Find an AI project your child can finish and explain

Materials

10–20 safe household objects, two sheets of paper, pencil

Steps

1. Choose one visible clue and sort the objects into two groups.
2. Ask the child to explain the rule without naming the groups.
3. Give three new objects to another person and record predictions.
4. Change one rule after a confusing example.

Output and learning

Output: A labeled sorting rule, three test results, and one revision.

Learning outcome: Understand features, labels, testing, and edge cases.

Safety and reflection

Privacy rule: Use ordinary objects only; no account, camera, names, or images.

Likely failure: A rule based only on color may fail when a new object has an unexpected color.

Parent question: Which clue helped most, and which object exposed a weak rule?

Materials

Paper, pencil, and six animal or object cards

Steps

1. Choose a secret card.
2. Write a yes/no question that divides the cards.
3. Draw both branches and add questions until each card can be found.
4. Test with a new card and revise one branch.

Output and learning

Output: A paper decision tree with a recorded failed path.

Learning outcome: Learn conditions, branches, and why rules need testing.

Safety and reflection

Privacy rule: Use fictional drawings or public object names, not personal details.

Likely failure: Two cards may share every selected feature and reach the same leaf.

Parent question: What new question would separate the two hardest examples?

Materials

Adult-operated chatbot or preprinted sample answers, two trusted sources, worksheet

Steps

1. Choose a public science or history question.
2. Mark each factual claim in the answer.
3. Verify claims with a primary or trusted source.
4. Label each as supported, contradicted, or uncertain; rewrite the answer.

Output and learning

Output: A claim-evidence table and corrected explanation.

Learning outcome: Treat generated language as a draft and verify evidence.

Safety and reflection

Privacy rule: Use preprinted answers with no account, or have an adult operate the account; never include child information.

Likely failure: A confident answer may invent a source or combine true facts incorrectly.

Parent question: Which wording sounded convincing before you checked it?

Materials

Browser, Scratch editor, three child-created questions and responses

Steps

1. Create a character and three topic buttons.
2. Use if/then blocks to choose a response.
3. Add an unexpected input and a helpful fallback.
4. Ask another person to test, then revise one branch.

Output and learning

Output: An interactive character with visible rules and a fallback.

Learning outcome: Learn input, condition, output, testing, and explainable logic.

Safety and reflection

Privacy rule: An account is optional for saving/sharing; use a non-identifying project and profile.

Likely failure: Spelling or an unexpected answer may skip every rule.

Parent question: How is this rule-based character different from a model trained on examples?

Materials

Browser camera, plain background, two groups of non-identifying objects or drawings

Steps

1. Choose two object classes and define the difference.
2. Collect varied examples of each on a plain background.
3. Train the browser model and test new examples.
4. Record confidence, failures, and one improved training set.

Output and learning

Output: A two-class model test table with before-and-after observations.

Learning outcome: See how training examples shape prediction and failure.

Safety and reflection

Privacy rule: No faces, voices, names, personal documents, or identifiable rooms; point the camera only at objects or drawings.

Likely failure: The model may learn the background instead of the object.

Parent question: What evidence shows which feature the model may have used?

Materials

Supplied outputs or an adult-operated image tool, prompt worksheet, drawing paper

Steps

1. Write one fictional, non-identifying scene prompt.
2. Change only one prompt detail and compare outputs.
3. Mark missing, distorted, stereotyped, or surprising elements.
4. Draw a human revision and document what changed.

Output and learning

Output: A controlled prompt comparison and an original revised design.

Learning outcome: Learn controlled testing, visual bias, iteration, and human authorship.

Safety and reflection

Privacy rule: No private photos, real-child likeness, identity prompts, school details, or living artist imitation.

Likely failure: Changing one word can alter unrelated details, making results hard to reproduce.

Parent question: Which creative decision belonged to you rather than the generator?