What is AlphaGo?
“AlphaGo is a narrow AI (Artificial Intelligence), a computer program developed by Alphabet Inc.'s Google DeepMind in London to play the board game Go.” In October 2015, the team claimed that AlphaGo had beaten its human rival and won by five games to zero. This is the first time that a computer program has beaten a human professional Go player on a full-sized 19x19 board without handicaps. From then on, the program starts to play thousands of games between its neural networks to make evolutions. It uses a technique called trial-and-error process which is also known as reinforcement learning.
Why Go?
Go is a board game which is believed to be invented in ancient China over 2500 years ago. It is regarded as the oldest board game, and people still play it today. Go is played on a grid of black lines (usually 19×19). Game pieces, which called stones, has two different colors – black and white, and are played on the lines' intersections. Unlike cheese, each move of Go has about 200 possibilities. As a result, there is so many possible configurations on the board – even more than there are atoms in the universe. Go is too complexity for computer programs to play. Before AlphaGo, there is no program can beat human professional players. Therefore, Google’s DeepMind team believes that it is an excellent opportunity for them to generate an AI to beat human players in the Go, in order to make further study and progression in the AI field. The ultimate goal of the team is to use AI techniques solve real-world problems, such as climate change, healthcare, starvation, and so on.
Competition
In March 2016, Google invested $1 million in a five-game Go match, to set up a contest between AlphaGo and 18-time world champion Lee Sedol. Before the game, most of the professional Go players did not believe that a computer program can beat human on the Go board. On the other hand, most of the AI developers declared that they thought there was a great chance for the AlphaGo to win the game. Especially that some of the Google engineers said that they believe there was no possibility that Lee Sedol could win the game. When the time came to March 15th, 2016, the last game had ended, AlphaGo had won 4 out of 5 of the match, and gained $1 million rewards. It became the first computer program ever beat a 9-dan professional human player without handicaps. Google donated all the money prizes to charities. The Korea Baduk Association also awarded AlphaGo an honorary 9-dan after this competition. During the rivalry, Lee and the team found that AlphaGo had a longer respond time when it is playing the black. AlphaGo had also made a mistake on the 79th move in game four which resulted it in losing the game. However, the DeepMind team claims that they have already fixed these problems on June 29th.
“Master”
On December 29th, 2016, there was a mysterious player called “Magist” had appeared on the Tygem server – one of the online Go game’s website. On December 30th, the account shifted its name to “Master”, then moved to the FoxGo server on January 1st in 2017. As of January the 5th, AlphaGo won all of the 60 games, which included three victories over Ke Jie, the top-ranked Go professional players in the world. “Master” was suspected to be an AI player as it had little to no quiet time between games. And on 4th January, the Google’s team admitted that both “Magist” and “Master” were played by the advanced version of AlphaGo. "Its adversaries included many world champions such as Ke Jie, Park Jeong-hwan, Yuta Iyama, Tuo Jiaxi, Mi Yuting, Shi Yue, Chen Yaoye, Li Qincheng, Gu Li, Chang Hao, Tang Weixing, Fan Tingyu, Zhou Ruiyang, Jiang Weijie, Chou Chun-hsun, Kim Ji-seok, Kang Dong-yun, Park Yeong-hun, and Won Seong-jin; national champions or world championship runners-up such as Lian Xiao, Tan Xiao, Meng Tailing, Dang Yifei, Huang Yunsong, Yang Dingxin, Gu Zihao, Shin Jinseo, Cho Han-seung, and An Sungjoon." All60 games except one were fast paced games with three 20 or 30 seconds byo-yomi. AlphaGo team extend one minute when playing with Nie WeiPing due to his old age. After finished the game, the “Master” even typed “Thank you professor Nie!” in Chinese to show its respect to Nie. One of the professional players Gu Li even offered a 100,000 yuan (14,400 USD) reward to the first human player that can defeat the “Master”.
Nonetheless, it is believed that human players tend to make more mistakes in fast-paced games, as they have less thinking time. We still don’t know if AlphaGo can make it as successful as in a slow-paced game (tournaments).
In the Future
On April 10th in 2017, the China Go Association, Google Corporation, and Zhejiang Provincial Sports Bureau held a news release conference in China Qiyuan jointly. They announced that AlphaGo would compete to several Chinese professional Go Players who are led by Ke Jie -- the world number 1 Go player. The competition will be held on May 23rd – 27th on Wuzhen Future of Go Summit, which includes:
- · A best of 3 matches versus world number 1, Ke Jie
- · AlphaGo versus a collaborating team of top Chinese professionals
- · Pair Go: human plus AlphaGo versus human plus AlphaGo
The whole Go community is looking forward to this competition. Some of the Go experts believe that it is very likely for human players will use AI technology as training in the future.
Source：
Hassabis, D. (2016, January 27). AlphaGo: using machine learning to master the ancient game of Go. Retrieved April 20, 2017, from https://blog.google/topics/machine-learning/alphago-machine-learning-game-go/
About Us. (n.d.). Retrieved April 20, 2017, from https://deepmind.com/about/
Artificial intelligence: Google's AlphaGo beats Go master Lee Se-dol. (2016, March 12). Retrieved April 20, 2017, from http://www.bbc.com/news/technology-35785875
Google achieves AI 'breakthrough' by beating Go champion. (2016, January 27). Retrieved April 20, 2017, from http://www.bbc.com/news/technology-35420579
Google reveals secret test of AI bot to beat top Go players. (n.d.). Retrieved April 20, 2017, from http://www.nature.com/news/google-reveals-secret-test-of-ai-bot-to-beat-top-go-players-1.21253
Exploring the mysteries of Go with AlphaGo and China's top players. (n.d.). Retrieved April 20, 2017, from https://deepmind.com/blog/exploring-mysteries-alphago/
When I was a kid, I was very into playing both Go and Chess. I remember that Chess was almost a solved game, as the best Chess players in the world could no longer beat a computer. But at the same time, the game Go had not been solved, and AI was still surprisingly bad against high level competition. It's very interesting to see that about 10 years later, Google has broken through and beaten a human at the most complex board game in the world.
ReplyDeleteI too like Benjamin was very into chess as a kid. I have never played Go before. I find it super interesting that it took $1M for Google to develop a program that could beat a human, but not just any human, a 18 time world champion. That was impressive to me. I think it was really funny that one of the professionals was willing to pay 100,000 yuan for someone to beat the AlphaGo.
ReplyDelete