Makes you think about how the Transformers learned English

Aug 25, 2014 07:21 GMT  ·  By

Regardless of how much critics hate the Transformers movie series, the films have lots of fans, and consequently, they have given today's scientists a lot of ideas. One of them in the field of intelligent robotics seems to have borne fruit.

Admittedly, allowing robots to learn data from the Internet isn't a new idea, especially since many men and women have dreamed about having the ability to do the same.

A certain cartoon called Freakazoid comes to mind, about a nerd that gains electricity-based powers and all the knowledge on the Internet.

But that's not what we're here to talk about today. No, what caught our eye is the latest advancement at Cornell University.

You see, a team of researchers there have created a computational system that can learn from publicly available Internet resources. The computational system, is called Robo Brain.

This means that future robots equipped with access to such a thing will be able to learn languages and everything else of import from the World Wide Web, just like the Transformers, the Geth from Mass Effect, and so many other robots in Sci-Fi settings.

Until now, human-assistant robots, all robots actually, have had to be taught one thing at a time by their creators, unless they were also designed with the ability to learn from their surroundings and recognized patterns, which doesn't ultimately amount to much.

Sure, some assistant robots can find your keys, pour drinks and put dishes away after a meal, but they can't learn other things on their own.

With the Robo Brain, they will be able to do that and more, although depending on how much autonomy the process is given, the robot may or may not stumble upon less fortuitous information than would be ideal for a household helper.

Don't get the wrong idea though. Robo Brain is not a chip included in the heads of the robots themselves. Instead, it is a supercomputer / cloud server that can receive queries from such robots and send back answers, or download “skill packages” like the ones we mentioned before.

With how widespread 3D access is, robot designers should have an easy time integrating the necessary broadband antennas. The same goes for Wi-Fi.

Thus, in a year or so, you might see robots that can learn from Robo Brain how to handle a coffee mug and what liquids should be poured in it, just by looking at it and sending a query to Robo Brain through the cloud. Same for human languages and behaviors.

The key to the Robo Brain's functionality is “structured deep learning,” where info is stored in “levels of abstraction.” An easy chair is part of the class of chairs, but chairs are a subclass of furniture, which is why Robo Brain knows chairs can be sat on but are not the only things that can be sat on. All in all, it means that robots won't be stumped by “all that glitters is not gold.”