The battle has begun
When a historic match between Google DeepMind’s AlphaGo, the most powerful Go software ever developed, and the top living player, 9-dan ranked Lee Se-dol, was first announced, few had any doubts that Lee would win easily. But as the five-day tournament drew closer, more and more wagers were placed on the software program.
If Lee loses, the long-held mystique of the game with a history of 5,000 years could dissipate. We may no longer refer to the 19-by-19 grid on the Go board as the epitome of life’s complexities and the human mind’s capacity for subtlety. We are reminded of a futuristic sci-fi scene of a society in which robotic soldiers make up the masses while all other jobs become unimportant. Google is indeed a stunningly visionary company. But I still place my bet on a 5:0 sweep by Lee against AlphaGo.
AlphaGo surfaced as a formidable contestant in the game of Go after it became the first machine to defeat a professional player - three-time European Go champion Fan Hui - without any handicap in a match in October 2015. From its learning trajectory, the machine can hardly be considered to have reached a guru level. It is at best advanced for an amateur or in the lower cadre of masters at the professional level.
But the hidden power of the software is that it is capable of moderating skills and scoring against its opponent. It has been playing just enough to defeat its opponent - without exposing its real potential. What’s threatening is the speed in the progress of its skills. Google DeepMind co-founder Demis Hassabis gave a 50:50 chance of the program winning against the world’s human champion. When he arrived in Seoul, he said he was confident in his machine’s victory.
In the game of Go, computer-like memory and processing ability is essential. The player must be familiar with millions of configurations, possibilities, sequences, rules and the game strategies of the opponent. But a true guru does not stick to them, as they can get in the way of the imaginative and inventive world of Go. There must not be a signature or uniform way of playing the game. Such super memory power can actually be AlphaGo’s biggest weakness. Can it really awe the audience in a uniquely wondrous way as a human master does? Or are we underestimating a machine’s capacity?
The game of Go is challenging even to its best players. One must incessantly ask if it is time to compromise or fight and determine the optimal timing of an attack. One must choose between what’s best for now or later. Only a deity would be able to know the right decisions. A human must stick to his or her guts. Instinct sometimes can be a better guide than shrewd and precise calculation. Out of millions of factors, one suddenly must pop out. As a first impression can often shape the image of a person, what comes first to the mind of Go players can be decisive for the game. It is how each master builds a quintessential style.
Its developers boast that AlphaGo is not like any programmed computer, as it is designed to learn human instincts. The game it won over Fan Hui suggests it has a human quality. It is how it advances so quickly in learning configurations that are estimated to outnumber all the atoms in the universe. Moreover, AlphaGo is said to be in training at a pace that is equivalent to what a human player could do in 1,000 years.
Human geniuses, too, can master the game in a few years. A wiz learns the game at age 8 or 9 and can become a professional at 11 or 12. Around the age of 20, he or she can become a world champion. Instinct can be trained, but it mostly must be inborn. If AlphaGo was a born genius, it should have already reached the almighty deity level from the way it has trained.
But its record does not show AlphaGo to be all that intuitive. Its artificial intelligence can hardly be declared exceptional or extraordinary. People with modest intelligence can catch up with a master with training. But it is meaningless to compare human capacity with a definite lifetime to a computer capable of completing millions of games over a short period.
AlphaGo is said to be stronger toward the latter half of a game. The more stones there are, the more advantageous it can be. It does not know fatigue and has no risk of losing concentration.
Lee participated in the Nongshim Cup, a team competition among Korea, China and Japan that ended on March 5. Lee was the last remaining from the Korean team and defeated a Japanese and two Chinese players one by one. By the time he faced the last contestant, 9-dan Ke Jie of China, for a final round, he had been playing nonstop for four days. Lee’s three consecutive wins is an excellent score, but Korea had to surrender the champion cup to China after Lee’s defeat in the last round. Lee had only a few days to rest until Wednesday before entering a five-day tournament against AlphaGo, which already would have fully studied Lee’s strategies in his recent game in China.
Although the machine does not have emotions, psychology is crucial. Despite little difference in skills, Fan Hui looked too intimidated by the machine. A stoic - and yet sensitive - approach could be a style the machine may not have mastered. Lee is the best candidate to humble the machine. His unique wealth of intuition and sensitivity would gush out even if AlphaGo plays better than expected. I have faith in Lee that he won’t be intimidated by the machine in this war of nerves.
Translation by the Korea JoongAng Daily staff JoongAng Ilbo, March 8, Page 29
The author is vice president of the Korea Baduk Association.
by Park Chi-moon