Artificial Intelligence (AI) seems to be conquering new grounds as it’s begun to beat professionals at popular science-fiction based strategy game, StarCraft. Prior to this, the artificial intelligence helmed by Google’s DeepMind had aced the game Go.
As reported by DeepMind on Thursday, two professional gamers of StarCraft II lost all 10 matches they played against an AI program named AlphaStar. This event is a proof of computers improving at solving complex problems – what London-based AI company, DeepMind, has been focusing on for quite some time.
Previously, the company had created the AI named AlphaGo, which defeated professional gamers of the age-old game Go. Since Go involves players who alternate placing white and black stones on a 19 X 19 grid, it may be played with approximately close to an uncountable number of moves. As a result, computer scientists had long struggled to design an AI capable of dealing with such complexity.
Similarly, AI StarCraft II’s extremely tricky for computers to conquer, owing to its complicatedness and strategy-dependent structure. In this particular game, gamers can assume the roles of different galactic units, namely Zerg, Protoss or Terran, and combat to gain control of the galaxy. As outlined by DeepMind, the major challenges faced by its AI during playing StarCraft II were the players being unable to witness everything happening simultaneously and the feature of continuous gameplay.
The training program for AlphaStar consists of the first three days spent on the neural network being trained on replay of games of StarCraft II played by human players. From thereon, this network helped create computer-based competition. These competitors played the game against each other over the span of 2 weeks and learnt from their experiences. Out of the best AI, 5 were used by DeepMind to play every game against Wünsch and Komincz.