Google is taking the machine learning of its DeepMind artificial intelligence to a whole new level — by placing it directly in a whole new space.
Having already developed algorithms to learn from 2D video games, DeepMind has now developed a new algorithm, giving it the ability to navigate through 3D mazes, as reported by PopSci.com.
A video called "Asynchronous Methods for Deep Reinforcement Learning: Labyrinth" was uploaded to YouTube by DeepMind on Thursday, showing the navigation of the Labyrinth, which looks like it's straight out of the first-person shooter video game Doom.
Watching this video, you can see the new algorithm hard at work, making directional decisions just as a human gamer would through the maze. What's amazing about DeepMind is its able to determine a dead end before it hits it, re-adjusting to navigate through the 3D Labyrinth.
Google made the development of this algorithm even more fun by awarding points for scooping up apples and channeling through open portals, while maneuvering through the 3D maze.
According to PopSci, DeepMind said that its program found a "reasonable strategy" for navigating through the maze but didn't provide a human max score for the game.
Still, this video is quite impressive for seeing the program in motion.