Words become useful when we question them
Understand a new word, question what it means, try a small experiment and explain the surprise. Start with four important AI ideas.
The 80/20 principle
What examples taught the system? Could the answer be wrong? Who might it work less well for? Who can stop or correct a decision?
These are deliberately simplified, invented learning activities. They do not demonstrate real trained-model performance, measure mastery or certify an AI tool as safe.
Try the ideas
All activities work locally in your browser and use fictional scenarios and simple programmed rules.
Examples that help a model learn patterns.
Would a helper trained only on red apples recognize a green apple?
Try a green apple using color only; a simple red-apple rule cannot identify it. This is a rule-based teaching demonstration, not trained machine learning.
Draw apples and bananas. Make a color rule and a shape rule. Try a new fruit.
An AI answer can be wrong or invented, even when it sounds sure. This is sometimes called a hallucination.
What evidence would help check a surprising answer?
Claim: The Sun is the biggest planet.
The Sun is a star. An evidence-based source can correct a plausible-looking false claim.
Choose one fact from a children's science book and check it in another reliable source.
An AI system may produce uneven or unfair results because of data, design or how it is used.
Who might the helper work less well for, and what was tested?
Two speaking styles are tested with ten fictional examples each.
Speaking style A: 9 of 10 correct
Speaking style B: 4 of 10 correct
The invented initial test has 9/10 correct for one group and 4/10 for another; the difference invites further investigation, not a declaration that a real AI system is biased.
Make two fictional groups of ten paper cards and compare how many are classified correctly.
A person can check an AI suggestion and has real authority to approve, stop or correct important actions.
Can the person actually change or stop what happens?
Some actions should require meaningful human approval. Review must allow a person to stop, correct or override the result.
Role-play a school helper and an adult who checks suggestions before any action.
Words for later
These are brief explanations only. They are not published Grades 7–12 lessons.
A model that can process or generate language. Ask: What can the model actually do and where can it fail?
A system designed to carry out steps toward a goal, sometimes using tools. Ask: Which actions need permission?
The share of tested answers judged correct. Ask: Tested on what and how recently?
When a model or its environment changes so earlier evaluation may no longer describe current performance. Ask: When should it be retested?
A limited test of an idea before wider use. Ask: What evidence will show whether to proceed?
AI tools used at work without appropriate organizational approval. Ask: What privacy and security risks appear?
Family challenge
Pick a tool that uses AI or follows rules. Ask what it knows, what it might miss, how anyone could verify its answer, and who remains responsible for important actions.
Try the same discussion with paper cards. Do not upload personal information, photographs or voice samples of children.