Fate of Humanity in the Hands of a Korean

(It did not take long to find a topic that excited TK to get back to writing!)

Yesterday, the world witnessed history:  Google's AlphaGo, an AI program for go, defeated the human champion Lee Sedol for the first time in the history of the game. 

The match to determine the fate of the human race.
(source)

TK is a hobbyist go player--around 14 kyu, which means I know just enough to understand the professional commentators when they explain the games played by the championship level players. I will spare you the stuff about how a game of go is played, how complex the game is, etc. Instead, for those who have no idea what it is like to play go:  from the player's perspective, what does the game feel like?

The best analogy I could come up with is:  go is like basketball at 4 a.m. The "flow" of the game is more like basketball than any other commonly played sport. In basketball, each team takes turn to run a play; the play may result in two points, three points, or no points. Ideally, the team would try to make a three point play every time, but of course that is not possible because the opponent would try to defend the three point line. In that case, you would try to win by finding the best two point shot. And of course, you would try your best to limit your turnovers, or possessions that result in zero points.

Go is similar, in that each move is worth variable amount of points. Like basketball, players usually trade points--you score a basket, I score a basket. But over the course of the game, you build a series of small advantages. In basketball, you would have a stretch of a few minutes where a team goes on a 6-0 run, and other team may respond by going on a 12-4 run. In go, like basketball, the player that puts together more runs, however small, ends up winning the game. But here is the twist that makes go such a complicated game: in the early and mid-game of go, it is almost impossible to know exactly how many points you have scored with each move. All what you have is a vague sense of how the game is going, based on certain recognized patterns. It is not until the late game where it is even possible to count the points.

That's why go is not just basketball, but basketball played at 4 a.m. At 4 a.m., the sun has not yet risen. In the dark, both teams can sort of make out the basket, but cannot really see exactly how far they are from the basket. Nor can they tell where the out of bounds lines are, or where the three point line is. The best they can do is to guess where they stand in relation to the basket and to the court. 

Each shot they take would be recorded. As time passes, the sun begins to rise. Around halftime, the sky is deep blue instead of pitch dark. The lines slowly become visible. Then the teams get the more concrete sense of where they stand: some of their shots in the dark were worth three points because they managed to correctly shoot the 3-point shot, and some of their shots were worth zero points because they were actually out of bounds when they shot the ball. A team with better night vision would realize sooner where they stand. By the time the sun rises and everything is visible, the game has about three minutes left in the fourth quarter. At this point, both teams finally know exactly what their scores are. Sometimes, with three minutes left in the fourth quarter, it is a 20-point basketball game (or around 7-point game in go.) In such a case, the trailing player would usually forfeit because it is virtually impossible to make up 20 points in three minutes. Sometimes, the game reveals itself as a 3- to 4-point game (or around 2-point game in go)--which leads to a furious finish between the two players until there is no possible move remaining on the board.

*                           *                             *

Google could not have picked a better opponent than Lee Se-dol. Technically, Lee is not currently the world champion; that would be China's Ke Jie. The world go circuit operates similarly to tennis or golf, in which individual players amass "points" for playing in various major and minor tournaments. Lee Sedol, in fact, is not even the current champion of Korea; that title belongs to Park Jeong-hwan. But Lee Sedol, nonetheless, is the master. He is akin to Roger Federer--although they may or may not be technically the world champions at any given moment, there is no doubt that they are the very top of their respective games if one considers their body of work for the last decade.

Lee Sedol is the perfect opponent for another reason: unlike the machine, he has personality. Lee is brash and confident, bordering on arrogant. A respectable football team may win a game by the score of 10-3, employing stiff defense and reliable special teams play. Lee Sedol's play style is the opposite of that respectable team. At every other possession, Lee would fake a punt, run an end-around and throw a hook-and-ladder pass like a drunk Boise State football team in 2007 Fiesta Bowl. He would fling himself to seemingly hopeless battles, contemptuous of the idea that such risk-taking may backfire. Against the longest odds, somehow Lee more often than not pulls it through. This trait made him not just a champion, but a superstar in the go circuit.

Lee does not just let his game talk; he lets his mouth run. Lee particularly delights in tweaking Chinese fans who, like English soccer fans, often get upset at the fact that another country dominates the game they invented. In an infamous exchange, an interviewer asked Lee if he admired the great players of the previous generation, giving Korea's Lee Chang-ho, Cho Hun-hyeon and China's Ma Xiaochun as examples. Lee replied:  "They are all great players, but I admire no one. Oh, and Ma Xiaochun is not a great player." In 2010, after defeating Kong Jie (then-China's top player) in one of the greatest comeback victories in the game's history, Lee casually said: "I didn't even try very hard because I though I lost already, but I ended up winning anyway." Lee was likewise confident against AlphaGo, which he estimated to be around the level of top-flight amateur. Before yesterday's match, Lee declared: "I would consider myself defeated if I lost just one of the five rounds."

After the loss, Lee Sedol remained confident. Lee said AlphaGo "truly surprised" him, but said "I have won a lot of world championship. Losing the first round does not really bother me." But Lee's swagger is gone, and he is taking his opponent seriously: "Now I think my odds of winning is around 50-50."

*                           *                             *

Lee Sedol (black) versus AlphaGo (white), Round 1, March 9, 2016.
(source)

The first game is pure thrills-ville. Playing black, Lee Sedol almost immediately uncorks an unorthodox play in the early going. Perhaps to test AlphaGo, Lee Sedol deviates from the standard opening move, essentially daring AlphaGo to take advantage of his deviation. (B7 in the above diagram.) AlphaGo impresses right away by accepting Lee's dare. After several exchanges of complex battles, AlphaGo pulls slightly ahead. If go were a basketball game, AlphaGo finished the first quarter leading by four points.

Then AlphaGo makes a strange play (W58), taking a risk that is completely unnecessary (as it was leading.) Kim Chan-woo, a pro go commentator who is involved in AI go program development, surmised that this happened because AlphaGo does not make the "best move," but the "move most likely to lead to a win." Lee Sedol takes advantage of AlphaGo's risk-taking, and pulls even.

At W80, AlphaGo makes a straightforward, amateurish mistake. This is what makes watching AlphaGo such a trip. Pro players make mistakes too, but they make "pro mistakes." AlphaGo plays like a pro, but makes amateur mistakes. It's like watching Tiger Woods stringing together a series of incredible birdies, then suddenly seeing the golfer take out the five iron to hit the ball on the green. Lee Sedol duly punishes AlphaGo, and is now leading. AlphaGo begins thinking for a while. The game is entering the halfway mark. Things are looking good for the human race.

Then the moment of the match:  W102. The move was so original that Lee Sedol laughed upon seeing it. (Lee later said he though W102 move was "impossible.") Needing a change of pace, AlphaGo launched a daring, reckless attack--a Lee Sedol-esque attack. Lee is taken aback, and in response, makes a critical mistake of his own (B127). AlphaGo is now clearly ahead. The commentators are stunned: "If Lee Sedol was playing a human, he would have forfeited already because there is no way to make up this difference. He is now only playing in the hopes that AlphaGo would make a mistake."

At its highest level, go becomes a wordless conversation. With each move, you ask a question, convey your intent, send a message. Facing defeat, Lee Sedol becomes restless, poking fruitlessly at different corners to find an opening to attack. Each poke was a challenge: "Come on, AlphaGo. This game isn't over yet. Come out and fight me." AlphaGo would turn forbidding. With W154, AlphaGo seems to say: "I won. I don't have to answer to your challenges any more." 

Lee keeps at it for another 30 moves or so, and finally forfeits. History is made; the robot won.

Round 2 is at 11 p.m. EST on March 9, 2016. TK will be live-tweeting the game from 10:30 p.m., at Twitter handle @askakorean.

Got a question or a comment for the Korean? Email away at askakorean@gmail.com.