AlphaGo vs Lee Sedol

Matches and what it means

From 9th to 15th March the world’s most anticipated game of Go had begun. A series of 5 matches between 18 times world champion Lee Sedol and a computer called AlphaGo. The world’s very first Man Vs Machine Go battle. AlphaGo had defeated Lee Sedol 4 – 1. But everyone complimented Lee Sedol’s skill and were stunned by AlphaGo’s capabilities.

Go is a very popular board game originated and played in China and now promoted all over the world. Unlike chess, Go has very simple rules. Pick up a stone and place it on a 19 x 19 board. The aim is to capture as much territory as you can and to capture as many of your opponent’s stones as possible. A game of chess can last for 10 to 60 minutes depending on the player’s skill, but a game of Go can last for 45 minutes to an hour and when two professionals play the game can last up to 6 hours. Go may have the simplest rules but is a very complex game and requires deep thinking. The game had amazing outcomes with beautiful moves played both by AlphaGo and Lee Sedol.

AlphaGo was built by Google’s Deepmind, a company that builds artificial intelligence programs. AlphaGo Is designed to play Go on a 19 x 19 board and designed to be the very first computer program to have defeated the best players at Go. AlphaGo uses neural networks (inspired by biological neural networks) that allow the computer to think Go players. To really understand how AlphaGo works we need to understand how a game tree works. A game tree is a flow chart that shows the possible game states for a line of moves. So an AI can run through its game tree and look for a scenario that fits the situation and then plays the move that would give it a higher chance of winning. So in a game of Tic-Tac-Toe that has 3 ^ 9 (3 to the power of 9) moves, a game tree is really useful. But in a game of Go or even chess that has over 10 ^120 possible combinations using a game tree is not really efficient. So what AlphaGo does is that it plays with itself. It then uses the game tree to look for a possible move that will bring success. It will then use what it learned in a match.

Before the match had begun, Lee Sedol had claimed that he will beat AlphaGo 5 – 0.

Game 1 - Throughout much of the game Lee Sedol seemed to be in control of the game. However, his 7th move was commented by many professionals as unusual. Lee claims that this move was to test AlphaGo’s strength and to determine how AlphaGo plays. AlphaGo had taken the advantage which resulted Lee Sedol resigning (basically saying that he lost the game). Lee Sedol was stunned by AlphaGo’s strong play.

Game 2 – In game 2 Lee Sedol had commented that AlphaGo played a near perfect game. Michael Redmond noted that AlphaGo’s moves were “creative” and “unique” to which Lee Sedol took an unusually long time to make a move. Some of AlphaGo’s moves were regarded as mistakes by some professionals including Lee Sedol but later in the game those moves that were labelled as mistakes actually helped AlphaGo win. Thore Graepel (one of the researchers that worked on AlphaGo) explained “AlphaGo does not attempt to maximize its points or its margin of victory, but tries to maximize its probability of winning.” So far AlphaGo is winning 2 – 0.

Game 3 – Analysts were starting to doubt whether or not AlphaGo was truly a strong player. But the third game cleared that doubt as AlphaGo had pulled some unexpected moves that helped it regain control over the game. We saw Lee Sedol struggling and then saw him resigning.

Game 4 – This game was a turnaround for Lee Sedol. With him taking full control of the game and winning. There were some mistakes made by AlphaGo that caused Lee Sedol to win. Analysts say that this was an error in the program and it happens.

Game 5 – Lee Sedol followed a similar strategy in game 4 – trying to take territory in the corners. AlphaGo made a few mistakes throughout the game as it was unable to compute some of the moves by Lee Sedol. But in the end AlphaGo gained control and led its way to victory.

The Future

Everyone praised the way AlphaGo played. Developers will start to work on the algorithm to build better AI that works more efficiently and ones that work in different areas. As games such as Go and chess are taken over by AI, it is said that there won’t be any AI advancements in the near future. But we will start to see advancements in different areas.

As for the Go community, millions and millions of people around the world have started to analyse every single game and find out other possible moves. Professionals all around the world are starting to question and understand the moves made by AlphaGo in order to be better players.

Prateek Pillai - Year 10