wibbit
← All courses·Course 2 of 5·Available now
🧠

How AI Actually Learns

The five tribes of ML, neural networks, and what goes wrong.

A deeper second pass. Kids discover that "AI" isn’t one thing — it’s five competing philosophies that all aim at the same goal. They train tiny networks, watch overfitting happen in real time, and learn why fair AI is so hard.

5
Modules
22
Lessons
~5 hours
Length
8 and older
Ages
Course 1
Prereq
The big idea

"Five philosophies, one goal: learn from data."

That's the throughline. Every module reinforces it from a different angle — and every lesson ends with the kid being able to demonstrate it.

Source material
Pedro Domingos — "The Master Algorithm" (age-adapted)
We translate research-grade ideas into something a curious kid can play with.
The full curriculum

5 modules. 22 lessons.

Each module ends with a six-question challenge. Pass five, earn the badge.

📐
Module 1
Rules vs. Examples
Supervised learning

Some programs follow rules a human wrote. Others learn from examples. The difference changes everything.

🔁
Earn the badge
Loop Spotter
Lessons in this module
  1. 1Sort the shapes (rules edition)
  2. 2Sort the shapes (examples edition)
  3. 3What’s a label?
  4. 4The feedback loop
  5. 5When examples beat rules
  6. Challenge: Loop Spotter
🏛️
Module 2
The Five Tribes of ML
Symbolists, Connectionists, Bayesians…

There isn’t one way to make AI. There are five — each with a different idea about what learning even means.

🗺️
Earn the badge
Tribe Tracker
Lessons in this module
  1. 1Five ways to think about learning
  2. 2Symbolists: rules and logic
  3. 3Connectionists: brains in the machine
  4. 4Evolutionaries, Bayesians, Analogizers
  5. 5Who won? (Spoiler: it’s mixed)
  6. Challenge: Tribe Tracker
🎯
Module 3
Overfitting
Memorizing is not learning

A model that aces the practice test and bombs the real one isn’t smart — it’s memorizing. We learn to tell the difference.

🕵️
Earn the badge
Cheat Detector
Lessons in this module
  1. 1The student who memorized everything
  2. 2Curve fitting: too tight, too loose, just right
  3. 3Train, test, validate
  4. 4Generalization to unseen data
  5. 5Why bigger isn’t always better
  6. Challenge: Cheat Detector
🕸️
Module 4
Neural Networks
Weights, forward pass, backprop

Inside every modern AI: layers of tiny knobs called weights. Training is the slow art of turning all of them in the right direction.

🛠️
Earn the badge
Network Builder
Lessons in this module
  1. 1Neurons and layers
  2. 2Weights — the knobs of learning
  3. 3The forward pass
  4. 4Learning from mistakes (backprop)
  5. 5Build a tiny network
  6. Challenge: Network Builder
⚠️
Module 5
Failure Modes & Fairness
Smart systems, dumb mistakes

AI fails in weird, predictable ways. Knowing the failure modes is how you stop them hurting people.

🔍
Earn the badge
Fault Finder
Lessons in this module
  1. 1Garbage in, garbage out
  2. 2Adversarial examples (the sticker on the stop sign)
  3. 3Who gets hurt when AI is wrong?
  4. 4Bias in, bias out
  5. 5Auditing an AI
  6. Challenge: Fault Finder

Ready when they are.

The whole first course is free for everyone. No card up front. Take Module 1 in fifteen minutes.

Start free →See plans