Life In 19x19
http://lifein19x19.com/

A further AlphaGo/Go Seigen parallel
http://lifein19x19.com/viewtopic.php?f=13&t=14129
Page 1 of 1

Author:  John Fairbairn [ Tue Mar 28, 2017 12:23 pm ]
Post subject:  A further AlphaGo/Go Seigen parallel

Ohashi Takufumi is continuing his analysis of AlphaGo (in its "Master" embodiment), and makes a remarkable point about the following position - master is White against Pak Cheong-hwan.



Looking at the GoGoD database, by far the most popular choice for White 6 is the wariuchi (splitting move) at A - 160 games out of 354 prior to AlphaGo. The second choice (139 games) is B. A further 8 games continued with moves on the right side, so in 87% of cases the pros deemed the right side to be the area to play. As regards A (45%) Ohashi says this move has been regarded as just "common sense" and "natural."

But in the 60-game series Master played C FOUR times, which is a move no pro had ever played before.

What is remarkable is that, according to Ohashi (without citing a source, but I believe him), Go Seigen insisted, "The wariuchi is bad in this position."

It seems obvious that AlphaGo is not just training itself by copying pro moves in databases (although it might just be the case that it's learning from Go's games, given all the parallels!).

But the most parallels are actually with Shin Fuseki in general. Go is forever associated with that but he was far from its main proponent - pros of all stripes played it enthusiastically, and the following is a game from 1935 (when Shin Fuseki was actually waning). Note the unusual emphasis White put on the centre, but Black was alert to the problems that might cause and the game ended up with both sides splitting the centre after a sort of role reversal. The main signal of that role reversal was, of all things, an unusual shoulder hit (Black 51) in true AlphaGo style!



Just like AlphaGo, White put his faith in the centre and was rewarded with a 1.5 point victory which apparently could have been a fraction more if he had played the endgame with more care.

Author:  oren [ Tue Mar 28, 2017 2:15 pm ]
Post subject:  Re: A further AlphaGo/Go Seigen parallel

Minor nit... author is "Ohashi Hirofumi"

Author:  handa711 [ Thu Mar 30, 2017 1:15 pm ]
Post subject:  Re: A further AlphaGo/Go Seigen parallel

AlphaGo defying all common sense.

Author:  wineandgolover [ Fri Mar 31, 2017 8:12 am ]
Post subject:  Re: A further AlphaGo/Go Seigen parallel

Can I assume that "wariuchi" is C in the OP?

Do we know why Go Seigen thought it bad? I'd guess the lack of development opportunities on the bottom, and the open skirt on the left.

Author:  Uberdude [ Fri Mar 31, 2017 8:29 am ]
Post subject:  Re: A further AlphaGo/Go Seigen parallel

wineandgolover wrote:
Can I assume that "wariuchi" is C in the OP?
Do we know why Go Seigen thought it bad? I'd guess the lack of development opportunities on the bottom, and the open skirt on the left.

John Fairbairn wrote:
the most popular choice for White 6 is the wariuchi (splitting move) at A


Go Seigen and AlphaGo both dislike splitting the right side. AlphaGo likes the high shimari at C. I'm guessing Go might like it too if he doesn't like splitting (I don't know if he liked the small high shimari particularly (but possibly yes as it's not vulnerable to the shoulder hit he liked to play against the small low one), but he certainly played the large low one quite a lot, as does AlphaGo/Master).

P.S. Actually something related I remember reading is that according to some opening theory judgements of pros, the exchange of 1 for 2 here is slightly good for black (basically it's corner better than side, though maybe Master or Go Seigen would disapprove of the small shimari):
Click Here To Show Diagram Code
[go]$$B
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . b . . . 2 a . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

It is played in pro games though (221 games in waltheri). Interestingly white's further extension at a, which is presumably made partly in respect of this thinking (and makes black's checking extension less good, but on the flip side is further from the white 4-4) is slightly more common (267), but has a lower winning percentage (45.3% vs 50.2%). The more modern style of the tight knight's move at b has 96 games and 53.1% win. So given this theory, in the original position I can see some logic for preferring white making his own shimari instead of splitting even if that does allow black the good move in the middle which is a good extension from both the shimari and the top right 4-4.

Author:  Bill Spight [ Fri Mar 31, 2017 10:22 am ]
Post subject:  Re: A further AlphaGo/Go Seigen parallel

wineandgolover wrote:
Can I assume that "wariuchi" is C in the OP?

Do we know why Go Seigen thought it bad? I'd guess the lack of development opportunities on the bottom, and the open skirt on the left.


I doubt if Go Seigen had a firm opinion about that wariuchi. Rui Naiwei once told me that in his study group he often changed his mind from week to week. That speaks wonders for his genius and creativity at an advanced age, but not to his having firm opinions. :) In his ten volume set, Go Seigen: 21st Century Go he sometimes likes the wariuchi as White, sometimes not. There is even an example where he contradicts himself on that point in two different places.

But my impression is that in general he did not like such a wariuchi for White. It is also my impression that a major reason was that it is placid. With current komi (both 6.5 territory komi and 7.5 area komi) he thought that White should make the game difficult for Black in the opening. At first I thought that maybe he was an old man living in the past, but now I think that he was right. :D

Author:  John Fairbairn [ Sat Apr 01, 2017 1:09 am ]
Post subject:  Re: A further AlphaGo/Go Seigen parallel

Quote:
At first I thought that maybe he was an old man living in the past, but now I think that he was right.


It's never too late to be an opsimath, Bill :)

Page 1 of 1 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/