This site is no longer being updated all posts are still accessible in this archive.

Sign in with your favourite social login to share, comment and pin your favourites.

Future Tech: 

This AI will beat your Space Invaders high score

Are you sure you want to play, Player 2?

Are you sure you want to play, Player 2? I will destroy you.

Anyone who has ever played Mario Hoops 3 on 3 on the Nintendo DS won’t be surprised to hear that computers can get unrealistically good at games. That’s right: we’re still bitter about the cheating AI in an otherwise perfectly good sports title, nine years later.

But while artificial intelligence being good at games is nothing new – IBM’s Deep Blue beat world champion chess-master Garry Kasparov in 1997 – this development from the Google-owned DeepMind Technologies is a bit different. The AI has LEARNED to play 49 classic Atari games with only the most basic instructions.

DeepMind’s vice president of engineering, Dr Demis Hassabis explained to the BBC, “The only information we gave the system was the raw pixels on the screen and the idea that it had to get a high score. And everything else it had to figure out by itself.”

The system learned how to play 49 arcade classics, including Space Invaders, Pong, Pac-Man, Pinball and Breakout. Terrifyingly, in 29 of these it was of a similar standard or better than human player. It aced Video Pinball, Boxing and Breakout, smashing the human’s puny reflexes, but it found Pac-Man, Private Eye and Montezuma’s Revenge tricky. Score three for humanity!

“On the face it, it looks trivial in the sense that these are games from the 80s and you can write solutions to these games quite easily. What is not trivial is to have one single system that can learn from the pixels, as perceptual inputs, what to do,” Hassabis explained.

“The same system can play 49 different games from the box without any pre-programming. You literally give it a new game, a new screen and it figures out after a few hours of game play what to do,” he added, making us wonder how many quarters these AIs would need to learn their trade the old fashioned way: in the arcade, where it counts.

The video below shows exactly how the system learns, in this case with Breakout. It’s stunning how quickly it goes from missing the ball every time, to learning the old ‘bounce it off the backboard’ trick.

So what use is this development outside the 1970s? Quite a lot, as it turns out. Learning from the environment could be invaluable to aspects of life where dealing with the unexpected usually puts things beyond robotic comprehension, from driverless cars to factory work.

“One of the things holding back robotics today, in factories, in things like elderly care robots and in household-cleaning robots, is that when these machines are in the real world, they have to deal with the unexpected,” explains Hassabis.

“In some sense, these machines need intelligence that is adaptable and they have to be able to learn for themselves.”

Insert your own warnings about the robots enforcing humans into slavery here. Especially if we’re stupid enough to put a high-score bounty on our heads.

Related Articles

[ccode]
Yes, send me the latest
ESET news

Want to receive the best stories from Go Explore on a weekly basis? Enter your email address here to subscribe

Seen something great online?
Seen something great online?