Google’s DeepMind AI learned how to play Montezuma’s Revenge

Last year, Google’s DeepMind company wanted to see if their AI could learn how to play old Atari games. The AI was able to figure out most of the games, but it had trouble with the more complicated Montezuma’s Revenge. It’s been over a year since, and now the AI has figured it out. The key was curiosity.

DeepMind made the AI “curious” enough to want to win the game. That’s right, the AI can now be curious. This lead to the AI exploring 12 out of the 24 rooms and beating the first room in just four tries. Before it was curious, the AI only explored two rooms. If you’re reading this you’re likely thinking one of two things: 1) that’s awesome! 2) how long until it’s curious about world domination? Just kidding. Kinda.

Check out the video above to see the DeepMind AI play Montezuma’s Revenge. It’ not the most entertaining Let’s Play on the internet, but when you consider it’s a computer figuring things out on the fly, it’s pretty interesting.

[via Engadget]