AI has mastered the complex game Go, and if you're willing to teach the machines to beat you at Minecraft, too, Microsoft wants your help. Researchers from Project AIX want to use the open-world game to improve its artificial intelligence systems. Unlike Go, which is very rule-specific, Minecraft requires what researchers call "general intelligence," a formidable challenge for deep learning systems. "Minecraft is the perfect platform for this kind of research because it's this very open world," says Katja Hofmann from Microsoft's Cambridge labs.
To help its AI systems master the block-building game, the team made its AIX platform available to researchers in a small, private beta. For now, the experiments run on users' local machines, sectioned off from general users. "People build amazing structures that do amazing things in Minecraft and this allows experimenters to put in tasks that will stretch AI technology beyond its current capacity," Hoffman told the BBC.
Microsoft's AIX research team playing Minecraft
However, Microsoft aims to make the AIX platform, including the Minecraft code, available to anyone via an open-source license. It's not clear what mainstream users will be able to do with it, other than perhaps let it observe their moves. However, it could one day become a digital assistant that helps you out. "Eventually, we will be able to scale this up further to include tasks that allow AI agents to learn to collaborate with humans and support them in a creative manner."
Another attraction to Minecraft is the endless situations it simulates from a first-person perspective. "It can actually be inside, looking out through the eyes of something that is living in that world," says chief engineer Matthew Johnson. It's also ideal for reinforcement learning, where machines combine experimentation with previous knowledge. All told, Microsoft hopes to use its popular game to accelerate AI at a pace that may make Elon Musk and Stephen Hawking uncomfortable. "This provides a way to take AI from where it is today up to human-level intelligence, which is where we want to be, in several decades time."