The network won in all rounds.
In the course of the game StarCraft II the players have to control armies and to develop the infrastructure, given the short-term and long-term benefits. Players can’t always consider the full map of the battle, so they have to act relying on intuition rather than explicit data, and respond quickly to enemy action.
Artificial intelligence played against Dario Vunisa and Gzhegozhem of Cominch. It should be noted that AlphaStar could see the whole map, and always ran only one of three possible armies. To train the neural network developers have used the game of real people. Various agents of the program competed with each other in the shootout. The tournament took place at an accelerated rate, so each one is enough for about 200 years.
The developers admitted that they did not expect the success of their artificial intelligence, because the competition neural network with real players, conducted earlier, was far from winning first.
© 2019, z-news.link. All rights reserved.