Artificial Intelligence from Google itself has learned to play the game – Money.pl

2015-02-26 15:21

Author: Sylwia Zimowska

Last year, Google has invested $ 400 million in a little-known company DeepMind a research on artificial intelligence. I have not had to wait for a breakthrough.

The software invented by DeepMind uses two techniques – deep learning, which is a process of deep allowing for machine learning and pattern recognition of objects by analyzing the data, and which was not so far the technique of “reinforced” learning which is a kind of incentive system.

It sounds complicated, but suffice it to take a look at this example. Program called Agent ultimately got to work out 49 classic Atari games, among them was shown to the above version of breakout. His goal was to self-development of concepts of each game by trial and error. By analyzing how the move affects the scoring system Agent understood what he should do in order to achieve the highest possible score. In the case of a breakout took him 600 sessions. Of the 49 games in 29 Agent reached 75% as a result of a professional player. That’s impressive.

And at the same time a little frightening. Indeed all walks toward the machines endowed with self-awareness.

“ If you can not see directly how scientists are working on developing DeepMind like artificial intelligence, there is no idea how fast progress. Risk serious danger comes in 5, maximum of 10 years “- says Elon Musk, who is one of the first investors DeepMind.

Denis Hassabis, founder of the company, however, is more optimistic: – “ We agree that the risk exists and we will treat it with care, but we are still decades from the point at which such technology will be able to threaten us .”

Combining this with the achievements of the Boston Dynamics is also owned by Google is not hard to imagine at this time marching, intelligent robot army, which is against humanity;).

>