Deepmind Builds a Picture of its World

Building a model of existing and changing context is a key part of our intelligence. 

DeepMind, a subsidiary of Google parent company Alphabet focused on artificial intelligence (AI), has created a computer program that independently builds a mental picture of the world. 
The system, which uses a generative query network, looks at a scene from several angles, then describes what it would look like from another angle. This requires the relatively sophisticated ability to learn about the physical world and interpret a scene similarly to a human. Such technology could lead to deeper AI that allows machines to describe and reason about the world. The system uses two neural networks, with one learning while the other "imagines" new perspectives. Harvard University's Sam Gershman says the DeepMind work combines important ideas about the mechanics of human visual perception. .... "