Dumb Question: Unsupervised Learning in Science ?

The vast majority of ML related work and investigation revolves on some amount of supervised learning where there are examples of known meanings. Does anybody know of an example where the learning was unsupervised. For instance someone (re-)learning the Egyptian hieroglyphs without the Rosetta stone or somebody (re-)discovering the Mayan alphabet using strictly machine learning techniques ? The example does not need to be about human languages.

Edit: added "need"