A new tool allows artificial intelligence to use a PC as a human
Artificial intelligence can be found in virtually any digital aspect that we interact with on a daily basis. Whether the self – labeling Facebook, or bots of Telegram that allow us to insert GIFs in conversations, work with intelligent algorithms to daily.
So far we have seen how these algorithms allow us to perform tasks at a very basic level, we find this part of the code embedded within the software. Now a new tool developed by OpenAI allows artificial intelligences use a PC as human reportedly on Quartz.
Do they grow eyes and hands? Something like that. To give a brief description, we can say that will use a mouse and virtual keyboard in addition to “look” to the screen in the same way that we do.
This is Universe, the tool that makes it possible
The idea behind this solution called Universe is that it is a major resource of a particular form of artificial intelligence research known as “reinforcement learning”. In this way the algorithm learns from its mistakes through billions of repetitions.
The idea behind the use of this technique is to “teach” any artificial intelligence so you can complete any task that a human would on a computer. This means that, for example, machines could play video games or browse the Internet for information. In the process it may be one of the custom web IA generated by using a tool called Wix.
The goal of its creators is to be able to write software which equates to general intelligence , with the ability to do anything with a computer. From OpenAI hope that once the algorithms have learned to solve a few tasks with Universe can transfer that knowledge to overcome new challenges.
Ilya Skutsever, research director OpenAI, the ideal version of the algorithm sounds very similar to what a human would be :
The key skill that we would like our artificial intelligence systems to have is general problem solving. This way you could take the algorithm and place it in a new world, where you would understand what is going on, find out what to do and then get it.
More than a program to “educate” machines
The practical cases in which Universe can be used compose a very wide range. The idea is not limited to train and “teach” artificial intelligence to solve problems, but can also be used to objectively measure the performance of their algorithms as a benchmarktarget.
Universe in any case is not the only program of reinforcement learning OpenAI available. They also have other techniques compiled within the suite OpenAI Gym:
Researchers could also give a slight edge to his creations recording their own shares in video games or surfing the Internet, so that the algorithm knows what the correct behavior .
In a sense we can say that Universe complements another tool “training” of artificial intelligence called ImageNet. Both can be used to cope with the resolution of a problem: ImageNet checks whether an artificial intelligence understands the world around it, while Universe challenges it to act in that world.
A technology still under development
The team that deals with the development of Universe is developing hundreds of different environments , each designed for a different application. The test process should last for a few years, but for now one of these algorithms has already discovered how to win Pong in less than 10 minutes.
This project is part of the philosophy of OpenAI to make research projects on artificial intelligence are available for everyone.