It is currently Sat Aug 19, 2017 9:58 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 6 posts ] 
Author Message
Offline
 Post subject: Ohashi Hirofumi explores AlphaGo's insights on thickness
Post #1 Posted: Wed Mar 01, 2017 11:58 am 
Gosei

Posts: 1830
Liked others: 11
Was liked: 2724
Ohashi Hirofumi has published excellent commentaries in Go World on several of the games played by AlphaGo as Master in the 60-game series at the turn of the year - and promises more to come. The commentaries are decent sized, easy to understand and above all insightful as regards the ways AlphaGo is making pros rethink the game. We are lucky to have had already some similar Chinese and Korean commentaries, but this is the first substantial set I have seen in Japanese. A notable feature of Ohashi's version is that he compares games with similar openings very well.

One position, below, especially caught my eye. It is from Game 23. AlphaGo is Black, playing Kim Cheong-hyeon.

Click Here To Show Diagram Code
[go]$$W
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O O . . . . . . . . . . . . . . . |
$$ | . X O X . O O . O . . . . . . . X . . |
$$ | . X X O . . X . . , X . . X . X . . . |
$$ | . X O O . . . . . . . . . . . O . . . |
$$ | . . X X . . . . . . . . . O O . O . . |
$$ | . . . . . . . . . . . . . . X . . . . |
$$ | . . . . . . . . . . . . . . . X . O . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . Q . . . . . . . . . . . . . . . . |
$$ | . X Q . . . . . . . . . . . . . . . . |
$$ | . X Q . . . . . . . . . . . . . . . . |
$$ | . X Q . a . . . . . . . . . . . . . . |
$$ | . . X Q . . . 2 . 1 . . . X . , X . . |
$$ | . . X Q . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


My fascination with this position can be traced in previous threads here. We were discussing a position where I raised a query about a shape that everyone else was treating as thick. I noted that I had recently seen several commentaries where a shape I (and evidently just about everyone here) would consider to be thickness was regarded by pros as "not really thickness." I'm not sure I convinced anyone, and it seems only a very few even paused for thought, but I have since seen further examples and now we have a case where the goddess AlphaGo apparently thinks something is "not really thickness."

Here is the relevant portion of Ohashi's commentary:

Quote:
White extended to 1 only to see Black invade nevertheless at 2. Since overall Black is ahead on profit, it is common sense for White to want to extend as far as possible to make his triangled stoneswork, but faced now with the Black peep at 'a', it's impossible for him to know which groups are the strong ones. Black is asserting that, "The white triangled stones are NOT thickness!" Even so, it would have been hard to make a narrower extension than 1 and so end up with over-concentrated shape. This sort of intuition (!?) about thickness is, I believe, the secret of Master's strength.


Needless to say there are very similar insights about thickness elsewhere in his commentaries, and likewise in comments by other pros about AlphaGo. Since, as I have said, I have seen an increasing number of comments of this tenor recently, even outside the AlphaGo ambience, it may be that pros had already discovered truths about the phenomenon of a doorstep sandwich of thickness dissolving into panada even before AlphaGo, and they are welcoming AI's confirmation of their suspicions.

Incidentally, the human in Game 23 has previously been given in this forum as Kim Chi-seok. Ohashi is a pro well plugged into the internet scene and he also writes regularly for Go World on AI in go. I'd be surprised if both he and Go World got the name wrong. In addition, the handles used in the two games so far attributed here to Kim Chi-seok were quite different. But that just stacks the odds in favour of Cheong-hyeon - it's not proof. Does anyone have hard evidence in favour of Chi-seok?


This post by John Fairbairn was liked by 4 people: Drew, jeromie, jptavan, swannod
Top
 Profile  
 
Offline
 Post subject: Re: Ohashi Hirofumi explores AlphaGo's insights on thickness
Post #2 Posted: Wed Mar 01, 2017 1:34 pm 
Tengen

Posts: 4405
Liked others: 0
Was liked: 585
The white wall is 0-connected - not too bad. The wall and the extension seem to be -1-connected (minus!), which is sometimes possible for a wall and its extension. However, this wall also needs life potential. Life by running is unattractive here so life by eyespace in between wall and exrension is necessary here. Reading must study whether the black invasion stone is 0-alive or -1-alive but it is flexible and can be sacrificed for some separating or white territory preventing group. It is not appparent that White could confine the then live black invasion group and build a white wall above as sufficient compensation in sente.

For the white extension to be good at the wide space, it must be an indirect connection but, as explained, the black invasion leads to White's disadvantage when White tries to justify the extension at least as indirect connection.

Hence studying AlphaGo is not necessary to assess the extension as improper.

Top
 Profile  
 
Offline
 Post subject: Re: Ohashi Hirofumi explores AlphaGo's insights on thickness
Post #3 Posted: Wed Mar 01, 2017 3:48 pm 
Judan

Posts: 7464
Liked others: 1358
Was liked: 1147
KGS: Kirby
Tygem: 커비라고해
Each situation is different. The more skilled the player, the better they know how to exploit aji and weaknesses in a position that a weaker player considers "thick".

Some positions are truly thick, having no refutation. The more skilled the player, the better they can recognize the difference.

It's the same with other concepts on the go board - life and death, for example. Maybe a player thinks a group is alive. A stronger player may know a way to kill it.

Thickness is a tougher concept, but similarly, players can better assess as they gain strength.

_________________
Discipline is remembering what you want. -David Campbell

Top
 Profile  
 
Offline
 Post subject: Re: Ohashi Hirofumi explores AlphaGo's insights on thickness
Post #4 Posted: Wed Mar 01, 2017 4:14 pm 
Tengen

Posts: 4068
Location: Cambridge, UK
Liked others: 145
Was liked: 2000
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
As the old proverb says "Walls may have ears, but they don't have eyes!"

Top
 Profile  
 
Offline
 Post subject: Re: Ohashi Hirofumi explores AlphaGo's insights on thickness
Post #5 Posted: Wed Mar 01, 2017 4:19 pm 
Judan

Posts: 7464
Liked others: 1358
Was liked: 1147
KGS: Kirby
Tygem: 커비라고해
For what it's worth, various Korean sources suggest that AlphaGo played against Kim Junghyun in game 23 (abc2080(P) aka 야왕).

I don't know why it was thought to be Kim Jiseok. I think Kim Jiseok probably played against AlphaGo in game 41 as 'airforce9'.

Although, it's difficult in general to have hard evidence on the true identities of these users, unless you work at Tygem or something.

_________________
Discipline is remembering what you want. -David Campbell

Top
 Profile  
 
Offline
 Post subject: Re: Ohashi Hirofumi explores AlphaGo's insights on thickness
Post #6 Posted: Wed Mar 01, 2017 4:48 pm 
Dies in gote

Posts: 60
Liked others: 5
Was liked: 19
John Fairbairn wrote:
Incidentally, the human in Game 23 has previously been given in this forum as Kim Chi-seok. Ohashi is a pro well plugged into the internet scene and he also writes regularly for Go World on AI in go. I'd be surprised if both he and Go World got the name wrong. In addition, the handles used in the two games so far attributed here to Kim Chi-seok were quite different. But that just stacks the odds in favour of Cheong-hyeon - it's not proof. Does anyone have hard evidence in favour of Chi-seok?
Thanks for reminding me. My SGFs have always been as correct as possible, but I did not always update the web page for every SGF change. Now this data has been stable for a while, and I made the HTML reflect the SGF. See AlphaGo. The README there says
Quote:
The opponent names here are conjectured. We follow reddit.
Other lists have Kim Jiseok instead of Kim Junghyun for game T23.sgf and Lee Donghun instead of Yun Chanhee for game T26.sgf.
Note that the numbering here is the chronological one. Elsewhere games #12 and #13 are often interchanged.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 6 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: Bing [Bot] and 5 guests


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