"Minecraft" has gripped the gaming world since its inception in 2011, and it looks like it may have caught the world of artificial intelligence in its grasp as well. Now, Microsoft is planning to open-source a platform based on the popular sandbox game to help conduct AI projects.
As noted by Microsoft, the AIX platform is already in use by Microsoft Research and is already available under a private beta to a few academic researchers. However, come July, the platform will be open-sourced, allowing computer scientists and amateurs to run tests of their own. All they need to do is install AIX - thus allowing them to connect to "Minecraft" and use the AI code to control a character and get feedback about the consequences of its actions.
The reasoning behind using "Minecraft" to test AI is simple for Microsoft: it's more "sophisticated" than existing AI research simulations and its a cheaper alternative to building a robot.
For example, researchers trying to teach a robot to climb a hill would have to replace or repair it every time it fell into river. On the other hand, if the same happens in "Minecraft" all the user has to do is to re-do the test after receiving feedback from the AIX platform about what went wrong in the previous attempt.
This new approach is being heralded as one with great potential due to the platform still being in its relative infancy.
"This is the state-of-the-art," said Prof. Jose Hernandez-Orallo from the Technical University of Valencia. "At this moment there is nothing comparable, and this is just in its beginnings, so I see many possibilities for it."
For the time being, the experiments will run on researchers' own computers and be "ripped off" from normal players. The end goal however, is to allow people to interact with the code itself.
"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," said Katja Hofmann, who leads the project at Microsoft Research's Cambridge lab in the U.K. "But 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.
"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," she said.
This isn't the only example of developers using games to test artificial intelligence in recent weeks. In related news, Google's AlphaGo AI program, DeepMind, beat world-renowned Go player Lee Sedol in three straight games of a five-game match in Seoul.
