Life In 19x19

In another thread the issue of time limits of professional games, and the effect on the quality of play and amount of mistakes came up. JF lamented the shorter time limits of modern Go (particularly outside Japan) and suggested they led to a more risky, probabilistic style of Go than "soba" Go (more compromises and trades for 'market price'), and that a soba-style player (I don't know what modern top pros are such, a Japanese player like Takao perhaps?) with 6 hours might have done better against AlphaGo than Lee Sedol did (with 2 hours). (Whilst I agree Lee Changho in his prime, or Go Seigen could well have done better against AlphaGo, I doubt any of the current top Japanese players would even though they still play 8 hour games). Anyway, I thought I would actually look at some pro games with in-depth commentaries and see how many mistakes they make to get some real evidence to inform the discussion.

For starters I am currently reading GoGameGuru's book Relentless on the Lee Sedol vs Gu Li Jubango in 2014. Those games had 4 hours each. Next up I will look at a Go Seigen game from one of JF's books (I have Kamakura (vs Kitani) and 9-dan Showdown (vs Fujisawa)). Then I will try to find some modern fast game with an in-depth commentary (maybe from gogameguru, could be hard to find), and perhaps a recent 8 hour Japanese title match. Of course some commentaries will be more critical than others, mistakes have different degrees so you can't just count them, but I think this will be an interesting exercise.

So let's start with game 3 of the Lee Sedol vs Gu Li Jubango, which was played with 4 hours each (pretty long for non-Japanese tournaments, most are 2-3 hours these days). Thanks for the analysis of course go to authors An Younggil and David Ormerod, which I am summarising.



Mistakes:
1) : Driving tesuji usually good but here directly 25 better. Then white extends at black 23 and black hanes underneath at o18. White is then not as thick as in the game. That thickness made the left side black thin.
2) : Lee probably wants to exchange this for o3 before playing r5, but white's resistance was powerful and he should have just played r5 directly.
3) iron pillar at c10 better.
4) b101: should push at 102 first (Note Gu's perfect opening play to here!)
5) w112: slightly passive, just connect is better but white is ahead so he fell back.
6) w116: should atari d7 then g8 then d12.
7) w124: should be e6, Gu was playing too slackly in the fighting due to his lead, Lee is in his element in this fighting in disadvantageous areas. The game was even by b131.
8) b135: big mistake, should b11 first, if white snips off the tail with e5 like in the game then black can get g9 in sente to seal the centre and then tenuki to play o12 (that's if white answers g9 at j7, if j8 then no seal and tenuki is to e14). Gu regained the lead to 146.
9) b147: losing move, should h9 cut. Lee's self-indulgent plan was that white would fall back to h3 and then he can cut and has gained 2 points.
10) w178: f13 would kill to end the game quickly and white can cope with h9 cut. This cut is bad because k12 in sente allows black k10 after h9 cut to make a ko shape to capture the right side in sente.
11) b185: d11 is a better way to live as black then has ko aji with the dead stones in the bottom left.
White won by Resign, 222 moves.
"We saw some careless moves in this game, which is unusual for Lee and Gu". Also Relentless mentions Lee said the tournament schedule was very too tight (this was the 3rd game between these 2 players in a week! so Gu had the psychological advantage).

Brilliances (according to an ill-defined threshold!):
- - : white flexibly compromises as black has bad aji with a ko in the corner, and another ko shape outside that's hard to fully digest.
- delicately reaching out to the right side stones (even without this the 6 white stones at s4 aren't totally dead despite black's s1 defence: with s6 white can either create a ko or seki in the corner, or if black allows t8 connection and tries to kill all white can live with a 30+ move long sequence using the n14/o13 stones!).
- b123: clever multi-purpose move to manage the left

So we have 11 mistakes in this 4 hour game, next up a Go Seigen game...

Hey Andrew,

I like this idea, but unless you find a way to randomize the games, your conclusion will likely suffer a selection bias. For example, why did you choose game three of this series?

Would you like me, without looking at the games first, to tell you which games to search for mistakes?

wineandgolover wrote:Hey Andrew,

I like this idea, but unless you find a way to randomize the games, your conclusion will likely suffer a selection bias. For example, why did you choose game three of this series?

Because I had just finished reading about game 2, and 3 comes next!

wineandgolover wrote:Would you like me, without looking at the games first, to tell you which games to search for mistakes?

If you like, but my determining factor in choosing the games will not involve looking through them beforehand, but trying to find ones with a similar level of detail of commentary. Relentless has ~60 pages on this game, though it is quite generous in diagrams and showing a lot of variations without too many moves per diagram (this game has 8 pages showing standard continuations of the Chinese opening, so that's wasted pages in respect of talking about their mistakes).

I also like that this thread has been started, and wonder if it is possible to link specific mistakes to the time settings themselves. (Right now we can't see what is happening with Lee Sedol and Gu Li's respective clocks when they are making mistakes.)

Reviewing faster games on YouTube - e.g., the Japanese TV games with 10 periods of 30 seconds - could be valuable. This way, you could see at what stage of the game the players were using their time, whether their faces indicate they are struggling to find the move, etc.

So onto a Go Seigen game against Kitani, from JF's Kamakura book. I was going to do game 3 of that match as well, but it only had 19 pages of commentary so went for game 4 with 29 pages (the first game has the most commentary). Time limits were 13 hours each, and no komi. JF is generally less blunt in calling moves mistakes so there is a bit more interpretation on my part than with the GGG book.



Mistakes:
1) : Kitani was afraid of the avalanche if he attached (it wasn't as well explored back then, Go's innovative inwards turn hadn't appeared yet) so played this full-wll knowing it was unusual, but if I would call it a mistake in a 10 kyu game review I have to call it a mistake even when played by Kitani.
WIP
2) Go said it should be kosumi to j4 to avoid giving white momentum to settle, as wedge and then hane were sharp, and this was the cause of black being outplayed in the opening.
3) minor, and others might disagree, but Kitani says this should be capture to avoid the peep at L7. Also is not so great as overconcentrated pushing from behind, but not much choice given earlier mistake. Cancel out?
4) Go says "questionable but perhaps inevitable" as induces white to fix thinness with 48. Fujisawa Hosai preferred c13. Go later said d13 better, but rejected that for weakening top. Kitani thought 47 as played was good.
WIP

but if I would call it a mistake in a 10 kyu game review I have to call it a mistake even when played by Kitani.

That's a very interesting view, and one I have a lot of sympathy for.

But in this particular case, a handful of other top pros (e.g. Seo Pong-su, Kono Rin) have followed Kitani's example, and the surely very relevant similar position (by transposition) where Black kicks into a White stone on the third line and White nobis to the centre is actually rather common. And at this stage of a game there is clearly no time pressure, even when the paw of our good friend Mickie is operating the buzzer.

I think this might be better classed as a stylistic choice (and here a psychological ploy, too) and not a mistake.

Since you are relying on actual commentaries, I'd suggest that mistakes here should be limited to moves either criticised by someone else or regretted by the perpetrator. Otherwise you get into tricky waters not just as above but in the case of e.g. apparently "good" empty triangles that escape commentary.

Interesting thread, but how do we know that the things flagged as mistakes aren't mistakenly so identified themselves? The commentator wouldn't necessarily be a better judge of move quality than the players.

gowan wrote:Interesting thread, but how do we know that the things flagged as mistakes aren't mistakenly so identified themselves? The commentator wouldn't necessarily be a better judge of move quality than the players.

But the commentators have more time to think about the plays.

Oh, wait!

Uberdude wrote:if I would call it a mistake in a 10 kyu game review I have to call it a mistake even when played by Kitani.

Quod licet Jovi non licet bovi.

It's common sense that shorter time limits tend to produce games with more mistakes. This is hardly controversial. But the contention that somehow Japanese players would do better against AlphaGo (or against Korean and Chinese pros), if only time limits were longer, is simply absurd.

One often hears this from the 'pro-Japan' camp, along with other drivel such as Japanese Go is more 'beautiful' and more 'deep' and so on. This is all complete and utter partisan nonsense.

Stronger players are almost invariably stronger no matter the time limits. Time limits don't diminish their relative strength vis-a-vis an opponent because BOTH would be playing with the SAME time control!

Now of course, if someone contends that Japanese players would do better against Korean and Chinese pros if the Japanese side were allotted twice the amount of thinking time as their opponents, then yes, I agree, Japanese players would do better. I'm certain that Iyama would fair better against Ke Jie if Iyama were allotted 8 hours and Ke Jie only 4 hours.

You know what, even better, I think Iyama would prevail over AlphaGo in a 10-game match, if Iyama were given 16 hours of thinking time per game and AlphaGo only ONE MINUTE per game.

The fact is, shorter time limits actually favor WEAKER players. Many over-the-hill pros can still be dangerous in shorter games (especially blitz Go) against opponents who are in their prime. I'd expect the Japanese to fair even worse if the time limits in international tournaments were significantly extended.

One last point, and I know aesthetics is a purely subjective matter, but to me, winning is more 'beautiful' and 'deeper' than losing. Korean and Chinese Go styles to me are more 'beautiful' and 'deeper' than Japanese Go.

Or maybe we can all just grow up and cut the partisan bullshit and apologist drivel and simply appreciate the Game of Go for its magnificent beauty and depth, regardless if it's played by the Chinese, Koreans, Japanese, apes, cats, donkeys, zebras, or even a computer.

End of rant.

P.S. I favor longer time limits.

gowan wrote:Interesting thread, but how do we know that the things flagged as mistakes aren't mistakenly so identified themselves? The commentator wouldn't necessarily be a better judge of move quality than the players.

You don't know for sure, but you critically appraise the sources and reasons given. Relentless is written by An Younggil 8p, who isn't as strong as Lee Sedol or Gu Li, that much is true. It worth noting though that Younggil actually has a 4-1 win record against Lee Sedol from 2001-2003 (An is 3 years older, was 4-5p to Lee's 3p, that is around the time Lee catapulted onto the scene and won the Fujitsu cup) so he can't be dismissed as some weak pro (of course there are many promising young pros who don't then mature into international title winners, I don't know why Younggil moved abroad). However, he has the benefit of hindsight, lots of time for analysis and reference to other pro's studies of the game. He received support in the analysis from Lee Hajin 3p, Han Jongjin 9p, Kim Jiseok 9p, Kim Juho 9p, Lee Jungwoo 9p, Mok Jinseok 9p, Park Jungsang 9p, Song Taegon 9p, Park Seungchul 7p, Choi Moonyong 6p and Na Hyun 4p, which suggests high quality. And of course during the game Lee Sedol is trying to find mistakes in Gu Li's moves and vice versa. So mistake #9 (b147) is pretty obvious because Gu's subsequent moves clearly punished the mistake. The player's thoughts are sometimes given too (moreso Lee than Gu, presumably because he and the author are Korean).

So to examine the veracity of the mistake claims from Lee vs Gu game 3 in more detail:
1) ":b23: Driving tesuji usually good but here directly 25 better. Then white extends at black 23 and black hanes underneath at o18. White is then not as thick as in the game. That thickness made the left side black thin"
There is a variation diagram to accompany that reasoning which is almost verbatim from the book. It's certainly true white is less thick. The question is does that reduced thickness mean the fighting variations (of which there are a lot, it's great in depth commentary) given later in the book for what happens if black attached on top of white's c12 invasion no longer favour white so crushingly. Also the way black killed the upper right corner left bad aji which white exploited with n14/o13 which then had a subtle but profound effect on the status of white's group in the lower right later. Killing the corner with the o18 hane would probably have not made such tactics possible (my analysis). So this one involves some judgement of unknown future possibilities, I would classify it as "probably true".
2) ":b37:: Lee probably wants to exchange this for o3 before playing r5, but white's resistance was powerful and he should have just played r5 directly.". There are quite a few variation diagrams showing the development of the fight if white meekly answered, some move order subtleties and so on, which I find persuasive. "probably true"

I have other things to do this morning and this analysis is slow so I recommend you buy these excellent books if you want to come to your own conclusion if the moves I say are mistakes really are. But I will just mention that Lee himself was the source for mistake #4 and maybe some others. And for #5 it shows a sequence using the j3 peep which could be why Gu didn't connect, but Lee's 135 showed that was the lesser of 2 evils.

As for Kitani's move 6, I expected that to cause comment . But I am sticking to my guns in being undeferential and calling it a mistake. Searching my database (mostly GoGoD) for the below pattern (with a black 4-4 behind so that black's extension from the wall of 2 is a double purpose extension from the 4-4 as in Go's game) I find only 2 hits: one is Kitani in that game and the other is a KGS 17k against an 8 kyu me which is hardly illustrious company. (Of course there could be games with it not in my database but it's clearly off-piste).

Click Here To Show Diagram Code

[go]$$W
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . O , X . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |[/go]

Reducing the search area to just a quadrant or smaller does throw up a few more hits, including a Kono Rin one in which the hoshi behind is a white rather than black one. The other hits are not on open boards but typically have opposing thickness nearby so the attach may cause worry the opponent would inside hane and fight, or white wants to extend on the left side to reduce some thick moyo there but first does the corner kosumi, presumably as a bodgy double sente to prevent the 3-3 attach.

John Fairbairn wrote: and the surely very relevant similar position (by transposition) where Black kicks into a White stone on the third line and White nobis to the centre is actually rather common.

But those kicks are typically played where the wall cannot make an efficient extension (which is not the same as a 3-space, in the related kick with a 4-4 (but also quite different with the bigger corner, 3-3 aji etc) you can kick even if they can 3 space extend, for example if you plan to then invade the gap, they attach on top and then get the corner 2nd line hand connect in sente, as in AlphaGo game 2, which of course led to Lee's 2 space extension and then AlphaGo's famous shoulder hit to overconcentrate from the other direction), such as if you low approach inside a Chinese formation.

Bill Spight wrote:

gowan wrote:Interesting thread, but how do we know that the things flagged as mistakes aren't mistakenly so identified themselves? The commentator wouldn't necessarily be a better judge of move quality than the players.

But the commentators have more time to think about the plays.

Oh, wait!

Clearly the solution is to only use commentaries published within three hours of the game's completion.

Clearly commentating a game is the same as playing in it.

Kirby wrote:Clearly commentating a game is the same as playing in it.

Actually it might sometimes be the other way. I heard from a 9p player that the players usually look at far more variations than the observers. Also, in a two-day game, the players might spend more time concentrating on the game (16 hrs. plus) than the people doing the comments. Of course, the comments might come from a number of people who were observing and studying the game together (parallel processing vs. serial processing ). There is that phenomenal example of Kitani, in a game with Sakata (I think). During the post game review he announced that at a certain point he read ahead 40 moves, saw that an approach ko would develop, and analysed the ko threat situation. Neither Sakata nor any of the observers saw that the ko would be there. My point would be that we can't say exactly who sees more in the game.