When Microsoft acquired Mojang in 2014 for $2.5 billion, the idea was for the tech giant to break into mobile gaming with the developer’s massively popular title, Minecraft.
Now, researchers at Microsoft are using a new platform called AIX to train an artificial intelligence agent to learn how to climb a hill in the open-world game.
The goal is to teach what’s called ‘general intelligence’, the ability to learn lessons from one’s environment and make decisions. In a blog, Microsoft describes it thusly:
A computer algorithm may be able to take one task and do it as well or even better than an average adult, but it can’t compete with how an infant is taking in all sorts of inputs – light, smell, touch, sound, discomfort – and learning that if you cry chances are good that Mom will feed you.
To that end, the agent needs to figure out how to complete it task, try a lot of different approaches and encounter problems like falling into rivers and lava pits.
The team is using Minecraft because unlike simpler games that have traditionally been used to train AI, it offers a lot more opportunities and flexibility for agents to learn. It also allows researchers to collaborate on complex experiments within its virtual world.
The AIX platform, created in-house by Katja Hoffman at Microsoft’s lab in Cambridge, allows users to work on AI projects in Minecraft. The company has been beta testing it a small group for some time now and will make it available to the public for free this summer.
It’s the latest in tools shared by tech heavyweights with the developer community to further AI and learning-based computing. Earlier this year, Microsoft open-sourced its toolkit for speech recognition based on human neural networks. Last November, Google released TensorFlow, its machine learning system for assisting in search in photos.
➤ Project AIX: Using Minecraft to build more intelligent technology [Next at Microsoft]