The New York Times: Researchers from Google's X Lab connected 16,000 processors into one of the largest neural networks for machine learning ever created. Modeled on the human brain, the system was capable of teaching itself new things, so the researchers turned it loose on images from YouTube to see what it would learn. Like a lot of humans, the neural network spent a lot of time looking at cats. Eventually, it taught itself to recognize cats--and 19,999 other items--with surprising accuracy.
"We never told it during the training, 'This is a cat,'" explained Google fellow Jeff Dean. "It basically invented the concept of a cat."