Sometimes it is disheartening that can take as long to get a robot to do something as basic as walking up stairs. Sometimes, as in this case, progress is one that remind villains like artificial intelligence system Skynet from the Terminator saga.
Scientists at the Tokyo Institute of Technology have just presented a robot that can learn from their environment, to deduce the course of action based on each new learning and, almost more frightening, ask for the Internet to other robots how to do something he does not know do.
This is possible thanks to a new scheduling algorithm called SOINN (Self Organizing Incremental Neural Network). For now, SOINN serves our artificial buddy is able to deduce that, if you ask for a glass of cold water, should momentarily leave the glass in the liquid that has taken the ice to catch and throw. This seems so simple, requires a series of deductive processes that are not easy to implement in an artificial intelligence system without having to program one by one.
[video=youtube]http://www.youtube.com/watch?v=OC2TTslf_YM[/video]
The best thing is that SOINN algorithm is incremental. In other words, the robot learns from situations that have already experienced. The platform even allows a robot to learn from the experiences of another. An Internet connection is enough to make a robot that has never made ​​tea can learn to do it, 'talking' to other robots.
For now, the goal of this new algorithm does not seem to get our artificial replicas reach the conclusion that we are a nuisance to them, but to get robots to perform tasks in production plants are able to continue their work even if unforeseen occurrence for non-scheduled. It also opens the doors for these devices are able to understand human language commands in colloquial.