Skip to content
Curiotide

AI for Science Experiments and STEM Projects at Home

  • Ages 5-7
  • Ages 8-9
  • Ages 10-12
Illustration of a white conical flask holding orange liquid with bubbles rising, above deep blue ocean waves

Science at home usually stalls in one of two places: nobody knows what the bug is, or nobody knows what to do next. AI is good at both of those and bad at being right, which makes it a decent lab partner and a terrible authority.

That distinction is the whole article. Use it to get started and to identify things. Check it against the physical world.

Identification, which is where kids get hooked

Seek

Seek, from iNaturalist, names plants, insects, and fungi from a phone camera. Identification happens on the device, with no account needed and no photos uploaded by default. Kids collect badges for the species they find, which is the rare gamification that sends a child outdoors.

Merlin Bird ID

Merlin, from the Cornell Lab of Ornithology, listens to birdsong and tells you who is singing. Free, no ads. Stand in the garden for five minutes and your child will learn four birds by voice.

Both apps give you the best question in the whole subject: how sure is it, and how would we find out? Take the phone’s answer to a field guide and check. When they disagree, you have a real afternoon.

Building a model, which is where they understand it

Google Teachable Machine

Teachable Machine trains an image, sound, or pose classifier in a browser in about ten minutes. No account, nothing uploaded by default.

Turn it into an experiment rather than a demo. Train it to tell apples from tomatoes with thirty photos, then test it in a different room, at night, with a green apple. Have your child predict what will break before testing it. The write-up falls out of it: hypothesis, method, result, explanation.

Machine Learning for Kids

Machine Learning for Kids connects model training to Scratch, so a trained classifier ends up inside a game your child built. Longer project, bigger payoff, and free. Suits eight and up.

The lab partner that gets things wrong

A general chatbot on your account is useful at two moments: before an experiment, and after it goes wrong.

Prompts worth having:

  • “I have baking soda, vinegar, food coloring, and a muffin tin. Give me three experiments and tell me what each one demonstrates.”
  • “My plant in the dark cupboard grew taller than the one on the windowsill. What are three possible explanations?”
  • “What would I need to change to test which of those explanations is right?”

That third one is where the science lives. A model is good at generating candidate explanations and unreliable at telling you which is true.

Where it fails: numbers, units, and anything about safety. Chatbots produce confident chemistry that does not work and occasional combinations nobody should mix. Every hands-on suggestion gets checked against a real source before it reaches the kitchen. Teaching kids that AI can be wrong has a demonstration built around exactly this.

Checking the numbers

Wolfram Alpha computes from structured data instead of predicting text, which makes it the right place to verify a figure a chatbot produced. Unit conversions, physical constants, distances, dates. Teach your child to use one for ideas and the other for facts.

A project that works for most ages

Sound classifier, one afternoon:

  1. Record ten examples each of two sounds around the house. Doorbell and microwave, or two family members saying “hello.”
  2. Train a model in Teachable Machine.
  3. Test it. It works.
  4. Now break it. Whisper. Stand across the room. Bring in a third person. Record with the tap running.
  5. Ask your child why it failed and what data would fix it.

Step five is the entire field of machine learning, and a nine-year-old can get there in an afternoon.

Keeping it honest

Science with kids works when the physical world gets the last word. The app said it was a red admiral; the wing pattern in the book says otherwise. The chatbot said the reaction would fizz; it did not. Those moments are worth more than any correct answer, because they teach a child that a confident machine is still a machine.

For where the Scratch projects lead next, learning to code with AI picks up from here, and our guide for ages 8 to 10 covers the wider set of tools that suit this age.