Google's Deep Mind AI System Learns How to Play Atari, Inevitably Enslave Human Race

Google's Deep Mind AI System Learns How to Play Atari, Inevitably Enslave Human Race

Photo Removed
Complex Original

Blank pixel used during image takedowns

Photo Removed

Google is hell bent on ushering in the end of man.

Google has hooked up an artificial intelligence system and is teaching it how to beat real, red-blooded, opposable thumbed humans, at old Atari 2600 games. As part of the Google Deep Mind project, researchers have tasked an AI system with playing 30 year old video games on an Atari 2600 to see if the system was capable of defeating human scores. Turns out it could. Often considered some of the most difficult video games ever made, thanks largely to a severe lack of quality control in the nascent video game industry, old Atari games border on the impossible. According to NBC News, the AI system beat its human opponents score 23 out of 49 times. Space Invader, Pong, and Krull were all hard enough to begin with, but are all but unplayable from a contemporary standing.

The Deep Q-network agent, or DQN as the team at DeepMind call it, doesn’t learn the rules of a game the same way a human does–obviously–but instead simply finds the best way to maximize its score each round. The system then develops unique strategies for every engagement; watching a game and observing which actions will increase its score. It's terrifying if you think about it and we're more than justified in our fears of an inevitable robot monarchy.

Latest in Pop Culture