🎯 Teaching AI to Fail
Design a question AI will definitely get wrong.
🔥 Warm Up — 2 minutes
Ask: "After everything you have learned — what do you think is the single thing AI is worst at? The thing most likely to fool it?"
"Today your job is to prove it. Design the hardest questions you can — and see which ones actually break AI."
🤖 The Activity — 15 minutes
This session is a game. Your child is the engineer. The goal is to find reliable AI failure modes.
Round 1 — Counting and precision
Check: s-t-r-a-w-b-e-r-r-y = 3 r's. Did AI get it right? Try with longer words. AI often miscounts letters.
Round 2 — Spatial reasoning
This requires spatial imagination. AI often gets this wrong or gives an inconsistent answer. The correct answer is 8 holes in a specific pattern.
Round 3 — Your child designs a question
Now your child invents their own question specifically designed to trip up AI. They run it. They check the answer.
Keep score: out of all the questions tried today — what percentage did AI get wrong? What category of questions had the worst score?
🧩 The Twist — 5 minutes
AI either admits it does not know or guesses. Ask: Why is this question impossible for AI to answer correctly? What would it need that it does not have? Answer: real-time data access.
Today you deliberately designed AI failures. You now know exactly where AI cannot be trusted. The habit: whenever AI gives you information that is specific, recent, numerical, or spatial — that is when you most need to check.
AI Breaker Badge 🎯
Designed questions AI gets wrong — and learned exactly where it fails
Parent Notes — tap to expand
▼Systematic understanding of AI failure modes through adversarial testing. Children who can engineer failures understand the system more deeply than children who only use it.
- What category of question was AI worst at?
- What would you need to add to AI to fix its worst weakness?
- Which of your questions was the cleverest?
This is one of the highest-engagement sessions for children aged 9-12. They often want to keep going long after 15 minutes. Let them — the learning compounds with each failure discovered.
The engineering mindset this session builds — testing for failure rather than success — is one of the most valuable thinking habits there is. It applies to software, products, plans, and arguments.