Campus Review Volume 26. Issue 3 | Seite 26

FACULTY FOCUS campusreview.com.au Rise of the machines An AI recently humbled the best Go player humanity had to offer; are the robots set to take over? We ask some experts about the fate of Homo sapiens. By Patrick Avenell T he first thing you need to understand is that Go is a much more complicated game than chess. More than 2500 years old and borne out of classical Chinese antiquity, Go quickly spread through Japan and Korea, and then other parts of East Asia. There are now professional Go players competing for purses worth hundreds of thousands of dollars, with each player credited with a ranking, called a Dan, between 1 and 9, not too dissimilar to belts in martial arts. Go is a non-chance abstract strategy game played on a 19x19 board. Players take turns placing black and white stones, attempting to surround their opponent and capture territory. There is no fixed end point of the game: only when both players elect not to keep placing stones are points tallied and a winner declared. There are 10761 different possible game variations, more than six 24 times the myriad ways a chess match can play out. When you play Go, there is no luck and all the information about the game is in front of you – tiles aren’t hidden like in Scrabble (another popular game of skill) and there are no dice as in Risk (another intellectual strategy game) – essentially, if you play the best game, you will always win. Go has only two rules, broadly speaking. Nevertheless, its enhanced complexity has meant computer programmers and software engineers have failed to write artificial intelligence (AI) programs to truly match wits with the best human players at the game, at least not in the same way IBM’s Deep Blue so fundamentally overcame Garry Kasparov across a series of contests in the late 1990s. Until the start of 2016, no AI had ever beaten a 9 Dan professional Go player with handicaps on a full-sized 19x19 board. It was too difficult to program in all the different combinations and machinations. To understand the sheer scope of the project, imagine a ball hits a branch right in the middle and will fall to either side; now imagine that happening billions of billions of times and having to map exactly where the ball will eventually end up. Leading experts recently thought an AI capable of vanquishing a champion human player was at least a decade away. All this changed, however, in March, when AlphaGo, an AI project