Deep learning algorithms have already beaten humans at games like Starcraft—now they have shown they can team up to beat us too.
The news: In a paper published in Science yesterday, DeepMind showed how it had let AI programs loose in a modified version of the 3D first-person video game Quake III Arena. The team used a reinforcement learning algorithm to train the agents on around 450,000 games of Capture The Flag. The AI agents were mixed up in teams with 40 human players - both as opponents and as teammates.
How to win: The teams of AI agents were consistently better than the other pairs, and developed strategies including: following teammates to outnumber opponents at key moments or waiting near the enemy base to grab a new flag when it appeared. There's a video of the agents in action here.
There’s no (A)I in team: The work is cool because it’s hard to get AI to cooperate. Something similar could help robots operate in the real world more effectively (with each other and with humans.) However, we must be careful not to extrapolate too much. It's likely it couldn't simply transfer to another scenario—never mind real life. Read more here and our explainer on machine learning here.
No comments:
Post a Comment