Why machine learning?

Some problems are easy for people but almost impossible to write rules for. Try writing an if-statement that recognises a cat in a photo from raw pixel values: the cat can be any colour, size, pose, lighting or angle. ML sidesteps this by learning the rule from thousands of labelled photos.

Other problems have rules that change over time. Fraud patterns, spam tactics and customer preferences shift every month. A learned model can simply be retrained on fresh data, while a hand-written rule set needs constant manual editing.

ML also helps when there is too much data for a human to inspect — millions of transactions, sensor readings or genomes — and when we want personalisation at scale: a different recommendation for each of a hundred million users.