'StarCraft II" (Photo : YouTube)
Google's Deepmind has already defeated a Go world champion using its AlphaGo program to win a series of the ancient Chinese board game. It was announced at the gaming convention Blizzcon that the British AI company is now teaming up with Blizzard to build its own AI agent to play "Starcraft II." The game developer wants to give the public the chance to build an AI bot using the new application program interface (API).
DeepMind created a complex Go-playing system. it is built on neural networks that function like the human brain, and machine learning.
Blizzard's "Starcraft II" is, meanwhile, one of the most competitive games played in the world. It involves skills such as resource management, battle strategy, and scouting.
A big difference compared with Go is that players cannot see the whole field so they have less information when planning out their moves. Players must use memory and long-term planning to succeed in the game.
The Google company's new API would improve AI in "Starcraft II." It could also be used as an indomitable AI bot that nobody could beat, or better AI player coaches, according to Engadget..
The new environment will be launched during Q1 2017. DeepMind will reduce the game's graphics into basic visuals, while the API will allow for bots to read from game data, according to The Verge. AIs will also learn from a dataset that includes replays from old "StarCraft II" games.
Oriol Vinyals is a research scientist at Google DeepMind and the former top-ranked "StarCraft" player in Spain. He shared it is possible an agent could play the BlizzCon champion in a man vs. machine match.
This is a big challenge for DeepMind. It has taught AI to play old-school Atari games, but it will be tougher to develop AI for playing "StarCraft" due to the complex graphics.
DeepMind's AI research could go beyond playing "Starcraft II." Vinyals shared in a blog post the new technology used to play the video game could also be used in real world applications in sectors such as science and energy.
Google DeepMind explained that the complexity of "Starcraft II" is a bridge between the gaming world and real world. Games have already been used as AI trainers. However, it added that AI tech is still far from challenging a human player in "Star Craft II."
Here's DeepMind AI learning "Breakout":
