Have you ever witnessed history? Been right there and realized that the things you see will be written down for centuries to come. People might not know your name, that you were there to see it, but they will definitely know the name of AlphaGo, the A.I. by Google. It has not only defeated the reigning European champion Fan Hui, but today it won the first match against the reigning world champion Lee Sedol.
Once again we’ve reached a milestone where the programmable computer is given a purpose by its owner and it excels in the work given, exceeding the human skill and capabilities. We’ve moved from the cogs and whistles to the silicate microprocessor supercomputers. Add thousands lines of code, tether some hardware together and you’ve got yourself a machine which can think a million things at a time. And so it happened once again that a computer game overtook its creator.
Of course, the idea of a machine replacing the human has existed since the industrial revolution. Even the idea of a computer game being pretty hard to beat was proven back in 1997, when IBM’s Deep Blue defeated Garry Kasparov in chess despite the optimism displayed on the third day.
Chess, of course, is a very simple game compared to go. Unlike go, chess doesn’t really have much to think about. The pawns move forward, the knight can jump over pieces while the other pieces can’t and that’s about it. If you can memorize how the pieces move, you can already predict your opponent’s possible moves and use the possibilites (and the hindrances) to your advantage. One of my typical winning endgames involve exploiting the fact that you can’t capture your own pieces. So when the opponent’s king is surrounded by pieces, all it takes to make one check where the king can’t escape or hide behind a piece and that’s a checkmate right there.
(Picture from senseis.xmp.net)
In go, the simplicity is what makes it so compelling. All you do in the game is place a stone on the intersection. Then your opponent will respond to your move by placing their own stone. You can either skip your turn or place a stone again. If you want to capture your opponent’s stone, you have to cover it from all four sides. Bam, that’s it. One stone per turn, anywhere you want as long as it’s free.
And as you play the game, the empty checkered board becomes a painting of territories and vicious attacks. A patchwork of threats, fights and defences. “I’ll play over here, then he’ll probably play over there. Oh no, he instead chose to attack me from the rear.” That’s what’s happening each and every turn. For a computer, it’s a disaster to just bruteforce every single combination outcome and every position after it, because every move will have greater consequences in the future. But the computer has to think, see forward and finally choose a move that would guarantee its victory. That’s where the challenge arises.
If you’re wondering how many different play positions are possible, then it’s not in millions or billions. It’s in hundreds of decimal points. If a million is 10^6, then go is 10^761 – that’s times more than there are stars in the sky. While chess has “only” 288 billion (2.88*10^14) different positions, it’s apparent that it’s a miracle in silica that AlphaGo even exists. A game which doesn’t only need a sharp eye for detail, but a crazy imagination and a courage to go through the wildest plans. That’s probably where the human falls short: it begins to doubt their own genial moves, while the computer thinks a heartbeat longer and it plays the move anyway. It doesn’t worry about the pride of winning or the humiliation of loss – it plays.
We are the creator of the machine, but what will happen when our monopoly of creative imagination will be shared with computers? When you have 361 different places to put a stone in the start, the first moves become vital to the endgame. And so the machine waits for the opponent’s move, thinks and then answers. It takes a stone and wins the game.
(Picture from Newsweek)
One day it can lift stones, other day it could lift mountains.