shapesinfinity.
Learn AI · Beginner · 3 min

Teaching AI to Understand Pictures and Words

How does AI learn what a cat is? Someone must show it.

How Data Labelling Works
Raw data (photos, text)Human labels imagesLabelled dataset growsAI model trainsAI recognises new imagesLabels must be correct

Raw data becomes labelled data, then AI learns from it.

goes inthe AI workscomes outwatch out

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

on your phone

Photo recognition

A person marks dogs and cats in photos. A phone's camera then spots them automatically.

out in the world

Traffic sign reading

Engineers label thousands of stop signs and speed limits. Self-driving cars use this to read road signs.

at work

Email filters

You mark emails as spam or safe. Your email service learns to sort new messages.

Step by step

  1. 1

    Collect raw data

    Gather unlabelled photos, videos, text, or audio files.

  2. 2

    Create labelling guidelines

    Write clear rules about what each label means.

  3. 3

    Humans label data

    People add tags, descriptions, or boxes to each item.

  4. 4

    Train the AI model

    AI learns patterns from labelled examples to predict new data.

  5. 5

    Check for mistakes

    Verify that labels are correct and consistent.

Remember

Labels teach AI what things areHumans do the labelling workBad labels make bad AIMore labels mean better learningLabels are like answers on a test

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

quick check

Why do we label data before training an AI model?

Hint

Think about how you learn: you need examples and answers.

quick check

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?

think

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.