It is currently Fri Oct 18, 2019 12:12 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 5 posts ] 
Author Message
Offline
 Post subject: Pro comments on new go ideas (Game 1)
Post #1 Posted: Thu Sep 06, 2018 5:14 am 
Oza

Posts: 2379
Liked others: 15
Was liked: 3450
In a different thread, uberdude discussed some AI move and said he would like to see a pro's comment on it. That made me realise that with the demise of (English) Go World, not to mention the atrophy of the book market, pro commentaries on recent games in English are very hard to come by, especially in written form. In fact long commentaries seem to be getting rarer in Japanese magazines.

Although I'm sure this is mainly a matter of paper economics and the internet, I suspect a further strand may be that in the light of AI activities pros are now a little unsure what to say. And if they think they have discovered something new, I imagine they want to keep it closely guarded.

But odd comments do emerge, and I thought it might be useful to present examples as I come across them.

For this to work, we will need to have people here embellishing what the pro said. As I say, the pros seem reluctant to go into great detail, for whatever reason, and we need to tease out for ourselves where the new ideas are going. The pros are just giving us signposts.

The first example is from the recent (June 2018) 5th Aidu Chuo Hospital Women's Tachiaoi Cup title match, Game 3. Fujisawa Rina was Black here (and went on to win both the game and the match). Her opponent was Xie Yimin. Time limits were a decent 3 hours each. Source is (Japanese) Go World, which of course has a full commentary, and much else besides.



TRANSLATION

The lower left joseki is a modern one. Previously the result up to Black 9 was considered excellent for Black, but nowadays the thinking has changed. Against Black's approach at 13, "I expected she would pincer" (said Fujisawa) and as expected Xie did indeed pincer at 14. If White 14 is at A (E17), Black 16 and White 15 will follow, then Black will have a perfect extension to B (J17).

Fujisawa made another approach on the other side at Black 15. If Black 15 invades at 17, the joseki of White 18 then Black 19 can be expected, but then there will be a feeling that the extension on the side (Black 9) is too low, and Fujisawa said, "If we can continue up to Black 21, the line in the actual game seems better." At any rate she had made a steady start.

White's invasion at 22 was unexpected. Fujisawa said she had expected C (P3) instead of White 22, and then Black D (Q5), White E (N4).

Black's construction move at 25 was on a good spot. As regards the alternative sequence of Diagram 1 [see variation line], Fujisawa said, "Answering the 3-3 invasion felt like being forced and so I couldn't play that."

TRANSLATION ENDS

My own comment is that Black seems to have thought she needed a less usual way to play in the upper left because the joseki in the lower left was not as good as previously thought (though still not bad).

Calling Black 25 (an AI-ish looking move) a construction move (kamae) instead of a shimari suggests new thinking to me (but see below).

LeelaZero does indeed opt for the new joseki for White 6 in the lower left, but does not show Black 9 as one of its several candidate moves. It prefers several approaches in the upper right, or the kamae at 25. The actual Black 9 registers a winrate drop, but only of 0.6%. Then although LZ likewise moves to the top right for move 10, it looks only at the 3-3 invasion there (the game choice registers a 0.4% drop). LZ disliked Black 13 a little but once it was played it agreed with the choice of Black 15 to follow, but White seem to have fallen behind by blocking on the wrong side with 18. LZ also frowned a little on White 22 and, like Fujisawa, thought White should have approached in the lower right. But for Black 25, LZ thinks Fujisawa should have either played the variation way, or G16 (and Black is then ahead with a winrate of about 52%). This seems to go against recent thoughts here that AI puts a heavy emphasis on empty or half-empty corners (i.e. on kakaris or shimaris), but in fact LZ thinks 25 is a significant mistake: -5% (and then only after 15 minutes pondering - initially it was over -8%).

That's where would really hear Fujisawa's thoughts, I imagine, but for the time being we will have to make do with what we've got. If these bite-sized pro comments do encourage others to join the discussion, I'll try to post more when I see them, but with each game in a different thread.


This post by John Fairbairn was liked by 5 people: Calvin Clark, Gomoto, sorin, Zenit, zermelo
Top
 Profile  
 
Offline
 Post subject:
Post #2 Posted: Thu Sep 06, 2018 5:32 am 
Honinbo
User avatar

Posts: 8678
Location: Santa Barbara, CA
Liked others: 323
Was liked: 2011
GD Posts: 312
Thanks, John.
Quote:
may be that in the light of AI activities pros are now a little unsure what to say.
( my emphasis )
Seems to me this is being very generous to certain pros.
Since AG emerged in 2015, there have been numerous live and recorded streams ( unfortunately, not on youtube which is blocked there, but on proprietary software ) where top Chinese pros frankly admitted they had no idea what's going on with some AI moves and board positions. ( I don't know the situation with Korean and Japanese top pros. )

My take is these top Chinese pros are being honest and sincere;
the corollary: pros who have issued disparaging remarks against top AI moves are also being honest, though IMO suffering from the Dunning-Kruger effect.

Top
 Profile  
 
Offline
 Post subject: Re: Pro comments on new go ideas (Game 1)
Post #3 Posted: Thu Sep 06, 2018 9:11 am 
Tengen

Posts: 4803
Liked others: 0
Was liked: 649
Already during the EGC 2017 (last year), professional players often lacked courage to express their own opinion and resorted to speculating what AI might choose. OTOH, this mostly affects pros relying much on subconscious thinking. Pros or I who use reasoning, careful reading, careful analysis etc. have not been discouraged by AI.

(Book market is an interesting topic but for the Books forum.)

Top
 Profile  
 
Offline
 Post subject: Re: Pro comments on new go ideas (Game 1)
Post #4 Posted: Thu Sep 06, 2018 9:56 am 
Tengen

Posts: 4277
Location: North Carolina
Liked others: 448
Was liked: 702
Rank: AGA 3k
GD Posts: 65
OGS: Hyperpape 5k
John Fairbairn wrote:
LZ also frowned a little on White 22 and, like Fujisawa, thought White should have approached in the lower right. But for Black 25, LZ thinks Fujisawa should have either played the variation way, or G16 (and Black is then ahead with a winrate of about 52%). This seems to go against recent thoughts here that AI puts a heavy emphasis on empty or half-empty corners (i.e. on kakaris or shimaris), but in fact LZ thinks 25 is a significant mistake: -5% (and then only after 15 minutes pondering - initially it was over -8%).
Perhaps this too obvious to state, but it strikes me how rapidly the AI changes its focus. Saying the top right has become hot is very natural, but it's a bit of a surprise to me that G16 becomes more important than the lower right.

Side note for EdLee:
EdLee wrote:
My take is these top Chinese pros are being honest and sincere;
the corollary: pros who have issued disparaging remarks against top AI moves are also being honest, though IMO suffering from the Dunning-Kruger effect.
I believe it's not Dunning-Kruger. Dunning-Kruger applies to people who are among the worst at a skill, not those like professionals who know quite a bit, but not as much as someone else (or some computer).

_________________
Occupy Babel!

Top
 Profile  
 
Offline
 Post subject:
Post #5 Posted: Fri Sep 07, 2018 3:12 pm 
Honinbo
User avatar

Posts: 8678
Location: Santa Barbara, CA
Liked others: 323
Was liked: 2011
GD Posts: 312
Hi hyperpape,
Thanks for the POV.
The DK studies also looked at bias or blindspots on the high-end side, but that's another story. When AG first came out, many pros ( and amateurs alike ) definitely exhibited the DK effect in that they didn't know the first thing about coding but was very happy talk trash about AG. If only someone taped the Kasparov-Letterman interview where Kasparov said he didn't think the computer would ever beat top humans. Fast foreward to 2014 and we got verbatim quotes from top go pros. Some Go pros today are still talking nonsense about the bots, which is classic DK. ( Though usually not from the newer, younger generations. )

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 5 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