Advancing artificial intelligence ... with Minecraft

A group of computer scientists have found an unorthodox but potentially groundbreaking new platform to study artificial intelligence: the wildly popular video game Minecraft.

Microsoft bought the maker of the popular video game, in which players use characters to explore and build projects in what amounts to a boundless virtual sandbox, for a reported $2.5 billion in 2014, according to PC Mag, and researchers for the tech giant are now using the game to try to solve one of AI’s toughest challenges: what researchers call "general intelligence," the complex way humans learn, principally through trial-and-error, and make decisions.

AI research has reached the point that computers can be programmed to carry out specific, complicated tasks, like understanding and translating speech, very well – even better than an average adult. Getting computers to learn a new skill from scratch, however, has proved more difficult.

Researchers working on Microsoft's Project AIX in New York, for example, have been working to teach an AI agent in Minecraft to learn from scratch how to climb to the top of a hill, navigating hurdles like rivers and lava pits and learning from its failures.

"We're trying to program it to learn, as opposed to programming it to accomplish specific tasks," said Fernando Diaz, a senior researcher on the project, in a Microsoft blog post.

AI researchers, the blog post adds, "are able to take tiny slices of that total awareness and build tools that do one thing, like recognize words, but they haven’t been able to combine them in the way the humans do effortlessly."

Robert Schapire, a principal researcher in Microsoft's New York lab, added in the blog post: "The things that seem really easy for [humans] are actually the things that are really difficult for an artificial intelligence."

Besides the five-person Microsoft research team, the company has also made the platform available, via a private beta, to a small group of academic researchers. In Minecraft, AI researchers have a cheap and simple, but endlessly vast and complex, platform on which to experiment. 

Jose Hernandez-Orallo, a professor at the Technical University of Valencia in Spain and one of the academics given early access to the software, told the BBC that at the moment "there is nothing comparable."

"This is state-of-the-art," he added. "And this is just in its beginnings, so I see many possibilities for it."

In the summer, the software will become available to the public via an open-source license, then everyday gamers will be able to participate in the research, with AI agents able to work with and, yes, learn from human-controlled agents.

"We're looking for opportunities where we can really help accelerate the pace of artificial intelligence innovation in a way that is going to be very close to the real world, with real experiences and real data," said Evelyne Viegas, director of artificial intelligence outreach at Microsoft Research, according to the company’s blog post.