Grandmaster: Deepmind AI Can Now Take on Starcraft II's Top 0.2%

DeepMind chose Starcraft II because it "stretches the limits of human intelligence."
Chris Young

In another bid to get Skynet ready by testing it on battle strategy videogames, the folks over at Deepmind have announced that their Artificial Intelligence has now risen the ranks to the level of Grandmaster in Starcraft II.

To the uninitiated, that means it can beat 99.8 percent of human players at a game that is well known for its strategic complexity.

RELATED: AN AI DEVELOPED BY FACEBOOK AND CARNEGIE MELLON BEATS CHAMPIONS OF POKER

Beating humans at their own game

Deepmind's Starcraft II AI has already shown its abilities against human players on several occasions. The company's latest update, however, shows that it is now capable of taking on the best of the best.

The team says that its AlphaStar AI can play one-on-one matches in the complex real-time strategy game at a Grandmaster level. What's more, it is in the top 0.2 percentile of players on Battle.net, the researchers say.

This isn't just smoke and mirrors either. The team at Deepmind made sure their AI faced the same restrictions human players do — limited map info and virtual camera views.

The AI even had its number of actions per minute restricted in order to mimic that of a human player.

Real-world applications

While this is an impressive display, the AI does have its limits. For one, it needs more training than a human to reach the same level of skill. However, it does show great promise for the future.

Most Popular

"One of the key things we're really excited about is that Starcraft raises a lot of challenges that you actually see in real-world problems," Dave Silver, who leads the lab's reinforcement learning research group, told the BBC.

"We see Starcraft as a benchmark domain to understand the science of AI, and advance in our quest to build better AI systems."

Grandmaster: Deepmind AI Can Now Take on Starcraft II's Top 0.2%
Dave Silver (middle). Source: Deepmind

As with Deepmind's other research into games, which has seen it successfully tackle chess, Go, and Quake III, the ultimate goal is to use its findings for real-world applications.

An expert AI, for example, could help self-driving cars to drive more safely or make sure machines operate efficiently.

For these kinds of operations, we would want an AI that is able to surpass human beings in terms of intelligence and strategic thinking — as scary as that might sound.

message circleSHOW COMMENT (1)chevron