Adapting Play To Score
- I would like to start a little discussion...
It is often said that professionals are very good at adapting their play to the score. As a result, the introduction of komi and changing its value have caused dramatic changes in fuseki theory. Sometimes people point out that Shusaku won all his games as Black because there was no komi, and the reply to that is that he played perfectly given the absence of komi and that he would have played differently had there been komi.
How does this work? Surely someone can't just choose to play at a stronger level, so what exactly do professionals do to overcome komi?
Also, is it possible that now that komi is shifting towards 6.5, styles of play will change and in a few years it will be necessary to raise it again to 7.5? If it was "plain as a pikestaff" that the right value of komi is 7, why was it initially set at 4.5?
Sorin Gherman It's not about playing at a different level (like: "I am Black, no komi, so I can be stupid" :-) ) but it's about playing more aggressively or more conservatively. As you said to start your discussion, it's exactly like the adaptation to the score estimation in the middle of the game: if one plays Black, no komi, it's like being already ahead, so he will choose the safe path; his opponent will feel behind, so she's the one that should assume more risks, and play more aggressively.
The problem is that both variables (komi and style of play) are changing, each being dependent on the other one: changing komi enforces innovation in fusekis and josekis, and those new discoveries can change the winning percentage for one side or another, which leads to a change of komi....
Theoretically, the komi can come back to 4.5 in the future, who knows?
Historically, the initial value of 4.5 was chosen because that, and not another one, was believed to be the proper compensation.
Andrew Grant In fact, when komi was first experimented with in the early 20th Century, values as low as 2 or 3 points were used. It only became "plain as a pikestaff that it was 7" after many years of experience and statistical analysis. And as Sorin Gherman says, there's no guarantee that someone won't in the future find a superior way of playing for White or Black that will force another change in the komi.
Bill: I did a pre-publication review of the "plain as a pikestaff" article for the AGA Journal, and made the same point as Sorin about changing styles of play and komi. However, I also got to see some raw data, with 1400 recent Japanese professional games played at 4.5 komi and 1400 played at 5.5 komi. Each set of games indicated a proper komi of 6.5 to divide the results most evenly. For those games, at any rate, the players did not adjust their play significantly according to the difference in komi.
Jan: I only adapt my play when I think I'm far ahead or behind. In my book, that's 30 points or more (since I'm not very good at estimating the score). When I'm just starting a game, all bets are off, so I consider komi to be some kind of magic bonus - it's not really a factor in how I play. This should change as I get stronger :-)
HolIgor: The best thing that can happen playing online is when your opponent loses the connection and you have time to calculate the score and find out that you are 10 points ahead. This becomes quite a different game.
Migeru I have a little theory of why playing more aggresively increases your chances of winning the game. Feel free to punch holes in it. The question is, why should playing more aggresively help? And the answer is: because slight increases in the variance lead to disproportionately large increases in the probability of extreme events.
The idea is to regard each play as a random change in the final score. One way of playing conservatively is to choose plays that lead to non-branching lines of play. This means that the uncertainty in the final result is reduced. One increases the variance of the final score by choosing open-ended moves. Some people say that the reason something is sente is not necessarily that the followup is big, but that the continuation has so many branches that it becomes "unacceptable".
Alex Weldon: That's basically it, Migeru. If you're winning by, say, 10 points, you want to keep the game calm, so that any fluctuation in the score will be less than that. In that situation, if you were given a choice between a situation that has a 70% chance of gaining you 20 points and a 30% chance of losing you 20 points, and a situation that has a 20% chance of gaining you 3 points and an 80% chance of losing you 3 points, you'd go for the second, because even though the first will probably gain you points and the second will probably lose you points, the first gives you a 70% chance of a (big) win, while the second gives you a 100% chance of victory (of course, nothing is 100% in Go, but you get the idea).
Conversely, if you're losing by 20 points, you might as well start a complicated fight, even if there's a 90% chance you'll lose big in it, because a 10% chance of victory is better than nothing.
As the almost-proverb goes: "Why cut your losses when you can cut everywhere?"
Another way to put it is this; certain moves are simple enough that you can read out the continuation. Others are so complicated that you don't know what will happen. If you read something out and it leads to probable victory, you should play it. But if you read out everything that you can read out, and every road seems to lead to defeat, you should head for the path that neither you nor your opponent can properly read, and hope for the best (unless you're losing REALLY badly, in which case you should probably just resign to avoid irritating the opponent).
aceofspades: This may be overkill, but the way I see it is that Go is probabilistic for humans (and to a lesser extent for computers) because of things like limited reading depth, guessing the order in which to examine moves, and of course changes in mental state. So regard a point in a game as having a number of different continuations with different probabilities, each with their own score. In other words, the state of the game is a probability distribution over the possible score differences at the end of the game. Being behind or ahead corresponds to the distribution's mean, but what the pros do by strategy adaptation to score is change the variance. Thus if you're 10 points behind in the late midgame, you want to keep a big fight going so that all sorts of things could happen: big groups could die/live, big territories could be taken, etc. By increasing the variance you're increasing the probability of all those events that would give you enough points to overcome your opponent's lead. Increasing the probability that you would lose by an even bigger amount doesn't matter because a close loss and big loss are both still a loss. So, typically, one has to sacrifices a few points to increase the variance if you're leading or decrease it if you're behind. This may me simply because the optimal move is too risky or too solid, but also because your opponent will simultaneously be trying to get the variance to go the other way.