American experts showed the IMPALA algorithm, able to successfully play a computer game with a virtual maze. Thanks to applied professionals a new approach to artificial intelligence was able to cope with this task much better than previously existing algorithms.
At first glance, the significance of the results can be greatly underestimated — people playing computer games, often playing against real opponents, and “artificial intelligence”. However, in the case of such opponents this phrase not too useful and is a sort of a tribute to tradition. The behavior of “computer” in this case is programmed in any situation not provided for by the game developers, the program cannot “go it alone” to find a solution. At the same time, the artificial intelligence, which by today’s standards can be called “real”, is able to learn the different actions, getting “points” if successful or “penalty points” in case of errors. However, he does not have access to the internal code of the game and learns to play it by trial and error is almost the same as it would a man.
In the past, algorithms such as those presented recently, have already successfully demonstrated the ability to overcome labyrinths located in them portals — these videos have already appeared two years ago. Experts say that to demonstrate the algorithm’s ability to navigate in a virtual maze not too far to seek the way in the real world.
However, now experts report that artificial intelligence has reached a new level of game he is ten times more efficient, and gaining points, more than double the previous algorithms. This is due to the fact that the system consists of two students, analysing different information, and then exchange among themselves the results.
Get short formal newsletter the best in the “MK” – subscribe to our Telegram.