DeepMind's AlphaStar AI Achieves GrandMaster Status In 'StarCraft II'

Movies like "I, Robot," "Transcendence" and "The Terminator" have proven, time and again, the capacity of Artificial Intelligence (AI), to overpower human beings and take over the world. It's fictitious, yes, but it's still a scary thought if robots rule over humanity.

But all of these are made up by Hollywood. DeepMind's "StarCraft II" AI, however, is be a prime example of what they could be provided the good moral intentions of its human counterpart. Back in July, Alphabet and Blizzard tested AlphaStar against human players in Europe in a series of “blind trial matches.”

The intention to pit AlphaStar against some of the best "StarCraft II" players was part of DeepMind's “ongoing scientific research into artificial intelligence” and it recently proved that all of their sacrifices didn't go to waste.

According to DeepMind, AlphaStar recently became the first AI to reach the league's top honors and has proved that it can go head-to-head with "StarCraft II's" best and brightest names. But what made it stand out is the fact that the AI ranked above 98.8% of all active players on Battle.net.

What this means is that AlphaStar can play one-on-one matches at Grandmaster level for Terran, Protoss or Zerg. It's a feat that normal players would spend a lifetime achieving, but the AI did it with “real-world restictions” such as limited map info, camera views and the like.

Kotaku pointed that AlphaStar was first instructed by DeepMind to study human players on how to play "StarCraft II." It was then “forced to play against itself” as one of the three factions so as to calculate crucial information, including the race's weaknesses.

"StarCraft II" pro Dario Wunsch described AlphaStar's gameplay to be “incredibly impressive” and very skilled in ways that the AI knew how to properly execute strategies.

“While AlphaStar has excellent and precise control, it doesn't feel superhuman,” he added.

Engadget meanwhile pointed to a perennial fact that AI technology, no matter how smart they could get, still has its limits. As an example, the outlet said that it needs “much more training than a human to match a comparable level of skill.” The same can be said when DeepMind acknowledged the fact that self-play may lead to forgetfulness.

“An agent playing against itself may keep improving, but it also may forget how to win against a previous version of itself.”

Click here to see AlphaStar battle Team Liquid's TLO and Mana.