Translated and compiled by Shin-wei Chang and Bing-sheng Lee

Go, also know as Weiqi, is considered one of the hardest and most complicated board games in the world. As a result, it has always been a goal for scientists to create an AI to beat world champions in the game.

AlphaGo, an AI developed by Google’s subsidiary company DeepMind, has amazed people around the world with its accomplishments. Last October, it won five games over Fan Hui, a three-time European Go winner.

On March 10, AlphaGo stunned the world again by beating Lee Sedol, a South Korean professional Go player who ranks second in international title, and the AI went on to win two more matches out of the five-game match. As the news spread and was discussed by the technological community, Aja Huang was recognized as the key promoter of the research and development team of AlphaGo. Huang has been described as the one who “instructed” AlphaGo and “designed AlphaGo’s brain.”

Born and raised in Taiwan, Aja Huang completed his PhD in information engineering at National Taiwan Normal University (NTNU). Instructed by professor Rémi Coulom and Lin Shun-sii, Huang published his research, “New Heuristics for Monte Carlo Tree Search Applied to the Game of Go,” in 2011.

Based on the research results, Huang predicted that Go programs could beat top human Go players in 10 to 20 years. However, he failed his prediction six years after he published his paper.

In 2010, “Erica,” a Go AI designed by Huang, beat “Zen,” a program that was publicly recognized as the best program at Go. In the same year, Erica even won the gold medal in the 19×19 Go tournament at the 15th Computer Olympiad.

On January 28, Huang was listed as one of the first author of the research, “Mastering the game of Go with deep neural networks and tree search,” which appeared in the prestigious scientific journal, “Nature.”

As the senior research scientist at Google DeepMind, Huang has been keeping a low profile regarding the games against Lee Sedol starting in March. He not only refused to leak any information about the matches, but also gave credit of the work to the entire team.

In “Eweiqi,” a famous Chinese software that allows players to play Go with other users online, some netizens found an account named “deepmind.” They suspected the account was used for testing the skill level of AlphaGo. On January 29, Huang clarified that he has been using the account, which was created before the AlphaGo team existed.

“Although I’m an amateur 6-dan player, AlphaGo has a significantly higher level than me,” Huang said. In this statement, he also predicted the future for Go programs saysing, “Go software of professional standards will soon be widespread in the market within the coming one to two years.”

Lee beats AlphaGo after three consecutive losses

Lee Sedol. Photo credit: AP/Lee Jin-man/達志影像

Lee Sedol. Photo credit: AP/Lee Jin-man/達志影像

On March 13, Lee Sedol beat AlphaGo in the fourth match of the five-game set, after losing three contests in a row to the AI computer program. This victory is considered a significant one for “human beings” and proves that the AI program still has flaws.

In the post-match press conference, Lee was welcomed by a round of applause and cheers from the media and reporters. Lee jokingly said this was the first time in his career that he received so many congratulations by winning just one game. He stated that the victory means a lot to him and he will cherish this achievement.

Lee also pointed out that AlphaGo has two drawbacks. When playing with black stones, the computer program hesitates for a longer period of time to place a piece, and it becomes susceptible to mistakes if there is an unexpected move made by its opponent.

Michael Redmond, a commentator for the online broadcast of the match, says that the watershed moment of the game came in move 78, when Lee played a “wedge” in the middle of the board. The surprising move caused AlphaGo to commit a critical error in move 79, subsequently driving the AI’s chances of winning down several moves later. When AlphaGo calculated that its chances of winning had dropped significantly, it decided to give up the game at move 180.

The match took nearly five hours and AlphaGo suffered its first loss in the nine competitions it has played against human beings.

Prior to the game, some critics claimed that the competition between AlphaGo and Lee is unfair because AlphaGo possesses statistics on Lee’s previous matches while Lee does not have the same information on AlphaGo.

Yet, Demis Hassabis, founder of DeepMind, explains that AlphaGo is not designed specifically to defeat Lee. The Al program strengthens its skills via constantly simulating Go games against itself. It has taken millions of games to train itself to its current level of competitiveness.

The last match out of the five-game match between Lee and AlphaGo is to be held on March 15.

Edited by Olivia Yang

China Times
China Times
Tech Bang
“Mastering the game of Go with deeop neural networks and tree research” (Nature)
“Go Grandmaster Lee Sedol Grabs Consolation Win Against Google’s AI” (Wired)
The News Lens