Short Bytes: DeepMind has open sourced their AI agent training platform known as DeepMind Lab. It renders a 3D game-like environment in which the simulated agent has to float and do various tasks which enhance its cognitive abilities.DeepMind, Alphabet’s subsidiary which recently created a lip-reading AI that defeated a pro-human, has open sourced its latest project called DeepMind Lab. In a blog post, the company acknowledged the need of developing more complex training environments for AI agents and working to improve their cognitive skills.
DeepMind Lab is a platform enabling simulated environments which DeepMind says serve as “laboratories for AI research”. Basically, DeepMind Lab is a platform with 3D game-like graphics observed by a simulated agent as the first person. The agent – posed as a floating orb – navigates in the virtual 3D environment using available actions.
The agent has to perform different tasks like moving through tough paths, collecting items, prevent itself from falling off the cliffs, etc. The purpose of this project is to focus on agent’s planning, strategy, navigation memory, first person vision, etc.
DeepMind Lab is a 3d game-like platform to train AI agent.
The platform is scalable with appropriate tools available to add new levels which can be customized as per requirements. DeepMind says they have been using their platform internally and the impact it has made on their thinking about artificial intelligence is worthy of attention.
Training agents using 3D environments is vital to the improvement of their general intelligence. In fact, our rich 3D world has had a major influence on the intelligence of various living organisms including humans. In a nutshell, the thing is, to achieve human-like intelligence the AI has to perceive the things the way they’re i.e. in 3D.
“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!”
If you have something to add, tell us in the comments below.
Read Our Whole Coverage On DeepMind