AlphaGo is a computer program developed by Google DeepMind in London to play the GO game board. In October 2015, AlphaGo became the first computer Go program to win a flawless human player on the 19 × 19 board. In March 2016, this program won Lee Sedol in three First match of a total of five matches. For the first program that succeeded Go computer beat professional players 9 and without handicaps. Despite losing to Lee Sedol in the fourth match, Lee gave up in the final match so AlphaGo got a score of 4 approval.
The AlphaGo algorithm combines machine learning and tree search techniques. In addition, AlphaGo also opposes intensive training against matches against humans and computers.
Go came from China more than 3,000 years ago. Winning this board game requires several layers of strategic thinking. Two players, using white or black stones, take turns placing their stones on the board. The aim is to surround and capture their opponent’s stones or strategically create territorial space. After all possible moves have been played, both the stone on the board and the blank points are counted. The highest number wins. As simple as the rules look, but Go is very complex. There are 10 amazing powers out of 170 possible configuration boards – more than the number of atoms in the known universe. This makes the game Go several times more complicated than chess.
AlphaGo, a computer program that combines advanced search trees with deep neural networks. This neural network takes the Go board description as input and processes it through a number of different network layers containing millions of connections such as neurons. One neural network, “policy network”, chooses the next step to play. Another neural network, the “value network”, predicts the winner of the game.
AlphaGo was introduced to many amateur gamers to help him develop an understanding of human gaming that made sense. Then they made him play against a different version of himself thousands of times, each time learning from his mistakes. Over time, AlphaGo improves and becomes stronger and better in learning and decision making. This process is known as reinforcement learning. AlphaGo then defeated the Go world champion in a different global arena and arguably became the biggest Go player of all time. At the end of 2017, we introduced AlphaZero, a single system that taught itself from the start how to master the game of chess, shogi, and Go, defeating the world championship program in each case. AlphaZero takes a completely different approach, replacing craft rules with deep neural networks and algorithms that don’t know anything outside the basic rules. His creative response and ability to master these three complex games, shows that one single algorithm can learn how to find new knowledge in various settings, and potentially, all perfect information games. Although it’s still early, AlphaZero’s encouraging results are an important step towards our mission to create a general-purpose learning system that can help us find solutions to some of the most important and complex scientific problems.
The name AlphaGo became famous after defeating top South Korean Go player Lee Sedol. The battle between the two took place in 2016. Now, AlphaGo is eating a new “victim”. The AI defeated Ke Jie, the current Go champion, in the first match of the three scheduled rounds. Even so, AlphaGo did not manage to win hands down in the match. As KompasTekno summarized from Cnet, AlphaGo only managed to win half a point, the smallest possible margin. Jie also expressed his admiration for AlphaGo.
“Last year, he was still like a human (way of thinking) when playing,” said Jie. “But this year, he became the god of Go,” Jie said. AlphaGo does not immediately become good at playing Go. Initially, AI was still often inferior to humans. Then, AlphaGo learned a lot from the defeat. He studied the best steps that could be taken when facing a certain pattern.
Now, AlphaGo is getting smarter.