In productionAI certificate program
Machines That Learn
AI is not magic: it is a machine that finds patterns in examples. Students train simple models by hand, watch them make mistakes, and learn to ask where an AI's answers come from.
- 5 weeks
- AI
- Ages 9 to 14
Opening later
Start with Mission to MarsWhat students learn
Rules against learning
The difference between a program someone wrote and a model that learned from examples.
What training means
Examples, labels and patterns, done by hand with cards before any computer.
How a machine sees
Pictures as grids of numbers, and how patterns in them become 'cat' or 'rock'.
Why AI gets things wrong
Too few examples, unfair examples and overconfidence, and how people test for them.
Five weeks, week by week
This program is still being made. The outline below is the plan, and details may change before it opens.
Week 1
What is AI?
Machines that follow rules, and machines that learn.
- Where you already meet AI
- Rules written by people
- Learning from examples instead
Week 2
Learning from examples
Training, the way a model actually does it.
- Examples and labels
- Finding the pattern
- Testing on examples it hasn't seen
Week 3
How a machine sees
From pixels to patterns.
- Pictures as numbers
- Edges, shapes and features
- Recognizing things it was never shown
Week 4
When it goes wrong
Mistakes, and whose fault they are.
- Too few examples
- Unfair examples
- Checking an AI's answer yourself
Week 5
Your model
Plan an AI that helps with a real problem.
- Choosing the problem
- Deciding what examples it needs
- Saying what could go wrong
Things to build
From things already in the house. No kit to buy.
- A sorting machine made of cards
- A pixel-picture classifier on graph paper
- A decision tree your family can follow
Final project, week 5
Plan an AI that helps
Pick a problem, describe the examples the model would learn from, and list the ways it could fail and how to catch them. An instructor reviews the plan and sends feedback.
Finish every lesson, pass each quiz and get the project approved to earn the certificate, which downloads as a PDF with an ID anyone can check.