Google's DeepMind Made AI Play ‘Quake III’ to Learn Teamwork; It Won Against Humans

Games have always been a favorite of companies working on artificial intelligence for developing or refining their algorithms. In the case of Google’s AI division, DeepMind, researchers taught machines to work together by playing the classic arena shooter “Quake III.” A tournament was even held to pit AI against humans, which the machines won.

In a blog post, DeepMind shared the results of its experiment where AI bots were trained in “Quake III Capture the Flag” mode. The machines learned via trial and error, with the researchers making them run through maps in the game that would change regularly. By doing so, the AI was taught to cope with changing situations.

Apparently, the results were promising enough for DeepMind to decide to hold a tournament to test the capabilities of the AI. There were three types of two-player teams involved. One was made up of AIs, another was made up of humans, and the third was a combination of the two.

After numerous matches, the success rate of the AI-only team was at 74 percent. In comparison, the team composed only of humans scored a success rate of 43 percent.

As The Verge pointed out, however, the experiment also indicated that the AIs performed worse if there were more than two machines working together as a team. As such, the DeepMind AI will require a lot more sessions playing “Quake III” to get even better.

Then again, at the rate that AI advancement is going, it’s likely that machines will surpass even the best video game players in only a matter of years. It’s simply fortunate that the goal of tech companies isn’t to take over the eSports scene. If they did, viewers might only have teams of AI bots to root for when watching tournaments instead of actual human competitors. The DeepMind “Quake III” AI experiment can serve as a wakeup call.