A Teacher's Guide to AI Lessons

You do not need a computer science background to teach AI well. What you need is a handful of activities that make an abstract technology visible, plus a clear policy so students know where the line is. This guide gives you both, ready to use in an elementary or middle school classroom.

What to teach, and in what order

Aim for four ideas, in this sequence: AI learns from examples; AI predicts rather than knows; AI reflects its training data; AI output is a draft, never a final answer. Everything else — prompting technique, tool choice, ethics debates — hangs off these four.

Five classroom activities that need no accounts

1. The paper classifier (30 minutes)

Give groups twenty index cards with simple drawings and ask them to write rules that sort cats from dogs. Then hand them three tricky cards. Students discover on their own that rules break at the edges — which is exactly why we train models on examples instead of hand-writing rules.

2. Human autocomplete (15 minutes)

One student says a word, the class shouts the most likely next word, and you build a sentence together. It is silly, fast, and it makes the core mechanic of a language model physically obvious.

3. Prompt laboratory (40 minutes)

Project a single AI tool on the board. Ask the class for a vague prompt, then improve it together one constraint at a time: audience, format, length, tone, example. Students see quality climb with each addition and leave with a reusable prompt checklist.

4. Spot the hallucination (30 minutes)

Ask the model about a topic your class has just studied. Print the answer, hand it out, and have students annotate anything they can verify, anything they cannot, and anything that is simply wrong. This is the single highest-value AI lesson you can run.

5. Bias audit (30 minutes)

Generate ten images from a neutral prompt such as "a person doing science". Tally what the class sees. Discuss where those patterns came from and who might be harmed by them.

Assessment that survives AI

Tasks that are easy to outsource to a chatbot — summarise this chapter, define these terms — no longer measure much. Tasks that hold up include: in-class discussion, work grounded in a personal or local context, process portfolios showing drafts, oral defence of a written piece, and critique assignments where the AI output itself is the object of study.

A simple rubric addition works well: award marks for verification and for explaining what the student changed and why.

A classroom AI policy in four lines

• Tell me when you used AI, and for what.
• You are responsible for every claim you submit, including the ones AI wrote.
• Never enter personal information about yourself or anyone else.
• AI may help you think; it may not think for you.

Post it, explain it once, and enforce it consistently. Students respond far better to a clear line than to a blanket ban they will quietly ignore.

Safety and privacy essentials

Check whether your tools meet your jurisdiction's student-data rules before asking students to create accounts. Where possible, use a single teacher account projected to the room, which removes the data question entirely for younger grades. Never upload student names, photographs, or assessment records into a general-purpose chatbot.

Cross-curricular hooks

AI is not only a computing topic. In English, compare an AI-written paragraph with a student one and analyse voice. In science, use it to generate hypotheses and then design tests. In history, ask it about a local event and audit the gaps. In art, discuss authorship and style. In maths, have it solve a problem incorrectly and ask students to find the error.

Bringing Curious Dojo into your class

Our five-world courses are self-paced, quiz-gated, and readable aloud, which makes them workable as a station activity or a homework track. Progress is visible to the adult account, and the environment contains no outbound links or open chat. Browse the other guides or read the blog for lesson-sized ideas.

FAQ

Do I need to be technical to teach this?

No. Every activity above runs with paper, a projector, and ordinary curiosity. Your subject expertise is what makes the critique work valuable.

Should students be allowed to use AI for homework?

Allowed with disclosure is more workable than banned. Bans push use underground and remove your chance to teach good habits.

How do I detect AI-written work?

Detectors are unreliable and produce false accusations. Design assignments that require process evidence and personal context instead.

What age can I start?

The paper classifier and human autocomplete activities work from around age seven. Prompting and bias auditing land best from around age ten.