Life In 19x19

Click Here To Show Diagram Code

[go]$$W
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . , . . . . . X . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . 2 1 . . |
$$ . . . . . . . . . 3 . . . |
$$ . . . . . . . . . . X . . |
$$ . . . . . . . . . . . . . |
$$ . . . , . . . . . O . . . |
$$ . . . . . . . O . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ --------------------------+[/go]

Interested?

Damn, I am.

more to come ...

One has to dig a little bit deeper than 1 or 2 moves with AI if you want to sneak peak yourself

Only 3 games with this search window in my pro database. (Hint: But if these games are all by 9 Dans, perhaps a second look does not mean time spent in vain.)

Oh, yes indeed, I was also very surprised when I saw this for the first time, which was soon after the "AlphaGo Teaching Tool" was released. I haven't seen the pattern prior to the invasion much nowadays anymore in pro games.

The next moves:

Click Here To Show Diagram Code

[go]$$W
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . , . . . . . X . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . 2 1 . . |
$$ . . . . . . . . . 3 4 . . |
$$ . . . . . . . . . 5 X . . |
$$ . . . . . . . . 9 8 6 . . |
$$ . . . , . . . . . O 7 . . |
$$ . . . . . . . O . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ --------------------------+[/go]

from here it gets a little bit messy ...

First let us have a look at some pro games:

game #1


game #2


game #3

If black plays a or b, just the marked move is a very good option for white. (see also game #2, the marked move would have been a good alternative)

Click Here To Show Diagram Code

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

after black c or d, depending on the board position, it gets ... interesting

Click Here To Show Diagram Code

[go]$$B
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . , . . . . . X . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . X O . . |
$$ . . . . . . . . . O X . . |
$$ . . . . . . . . . O X . . |
$$ . . . . . . . . O X X . . |
$$ . . . , . . . . d O O c . |
$$ . . . . . . . O . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ --------------------------+[/go]

the next moves:

Click Here To Show Diagram Code

[go]$$B
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . , . . . . . X . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . X O . . |
$$ . . . . . . . . . O X . . |
$$ . . . . . . . . . O X . . |
$$ . . . . . . . . O X X . . |
$$ . . . , . . . . . O O 1 . |
$$ . . . . . . . O . . . 2 . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ --------------------------+[/go]

Click Here To Show Diagram Code

[go]$$B
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . , . . . . . X . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . X O . . |
$$ . . . . . . . . . O X . . |
$$ . . . . . . . . . O X . . |
$$ . . . . . . . 2 O X X . . |
$$ . . . , . . . . 1 O O . . |
$$ . . . . . . . O . . . . . |
$$ . . . . . . . . . . . . . |
$$ . . . . . . . . . . . . . |
$$ --------------------------+[/go]

Poor Takao, did Yoda and Kono conspire to embarrass him by beating him with 'bad' moves

The resulting shape reminds me of what can happen following this footsweep shoulder hit combo of 1 and 3 (sometimes black 2 is one line closer), where you squeeze around with a shape full of cutting points but they can't exploit them quite yet due to having to fix your own problems first. The idea of solidifying your corner in exchange for solidifying them (bonus if overconcentrating) seems to be similar.

Click Here To Show Diagram Code

[go]$$W
$$ . . . . . . . . . . |
$$ , . . . . . 2 . . . |
$$ . . . . . . . . . . |
$$ . . . . . . 0 . . . |
$$ . . . . . . 3 4 . . |
$$ . . . . . . 5 X . . |
$$ . . . . . 9 8 6 1 . |
$$ , . . . . . O 7 . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ . . . . . . . . . . |
$$ --------------------+[/go]

I also have an interesting LeelaZero move to share.
We are all familiar with white 1, mostly regarded as a "handicap go" move, but sometimes seen in pro games too.
In my database, hane at "a" is the most often seen reply (92% of the cases), followed by the more aggressive "b" (5%).

The first interesting bit is that playing 1 with white seems to be a favorite of LZ's, while in human pro games it comes up only in about 3% of the cases in the pattern below.
This seems to be LZ specific, since I haven't seen AG do that so often.

Click Here To Show Diagram Code

[go]$$W
$$ . . . . . . . . . . . |
$$ . , . . . . . , . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . a . . . |
$$ . . . . . . . 1 X . . |
$$ . . . . . . . b . . . |
$$ . O . . . . . O . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ ----------------------+[/go]

The really interesting move though is black 2 below, which never appeared in any pro game in my database! I already saw this in so many LZ analysis sessions, that I realize it is a fundamental new way for LZ to deal with this sort of situation, and it seems to be firmly convinced that it is a good move, on par with the most common human way to reply to 1, which is to play hane at "a".

On my computer, black 2 below is favored by LZ over "a" initially (by "number of visits") for a long while, and it's only when both "a" and 2 get about 30K visits that "a" and 2 get comparable with each other (and take turns turning blue ).

Click Here To Show Diagram Code

[go]$$W
$$ . . . . . . . . . . . |
$$ . , . . . . . , . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . a . . . |
$$ . . . . . . . 1 X . . |
$$ . . . . . . . . 2 . . |
$$ . O . . . . . O . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ ----------------------+[/go]

See below for LZ's favorite local continuation:

Click Here To Show Diagram Code

[go]$$W
$$ . . . . . . . . . . . |
$$ . , . . . . . , . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . .8. . . |
$$ . . . . . . . 6 . . . |
$$ . . . . . . . 1 X . . |
$$ . . . . . . . 3 2 . . |
$$ . O . . . . . O 4 . . |
$$ . . . . . . . 7 5 . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ ----------------------+[/go]

See the next diagram for what is typically LZ's line of play for the human hane with 2.
The most common human move is "a" instead of "4", but LZ seems to prefer kosumi at 4 because it's easier to make eyes locally (with a follow-up for black at "b").

Click Here To Show Diagram Code

[go]$$W
$$ . . . . . . . . . . . |
$$ . , . . . . . , . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . 2 . . . |
$$ . . . . . . 3 1 X . . |
$$ . . . . . . . . a 4 . |
$$ . O . . . . . O . . . |
$$ . . . . . . . . 5 b . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ ----------------------+[/go]

The really interesting move though is black 2 below, which never appeared in any pro game in my database! I already saw this in so many LZ analysis sessions, that I realize it is a fundamental new way for LZ to deal with this sort of situation, and it seems to be firmly convinced that it is a good move, on par with the most common human way to reply to 1, which is to play hane at "a".

I have 17 examples (Honinbo Shugen was the first), but 10 of those have appeared this year, so clearly under the influence of the bots. Well spotted!

John Fairbairn wrote:

The really interesting move though is black 2 below, which never appeared in any pro game in my database! I already saw this in so many LZ analysis sessions, that I realize it is a fundamental new way for LZ to deal with this sort of situation, and it seems to be firmly convinced that it is a good move, on par with the most common human way to reply to 1, which is to play hane at "a".

I have 17 examples (Honinbo Shugen was the first), but 10 of those have appeared this year, so clearly under the influence of the bots. Well spotted!

I see, if I reduce the search area to 9x9 (excluding the middle side handicap points, so ignoring white's extension on the lower side in my example), I got 5 matches. It looks like I don't have Honinbo Shugen's game in my collection.

Here are 2 examples of this bot joseki from my LZ adventures on Fox, the immediate angle play followup is a good shape to learn.

viewtopic.php?p=239162#p239162

Uberdude wrote:Here are 2 examples of this bot joseki from my LZ adventures on Fox, the immediate angle play followup is a good shape to learn.

viewtopic.php?p=239162#p239162

Oh, so it plays the attachment on the 4th line out of the blue, just based on the thickness in the opposite corner. I guess it tries to develop the whole side, without any mid-side stone present, if I were to guess a purpose for this move...

The immediate angle play is very interesting too, I saw that as a continuation even in the position I was analyzing. That seems part of LZ's joseki already.

sorin wrote: Oh, so it plays the attachment on the 4th line out of the blue, just based on the thickness in the opposite corner. I guess it tries to develop the whole side, without any mid-side stone present, if I were to guess a purpose for this move...

Yes, it's a move with wide applicability that's back in my "normal moves" vocabulary. The old lesson of "don't play the attach, 1-2 is a bad exchange" always felt a bit of an oversimplification to me as 4-5 also feels sad losing 3-3 options, and the way bots like the kosumi at a over 4 seems to emphasise this point. Actually I remember Guo Juan pre-AI saying something like "don't play that attach until you are 5d, as before then you can't identify the situations it's good so better not to play it". I quite like the attach and crosscut overconcentration technique though, I did it twice in one of my British championship games for the lols.

Click Here To Show Diagram Code

[go]$$W
$$ | . . . . . . . . . .
$$ | . . X , . . . . . ,
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . 2 . . . . . .
$$ | . . X 1 3 . . . . .
$$ | . a 4 . . . . . . .
$$ | . . 5 O . O . . . ,
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ +--------------------[/go]

Also I wonder if there's any bias with bots of white playing it more than black, as a move which settles the shape and simplifies the game it might make komi relatively more valuable.