Developing computer systems capable of understanding the world will require algorithms that learn patterns and high-level concepts without the extensive aid of humans. Though a great deal of progress has been made on applications by training deep artificial neural networks from human-provided annotations, recent research has also explored methods to train such networks from unlabeled data. These “unsupervised” learning methods attempt to discover useful features of the data that can be used for other machine learning tasks. In some cases, we find that neural networks trained in this way are able to detect meaningful patterns on their own without any prior knowledge. I will describe the basic motivations for this approach and explain how such goals can be achieved by certain learning algorithms. I will also overview several results that highlight the potential capabilities of the unsupervised learning approach.