AlphaStar is the first Artificial Intelligence program by Deepmind to defeat a top professional player. The Google-owned AI lab’s is now capable of defeating 99.8 percent of all human players in competition. Ever since computers cracked Go, chess and poker, StarCraft has emerged as the next grand challenge.
Goal of Research
The goal of this type of research is to demonstrate with enough time and resources AI can best human in any form of challenge. DeepMind trains AI programs against versions of themselves so everytime they turn into better version of itself. DeepMind Uses Reinforcement Learning Algorithm for this.
Why Strategy Game ?
Strategical thinking is considered a unique quality of human mind. For an algorithm to think strategically it has to take decisions based on multiple data points. This implies computer will be closer to decision making just as humans. StarCraft 2 is designed by Blizzard Entertainment in such a way that each level is targeted to challenge human thinking ability.
Logic behind AlphaStar
Inorder to master StarCraft 2 an AI system needs to be trained based on multiple inputs such as long-term planning, realtime inputs, exploration-exploitation balance, large-action space, imperfect information. For this AlphaStar was trained to play against itself. In that way the program became a better version of itself everytime.
Reference
https://www.theverge.com/2019/10/30/20939147/deepmind-google-alphastar-starcraft-2-research-grandmaster-level
https://www.youtube.com/watch?v=UuhECwm31dM https://towardsdatascience.com/the-science-behind-alphastar-714bd7824d4b
Comments
Post a Comment