A translator may not know a language, but can tell from the tone of voice whether a person is calm or agitated. In the same way, an algorithm that knew nothing about El Niño was fed 36 years of wind and water temperature records from the Pacific Ocean. It found two recurring states in those records. One turned out to be the ocean's warm phase, which scientists call El Niño. The other was the cold phase, La Niña.

The algorithm first learned to predict water temperature from wind. Then it grouped similar combinations of wind and temperature. This yielded four wind types and four temperature types. Two of them matched El Niño and La Niña exactly. Standard data clustering that ignored the link between wind and water could not achieve this. And when the data were shuffled, the algorithm found nothing, because the link had disappeared. So it truly captured how wind influences ocean temperature.

This method can now be used to uncover other hidden patterns in the climate. For instance, one could take pressure and precipitation and see what stable states they form. In this way, computers might help scientists hear new "intonations" in the ocean and atmosphere that we do not yet discern.