Skip to content
Session 03 / 04· Know the machine

How Machines Learn (and Why They Get It Wrong)

The mechanical why behind every quirk — so you know when to trust it.

Grades 8–1260 minuteslive clicker activities

The hook

AI isn't magic and it isn't a brain. It's a pattern-predictor trained on a giant pile of human stuff — which is exactly why it's brilliant, biased, and sometimes confidently, hilariously wrong.

What students learn

  • How machine learning actually works: training data, features, predictions — no maths required.
  • Why models hallucinate: they predict plausible text, not true text.
  • What “shortcut learning” is, and how it makes AI fail in surprising ways.
  • How bias gets written straight into an algorithm — garbage in, garbage out.
  • The “centaur” idea: you plus a tool you understand beats either alone.

The promise

Students understand training data → patterns → predictions well enough to know where an AI is strong, where it's thin, and where to double-check it hardest.

Lands on

Match the tool to what it actually is — a confident pattern-predictor — and verify hardest where its training is thin.

The interactive AI demos

Every idea is a Mirror Move

We run it on the room, show it inside the machine, prove it on a live chatbot, then name the skill. Here are this session's loops.

The hour, beat by beat

  1. 1

    Not a brain, a predictor

    Training data → patterns → predictions, told as a story.

  2. 2

    AI weirdness

    Giraffing, recipes with broken glass, the tank-detector that learned the weather.

  3. 3

    When algorithms judge people

    COMPAS and the ProPublica investigation; bias you can't see.

  4. 4

    Train it live

    Teachable Machine with the room, then deliberately break it.

  5. 5

    The centaur

    Human plus a tool you understand — the optimistic note.

Drawn from

  • Shane — You Look Like a Thing and I Love You
  • Fry — Hello World
  • O'Neil — Weapons of Math Destruction
  • Google Teachable Machine

All sources are paraphrased; quotes are short and attributed. We teach the same skepticism toward AI that we teach toward any confident claim.

Next session

Truth in the Age of AI

Verify the machine

Ready to book?

Run this session standalone, or as part of the four-session package.