Google DeepMind AI teaches itself using snippets of video

Google's DeepMind has developed an AI that identifies pictures and sounds in snippets of video before being told by a human what it is watching. The incredible system uses an algorithm to recognise things such as crowds, people dancing and water, without seeing a specific label for any of them. Currently most algorithms need to be told what images are to tell them apart - for example they need to be told something is a dog before understanding it as such. This research takes AI closer to being able to teach itself by watching the world around it - just like humans do everyday. This advanced algorithm was created by combing one network that was expert at recognising images and another that could identify audio.