Last week, to great fanfare in the world of Artificial Intelligence, it was announced that the game of checkers (better known as draughts in the UK) had been solved by the University of Alberta’s Chinook project—the most complex game at which a computer has ever become unbeatable. As some commentators have observed, even God could only draw a game against Chinook, something that may seem slightly alarming, or even blasphemous, to amateur theologians.
Today is another interesting day for AI, as it sees the most sophisticated poker-playing programme yet developed taking on two of the world’s finest human players in an eagerly-anticipated face-off. The humans, here, are clear favourites—machines have yet to demonstrate skills at the “soft” science of bluffing that can match our own. But for how much longer will we be able to win any games against the machines?
The ancient game of Go is a widely-cited example of the current limits of AI. Computers may be able to play chess at a higher level than any human in history, but they’re barely as good as a decent amateur at Go. This is partly due to the size of the playing-board (a massive 19 by 19 squares) and the open-endedness of game strategy (there are a possible 361 opening moves, and up to 300 options on every subsequent turn), but it’s also a more intangible business, related to the difficulty of articulating Go strategy as anything more precise than “play lots of Go.” This will certainly change, but it’s an important reminder of the ways in which AI remains limited: its problems need to be precisely quantified by humans.