I was going to write this article Monday as a precursor for the game to come but decided instead to wait until the game was completed before reporting on it. The game I’m speaking about is a game with a simple name and simple rules. A game that, while seemingly all to simple, is a game that can be impossibly complex. The name of this game is simply “Go”. Many people in the west are not familiar with this game but in the East Asia it is enjoyed by many. While Go has very simple gameplay, two players take turns placing tiles on a board, trying to gain territory by arranging those tiles in strategic shapes or patterns, the trillions of possible moves and almost limitless board positions have many Go players and enthusiasts claiming it is the most complex game ever to be created. CEO of Google’s Deepmind project, a 5 game series marked as the ultimate face off between man and machine, is even quoted as saying “There are more possible Go positions than there are atoms in the universe”.
This afternoon, in Seoul South Korea, Deepmind’s Alphago program took on the world champion Go player Lee Sedol in the first of 5 games to be played to see if an AI program can truly beat a human in the most complex game in history. Alphago’s creation was something of a marvel as well. Rather than simply re-programming the system when it made mistakes, Deepmind built “reinforcement learning” into Alphago’s program. This means that Alphago is able to realize it’s mistakes and make the necessary corrections. Alphago is also capable of looking into the future to predict the best possible moves, and the moves of it’s competitor, to find the best way to gain victory. In essence, Alphago, an artificial intelligence, is learning.
But an AI certainly can’t beat a human being in a game that is largely based on intuition, can it? There is some precedent coming into today that it, in fact, can. Back in October of 2015, Alphago took on the European champion and won. While many people believed a success in AI like this was years, if not decades, away, the win helped to spring board a head to head match up with the best of the best, Lee Sedol. Because of this win, and in fact perhaps even without it, Alphago seems to be a clear favorite. If Lee Sedol is able to beat the Alphago system, he will win $1 million. During the press conferences leading up to his match up, Lee exuded the confidence of a world champion, however that confidence seemed to falter a bit just before the first move of the game was made. Lee’s eyes could be seen darting from side to side and he took many sips of water before getting the game underway. Here in lies a huge advantage for AI…..it doesn’t get nervous.
Sitting across from Lee was Deepmind developer Aja Huang who would make the moves that Alphago chose. While the game was very close, and both sides made their share of mistakes, Lee eventually conceded that Alphago was simply too dominant in this game. Lee had lost and Alphago took a 1-0 lead in the 5 game series. Lee’s confidence had faded away in the post game press conference where he, along with many other Go aficionados, were stunned at the outcome. Lee was quoted as saying “I didn’t know Alphago would play such a perfect game”. Looking dejected and on the verge of tears, Lee will have another chance to beat Alphago tomorrow, March 10th. It will be streamed on Youtube for any person who is curious about the result, though a warning, Go is not the typical spectator sport many are accustomed to watching. There are often long pauses and movements are on the inches scale, rather than feet or yards. Regardless of its spectator appeal, this match up between AI and human is of the most extreme importance and could truly serve as a milestone in artificial intelligence creation and development.
If you would like to read articles on the match or about Google’s Deepmind program you can find information here: http://www.theguardian.com/technology/2016/mar/09/google-deepmind-alphago-ai-defeats-human-lee-sedol-first-game-go-contest