Course 5, lesson 47 of 100, Ages 11+
Who labels the data?
The people behind AI
Like I’m 5
Lots of AI learns from examples that people have labelled by hand. Behind many clever AIs are thousands of people carefully tagging pictures, sounds and words.
The big idea
Labelling can mean drawing boxes around cars in street photos, transcribing audio, or rating which chatbot answer is better. It takes time, skill and attention, especially for medical or legal data.
These workers deserve fair pay, clear instructions and protection from upsetting content. Label quality also depends on good guidelines: if two people label the same example differently, the instructions may need to be clearer.
Examples
- Street photos: Workers draw boxes around pedestrians for self-driving car training.
- Medical images: Doctors label scans because it needs expert knowledge.
- Chatbot ratings: People compare answers and pick the more helpful, honest one.
How it works
- Write clear labelling instructions with examples.
- Have trained people label the data, sometimes twice.
- Compare their labels and fix disagreements.
Check your understanding
- Who labels medical scans for AI?
- Options: Trained medical experts; Anyone with no training; The AI itself, with no help.
Answer: Trained medical experts. Expert data needs expert labellers. - What might it mean if two labellers often disagree?
- Options: The instructions may be unclear; The data is perfect; Labellers are always wrong.
Answer: The instructions may be unclear. Disagreement is a signal to clarify the guidelines.
Remember
Behind much AI are people labelling data. Clear instructions and fair treatment matter.
Talk about it
Would you want a job labelling AI data? What would make it a good job?
Go deeper
Inter-annotator agreement (for example Cohen's kappa) measures label consistency. The working conditions of data workers are an important AI ethics issue.