AI defeated professional players in StarCraft II: Developers had to succumb

AI defeated professional players in StarCraft II: Developers had to succumb
03:02 Today’s Telegraph naked-science.ru Photo: from open sources news technology: neural Network DeepMind has defeated professional players in StarCraft II

Artificial intelligence AlphaStar, developed by DeepMind (belongs to holding Google Inc.), defeated two professional players in the strategy StarCraft II, taking place in real time. This is stated on the website of the company.

Matches algorithm with two professional players — Dario Wansem (Dario “TLO” Wünsch) and Comin Grzegorz (Grzegorz “MaNa” Komincz) — took place in December, but this week, 24 January, was the last fight, before which properties of the program changed. As a result, the first victory went to the man.

Earlier algorithms DeepMind defeated the strongest go player and much better than predicted form of protein, compared to professional biologists . However, in each of these cases for artificial intelligence have been opened all the available information. With real-time strategy such as StarCraft II, everything is different: it is a game with closed information. The machine does not know what is happening behind the fog of war, and can not predict the actions of the player, so she has to adapt to his maneuvers in the moment.

The developers of the program used for learning games real people that they provided by the company Blizzard. Adopting the behavioral patterns of the players, she learned to defeat the in-game artificial intelligence of the highest complexity in 95 percent of cases. Then the agents began to compete with each other in the tournament for the knockout. It took place at an accelerated rate, so each one is enough for about 200 years.

Selecting the most successful agent, the authors put him against TLO, and MaNa. First, he won five of five games from TLO, and then another version of the neural network many times defeated MaNa. In this case the algorithm made fewer steps per minute than his opponents. This is probably due to a slight advantage, which he still possessed. Unlike man, he saw no part of the known card (the same screen), and the whole of it, so he did not have to switch between the different areas. Moreover, the audience noticed that he gave the orders to three groups located in different zones at the same time, what exactly is not an able man.

In the last game the developers made concessions and turned off this ability AI this time won MaNa. However, the authors noticed that the version of the algorithm, which you need to move the camera, was educated in the tournament just seven days, which is half of the previous training period. MaNa himself praised the actions of the program: “I Play with a real person, he’d be filigree to manage their stalkers (unit one Russ. — Approx. ed.)”.

The developers have said that games like StarCraft II are the only training ground for artificial intelligence. In the future it will be accepted for other purposes. For example, earlier in DeepMind has stated that their algorithm will diagnose eye disease.

Go to the Main page to Add a comment

Leave a comment

Your email address will not be published. Required fields are marked *