How far ahead do pros see?

[John Fairbairn]

There was a recent discussion of how strength of amateurs related to how far you regularly looked ahead. Pros tend to avoid the constant questions of how far they can see ahead, usually by making jokey exaggerations, but I have just been reading a longish article by Takagawa where he takes the question seriously.

He says Sakata once said he saw 30 moves ahead at a glance, and that Ishida Yoshio had said something very similar. For himself, he said he didn't see very many moves at a glance, but given time he could get to that sort of level easily enough. This was, incidentally, for fairly difficult positions. For easy positions such as those involving a ladder, the number would rise to 50 or more.

Unofrtunately he didn't explain what the pros like Sakata and Ishida did with all their time once they had seen so much instantly. Presumably the tactical genius Sakata looked even deeper, but if so no doubt things got exponentially difficult even for him, and that might be how he used his time. Ishida perhaps spent his spare time counting - that might be the real reason he got his nickname "Computer". That would be borne out by Takagawa's later comments.

Before that, I can just interpose that 30 moves at a glance for Ishida seems accurate enough to me, as I've been in press rooms where he's rattled off sequences that long instantly in response to questions from other pros. And, incidentally, not only can he lay down 30 or so variation stones at high speed, he can pick them all up again almost as fast to get back precisely where he started. Once or twice, he hummed and hawed once the stones were on the board, but not about the stones played - he was more concerned that the result didn't "compute", i.e. he evaluated the result almost as quickly as he saw the variation. I tend to look at the players' eyes in these situations and what I've noticed is that usually they are focused only on the local position. We amateurs are often enjoined to consider the whole board, but pros seem to have the rest of the board stored away in their brains and so can concentrate on local affairs.

Although Takagawa doesn't say so specifically, he implies that at least pros like him spend a lot of time counting. That is a form of lookahead, but he scoffs at the famous story about how Honinbo Jowa worked out at move 101 that he would won by 2 points - and he did. Takagawa says it was a bit of fake colour added by Sekiyama Sendayu. But he doesn't dismiss the idea that Jowa was working out the complete game. He is just saying that that level of precision is an "exaggeration".

To illustrate that, he then shows a position from one of his title matches in which he worked out, in a position at almost the same stage, that he would win by 0.5 or 1.5 points. The main difference was that the position in his game was fairly easy, he said, whereas the Jowa game was a difficult position (and even I could see the justification for that assessment). Also, in his case he couldn't confirm that his prediction was accurate because his opponent made a mistake and resigned!

But he still felt confident about his count, and the reason was that at that time he was in his thirties, in his prime, and he was playing quite a lot of games each year which involved lots of boundary-play counting. He naturally became good at it. His games developed this way because he (as a player who needed time to go deep) would carefully ration his time. In a 9-hours-each game (then common), he would try to allow two hours each for each 50 moves (both Black and White), which meant 200 moves in 8 hours. By then, he said, they were likely to be down to the small endgame plays, for which he would allow an hour. He stressed that time management was a big part of his game.

[snorri]

Thanks for sharing this. Maybe flowing like the river as Takagawa sounds easy, but does take a lot of time and is not a way to avoid reading. Do you think there are modern players who use that style? I know that sounds like a troll, but I'm really curious whether modern pros admire Takagawa and say, "well, I wish I could play like that but it's very hard to today for this or that reason."

[RobertJasiek]

1) Except for the trivial pure lookahead depth question, statements about how far somebody can read ahead are rather meaningless because of the tree breadth.

2) I can confirm the pro statement about quick count while only looking looking locally where currently things are changing. The underlying method is to store the global count together with assumed territory boundaries and to update both iteratively on the local scale. This saves much counting time! An advanced version stores also the local values of every territory region as separate numbers so that one does not need to count again the whole region when a move is played there.

3) In a peaceful, easy endgame, I can predict my game's score by ca. +-1 point 100 or 200 moves before the end. Presumably the trick here is that I make yose sequence mistakes for both players and they cancel out each other fairly. So I am, of course, not surprised that quite some pros game do likewise. However, Edo / Meiji endgames were often much more complicated. The much greater thinking time then compensated for that to some extent. Of course, nobody can read the entire endgame, see trivial test problems in Mathematical Go Endgames. But when making yose sequence mistakes for both players is allowed, then Jowa & Co surely must have had a more than impressive endgame reading.

[tapir]

Why is it people exaggerate so much? Danigabi wins 80% against 2 dans giving 6 stones handicap. Robert can count peaceful endgames 200 moves ahead - as if amateur games at his level ever have a peaceful endgame. If so, Robert, please use this strength in the next german championship, play accordingly towards a peaceful endgame and win your games by 1.5 or 2.5 margins. You know I write forum posts with a speed of 2000 characters per minute. Amazing, innit?

[RobertJasiek]

Maybe I should try again. It is how I won a former best-of-5 Berliner Championship: all three wins by 0.5 points :)

[Oroth]

Interesting quote from an online Lee Changho-ho interview (or at least the exerpts of one):

Q: How many moves ahead you read before you play a move?
A: Usually professional players, including me, read around 100 moves ahead. But that's not the case for every move. First select 10 candidate moves and then read ahead for each of them. After reading ahead 20 to 30 moves for a candidate move, one could reach a tentative conclusion like "this is a bad shape" or simply "this is not it." At that point, I stop any further reading for that candidate move and look for another. This is a process of elimination that ususally leaves one or two candidate moves. For each of these final candidate moves, I read ahead about 100 moves. This might surprise amateur players, but the more difficult thing is not reading ahead 100 moves, but deciding which of the final cadidate moves gives a better result. .... The most painful moment is when I realize that I am on the wrong way a few move after my original decision. That gives me an agony beyond description. People call me "Stone Buddha" for my lack of facial expression during games. But you will notice some changes in my face when I am in a bad situation. You have to look at my face carefully...

[Li Kao]

Redmond explained that “The problem is that you can have two endgame moves that are about the same size, but they each lead to a different endgame.” He launched into an analysis involving calculations of moves as small as 1/6th or 1/12th of a point, “so you have very fine points implicit in the seemingly simplest yose moves, including follow-ups and ko threats, which complicate the calculation.” And, he added, “calculating is not good enough; in fact it’s confusing, because there’s no way to see which move is bigger, you just have to read it out, and then it’s very clear. Right now I can do 30 moves, and I have done a 50-move yose.”

Eventually Redmond expects to be able to read out the last 100 moves, “because top players are capable of reading out the last 100 moves in less than an hour. If I can have a picture of what’s happening when I come to the last 100 moves, it’ll make a big difference.”

http://www.usgo.org/news/2010/06/michae ... ros-train/

[Kirby]

I'm sure that pros are good at reading many moves ahead. But I think they're also good at pruning bad moves so they don't have to read "silly options".

I'm sure that I read many situations that pros wouldn't consider, because they are poor choices, inherently.

[snorri]

This thread is getting a little nutty.

The-Most-Interesting-Man-in-the-World.jpg (55.32 KiB) Viewed 13352 times

[Mef]

I'm sure that pros are good at reading many moves ahead. But I think they're also good at pruning bad moves so they don't have to read "silly options".

I'm sure that I read many situations that pros wouldn't consider, because they are poor choices, inherently.

I think another (similar) issue is properly evaluating what you read. I have seen commentaries, etc where they may show a variation or two, and personally looking at the position I might have read out the same sequences. Of course after the options have been shown, the variation decided on as better or best is different that what I would have thought. In this case it's not even necessarily due to a difference in depth or breadth of search, but instead in evaluation of the final leaves in the tree. Perhaps it could be due to even deeper non-demonstrated reading, though I would imagine much of it is owed to better counting and more sophisticated understanding of the nuance in each position. What "feels" right may not necessarily be borne out by strict calculation (which may be why some pros spend so much time counting).

[jts]

I think another (similar) issue is properly evaluating what you read. I have seen commentaries, etc where they may show a variation or two, and personally looking at the position I might have read out the same sequences. Of course after the options have been shown, the variation decided on as better or best is different that what I would have thought. In this case it's not even necessarily due to a difference in depth or breadth of search, but instead in evaluation of the final leaves in the tree. Perhaps it could be due to even deeper non-demonstrated reading, though I would imagine much of it is owed to better counting and more sophisticated understanding of the nuance in each position. What "feels" right may not necessarily be borne out by strict calculation (which may be why some pros spend so much time counting).

I have the exact same feeling. It's especially irritating when the solution to a "black to capture" or "black to connect" tesuji problem actually allows white to escape/cut/whatever but they show a diagram that says "clearly white cannot allow this." Partly the problem is that when I get the sense that "obviously" white has escaped or cut or whatever, I stop reading, but even if I had read out the answer, I wouldn't have been able to confidently say that this was better for me than the variation where black accomplishes his tactical goal.

[hailthorn011]

This thread is getting a little nutty.

The-Most-Interesting-Man-in-the-World.jpg

Very funny.

I've always thought Pros could read 100 moves ahead because that's what I learned when I came into the game originally. Personally speaking, I think that's incredible. I'm not even sure I can read 1 move ahead, but there are players who can read 100 moves ahead.

However, if you think about it, it makes sense. I assume a professional player has played a lot of games in his/her lifetime, so situations on the board might become easier to analyze. Eventually, one could read the moves out based on their previous experiences and see if it will result in a positive situation at the end.

From my own experience, I suppose there are situations now that I recognize based on countless experiences with them.

That's just an idea that came to my head though. I'm probably way off the mark though.

[Tryphon]

I'm not even sure I can read 1 move ahead, but there are players who can read 100 moves ahead.

I don't remember where I read that your number of dans is the number of moves you can read properly.

And if you are a kyu, your number of kyus is the number of mistakes you make before reading properly the next move

[Bill Spight]

The title of this thread is about how far ahead pros see, not how far ahead they read. There is a difference. There is also a question of chunking. Any experienced player can "see" several moves ahead when the sequence of play is something that has been learned as a unit. That does not mean that the player actually visualizes the whole sequence consciously, although if the situation demands it, that is easy to do. Killing by nakade is an example. A long, but simple ladder is another.

When I was learning go I saw a 40 move variation in a magazine aimed at kyu players. Such a long variation was unusual in that magazine. The pro author commented that it was easy for pros to read 40 moves ahead, if it was a one lane road, and encouraged his readers to do so, too. (It didn't look like a one lane road to me, but that was my problem. )

After the first Environmental Go Game, between Jiang Jujo and Rui Naiwei, I found what I thought was a mistake by Jujo, who had played Black. As he and I were discussing it, Rui came and looked over our shoulders for a second or two and then rattled off an 8 move combination for White. She obviously saw that sequence at a glance, but she had missed it during the game, as had he.

It is very likely that pros' brains are wired differently from ours, given that they develop their abilities in childhood, and that they indeed see further ahead than we do, whatever that exactly means. But they are still human.

[hailthorn011]

I'm not even sure I can read 1 move ahead, but there are players who can read 100 moves ahead.

I don't remember where I read that your number of dans is the number of moves you can read properly.

And if you are a kyu, your number of kyus is the number of mistakes you make before reading properly the next move

Oh great, that means I make 9 mistakes But that's interesting. I'd never heard that before.