AI
KidsAiTools
Learning GamesAI ProjectsAI ToolsLearning Guides

Teachable Machine · Ages 9-12

Teachable Machine for kids: learning how AI learns from examples.

Teachable Machine is the most advanced of the three external tools we use in Chapter 1. Instead of using a pre-trained model, the kid trains one themselves by giving it examples. The concept is profound — but the tool requires camera, microphone, or file uploads to work, which means it needs grown-up co-play. KidsAiTools Day 3 starts with an internal demo (no upload) so the concept lands first; Teachable Machine is then optional with a grown-up.

Start Day 3: AI Learns from ExamplesSee the Teachable Machine review
By KidsAiTools Editorial Team·Reviewed by KidsAiTools Editorial Team·Published ·8 min read

Teachable Machine is a Google research experiment. KidsAiTools is not affiliated with Google. We have no paid relationship with Google or with Teachable Machine. We selected it for Day 3 (optional with a grown-up) based on our review criteria.

What Teachable Machine is

Teachable Machine is a Google experiment that lets anyone train a small AI model in a browser. The user creates two or three categories — say, 'apple', 'orange', 'banana' — gives the AI a handful of examples for each, and the AI learns to tell them apart. The training happens entirely in the browser; nothing is uploaded.

The default input is the device's camera: the kid holds up an apple, clicks 'add example' a few times, switches to an orange, repeats, and so on. Once trained, the model can recognize new fruit the camera sees. Microphone and file-upload modes also exist for sound and image classification.

Why it is more advanced than Quick, Draw! or AutoDraw

Quick, Draw! and AutoDraw use models that someone else trained. The kid is a user. Teachable Machine flips it: the kid is the trainer. They have to decide what categories they want, what counts as a good example, and what happens when an example is ambiguous. This is a meaningful jump in cognitive load.

It is also why we recommend Teachable Machine for ages 9-12 with a grown-up alongside, not for ages 6-8 alone. The training-side concepts (how many examples, how varied) take a few rounds to feel, and a 7-year-old will get frustrated before that frustration turns into insight.

Why ages 6-8 should not use it alone

There are three reasons. First, the conceptual jump (from 'AI guesses' to 'I am training the AI') is too big without scaffolding. Second, the tool defaults to camera input — and a 6-year-old left alone will likely include their face, the room, and possibly a sibling in the training data. Third, frustration when the model classifies poorly will land before the insight that 'mixed-up examples confuse AI'.

For ages 6-8 we strongly recommend doing only the internal Day 3 examples demo — no Teachable Machine — and saving the actual tool for when they are 9 or 10.

How KidsAiTools handles Day 3

Day 3 has two parts. The first is an internal examples demo that runs entirely inside our page: the kid sees how clear examples teach a tiny model well, and how mixed-up examples confuse it. No camera, no microphone, no upload. This is the part 6-8 year-olds also do.

The second part is optional and only for 9-12 with a grown-up: open Teachable Machine itself and train a small image classifier. We walk through the camera permission, what categories to pick, what kinds of examples to avoid, and how to read the results. After training, the kid does a parent recap on what worked and what did not.

A 20-minute parent activity idea

If your child is 9-12 and you want to do the Teachable Machine portion of Day 3 with them, here is a clean 20-minute activity that avoids most of the failure modes we have seen.

What kids can learn

Three concepts that Teachable Machine makes physically visible.

  1. Lesson 1

    AI needs examples

    A model that has never seen an apple cannot recognize one. The kid feels this immediately the first time they try to test the model with no examples loaded — the AI just guesses random categories. Examples are the only way AI learns.

  2. Lesson 2

    Clear examples help

    A few clean examples — apple at 3 angles, on different backgrounds, all clearly an apple — produce a model that classifies new apples well. This is the single biggest "aha" in Day 3.

  3. Lesson 3

    Mixed-up examples confuse AI

    If the kid puts a half-eaten apple, an orange, and a banana all in the "apple" category, the model gets confused. Doing this on purpose for one round is the most useful thing the kid can do, because they then feel why training data quality matters.

Camera, microphone, and upload — what to know

Teachable Machine processes inputs locally in the browser, but the tool still needs camera or microphone access to use the default modes. Some adult guidance is required so a child does not record things they should not.

Camera mode

Default mode for image classification. The kid will need to point the camera at training examples. Avoid pointing the camera at faces, identifying documents, or anywhere that captures private information by accident.

Microphone mode

Used for sound classification. Avoid recording family conversations, identifying voices, or speaking names. Use simple sounds — claps, taps, "yes/no" — instead of personal speech.

Upload mode

Lets the user upload image files instead of using the camera. Avoid uploading family photos, identifying photos, or anything you would not be okay sharing publicly. Stick to neutral objects.

Do not use Teachable Machine for these

  • Faces — your face, family member faces, friend faces
  • Private photos from your camera roll
  • School name, school logos, school IDs, or anything identifying your school
  • Home address, mailbox photos, or any identifying location
  • Personal data — names, IDs, anything that identifies a real person

A 20-minute parent-guided activity (ages 9-12)

Open Teachable Machine on the parent device first. Walk through the permission prompts together. Then hand off and stay nearby.

  1. 1Decide on three categories that involve neutral objects only — for example, three different fruits, three coins, or three small toys. Avoid anything personal.
  2. 2Train each category with 5-10 examples. Show the camera the same object from different angles, on different backgrounds.
  3. 3Test the model with a new example you did not train on. See if it gets it right. If not, ask the child why.
  4. 4Run a deliberately bad round. Mix up the categories on purpose for one class. Test again. Notice that the model now performs worse.
  5. 5Recap with the kid: did clear examples or mixed-up examples produce a better model? This is the lesson.

Who is this article for?

Teachable Machine has the highest cognitive bar of the three Chapter 1 tools. This guide helps you decide if it is the right time.

You will get value if…

  • Your child is 9-12 and has already done Day 1 (Quick, Draw!) and Day 2 (AutoDraw)
  • You can do a 20-minute parent-guided activity together
  • You want to understand the camera/microphone/upload trade-offs before starting
  • You are okay treating the first attempt as practice and trying again

Skip this if…

  • Your child is under 9 — do only the internal Day 3 examples demo
  • You do not have 20 minutes to walk through the activity together
  • Your child has not done Day 1 and Day 2 yet — start there

Start with the internal Day 3 examples demo?

Day 3 begins with a no-upload internal demo. Teachable Machine is optional with a grown-up. Free.

Start Day 3: AI Learns from Examples

Frequently asked questions

Is Teachable Machine free?+
Yes. It runs entirely in the browser, no signup, no payment. Training happens on the device — no upload required by default for camera or microphone modes.
Are camera and microphone inputs sent to a server?+
No, training data stays on the device. Teachable Machine runs the training in the browser using TensorFlow.js. Even so, we still recommend grown-up presence the first time so the child does not accidentally include private content in their examples.
How long does it take?+
A short Teachable Machine session is 15-20 minutes including setup, training, and a deliberately-bad round. With our Day 3 wrapping (internal demo first, then optional Teachable Machine, then recap), expect 20-30 minutes total for the optional path.
My child wants to skip the internal demo and go straight to Teachable Machine. Should I let them?+
We recommend doing the internal demo first. The internal demo lands the 'clear vs mixed-up examples' concept in 5 minutes. A child who arrives at Teachable Machine without that concept often spends the whole session frustrated by a bad model rather than learning from it.
My 7-year-old wants to do it. Is it okay?+
We do not recommend the Teachable Machine portion for ages 6-8. Do only the internal Day 3 examples demo at this age. Save the actual tool for 9 or 10.
How is your Day 3 different from just opening Teachable Machine directly?+
Day 3 starts with an internal demo so the concept lands without needing camera or upload. Teachable Machine is then optional with a grown-up. We also walk through which inputs to avoid and end with a structured recap. Without the wrapping, kids often have a frustrating first session.

Related reading

Continue with Chapter 1

Day 3 is the last day of Chapter 1. After this you have done the full foundational AI literacy loop.

Continue with Chapter 1

Disclosure

No paid relationship. KidsAiTools is not affiliated with Google. Teachable Machine is a Google research experiment. We selected it for Chapter 1 Day 3 (optional with a grown-up) based on our review criteria, with no commercial arrangement.