Data and Patterns

If AI learns from data, what exactly does "data" mean? And how does learning from data actually work? This lesson breaks it down without the maths.

Data is just recorded experience

Every time you search online, buy something, or stream a song, a record of that action is created. Multiply that by millions of people and years of activity, and you have an enormous collection of recorded human behaviour — data. AI uses this as its raw material for learning.

Patterns are the point

AI systems do not memorise individual records. Instead, they look for patterns — things that tend to go together or follow each other. Some patterns are obvious: people who buy nappies often buy wipes. Others are subtle: the exact combination of words in an email that makes it likely to be spam.

The key insight is that patterns in past data are often good predictors of future behaviour. If ten thousand people with your browsing history ended up buying a particular product, it is reasonable to suggest that product to you too.

More data, better patterns

The quality and quantity of data matters enormously. An AI trained on New Zealand medical records will make better predictions for New Zealand patients than one trained only on data from the United States, where genetics, diet, and health conditions differ. This is why the source of training data is one of the most important — and often least discussed — aspects of any AI system.

Garbage in, garbage out

If the training data is biased, incomplete, or just wrong, the AI will learn to reproduce those flaws. A hiring tool trained on historical recruitment data will likely learn to replicate the biases of past hiring decisions. This is not a bug — it is the system working exactly as designed, which is what makes it a problem.

Data is not neutral. It reflects the world that created it — including all its inequalities and blind spots.