Foundations · Module 3
Supervised and unsupervised learning
When we say a model learns, we mean it changes its internal settings so it can make better guesses.
Previously
Data and representation
In AI, the word data sounds fancy, but it is usually boring.
This module
Supervised and unsupervised learning
When we say a model learns, we mean it changes its internal settings so it can make better guesses.
Next
Responsible AI basics and limitations
AI systems can cause harm even when everybody is trying to do the right thing.
Progress
Mark this module complete when you can explain it without rereading every paragraph.
Why this matters
Glossary Tip.
What you will be able to do
- 1 Explain supervised and unsupervised learning in your own words and apply it to a realistic scenario.
- 2 The learning paradigm follows from what you have, what you want, and what error costs you.
- 3 Check the assumption "Success can be defined" and explain what changes if it is false.
- 4 Check the assumption "You can tolerate mistakes" and explain what changes if it is false.
Before you begin
- No previous technical background required
- Read the section explanation before using tools
Common ways people get this wrong
- Optimising the wrong target. A high score can hide harm. If the metric is not aligned with the goal, the model gets good at the wrong job.
- False certainty. A model output is not a fact. When uncertainty is high, the system should slow down and ask for a better input.
Main idea at a glance
Two ways models learn from data
Supervised has answers. Unsupervised searches for structure.
Stage 1
Do I have labelled answers?
I start by asking whether my training data has correct answers attached. This choice splits my approach entirely.
I think this is the most important branching point in model building.
When we say a model learns, we mean it changes its internal settings so it can make better guesses. It is not learning like a person learns. It is closer to practice. You show examples, it adjusts, and it gets less wrong over time.
Interactive lab
Glossary Tip
This module includes an interactive practice component. Open the deeper tool or workspace step when you want to test the idea rather than only read it.
Interactive lab
Glossary Tip
This module includes an interactive practice component. Open the deeper tool or workspace step when you want to test the idea rather than only read it.
In supervised learning, you give the model an input and an answer. The model tries to guess the answer, then it is corrected. Over many examples, it learns a pattern that can generalise to new cases.
Email spam filtering is a classic supervised example. You have emails, and you have labels like spam and not spam. Image classification is another. You have images, and you have labels like cat, dog, or receipt. House prices are supervised too, but the answer is a number. The same pattern applies. Inputs in, answer attached, model learns to predict.
There are two common supervised shapes.
Interactive lab
Glossary Tip
This module includes an interactive practice component. Open the deeper tool or workspace step when you want to test the idea rather than only read it.
Interactive lab
Glossary Tip
This module includes an interactive practice component. Open the deeper tool or workspace step when you want to test the idea rather than only read it.
The difference matters because the mistakes feel different. A wrong category can block a real email. A wrong price can cost real money.
Interactive lab
Glossary Tip
This module includes an interactive practice component. Open the deeper tool or workspace step when you want to test the idea rather than only read it.
Instead of asking "what is the right label", you ask "what patterns exist". This is useful when labels are missing, expensive, or not even well defined.
Grouping customers by behaviour is a common unsupervised example. You might discover that one group buys weekly and another group buys once a year. Topic discovery in documents is another. You might find clusters of themes in support tickets without anyone labelling them by hand. Anomaly detection is a third. You look for unusual behaviour that might signal fraud or intrusion.
Unsupervised learning is harder to evaluate because there is no single correct answer waiting in a spreadsheet. If you change your settings, the groupings can change. Sometimes both results are reasonable. You have to judge usefulness, not just score points.
Imagine a bank clustering transactions to find “normal” behaviour. If the system learns that weekend spending is “unusual” for a certain group, it might flag normal customers as fraud. Unsupervised results still need human judgement and context.
In practice, teams use clustering to create segments and then make decisions based on those segments. That means errors in the clustering can become policy, pricing, or access decisions. Treat cluster labels as hypotheses, not truth.
Here are a few beginner misconceptions to avoid. First, more data is not always better data. If it is biased or messy, you scale the problem. Second, unsupervised learning is not a free shortcut. It still needs careful interpretation. Third, a model learning a pattern does not mean it understands a reason. It means it found a shortcut that worked on the training data.
Worked example. A classifier that is "accurate" and still useless
Worked example. A classifier that is "accurate" and still useless
Suppose only 1% of transactions are fraud. A lazy model that always predicts “not fraud” gets 99% accuracy. It is also worthless. This is why I keep saying: one metric can lie to you.
In real systems, you usually care about questions like: how many real fraud cases did we catch (recall), how many innocent people did we annoy (false positives), and what is the operational cost of review.
Common mistakes in learning paradigms
Learning paradigm mistakes
-
Assuming supervised labels are ground truth
Labels can be noisy, biased, or outdated, so supervised training still needs judgement.
-
Treating cluster names as real categories
Unsupervised groupings are hypotheses, not fixed truth about people or behaviour.
-
Rewarding one metric and ignoring side effects
The model optimises what you score, even when it harms other outcomes.
Verification. Prove you understand the evaluation question
Evaluation reasoning drill
-
Choose the error that hurts most
For spam, fraud, or triage, explain which failure has the larger real-world cost.
-
Set a borderline-case policy
Define review, escalation, or defer rules before the model goes live.
-
Distinguish uncertainty from confident error
Uncertainty needs support signals. Confident error needs faster controls and correction.
After this section you should be able to
Section outcomes
-
Explain when supervised learning is appropriate and what it optimises for
Map labelled tasks to the correct metric and decision boundary.
-
Explain when unsupervised learning is useful and why evaluation is harder
Judge usefulness instead of assuming there is one objectively correct grouping.
-
Explain what breaks when people treat model outputs as understanding
Recognise shortcut learning and prevent over-trust in fluent outputs.
Mental model
Choosing a learning setup
The learning paradigm follows from what you have, what you want, and what error costs you.
-
1
Do you have labelled outcomes
-
2
Supervised learning
-
3
Unsupervised learning
-
4
Do actions change the world
-
5
Reinforcement learning
Assumptions to keep in mind
- Success can be defined. If you cannot say what good means, models will optimise the wrong thing. The best metric is the one tied to a real decision.
- You can tolerate mistakes. Some errors are acceptable and some are not. Safety comes from deciding this early, not after a failure.
Failure modes to notice
- Optimising the wrong target. A high score can hide harm. If the metric is not aligned with the goal, the model gets good at the wrong job.
- False certainty. A model output is not a fact. When uncertainty is high, the system should slow down and ask for a better input.
Check yourself
Quick check. Supervised and unsupervised learning
0 of 12 opened
Scenario. After training, the model gets better at predicting from examples. In plain language, what changed
It adjusted its internal settings so it makes better guesses from examples.
Scenario. You accidentally train and test on the same emails. Why is that a problem
You are grading your own homework. It makes results look better than real performance.
Scenario. You have thousands of emails labelled spam or not spam. What makes this supervised
Examples include the correct answer the model should learn to predict.
Scenario. You have no labels, but you want to group customers by behaviour. What makes this unsupervised
Examples have no answers, so the model looks for patterns or structure.
Scenario. A model predicts 'spam' vs 'not spam'. What type of task is that
Classification: predicting a category.
Scenario. A model predicts a delivery time in minutes. What type of task is that
Regression: predicting a number.
Scenario. Why is unsupervised learning harder to evaluate
There is no single correct answer, so usefulness depends on interpretation and context.
Scenario. A clustering tool produces five clusters with neat names. What is a practical risk
Over-interpreting clusters as 'real categories' when they are just one grouping choice.
Scenario. You train a classification model and it learns patterns. What does 'training' actually do in technical terms
It adjusts internal parameters to minimise prediction error on training examples, learning a mapping from inputs to outputs.
Scenario. A model that predicts fraud has high precision but low recall. What does that mean in plain terms
When it says fraud, it is usually correct, but it misses many real fraud cases because it is too conservative.
Scenario. Why do we use validation data separately from training data
To make honest choices during model building without cheating by testing on the same data the model learned from.
Scenario. You discover your unsupervised clustering groups customers differently each time you run it. Is this normal
Yes, unsupervised results can vary with settings. You must judge usefulness and consistency, not assume a single correct answer.
Artefact and reflection
Artefact
A short module note with one key definition and one practical example
Reflection
Where in your work would explain supervised and unsupervised learning in your own words and apply it to a realistic scenario. change a decision, and what evidence would make you trust that change?
Optional practice
Adjust distance, grouping, and ambiguity. See how reasonable clusters can still mislead if you treat them as truth.
Also in this module
Train a tiny classifier
Adjust how much training data you have and how noisy it is. Notice how it changes expected accuracy and stability.
Threshold trade-offs playground
Move the decision threshold and watch false positives and false negatives change. Practise picking a threshold that matches the real-world cost.