Data & Patterns

📊 Data is AI's food

AI learns from data: photos, sounds, text, numbers. The more good, relevant examples it sees, the better it usually gets. A system that recognises maize leaf disease needs many photos of healthy and diseased leaves, taken in different light, on different phones.

🏷️ Labels

In supervised learning, every example comes with the right answer, called a label: this photo is "healthy", that one is "maize streak virus". The AI tries to find what separates the labels. Labelling is often done by people, and mistakes in labels become mistakes in the AI.

🔍 Features

Features are the details the AI looks at to make a decision. To tell mangoes from avocados, useful features might be colour, shape and skin texture. Choosing good features (or letting the AI discover them) is a big part of building a model.

🧪 Training data and test data

You don't test a learner with the exact questions they practised. In the same way, data is split: training data teaches the model, and separate test data, which the model has never seen, checks whether it really learned the pattern or just memorised the examples.

🗑️ Garbage in, garbage out

If the data is poor, so is the AI. Too few examples, wrong labels, or examples that only show one kind of situation (only sunny photos, only one region's accent) produce a model that fails in the real world. Good data is varied, correctly labelled and fairly collected.

Key points

• AI learns from data: photos, sounds, text and numbers.
• A label is the right answer attached to each example.
• Features are the details the AI uses to decide.
• Training data teaches the model; test data checks it.
• Test data must be examples the model has never seen.
• Garbage in, garbage out: poor data makes poor AI.

Try it for real: Be the data collector

• Imagine building an AI that tells ripe from unripe bananas. List the features you would use and how you would label 20 example photos.
• Explain why your photos should come from different shops, times of day and phones.