Machine learning lets a computer learn a task from examples instead of following rules written by hand. A spam filter, for instance, is shown thousands of emails already marked as spam or not spam, and works out for itself which patterns give spam away.
The term was coined in 1959 by Arthur Samuel, an IBM researcher whose draughts-playing program got better the more it played. A learning system has many internal settings. During training they are adjusted again and again until its answers on the examples are as accurate as possible, and the system then uses them on cases it has never seen. A turning point came in 2012, when AlexNet, a system from the University of Toronto, recognised objects in photos far better than earlier methods, using layered 'neural networks' loosely modelled on the brain. Machine learning now powers translation apps, voice assistants, film and music recommendations and the analysis of medical scans. Because it learns from data, gaps or biases in that data carry through into its answers.


