Meet the new StarCraft II Grandmaster: DeepMind’s AlphaStar

Artificial Intelligence company DeepMind has announced that its AlphaStar AI has reached the highest level of play in esport StarCraft II.

The latest in a series of milestones achieved by the company, the news comes after DeepMind’s AlphaGo won a best-of-five against Go world champion Lee Sedol, before AlphaStar beat a professional StarCraft II player earlier this year. This latest development has seen AlphaStar playing against versions of itself to achieve Grandmaster ranking, denoting the top competitive league of StarCraft II when played as an esport. The AI is now ranked above 99.8% of other players.

The result is remarkable owing to the complexity of a game such as StarCraft II when compared with Go, and was accordingly published in the scientific paper Nature. For one, the game is played in real time rather than in turns, with there being no steady state of play to consider and then act upon. There’s also a vast array of actions possible at any time - up to 1026 according to DeepMind.

SEE ALSO:

• Google's AI DeepMind might be able to spot acute kidney injury 48 hours before doctors spot it

• Google’s DeepMind unveils self-teaching AI software, AlphaGo Zero

• Royal Free hospital breached data protection in deal with Google's DeepMind

• Read the latest issue of Gigabit here

The AI was trained via an array of machine learning techniques including neural networks and reinforcement and imitation learning. Breaking down the jargon, AlphaStar essentially played versions of itself in different configurations, some intentionally weakened and known as “Exploiters”, learning each time.

AlphaStar does not win owing simply to an inhuman speed of operations; indeed, the AI is limited to a human-comparable frequency of actions. Neither is it able to see more of the game at any moment, restricted by a moveable camera viewpoint just as a human player is. Instead, AlphaStar competes on a level playing field.

Dario “TLO” Wünsch, a professional player of the game, described AlphaStar’s play as follows: “I’ve found AlphaStar’s gameplay incredibly impressive – the system is very skilled at assessing its strategic position, and knows exactly when to engage or disengage with its opponent. And while AlphaStar has excellent and precise control, it doesn’t feel superhuman – certainly not on a level that a human couldn’t theoretically achieve. Overall, it feels very fair – like it is playing a ‘real’ game of StarCraft.”

(Image: Blizzard)