3 · Pick a question AI can help answer
Your whole project grows from one good question. The best science-fair questions are ones you actually wonder about — and that you can really test with the time, tools, and AI you have.
A strong question for an AI project usually asks the model to sort, recognize, or predict something — because that's what the AI you'll train (Google Teachable Machine) is good at. Some examples:
- "Can an AI tell ripe bananas from unripe ones by photo?"
- "Can an AI hear the difference between me clapping and snapping?"
- "Can an AI sort photos of recyclables from trash?"
- "Can an AI recognize my good posture vs. slouching at my desk?"
A good capstone question passes the "Goldilocks" test — not too big, not too small, just right:
| Too big | Too small | Just right |
|---|---|---|
| "Can AI cure diseases?" | "Is this one photo a cat?" | "Can an AI tell cats from dogs in 20 new photos?" |
| (No way to test it) | (Nothing to discover) | (Testable, finishable, real) |
The "can I actually test it?" checklist — your question is ready when you can answer YES to all of these:
- ✅ Can I collect examples for it safely? (photos of leaves, sounds of instruments — not people's private info)
- ✅ Can I measure the answer with a number? (right how many times out of how many?)
- ✅ Can I finish it with my time and tools?
- ✅ Could the answer turn out to be "no"? (If it can only ever be "yes," it's not a real test.)
A note on what NOT to study: avoid questions that need other people's faces, voices, names, or any private information — those raise privacy and consent problems you don't need. Study objects, sounds, plants, or your own poses instead. We'll dig into ethics in Lesson 6, but it starts here, with the question you pick.
You'll write your real question in the Lesson 4 checkpoint. Take your time — a clear question makes every later step easier.
Think about it. Write one question you genuinely wonder about that an AI could help test by sorting or recognizing things. Does it pass all four checklist items?
Sources
- Science Buddies. Steps of the Scientific Method — start with a clear, testable question. https://www.sciencebuddies.org/science-fair-projects/science-fair/steps-of-the-scientific-method
- MIT RAISE. Day of AI — free K-12 activities, including training a model on your own examples. https://raise.mit.edu/