Data & labels, explained simply
Most of the AI you meet, from photo apps to voice assistants, learned from examples rather than from rules someone typed in. Those examples are called training data.
Each example comes with a label: the right answer. A picture of a strawberry is labelled "berry", a picture of a quartz stone is labelled "crystal". The AI studies lots of labelled examples to spot what the members of each group have in common.
Whoever collects and labels the data is the AI’s teacher. If the labels are careless or the examples are too similar to each other, the AI learns the wrong lessons. Good data is the most important ingredient in machine learning.
Example: Labelling a fruit bowl
- Photo 1: red, round, bumpy → label: berry
- Photo 2: clear, pointy, shiny → label: crystal
- Photo 3: purple, round, smooth → label: berry
- After enough labelled photos, the model can guess the label for a photo it has never seen.
Try this
- Sort ten household objects into two groups and write a label on a sticky note for each. Could a friend work out your rule?
- Label the same objects a second way, such as "hard" and "soft". Same data, different labels, different lesson.
- Mix up one label on purpose and talk about how that could confuse a learning machine.
