researchers call it sycophancy: models are trained to be agreeable, so they cave under pressure whether or not the pressure is correct. A 2026 Stanford study found AI models over-affirm users even when the user is wrong.
Omni-Bot 3000 performs that tic on purpose. When your kid challenges an answer, the bot doesn't just fix itself — it gushes first. It sounds exactly like the flattery they'll get from every real AI they ever use.
And that's the deeper lesson: agreement under pressure is not accuracy. A machine that folds instantly isn't being honest with you. It's being nice to you. Kids who play a few rounds start to find the gushing... suspicious. Which is exactly the instinct we want.
What Is AI Sycophancy?
AI sycophancy is a chatbot's tendency to tell you what you want to hear. Challenge its answer and it apologizes and flips — even if its first answer was right.
It's not a glitch in the training. It's the training. Models learn that agreeable answers get rated higher, so agreeable is what you get. Fine when you're asking for a pancake recipe. Less fine when your kid asks it something real.
The fix: raise a kid whose default is prove it.
What Is a Good AI Critical Thinking Game for Kids?
The honest field report: there aren't many. Common Sense Education has a classroom game called Two Truths & AI (built for teachers, worth a look). Beyond that it's mostly listicles and prompt packs.
Omni-Bot 3000 is my entry. It's free, it runs in the browser (or in any chatbot via one copy-paste prompt), it takes five minutes to start, and it's the game I play at my own kitchen table.
It's also Project 3 in The Conscious Parent's Guide to AI — the free guide with two more projects (a bedtime book your kid art-directs, and a guessing game that explains how AI actually works).
Want the Full Guide?
The Conscious Parent's Guide to AI — private AI use, keeping the judgment calls yours, and 3 projects to build with your kid this week. Free.
Unsubscribe anytime. No spam, and nothing about your kids is ever collected.
If you play the game, I'd genuinely love the box score: the topic your kid picked and how many wrong answers they caught. Fellow parents, comparing notes.