Starcraft II players will be able to face AlphaStar, DeepMind's AI

StarCraft proves to be a delicacy to determine the effectiveness of artificial intelligence to manage multiple tasks simultaneously and in real-time. 

Indeed, the game requires skills in areas as diverse as game theory, imperfect information management, long-term strategy, real-time actions, and actions to be carried out over large areas.

Sub-objectives of constructive play

For example, the only purpose of the game is to beat his opponent, of course, but to achieve this, the player must pay attention to several sub-objectives, such as the management of his resources, their collection, the construction of protective buildings or allowing to generate troops. 

Depending on the situations and the ways of playing, the game can last only a few minutes, or reach one hour, which implies that certain actions will have no direct consequence. Finally, the fact of not seeing the map in its entirety requires specific actions, such as recognition, and the decision-making that results from the information collected.

A scientific experiment

This year, the IA of Deepmind called AlphaStar had managed to beat professional players of StarCraft II. She had then won an overwhelming victory with a score of 10 - 1. 

Now it is possible for anyone to rub AlphaStar's artificial neurons, which will come in different experimental versions. Nevertheless, players will only be able to express their interest in facing the AI ​​via a pop-up or game options, but will not be warned when they are actually playing against this machine, so as not to distort the data, and offer an experience as close as possible to what would be a real one-on-one match. 

"  AlphaStar will play a small number of games for scientific purposes, but we will not reveal exactly when, or how often, the AI ​​will play these parts, to be sure they remain anonymous.  " 

So it is well anonymously and during a few games only that the AI ​​will endorse, in turn, the three races available in the game (Terrans, Zergs, and Protoss), so as to continue training AlphaStar. Nevertheless, replays of these parts should be available, accompanying a scientific publication, later this year.