Everything you need before lesson one
You don't need a technical background. You need this page, fifteen minutes twice a week, and a willingness to be a pretend robot occasionally. Here's the whole picture.
What is AI, in one honest paragraph?
Ordinary software follows instructions a human wrote: if this, do that. Machine learning — the engine behind almost everything called "AI" today — works differently: instead of writing the rules, people show the computer thousands of examples (photos labelled "cat" and "dog", say) and the computer works out its own rules for telling them apart. The result is called a model. Voice assistants, photo face-grouping, video recommendations and chatbots are all models trained on examples. That's the whole idea, and it's the idea every lesson on this site builds towards.
Why teach this to a five-year-old?
Not to make a programmer — to build accurate intuitions early. Children are already growing up surrounded by systems that learn from them. A child who understands "machines learn from examples, and they can be wrong or unfair" asks better questions for the rest of their life: What was it trained on? Who does it work for? Should I trust this answer? Those questions are taught here through sock-sorting machines and pattern hunts, not lectures.
How the courses are built
Same ideas, three translations. All three courses cover the same backbone — learning from examples → data and labels → training and testing → mistakes and fairness → building something real. Ages 5–6 meet these ideas as physical games. Ages 7–8 get the real vocabulary plus experiments and a lab notebook. Ages 9–10 look inside the machinery (decision trees, neural networks, chatbots) and finish with an engineering capstone.
Strictly sequential. Phases build on each other. Resist skipping ahead — the "boring" sorting games in Phase 2 are what make Phase 5 land.
Project per phase. Lessons teach; projects make it theirs. Never skip the project — it's where the learning becomes permanent.
Your role, by age
Ages 5–6: You are the teacher and the show. You read the lesson, you run the game, you operate any screen. Your child plays, shows, sorts, predicts and laughs.
Ages 7–8: You are the lab assistant. Read lessons together; let your child drive the mouse and the notebook while you supervise, fetch props, and ask "why do you think that happened?"
Ages 9–10: You are the project sponsor. Your child reads and builds; you discuss, challenge gently, control all accounts and uploads, and attend Demo Day as the proudest person in the room.
The tools (all free, no child accounts)
Teachable Machine (teachablemachine.withgoogle.com) — train real image/sound models in a browser. Used from the end of the 5–6 course onward. Nothing is uploaded unless you choose to; close the tab and it's gone.
Google Sheets (or any spreadsheet) — for the data-science lessons at 9–10.
Scratch (scratch.mit.edu) — MIT's block-based coding, used in the 9–10 capstone. Projects can be built and saved locally without any account.
Cardboard, socks, toys, paper, stickers — the most important lab equipment, already in your home.
Five teaching habits that make it work
1. Short and often beats long and rare. Two 15-minute sessions a week outperform a monthly marathon.
2. Praise the thinking, not the correctness. "I love how you tested it!" builds a scientist; "you're so smart" builds fear of mistakes.
3. Let mistakes stay on stage. A confused AI model is the best lesson on the page. Never rush to fix it — ask "why do you think it got confused?"
4. Use the magic question everywhere. "What examples did it learn from?" Ask it about every smart-seeming thing your family meets. It's the seed of lifelong AI literacy.
5. Stop while it's fun. End sessions one game before boredom. They'll beg for the next one.
A safety note, plainly
These courses never require a child to have an online account, chat with an AI unsupervised, or share photos or personal information. Where a webcam is used (Teachable Machine), it runs in the browser and you're sitting right there. The 9–10 course explicitly teaches privacy, deepfake awareness and verification habits — please don't skip Phase 4.
Mini glossary (for your back pocket)
- Data
- Recorded examples: photos, numbers, sounds, survey answers. The food AI learns from.
- Label
- The answer attached to an example ("this photo = cat"). Wrong labels teach wrong things.
- Training
- The process of a computer adjusting itself, example by example, until its guesses come out right.
- Model
- The result of training — the "brain" that makes predictions on new examples.
- Prediction
- The model's best guess. Clever, useful — and sometimes confidently wrong.
- Bias
- When training examples leave people or cases out, the model fails them. Fix: better, fairer data.
- Hallucination
- A chatbot producing fluent text that isn't true. Fluent ≠ factual; verify what matters.