Teacher guide

For teachers of students aged 14–16 (Grade 9–10 / Year 10–11). You do not need any AI background to teach this unit. Every technical answer you might be asked for is in the misconception table below.

Last reviewed: 2026-10-25

Unit overview

Lesson Core question Main activity Handout
1. How AI makes text Does a chatbot “know” things? Next-word interactive + human predictor game Recording sheet
2. When AI gets it wrong Why are wrong answers so convincing? Fact-check challenge in pairs Fact-check worksheet
3. Bias and fairness Whose perspective is missing? Compare two outputs; discuss training data — (discussion)
4. Using AI to learn, not to skip When does AI help my learning? Design personal AI-use rules My AI-use rules

Each lesson: 50 minutes, structured as Hook (5) → Main activity (25) → Discussion (10) → Exit ticket (5) → Buffer (5).

Prerequisites and requirements

  • Prerequisites: none. No prior AI knowledge for students or teacher.
  • Tech: one device with a browser per pair for Lessons 1–2 (the interactive runs offline, no accounts, no network). Lesson 3 works best if you can show two short AI outputs you prepared; Lesson 4 needs nothing.
  • No devices? Every lesson has an unplugged fallback (marked “No devices” in each lesson plan). The core of Lesson 1 works on paper: students play the predictor themselves.
  • Safety: the unit asks students to discuss AI, not to use it for assignments. Nothing in this unit requires students to have accounts with any AI service.

How to talk about students’ personal AI use — without judgment

Many students use chatbots for homework. The unit works only if students can say so honestly. Three rules for yourself as the teacher:

  1. Never ask “who here uses AI to cheat?” Ask “when does AI help you learn?” — the same behavior gets discussed without confessions.
  2. The target is thinking, not tools. The unit’s message: using AI is a decision about your learning, not a rule handed down. Lesson 4 lets students write their own rules — keep that frame even if a rule is laxer than you would like; the reflection is the point.
  3. No anecdotes about getting caught. Fear shuts the discussion down; you need it open for Lessons 2–4 to land.

Misconceptions to watch for (and what to say)

A student says… The misconception You say…
“The chatbot knows things.” Knowledge vs. prediction “It predicts likely next words from patterns in text it was trained on. It has no notes, no memory of facts — Lesson 1 shows you the machinery.”
“It cited a website, so it’s true.” Citations are proof “The citation text itself is predicted. Open the link. Does it exist? Does it say that? Lesson 2 is exactly this skill.”
“It’s just a computer, computers can’t be unfair.” Neutrality of technology “It learned from human text, and human text contains unfair patterns. If the training data leans one way, the output leans that way. Lesson 3.”
“AI is always getting better, so soon it’s fine to trust.” Trust as a dial that only goes up “Capability and trustworthiness are different axes. Fluent wrong answers are the normal failure mode — the check stays yours.”
“If it sounds sure, it is sure.” Confidence = truth “The model has no ‘sure’ — it has ‘likely’. You supply the certainty, by checking.”

If a student asks something you cannot answer: “Good question — I don’t know. Let’s find out or add it to the parking lot.” Modeling not-knowing is itself AI literacy.

Differentiation

Support:

  • Sentence starters on the fact-check worksheet (“This claim says __. The source says __.”)
  • Pair students deliberately: fluent AI users with careful readers, swap roles each lesson.
  • Reduce scope: two claims instead of three in Lesson 2; two rules instead of three in Lesson 4.

Stretch:

  • Lesson 2: find a claim where the source partially disagrees — partial truth is the interesting case.
  • Lesson 3: propose one concrete change to how a model’s training data could be picked to be fairer, and its trade-off.
  • Lesson 4: write a rule for AI use during exams vs. homework that is defensible to both teacher and students.
  • Early finishers: run the interactive at extreme temperature and explain what changes in one paragraph.

Answer keys

  • Lesson 1 (recording sheet): “surprise” words at high temperature are expected, not errors — the model picks unlikely words on purpose sometimes (that is what the slider does). A chatbot “knowing” things is false; it predicts.
  • Lesson 2 (worked example on the worksheet): the sample chatbot answer contains (1) an invented statistic “in 1842 the bridge collapsed twice”, (2) a plausible but non-existent book source, (3) a correct-sounding but unverified date. Students need two of three to meet the objective.
  • Lesson 3 (comparison pairs): there is no single right answer — assess whether the stated cause (training data, who wrote the sources, default perspectives) matches the example.
  • Lesson 4 (scenario sort): scenarios 1 and 4 → AI helps learning; scenario 2 (essay ghostwriting) → replaces thinking; scenario 3 (exam practice answers before trying) → replaces thinking unless the rule says “try first, then check”. Accept any rule with a reason.

Assessment in this unit

  • Exit tickets (Lessons 1–4) check the four objectives informally.
  • The fact-check worksheet is the Lesson 2 artifact.
  • Final project covers all four objectives, scored with the shared rubric (criteria adapted for school use as noted there).
  • The performance task has a school variant using a flawed “AI-written” essay, constructed for teaching purposes.

One-minute prep checklist (per lesson)

  1. Print handouts (or share the printable pages — they print cleanly).
  2. Open the interactive in a browser tab (Lesson 1) — it works offline.
  3. Re-read the misconception table above.
  4. Decide pairs. That’s it.