Blizzard and Google begin collaboration on using DeepMind to create an AI Starcraft II bot that will eventually be able to beat the best Starcraft players in the world using machine learning, as it has with so many games including most recently with beating the top Go player in the world.
A big wrench in the system includes the fog of war, not allowing the computer to know what the other player is doing without scouting.
Using inputs like the map, minimap, and others Deepmind will initially learn from watching years of top tier gameplay footage from Starcraft tournaments then move on to live human opponents.
In June I predicted the following “When will AI be able to learn how to play Esports games and school us all? In 5 Years. In the last few months Google DeepMind beats the top Go boardgame player in the world and learns curiosity and reward to complete the video game Montezuma’s Revenge. There are more variables in Esports titles than most of the simple games being played right now by DeepMind. Hearthstone could be the lowest hanging fruit and could be mastered in a year or two if focused on today given you plug in all the cards abilities and run the scenarios. A MOBA game like League of Legends will take much longer, there are so many variables that in order to run enough scenarios to effectively learn how to play it we would need to figure out a different approach or wait until quantum computing comes to fruition and badabing it would take a second to master not only any MOBA Esport but pretty much anything in the world.”
So 5 months not 5 years later DeepMind is already attacking its first Esports game. Because of that I will now adjust my estimate that DeepMind will defeat top players in MOBA games like League of Legends from June 2021 to March 2019.