Humans had a good run as caretakers of the planet (for better, or, as many would argue, for worse), but our time is slowly drawing to a close: This week, humanity took one step closer to obsolescence as Google announced (with a certain pride) that its DeepMind AI had skunked two pro gamers at Blizzard’s iconic StarCraft II.
The AI “AlphaStar” beat Team Liquid’s Grzegorz “MaNa” Komincz and Dario “TLO” Wünsch 5-0 in two separate five-game series back in December, both Google and Blizzard confirmed, which is a pretty sorry showing for our human representatives. You can watch the demonstration game by clicking the video above – though, be warned, it isn’t pretty.
Using games to test AI (Artificial Intelligence, also known as machine learning) systems, and their performance and abilities, is nothing new. Companies have, for decades, starting with MANIAC in 1956, pit computers against humans in games like chess. The most famous defeat was of world champion Gary Kasparov by IBM’s Deep Blue back in 1997.
AI has been playing computer games for years as well, and becoming increasingly adept at not only understanding the increasingly-complex, skill-based digital gameplay, but also decision-making within the game. AlphaStar, the first AI to defeat a top professional player, has now proven that we are going to need more sophisticated ways to test our AI: These games are simply becoming too easy for them.
But human hubris is a real thing as well, and will probably remain so for the foreseeable future: Top gamer Wünsch reportedly watched footage of his opponent, AlphaStar, playing, and still felt “extremely confident” going into the game.
The AI, in the end, was just too tough to touch.
All your games are belong to us
It would be easy to assume that the AI would have the upper hand due to faster-than-human reaction times, but (surprisingly) this wasn’t the case; in fact, the AI’s clicks and key-presses were actually slower than the humans’.
The wins, according to the Google DeepMind team, were instead secured by “superior macro and micro-strategic decision-making.”
Which means that AlphaStar was simply smarter.
Though, as is also very human, there have been some accusations of cheating – that the AI was operating at a level unattainable by humans in the first place. You can read more about that here: The DeepMind StarCraft AI May Have Been ‘Cheating’ After All.
But, ultimately and regardless, this is a truly significant victory for machine learning. While AI has also dominated in games like Mario, Quake III Arena Capture the Flag, and Dota 2, the difference is (until now, at least, so the difference was) that the StarCraft game, much more complex than those we just mentioned, had proven too big an obstacle for the machines to overcome. This conquering of human adversaries last December shows that is no longer the case.
All hail our new AI overlords.