DeepMind, the Google Inc.-owned machine learning firm behind AlphaGo, has announced that it is open-sourcing DeepMind Lab, a gamelike platform aimed at training artificial intelligences to navigate and interact with 3D environments.

According to DeepMind, the training framework’s 3D environment is better for developing AI agents because it more closely resembles the real world, which is more conducive to learning than more abstract AI frameworks.

“As well as facilitating agent evaluation, there are compelling reasons to think that it may be fundamentally easier to develop intelligence in a 3D world, observed from a first-person viewpoint, like DeepMind Lab,” the DeepMind team explained in a blog post. “After all, the only known examples of general-purpose intelligence in the natural world arose from a combination of evolution, development, and learning, grounded in physics and the sensory apparatus of animals.”

The team added that it’s possible that “a large fraction of animal and human intelligence is a direct consequence of the richness of our environment, and unlikely to arise without it. Consider the alternative: if you or I had grown up in a world that looked like Space Invaders or Pac-Man, it doesn’t seem likely we would have achieved much general intelligence!”

DeepMind has turned to a number of games in its AI research, from board games like Go to video games like Starcraft. In fact, the DeepMind Lab training platform itself is built on top of Quake III Arena, a first-person shooter released in 1999 by Id Software, the same studio behind Doom.

The training platform allows AI agents to perform a number of tasks that take advantage of the 3D environment, such as navigating mazes, collecting items, learning and remembering randomly generated environments and more. Apparently, DeepMind Lab even allows AI agents to practice their shooting skills with laser tag, because AI researchers love teaching robots how to kill things.

DeepMind Lab is not the only game-based AI training platform out there. Earlier this year, Microsoft Corp. open-sourced Project Malmo, an AI testing platform based on the popular block-building game, Minecraft. Both Minecraft and Quake III Arena are first-person 3D game environments, but each has something different to offer AI researchers.

For example, Minecraft has a well developed crafting system, allowing AI agents to practice building complicated structures and even logic systems. Meanwhile, Quake III Arena offers faster, more precise movement with significantly more speed and verticality, which could offer a better testing ground for highly mobile AI agents.

You can find out more about DeepMind Lab in the team’s in-depth research paper, and you can watch a video of DeepMind Lab in action here.

Image courtesy of Google