Data: turning the world into a table of numbers

You hear the word data all the time: big data, data protection, data centre. But what is it? And why does every AI lesson start with it?

What is data?

Data is information that has been written down or recorded, usually as numbers or words. The temperature at noon is a piece of data. So is the number of goals in a match, your height, or a list of songs you have played.

You have already seen that computers turn everything into numbers. Data is where those numbers come from.

Where does data come from?

• Sensors measure the world: a thermometer, a camera, a microphone, a step counter.
• People fill in forms, take surveys, or tap buttons.
• Programs keep records: the games you played, the pages you opened.
• Scientists run experiments and write down every result.

Data in a table

The most common way to arrange data is a table, like a spreadsheet. Here is a tiny one about fruit:

In a table:

• Each row is one example, one thing we measured. Here, one piece of fruit.
• Each column is one kind of measurement. We call these features. Here, weight, redness and sweetness.
• Sometimes there is a special column, the label. This is the answer we want to know. Here, it is the kind of fruit.

Look at the numbers. The apples are red and a little less sweet. The bananas are not red and are sweeter. Can you spot the pattern? You did it by looking at the data. That is exactly what a learning computer tries to do.

Words become numbers too

What if a column holds words, like colours? The computer turns them into numbers too, using a code: for example, 0 for red, 1 for green and 2 for yellow. Or it uses a separate yes-or-no column for each colour. Remember: if it is in a computer, it is numbers.

Features: choosing what to measure

Which measurements are useful? To tell apples from bananas, colour helps a lot. Which shop sold it does not. Picking good features is an important skill. Too few, and you cannot tell the answer. Too many unhelpful ones, and the useful pattern gets lost in the noise.

Good data and bad data

A computer can only find patterns that are in the data you give it. So the quality matters:

• Enough examples. Four fruits is too few to be sure. Thousands is better.
• Variety. If you only measured red apples, the computer will never learn that some apples are green.
• Accuracy. If some numbers are wrong or missing, the patterns blur.
• Fairness. If the data leaves out some people or situations, any system built from it will work badly for them.

People say garbage in, garbage out. You cannot get good answers from bad data.

Where this leads

Imagine you have a table with thousands of rows and the labels filled in. Could a computer work out the rule for itself? That is the idea of machine learning, and it is where this path is heading.