If you’re too lazy to do the dishes, call a robot. It’s something out of a science fiction movie, horror movie or a dream, depending on how you view artificial intelligence.
These robots are being trained to serve as helpers around the home and office by using the internet to research and understand how the world works and what social norms shape human behavior.
Ideally, all of the knowledge will come from a Robo Brain repository of knowledge downloaded from the internet. The robots will be able to draw upon the information as needed.
« Our laptops and cell phones have access to all the information we want, » said Ashutosh Saxena, a researcher at Cornell working on the project. « If a robot encounters a situation it hasn’t seen before it can query Robo Brain in the cloud. »
Saxena, along with his colleagues from Cornell, Stanford, Brown and UC Berkeley began compiling the information for the Robo Brain from the internet. Beginning in July, they downloaded approximately one billion images, 120,000 YouTube videos and millions of how-to documents and manuals.
The Robo Brain will first process a picture of the image to figure out what the object is and what might be in it. Then it will learn to recognize object based on usage as well as human behavior and language. For example, if a robot sees a coffee mug, it will know what it is, but it will also know that liquids can be poured in and out of it and it must be carried upright when filled.
This kind of knowledge is possible by what computer scientists call « structured deep learning. » In simpler terms, it means that information is stored in many levels. For example, an office chair is a type of chair, which is a type of furniture.
Robots store the information in a Markov model form, which means an object is represented as a set of points connected by lines. It calculates probabilities regarding objects, actions or parts of the image to see what it could be.
Computers aren’t always right, however, so the Robo Brain will have teachers through crowdsourcing. People will be able to make additions and corrections to previously established rules. The public will be able to indirectly teach robots and further the development of artificial intelligence.