It is currently Mon Apr 06, 2020 11:24 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 10 posts ] 
Author Message
Offline
 Post subject: Hazama - new AI insight?
Post #1 Posted: Tue Oct 29, 2019 6:53 am 
Oza

Posts: 2481
Liked others: 16
Was liked: 3587
We have to forage relentlessly for scraps if we are to delve into some understanding of the insights pros are getting from AI bots. Here is one such scrap from the current Oza title match, Game 1. Iyama is White against Shibano.



Early play on the left side was seen as evidence of home preparation (obviously AI enhanced!) because of the speed of play.

But the most interesting comment was from Ohashi Naruya, a 7-dan pro from the same stable as Nakamura Sumire. He said the triangled last move, rather then the usual press at A, was one that was difficult for a human to play - but was clearly AI induced! Humans are reluctant to play the hazama shape this creates. Maybe that reluctance explains why this shape is sometimes called an elephant jump. Sensible elephants can't jump. Sensible players can't make an ungainly hazama shape. The weakness at B (the elephant's eye) is just too debilitating.

Of course we already know that there are special occasions when the hazama is, if not necessarily good, at least interesting - a sort of sacrifice tactic to get the opponent to play on the eye and get rolled on by Nellie on at least one side.

However, the AI use of hazama seems somewhat different from that and it would further seem that Shibano may have worked out why. Or at least is confident of now testing his lab research in the field.

And maybe Iyama has seen the same light? His next move was at C. This hazama is an old joseki ploy, of course, but I'm guessing strongly that Iyama played it less out of respect for old joseki books and much more because of its overtones of AI strategy.

If you wonder what these are, I can offer another scrap or two by quoting from my own Go Wisdom entry on "Direction of Play":

Quote:
... Sometimes, during fighting, you wish to change direction. There are good and bad ways to do this. The good ways include mainly these tactical manoeuvres: cut, nobi, tsuke, cap, inducing moves (→ momentum), hazama, hasamitsuke, sabaki and peeps.


In the early fuseki, as distinct from fighting, the hazama may not actually change the direction, but it does seem to offer a better choice of future directions than many other moves


This post by John Fairbairn was liked by 3 people: Gomoto, gowan, xela
Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #2 Posted: Tue Oct 29, 2019 7:52 am 
Judan

Posts: 6362
Location: Cambridge, UK
Liked others: 369
Was liked: 3468
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Regarding Iyama's elephant jump at C, there was a very interesting AI insight on this. Shibano answered with the kosumi at p4, and to defend against cutting at q6 Iyama played the p5 attachment. This then became a ko, his o16 ko threat at the top wasn't good enough (Shibano and AI both ignored) and he fell behind in the ko trade. Bots say that the p5 attachment should have simply connected in the middle of the jump with the double kosumi at q6, a shape we tend to think of as rather flabby and low in eyeshape and inefficient. Alexander Dinerstein posted about this on facebook and there's quite some discussion there: https://www.facebook.com/photo.php?fbid ... 8134524144


This post by Uberdude was liked by: Kirby
Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #3 Posted: Tue Oct 29, 2019 8:45 am 
Gosei

Posts: 1539
Location: Earth
Liked others: 571
Was liked: 249
I have looked at the two space low pincer just yesterday. It can get quite complicated. I looked for any easy way off, but I think I just have to accept it develops into difficult fighting. A nice example is this game (KataGo thinks the Hazama looses about 2 points versus the press. To play at the eye of the Hazama is not the right response by the way according to Kata.):

Park Jinsol 9p - Zhou Ruiyang 9p, W+Resign (Komi 6.5)
24th Samsung Cup, preliminary, 2019-07-03




two variations (move 13)


This post by Gomoto was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #4 Posted: Tue Oct 29, 2019 8:53 am 
Honinbo

Posts: 9540
Liked others: 3016
Was liked: 3172
Gomoto wrote:
To play at the eye of the Hazama is not the right response by the way according to Kata.):


It seldom is. According to humans. ;)

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins
----
Many are cold, but few are frozen.


This post by Bill Spight was liked by: Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #5 Posted: Tue Oct 29, 2019 10:22 am 
Judan

Posts: 6362
Location: Cambridge, UK
Liked others: 369
Was liked: 3468
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Bill Spight wrote:
Gomoto wrote:
To play at the eye of the Hazama is not the right response by the way according to Kata.):


It seldom is. According to humans. ;)


This needs the caveat that the elephant jump was a good, or at least ok, move, which is generally the case if you pattern search pro games. When DDKs play the elephant jump it is usually a bad move, and playing in the eye to split it is good.

P.S. A herd of elephants from Lee Sedol and Mok Jinseok: http://ps.waltheri.net/database/game/53750/


This post by Uberdude was liked by 3 people: ez4u, Gomoto, xela
Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #6 Posted: Tue Oct 29, 2019 10:42 am 
Gosei

Posts: 1539
Location: Earth
Liked others: 571
Was liked: 249
Bill - Uberdude dialectic

I love it

Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #7 Posted: Tue Oct 29, 2019 3:45 pm 
Lives with ko

Posts: 156
Liked others: 16
Was liked: 32
Uberdude wrote:
Bill Spight wrote:
Gomoto wrote:
To play at the eye of the Hazama is not the right response by the way according to Kata.):


It seldom is. According to humans. ;)


This needs the caveat that the elephant jump was a good, or at least ok, move, which is generally the case if you pattern search pro games. When DDKs play the elephant jump it is usually a bad move, and playing in the eye to split it is good.

P.S. A herd of elephants from Lee Sedol and Mok Jinseok: http://ps.waltheri.net/database/game/53750/


As a SDK I certainly like the psychological implications of playing an elephant jump. Knowing that playing in the eye is bad if the jump was good, usually the opponents are too scared to play there, even if it happens to be the correct play. :)
Thanks for posting the elephant herd game. Now I'd be too scared to play like that.

_________________
If something sank it might be a treasure. And 2kyu advice is not necessarily Dan repertoire..

Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #8 Posted: Wed Nov 06, 2019 4:59 am 
Oza

Posts: 2481
Liked others: 16
Was liked: 3587
After herds of elephants, now a stampede:

https://www.youtube.com/watch?v=7DGNpwcH3h0&t=8

There's an American touch at the end.

Lang may yer lum reek!


This post by John Fairbairn was liked by: xela
Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #9 Posted: Wed Nov 06, 2019 6:04 am 
Lives in gote

Posts: 413
Location: Adelaide, South Australia
Liked others: 162
Was liked: 203
Rank: Australian 2 dan
GD Posts: 200
John Fairbairn wrote:
After herds of elephants, now a stampede...

I did not expect that! Interesting combination of songs. The music brought back a few memories.

Top
 Profile  
 
Offline
 Post subject: Re: Hazama - new AI insight?
Post #10 Posted: Wed Nov 06, 2019 7:45 am 
Honinbo

Posts: 9540
Liked others: 3016
Was liked: 3172
OT. Some of us remember elephant jokes. ;)

Why do elephants have flat feet?

From jumping out of trees.

Why do ducks have flat feet?

From stamping out forest fires.

Why do forest rangers have flat feet?

From stamping out burning ducks.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins
----
Many are cold, but few are frozen.


This post by Bill Spight was liked by 2 people: ez4u, gowan
Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 10 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group