Teaching AI to Understand Pictures and Words
How does AI learn what a cat is? Someone must show it.
Raw data becomes labelled data, then AI learns from it.
In simple words
Labelling is like adding sticky notes to pictures. You write what you see, and AI learns from your notes.
The real definition
Data annotation is adding descriptive labels to raw data. AI models learn patterns from these labelled examples to make predictions.
Like… Teaching with flashcards
You write a question on one side and the answer on the other. AI practises with these cards until it learns the answers.
But: Real learning involves understanding; AI finds patterns but doesn't truly understand like humans do.
You see it every day
Photo recognition
A person marks dogs and cats in photos. A phone's camera then spots them automatically.
Traffic sign reading
Engineers label thousands of stop signs and speed limits. Self-driving cars use this to read road signs.
Email filters
You mark emails as spam or safe. Your email service learns to sort new messages.
Step by step
- 1
Collect raw data
Gather unlabelled photos, videos, text, or audio files.
- 2
Create labelling guidelines
Write clear rules about what each label means.
- 3
Humans label data
People add tags, descriptions, or boxes to each item.
- 4
Train the AI model
AI learns patterns from labelled examples to predict new data.
- 5
Check for mistakes
Verify that labels are correct and consistent.
Remember
Memory trick
Think: RAW → LABELS → LEARN → PREDICT. Every step needs human care at the start.
Words
- annotation ann-oh-TAY-shun
- Adding descriptions or labels to data so AI can learn.
- label LAY-bul
- A word or tag that describes what is in the data.
- dataset DAY-tuh set
- A collection of data used to teach an AI model.
- raw data raw DAY-tuh
- Information in its original form, without any labels or organisation.
- model (AI) MOD-ul
- A trained program that makes predictions based on patterns.
- training TRAY-ning
- The process where AI learns patterns from labelled examples.
- pattern PAT-urn
- A repeated rule or structure that AI finds in data.
- crowdsourcing CROWD-SOR-sing
- Asking a large group of people to help with a task.
Check yourself
Why do we label data before training an AI model?
Hint
Think about how you learn: you need examples and answers.
Labels are examples that teach AI to recognise patterns in new data.
A hospital labels X-rays as 'healthy lung' or 'sick lung'. What role do humans play?
Hint
Who decides what 'healthy' means in a medical X-ray?
Humans provide the correct answers; AI learns from them to help doctors faster.
What could happen if someone labels data incorrectly? Give one example.
Show a good answer
If photos of cats are labelled 'dog', the AI learns wrong patterns. It will then call cats 'dogs' when it sees new photos.
Tell a friend
“Data labelling is when humans mark data to teach AI what things are.”
People also ask
Who does data labelling?
Companies hire people or use crowdsourcing platforms. Humans provide most labels because AI cannot label accurately without training first.
How long does labelling take?
It depends on the data size and complexity. Labelling thousands of images can take weeks or months with many people.
Can AI label data by itself?
No. AI needs human-labelled examples first to learn. Over time, AI can help speed up labelling. Humans verify quality.