STRONG POINTS



  • DeepMind AlphaStar AI has been under construction since the end of 2016.
  • He is trained with over 200 years of cumulative playing experience.
  • StarCraft II is perceived as a major challenge because of its complexity.


AlphaStar, an artificial intelligence bot created by the DeepMind artificial intelligence research company to play the famous sports game Starcraft II, beat two professional human players by a staggering margin in their first demonstration in situ real. In a series of eleven games, a human could only win once. The only match in which a human won was played with slightly different variables preventing Alphastar from seeing the game map in its entirety, which was the case for all the other matches as it interacted directly with the game engine rather than via the on-screen interface. Human players can not see the entire card during the game. DeepMind has not released detailed information about the exact functioning of the IA bot, but indicates that a peer-reviewed article will be published soon.

The experiment was conducted with human players Dario W√ľnsch, known as TLO, and Grzegorz Komincz, who plays the role of MaNa, both of the Liquid team. They faced the AlphaStar AI in five games, each using a ladder card in professional match conditions, and all lost. Each time, a slightly different version of AlphaStar was used to see how different formations and strategies might work. These ten matches were played in a few weeks and not all were shown during the event.

MaNa then played an extra match with a prototype branch of AlphaStar during the demonstration and was able to win. This particular branch used a camera interface similar to that seen by humans, rather than knowing the entire map as in previous rounds. He was also trained for only seven days.

TLO said in a comprehensive DeepMind article on the demonstration: "AlphaStar adopts well-known strategies and reverses them. The agent has demonstrated strategies I had not thought of before, which means there can always be new ways to play the game that we have not explored yet. Then MaNa said, "I realized that my gameplay depends on the compulsion of mistakes. and being able to exploit human reactions, so it put the game in a whole new light for me. We are all impatient to see what will follow. "

DeepMind forms AlphaStar since the end of 2016, in partnership with Blizzard Entertainment, the studio behind the StarCraft franchise. The main challenges in artificial intelligence are to be able to track several friendly and enemy units, plan moves well in advance, take into account defensive resources and tactics and adapt to the game's style of play. the opponent on a large map in real time. The company is committed because it represents a huge challenge, following the success of its AlphaGo AI, which has won decisive wins in the board game Go, because StarCraft II is considered one of the most important strategy games in the world. real time the most difficult.

AlphaStar has more than 200 years of cumulative gaming experience to its credit. He was trained to watch matches in real time and then play against the game's own robots. Later in its development, several versions of the AlphaGo robot were designed to compete in order to understand their experiences.

The DeepMind Artificial Intelligence Systems Research Company was purchased by Google in 2014 and is now a separate company under the Alphabet umbrella. used several tensor processing units from Google to train AlphaStar, but once that was done, he was able to run on a single GPU to play his matches. According to DeepMind researchers, the AI ​is a little slower than human players in terms of speed per minute, but its accuracy is much higher and its reaction time is much better.

Hey, Guy hope you like this article, please share it with your friends and family. Also, comment on your points of view and thank you for reading this article.