Five ways into how AI works.
Each learning path pairs short lessons with a long-read article and one hands-on idea. Everything is free, there is nothing to sign up for, and you can start anywhere.
How a path works
- Play
Start with a small experiment, like the pixel pad on our home page. You get a feel for the idea before any theory.
- Read
Four short lessons explain what just happened, one concept at a time, with everyday examples.
- Try
Each path ends with a "Try this" task you can do with your own phone or browser in a few minutes.
Path 01 · Vision
Machines that see
How a computer turns a grid of coloured dots into "that's a stop sign", and why it sometimes sees a stop sign where there is none.
- pixel
- feature
- layer
- label
- confidence score
-
Pixels are numbers
Every digital image is a grid of brightness and colour values. You will see how a photo looks to a computer before any AI touches it.
-
Edges, then shapes
Early layers of a vision model react to simple contrasts. Later layers combine them into corners, wheels, eyes and whole objects.
-
Labels and confidence
What "92% cat" actually means, why the number is not a probability you can bank on, and how a model picks between labels.
-
When vision goes wrong
Odd lighting, unusual angles, stickers on signs and people who were missing from the training photos.
Try this: point your phone's photo search at the word "dog" and count how many results are not dogs.
Path 02 · Foundations
How models learn
The core loop behind almost every AI system: show examples, measure mistakes, adjust. And why the examples matter more than anything else.
- training data
- parameter
- error score
- overfitting
- test set
-
Rules or learning
When a programmer writes every step, and when a model works the steps out from examples. Most apps mix both.
-
The training loop
Guess, measure the error, nudge the numbers, repeat. We walk through one round slowly so the idea sticks.
-
Testing on new data
A model is only useful on examples it has never seen. Here is how builders hold data back to check that.
-
Data shapes behaviour
Gaps, imbalances and labels people disagree on all leave fingerprints on what a model does later.
Try this: look at the photos you would use to teach a model "breakfast". Whose breakfasts are missing?
Path 03 · Language AI
Words and chat
What happens between your question and the answer on screen, and why a smooth, confident answer can still be wrong.
- large language model
- prompt
- context
- hallucination
- knowledge cut-off
-
Text in, numbers out
Before a model can work with a sentence, the text is split into small pieces and turned into numbers. Here is the plain version.
-
Predicting the next word
A chat assistant writes one small piece at a time, picking what is likely to come next given everything before it.
-
Fluent is not the same as true
Why made-up facts, quotes and references appear, and which kinds of questions make them more likely.
-
Checking what you read
Simple habits for important answers: ask for the source, open it yourself and compare with an official page.
Try this: ask a chat assistant about a tiny local street you know well, then compare its answer with what you know.
Path 04 · AI in finance
AI and everyday money
The ideas behind the payment checks you meet every day, explained as concepts. We do not review or recommend banks, cards, apps or any financial product.
- anomaly
- risk score
- false alarm
- threshold
- human review
-
What "unusual" means to a computer
A payment is compared with your usual pattern and with patterns seen across many accounts. Different from usual is not the same as wrong.
-
Scores, not verdicts
Models usually produce a score. Rules and people decide what happens next, from letting it pass to asking you to confirm.
-
False alarms and missed cases
Every alert system trades one kind of mistake for the other. Here is why a declined coffee and a missed fraud case are two sides of one dial.
-
Your side of the process
Why a bank may ask you to confirm a payment, and why you should reply only through the number on your card or the official app you already use.
Try this: list three things about a payment that would look unusual for you, such as the place, the time or the amount.
Path 05 · Privacy
Your data and AI
What AI apps may collect from you, how Canada's federal privacy law for businesses applies, and what you can ask for.
- personal information
- consent
- purpose
- retention
- PIPEDA
-
What AI apps may collect
What you type, files and photos you upload, voice recordings, device details and how you use the app.
-
Consent and purpose
The basics of PIPEDA: organizations should say why they collect personal information and use it only for reasonable purposes.
-
Your rights
Asking for access to your information, asking for corrections, withdrawing consent and complaining to the Office of the Privacy Commissioner of Canada.
-
Everyday habits
What to keep out of a chat box, how to read the settings that matter and when to delete a conversation.
Try this: open the privacy settings of one AI app you use and find out whether your chats are used to improve the service.
How to use the materials
All lessons and articles are free to read, share and come back to. No account, no paid version, no certificate to buy.
- New to AI?
- Start with Path 02, then Path 01. They give you the vocabulary for the rest.
- Teaching a class?
- You may link to any page. The pixel pad and the quiz work well on a projector, with no setup.
- Worried about a specific app?
- Path 05 explains your rights in Canada. For a real complaint, go to the Office of the Privacy Commissioner directly.
- Need advice?
- We do not give personal financial, legal or technical advice. Our paths explain how things work in general.