Google DeepMind AI Can Learn How To Play 49 Classic Atari Games

Google DeepMind AI can learn how to Play 49 Classic Atari Games

Photo Credit: Getty Images

  • Space Invaders and Pong, to boxing and tennis games and the 3D-racing challenge Enduro.

Artificial Intelligence is possibly making a comeback. Google scientists published a new paper that says that their AI project Deep Mind learned to play Atari video games.

Google has a machine learning project called DeepMind. Google researches have published today a paper in the Nature Journal that says that they managed to develop a software that is able to learn to play classic Atari Video Games. 

The software has learned 49 video games and is better at playing them in more than half the games than human players. Dr. Demis Hassabis, DeepMind’s VP Engineering, said: “Up until now, self-learning systems have only been used for relatively simple problems. For the first time, we have used it in a perceptually rich environment to complete tasks that are very challenging to humans.”

The DeepMind software is able to learn to play video games starting with just very basic instructions and prerequisite. The software knows the goal of getting a high score and how the game graphics.

The games the DeepMind AI is able to play include Space Invaders and Pong. The software is best at playing Video Pinball, Boxing and Breakout. It apparently sucks at Pac-Man, Private Eye and Montezuma’s Revenge.

The BBC published a video that shows how Dr. Demis Hassabis exposes the AI to Breakout. Watch below.

For years it looked like Artificial Intelligence has stalled. Is it time again to get afraid of AI to wipe out humanity? I remember researching AI when I studied Computer Science in the early nineties. The topic was red hot, but AI research got stuck around 2000. The complex solutions did not deliver any usable results. A big thing in the nineties have been neural networks based approaches but they turned out to be too limited. The new DeepMind research results are grounds for new hope that there is a way to make usable artificial intelligence. If robots and software are getting better at learning themselves how things work, we might not know the outcome of this. Computers could find way better solutions that humans never were able to come up with. This can be good and bad.

The paper has been published on Nature.

Google bought DeepMind back in early 2014 for $400 million.


Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *