AI Explorers · lesson 7 of 24 · 25 min
Imagine describing your best friend to someone who has never met them, using only a short list of facts. What would you pick? Computers face that same problem every time they work with the real world.
What is representation?
A representation is a way of describing something so a computer can work with it. Computers cannot hold a real sneaker or a real song. They hold descriptions made of numbers, words and categories.
The details we choose to describe are called features. For a sneaker, features might be:
| Feature | Example value |
|---|---|
| Brand | (any brand name) |
| Size | 8.5 |
| Main color | White |
| Price range | Medium |
| Style | High-top |
| Good for | Basketball |
A well-chosen set of features makes a problem easier. If an app wants to recommend basketball shoes, “Good for” and “Style” matter a lot. “Color of the box” probably does not.
Choosing features is a human decision
Picking features is powerful, and it is done by people. Say a school wanted a computer to suggest who should join an advanced math class. If the only feature is “grade on one test,” students who had a bad day get left out. Adding features like “homework completion” and “teacher recommendation” gives a fuller picture. But adding the wrong feature, like a student’s zip code, could quietly bring unfairness into the system.
So whenever you see an AI decision, ask: what features did it use, and what did it leave out?
Graphs: representing connections
Some information is about connections. For that, computer scientists use a graph, which is not a bar chart but a set of dots connected by lines. The dots are called nodes and the lines are called edges.
A map app treats intersections as nodes and streets as edges. Each edge can have a number, like how many minutes it takes to drive. A social app treats people as nodes and friendships as edges. Once the world is a graph, the computer can answer questions like “what is the shortest way from home to the rec center?” You will do exactly that in lesson 9.
Representation is also why the same thing can look very different to different computer systems. A music app might describe a song with features like tempo, genre and how often people skip it. A lyrics site might describe the same song by its words. Neither is wrong. Each one picked the features that fit its job. Choosing a good representation is often the first big step in solving any AI problem.
Try it
- Pick a category you know well: songs, sneakers, snacks, video games or sports teams.
- Choose six examples from that category.
- Make a table with the six examples as rows and four features you choose as columns. Fill it in.
- Now pretend you are a recommendation app. A friend tells you one example they love. Using only your table, which other example would you recommend? Circle it.
- Write one sentence: which feature mattered most for your choice, and is there a feature you wish you had added?
Check yourself
What is a feature?
A detail used to describe something in data, like size or color for a sneaker.
Why does it matter who chooses the features?
Features shape what the computer pays attention to. Leaving out important details or adding unfair ones can lead to bad or unfair results.
In a map app’s graph, what are the nodes and edges?
Nodes are places like intersections, and edges are the streets connecting them, often labeled with travel time.
Key words
- Representation: a way of describing something so a computer can use it.
- Feature: one detail that describes an item in data.
- Graph: nodes connected by edges, used to show connections.
- Node: a point in a graph, like a place or a person.
- Edge: a connection between two nodes.
Previous lesson: How computers hear: sound and speech
Back to AI Explorers
Next lesson: Rules and decision trees