It is currently Fri Mar 29, 2024 8:52 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 29 posts ]  Go to page 1, 2  Next
Author Message
Offline
 Post subject: More Shibano AI magic
Post #1 Posted: Mon Jun 22, 2020 2:54 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
I think this will appeal especially to Andrew with his AI Gospel, although it may mean him getting out his chisel again and some fresh slabs of stone.

In the beginning was the Word. Well, yes - but what if the word was the wrong one? It's a bit like setting off on a journey with the map upside down.

In his latest musings on why certain fuseki plays have disappeared almost overnight, thanks to AI, Shibano Toramaru has given me the first glimmer as to what may be going on with all these shimaris from the hoshi stone we keep seeing. It seems as if they may not be shimaris after all.

His starting point is the position below, and he is looking at why the triangled extension has disappeared. (In passing, I was surprised to see the one-space shimari in the lower right. I was expecting A. In fact when I checked with katago it seems this tight shimari is the new black. Bots really are fickle!)



I'm not going into Shibano's specifics on the extension, but, amongst other things, either (a) the extension is liable to end up on a "non-urgent point" once Black plays an approach move on the lower left side (i.e. White would be extending from a safe position) or (b) Black can invade at the 3-3 (D) and give White thickness. That (b) would have bothered pros pre-AI, and may still leave pros like Shibano a bit queasy (he calls this development "remarkable") but he notes that bots don't mind and instead then play moves like E. He sums this up in general terms by saying that when bots give the opponent thickness they respond by playing a fast-paced game.

But back to shimaris and non-shimaris. Shibano explains why moves like the triangled extension were played in the first place. They were following the mantra: first occupy the open corners, second play a shimari or kakari, and third extend.

But he also observes that this proverb arose in the days when nobody played on the 4-4 point. And it doesn't really tell you what to do with a 4-4. An idea has developed that a 4-4 finishes off the corner in one move, and so once you have played it you can extend. This is what is generally thought of as a fast-paced game. But actually it seems that what AI is teaching us is that a 4-4 needs a shimari. So, while we can still talk about a shimari from a 4-4 stone, we need to recognise that it's something quite different from a shimari from komoku. For one thing it's territorially deficient.

To highlight this difference, a term Shibano also uses of AI-favoured moves like B and C (instead of the wide extension) is kamae - a construction move. That's useful because it brings in a whole host of different associations. As one example, a kamae can of itself be thought of often as an extension.

Kamae moves are mentioned a lot in Japanese commentaries, though not usually this early in the game, but have not been highlighted in the west. I suspect that may be because each translator renders it in a different way, so it just disappears into the morass of other words. Robert Jasiek is the only person I can recall picking up on it here (apart from myself, of course) but I haven't actually read his books, and from what he has said on L19 I think his version of "construction" may be rather different.

In fact even Japanese players can require explanations of kamae. You may be surprised at the definitions in, for example, the Nihon Ki-in's terms dictionary. Two examples in the diagram below:



Starting with the definition of kamaeru as "to set oneself up in good order* by making shape" it says that the White triangled stone in the lower left is a one-space shimari (ikkenjimari), but given the idea that it is setting things in order, we call it kamaeru." For the example on the right it says the kamae triangled move is superior to the sagari down towards the edge because a White peep at the triangle wold be too severe. [*a bit of a wishy-washy phrase but the term behind it, seibi, refers to the sort of thing you do as preventive maintenance, for example servicing your car to keep it in good running order to avoid repair bills]

Hayashi Yutaka also gives the lower-right example but adds also the one in the upper left. And he gives the definition of kamae as "a move that settles oneself in order to avid incursion by the opponent." Of course kamaeru is a normal word in everyday Japanese and so is used in go in extended ways with all its other associations, but that's a good starting point.

Obviously Shibano is talking to Japanese people in their own language, and can make assumptions and use associations that don't apply to us directly. But for us my interpretation is that at the very least we have to recognise that there are different kinds of shimari. In Robert's style you might call them Shimari I and Shimari II (and maybe Shimari III if the Nihon Ki-in example above disconcerts you), and a shimari from a 4-4 stone is a unique type, possibly deserving its own name. In addition, whatever you call it, in practice you should try to view it through the prism of a kamae move when you see it in AI play.

Furthermore, there may well be a Fast Play I and a Fast Play II. The "fast play" referred to be E at the top seems to be quite different from the wide-extension kind of fast play, not least because one of its attributes may be an emphasis on combating thickness. Maybe that demands a new term, too.

Over now to our Master Chiseller!


This post by John Fairbairn was liked by 2 people: gowan, sorin
Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #2 Posted: Mon Jun 22, 2020 4:18 am 
Judan

Posts: 6087
Liked others: 0
Was liked: 786
Construction as an explicit class of move types / meanings I have not "picked up" but introduced on my own.

I have also explained that moves can have several meanings. Therefore, types I and II names might not be needed when two or more meanings can be assigned to some moves. Numbers are useful when words would not clarify subtypes. E.g., capturable-2 (or capturable (of) type 2, if you prefer a verbose form) I invented because text shorter than the full definition does not really clarify the term, let alone distinguish it from capturable-3. For capturable-1, "capturable of the type under the stones" would have worked but still be ambiguous and unnecessarily clumsy. When capturable-2 is needed, also using capturable-1 is consistent.

Hence, there are cases when numbered terms make sense but not all finer terms need numbers.

Fast versus slow or far versus narrow or fast versus territorial extension / enclosure / construction move will do.

What do we need Japanese (kamae) for? As a non-Japanese speaker, I dislike words hiding their English meaning. Of course, if you think to have met a concept with more nuanced meaning than construction, feel free to reveal it now or after any necessary further study of sources. But then please do not bury it under its Japanese name to immediately hide your discovery! Go terminology must not be alchemy but deserves speaking terms whenever possible.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #3 Posted: Mon Jun 22, 2020 5:05 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
John Fairbairn wrote:
In his latest musings on why certain fuseki plays have disappeared almost overnight, thanks to AI, Shibano Toramaru has given me the first glimmer as to what may be going on with all these shimaris from the hoshi stone we keep seeing. It seems as if they may not be shimaris after all.


That is my feeling, too. But people call them enclosures, so there you go.

John Fairbairn wrote:
But back to shimaris and non-shimaris. Shibano explains why moves like the triangled extension were played in the first place. They were following the mantra: first occupy the open corners, second play a shimari or kakari, and third extend. But he also observes that this proverb arose in the days when nobody played on the 4-4 point. And it doesn't really tell you what to do with a 4-4. An idea has developed that a 4-4 finishes off the corner in one move, and so once you have played it you can extend. This is what is generally thought of as a fast-paced game. But actually it seems that what AI is teaching us is that a 4-4 needs a shimari.


As a rule of thumb.

John Fairbairn wrote:
To highlight this difference, a term Shibano also uses of AI-favoured moves like B and C (instead of the wide extension) is kamae - a construction move. That's useful because it brings in a whole host of different associations. As one example, a kamae can of itself be thought of often as an extension.

Kamae moves are mentioned a lot in Japanese commentaries, though not usually this early in the game, but have not been highlighted in the west. I suspect that may be because each translator renders it in a different way, so it just disappears into the morass of other words.


Just a footnote. Volume 3 of the Sakata no Go series, which deals with the opening, is Ishi no Kamaekata. :)

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #4 Posted: Mon Jun 22, 2020 9:02 pm 
Lives in gote

Posts: 388
Liked others: 416
Was liked: 198
Personally, I don't like to think in too many categories (such as shimari 1, shimari 2, etc..).

The commonality I see between the 3-4 based shimari and the 4-4 based one is that they both prevent the opponent from playing a very desirable move (a keima approach).

In that sense, the 4-4 based shimari is very different compared to extending to the middle of the side from 4-4, since the large extension doesn't prevent the opponent from making a keima approach to a single corner stone.

_________________
Sorin - 361points.com

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #5 Posted: Tue Jun 23, 2020 1:41 am 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
sorin wrote:
The commonality I see between the 3-4 based shimari and the 4-4 based one is that they both prevent the opponent from playing a very desirable move (a keima approach).


That has a circular scent. Isn't the keima approach is a desirable move because it prevents the owner of the corner stone to make a very desirable move, the enclosure?

As Uberdude has pointed out in his "LeelaZero opening gospel", AI thinks corners are urgent, not only the first move in the corner but also the second move. Now the question is why?

Uberdude has also pointed out that a 3-4 point is a more urgent position than a 4-4 point. In his video/post he demonstrates that idea by showing that ignoring threats against 3-4 / 4-4 consecutively end up worse for the 3-4. Why? Because the 3-4 stone gets more tightly controlled while the 4-4 can still make life with a fair amount of points. It is - and will increasingly be - about efficiency.

There's also this somewhat mysterious asymmetry about the follow-up to the 4-4: while Black can follow up on his 4-4 with a keima "enclosure" to either side, White can not only approach from either side but also, as we know all too well, make a 3-3 invasion. Why is a 3-3 reinforcement not a good move for Black? Because it's not efficient! For White to 3-3 a 4-4 makes a good piece of territory in sente. The resulting Black influence needs to be "used" and so its efficiency is not granted. (For this reason, I wouldn't be surprised if the 4-4 fell out of grace with the next generation of AI )

An enclosure from 4-4 doesn't forestall a 3-3 invasion, but it may prevent it from being efficient. That is (my interpretation of) why an enclosure from a 4-4 is urgent (but not as urgent as one from a 3-4). Rin Kaiho said "Go IS the order of play". We may now say "Go IS efficiency".

As John points out, the 4-4 + 6-3 formation should probably not be called "enclosure" because it doesn't really enclose the corner. I believe the same can be said about the AI's preferred 3-4 enclosure, the large high enclosure, which can be invaded very much like the small knight's enclosure from a 4-4. In both cases however, the invader will end up rather cramped and the outside will be strong, not only influential. "Enclosure" is a popular term because of our visual, territorial approach to Go, seeing it as a game of surrounding open points. In my case it was Minue who relieved me from that mistaken idea, showing that Go is a game of creating stable groups, ending up with more alive stones than the opponent. Territory is merely an efficient way of doing so, a second order concept.

Therefore, "reinforcement" is probably a better term than "enclosure". But as Bill says, people call it an enclosure or shimari.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #6 Posted: Tue Jun 23, 2020 2:45 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
Therefore, "reinforcement" is probably a better term than "enclosure". But as Bill says, people call it an enclosure or shimari.


That sounds VERY good to me.

A point to remember is that Shibano was equivocating about the word shimari even though to a Japanese there is absolutely nothing in it to suggest the wrong ideas of enclosure, corners or territory. They have far, far less work to do than us in correcting misperceptions. Yet still he felt the need - though he was only talking about the 4-4 point, let me stress.

To a Japanese, the meanings and association of the verb shimaru range from closing something tightly, like shutting the door or slamming in the bolt. The associations are of something firm or tight - even like a six-pack on a body builder, but in general something that has been tightened. In a more dynamic sense in sport in can be used (e.g. as shimatte iku) of playing tightly, closing down defensively in soccer - the famous old InterMilan style. There are also extended senses of caution and frugality (tight-fisted). The opposites are often useful in understanding associations. If you lack shimari in Japanese you are slovenly, straggly, weak, flabby. One phrase I love is: shimari no nai kao. It is said of a person who has the vacant look of stupidity of the village idiot. So, to become better go players, we need to lose our shimari, too, and become a bit doolally. People of my age have a head start, of course.


This post by John Fairbairn was liked by: Knotwilg
Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #7 Posted: Tue Jun 23, 2020 5:06 am 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
John Fairbairn wrote:
... closing something tightly, like shutting the door or slamming in the bolt. The associations are of something firm or tight - even like a six-pack on a body builder, but in general something that has been tightened. In a more dynamic sense in sport in can be used (e.g. as shimatte iku) of playing tightly, closing down defensively in soccer - the famous old InterMilan style.


A lock? A grip? A hold? A clasp?

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #8 Posted: Tue Jun 23, 2020 5:35 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
A lock? A grip? A hold? A clasp?


Not for me. They would be too close to other words to do with capture (e.g. kakae, hasami, geta).

But, since there are various kinds of reinforcement, for the shimaris we normally talk about it might be best to say 'corner reinforcement' (after all some people already say - wrongly, I hope we have established - 'corner enclosure').

Incidentally, the Chinese is 守角 - (prophylactically defending the corner), which is much better than enclosing the corner but does not completely overlap with the Japanese (and no there's no reason why it should). The Korean is 굳힘 which means, usually with positive associations, hardening, consolidation, strengthening, reinforcement. I think that shimaris the case for the prosecution, M'Lud.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #9 Posted: Tue Jun 23, 2020 5:37 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Knotwilg wrote:
sorin wrote:
The commonality I see between the 3-4 based shimari and the 4-4 based one is that they both prevent the opponent from playing a very desirable move (a keima approach).


That has a circular scent. Isn't the keima approach is a desirable move because it prevents the owner of the corner stone to make a very desirable move, the enclosure?

As Uberdude has pointed out in his "LeelaZero opening gospel", AI thinks corners are urgent, not only the first move in the corner but also the second move. Now the question is why?


Maybe we have it backwards. Maybe the question is why we came to believe that the second move in the corner is unnecessary. (I leave the question as rhetorical because I don't want to do the research right now. ;)) That really happened in the 20th century and late 19th century, as the 4-4 and other 4th line moves came into vogue. As Shibano points out, quoting what JF says, "An idea has developed that a 4-4 finishes off the corner in one move, and so once you have played it you can extend." That's opening dogma as I learned it, and as I could have picked up from looking at professional games. OC, the extension was seen as desirable, but conversely, there was no efficient way to make territory out of the corner, as you point out below. Whether you add a second stone at the 3-3, the 3-4, or the 3-5, the 4-4 stone ends up being awkwardly placed. Not that we don't make those plays in certain situations, OC. ;) So if you are not going to use the 4-4 to make territory, why not use it to extend and make fast development?

Knotwilg wrote:
"Enclosure" is a popular term because of our visual, territorial approach to Go, seeing it as a game of surrounding open points. In my case it was Minue who relieved me from that mistaken idea, showing that Go is a game of creating stable groups, ending up with more alive stones than the opponent. Territory is merely an efficient way of doing so, a second order concept.


I now view territory as an emergent concept, a way to score the game. ;) (Under "territory" scoring we count dead stones as well as territory. Why? Perhaps because we started off counting dead stones and developed the idea of territory later.) But I now think that the main reason for the fact that the bots want to play a second stone in the corner before extending on the side is to make a more or less stable group. Takagawa was of the 4-4 is enough school, but in his Igo Reader series, aimed at kyu players, he states that the reason for playing corners first is not because it is easy to make territory in the corner, but because it is easy to make a base. I am beginning to think that a lot of AI moves can be explained by the idea of stabilizing groups (or sacrificing them, as the case may be ;)).

Why play the 4-4 at all? OC, it is more oriented towards the center than the 3-4. It is also flexible in the direction in which it develops. Why did Dosaku like the 5-3 and not the 4-4? There may have been some prejudice against the 4-4 as old fashioned or as a handicap stone, and I think that that is why the 5-4 was sometimes played but not the 4-4. I suspect that the 5-3 offered a balance between territory, being on the 3d line, and center, being on the 5th line. If the bots like the 4-4, why not the 5-3? The answer they provide is the 3-3 invasion. It undermines the 5-3 more than it does the 4-4. From the 4-4 the reply to the 3-3 is the block, from the 5-3 it is usually the keima.

Now the keima extension from the 4-4 is something that humans came up with before the AI era. I got curious what Go Seigen called it in his 21st Century Go series. So I looked in vol. 1, which came out in 1997, and found the first example. He didn't call it anything. ;) But while I was looking for that, I noticed that in the third lesson he shows :w4: as a 5-3 and :b5: as the 3-3 invasion, and :w6: as the 4-5 keima. :o As usual, Go Seigen got there first. :cool: However, he does say that he prefers the 5-4 high approach over the 3-3 or 3-4.

Now, one virtue, it seems, of the 5-3 approach to the 3-4 is the threat of the press on the 4-5, which is one of the bots' favorite moves. It obviously builds central influence, but, as JF points out, it ain't that thick. And usually an extension does not follow right away, and it is definitely pincerable. But the press does add that second (and third) stone to the 5-3, and that, as Fujisawa Hideyuki might have said, ain't chopped blowfish liver. (Don't eat it, BTW.)

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.


This post by Bill Spight was liked by: sorin
Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #10 Posted: Tue Jun 23, 2020 6:47 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
Maybe we have it backwards. Maybe the question is why we came to believe that the second move in the corner is unnecessary.


Excellent point, Bill!

The rest of your post inspired me to do a little digging as to when the small shimari from 4-4 was first played. Onoda Chiyotaro had the honour in the first official game in 1993, and Go Seigen followed him very soon after. Thereafter, it did appear several more times in this Shin Fuseki era, but not perhaps as much as we might expect in retrospect. I have an impression that tengen and sanrensei rather hijacked the theoretical debates at the time. There is no mention of the shimari in the book Shin-Fuseki-ho, but it does highlight the point you made about the flexibility of 4-4 (along with 3-3 and 5-5 in those days, of course.

But riffling through those few early games with the shimari I was astonished to see something I don't anyone has ever mentioned before. I saw it because I was primed to see it. You may recall I "invented" the concept of a Go Seigen group. It started off as a joke after spotting a remark by Go that the areas on the side in the "armpits" of the corners were unjustly ignored, but were actually good place to make strong groups. Checking that out, I noticed that in games where such a group (the GSG group) had been made, the whole of that quadrant became a no-go area, a Chernobyl area. There was a lot more of interest in that, as I explained in The Go Companion. But guess what! If you look at those early Shin Fuseki games (start with Go vs Hashimoto 1933-04-26b), you see the same pattern - the whoe quadrant goes into lockdown. In that first Go game, he played the shimari on move 10 and there was no other moves in that quadrant till move 47 and that was almost as far from the corner as it was possible to be. Then there was no other move in that quadrant till move 66 - and that was another shimari by Go!

That was a bit of an extreme case, but I think you can see the same pattern to some degree in perhaps all the other games. The simplest way of interpreting that behaviour, I think, is that the opponent was, per standard theory, staying away from strength. In other words, the shimari was indeed seen as a reinforcement.

Incidentally, in the Onoda game (1933-03-22), his opponent Miyasaka played an AI shoulder against the shimari on move 28, but that was a probe (AI again!) connected with a shape on the other side. The rest of the quadrant, the bulk, remained in lockdown until move 85.

Since Onoda is in the frame now, I have a feeling you might be the best to appreciate his eventual comment on Shin Fuseki, Bill: "If we assume that recent fusekis are rational, and everyone starts to use them, then a single form will result. Assuming that, go (as regards the fuseki) will become much narrower than existing fusekis. I think in the end it will go up a blind alley. The result of that will surely be that we will return to the old ways."

A bit of a non sequitur, perhaps, but a nice evocation of the human spirit!


This post by John Fairbairn was liked by 2 people: Bill Spight, gowan
Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #11 Posted: Tue Jun 23, 2020 7:47 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Thanks, John. Verrrry interesting. :)

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #12 Posted: Tue Jun 23, 2020 8:56 am 
Gosei

Posts: 1625
Liked others: 542
Was liked: 450
Rank: senior player
GD Posts: 1000
The discussion so far in this thread seems relevant for the so-called Kobayashi Fuseki, which was very popular10 to 20 years ago. In particular the second version:
Click Here To Show Diagram Code
[go]$$B Kobayashi fuseki (second version)
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


Now it seems White would reinforce the lower left corner but I think I recall in the past White often played below the star point on the lower side.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #13 Posted: Tue Jun 23, 2020 12:53 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Kobayashi II

Click Here To Show Diagram Code
[go]$$Wcm6 :w6: and alternatives
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . a . . 1 . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . k . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . h . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . O . j . . . d i . . . . , . . . |
$$ | . . . . . c g . . b e . . . f X . . . |
$$ | . . . . . . . . . . . . . . . l . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Elf's top choice is :w6:

Code:
Alternatives and estimated winrate loss by comparison with :w6:

[b]a[/b]     2%
[b]b[/b]     2½%
[b]c[/b]     3%
[b]d[/b]     3½%
[b]e[/b]     4½%
[b]f[/b]     4½%
[b]g[/b]     5%
[b]h[/b]     5%
[b]i[/b]     5%
[b]j[/b]     6%
[b]k[/b]     7½%
[b]l[/b]    10½%


OC, we do not know the margin of error for Elf's winrate estimates, or any other bot's estimates. My guess is that those for a - d are within the margin of error, those for j-l are large enough to indicate actual errors, and the rest are on the cusp.

Note that sanrensei is most likely a minor error. It is true that the 7½ pt. komi favors White, but, as is usually the case, correct play for White appears to have a narrower path. White is less likely to get away with sanrensei, for instance, than Black.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #14 Posted: Tue Jun 23, 2020 1:07 pm 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
sorin wrote:
The commonality I see between the 3-4 based shimari and the 4-4 based one is that they both prevent the opponent from playing a very desirable move (a keima approach).

In that sense, the 4-4 based shimari is very different compared to extending to the middle of the side from 4-4, since the large extension doesn't prevent the opponent from making a keima approach to a single corner stone.


These days, the knight move from 4-4 (I think I tend not to call it shimari, but English enclose or just knight move, whereas I do use Japanese shimari for 3-4 ones, guess that reflects I see them differently!) is not so much about preventing the approach, but the 3-3 invasion. I remember Li Zhe, the philosophical 6p, declaring soon after AlphaGo that approaching 4-4s was slack and 3-3 was more severe to take the base. So the knight move is more about making the 4-4 have a base and be strong / thick, whereas a 3-4 already has some base so the shimari is relatively more about making territory (and also denying the approach, and the big shimaris in a way that is not liable to overconcentration).

Also, the 4-4 knight move doesn't really stop them approaching, just means you deny them the choice of direction. We used to think approaching a 4-4 plus 6-3 at the 3-6 was bad, because of kick and pincer:

Click Here To Show Diagram Code
[go]$$W
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . a B . . |
$$ . . . . . . . . . . . |
$$ . 4 . . . 3 . X . . . |
$$ . . . . . 1 2 . . . . |
$$ . . . . . . . . . . . |
$$ . . . . . . . . . . . |
$$ ----------------------+[/go]


So instead 2nd line approaches at 2-4 or 2-5 were the standard moves against such a shimari (oops!). Whilst bots do sometimes used those submarine approaches, they (more?) often just approach at 3-6 anyway, as a 10 kyu would have been scolded about 5 years ago. My take on this is it's about making the initial knight move end up in the wrong place: if you had approached first and black kicked then he surely would have played the high extension at a instead for a more powerful attack on the white 2 stones, which locally speaking are in better shape than the black kosumi. The knight move being on the 3rd line gives white all sorts of aji for the 2 stones to sabaki, you can even tenuki again to another big opening point and still not die.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #15 Posted: Tue Jun 23, 2020 2:19 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Back to Shibano's point, that the high side extension from the 4-4 towards the Black one space high enclosure has largely disappeared from pro play because of the opinions of AI.

I just checked with Waltheri, and apparently that had already happened before this century. Maybe I misunderstood. :-| The low side extension, which was the most popular play, seems that it might be showing that effect, but has been played recently in high level play. Quien sabe? Maybe he would have done better to choose Kobayashi II as his example.

Be that as it may, both plays seem to be playable, with winrate estimates within the margin of error.

Click Here To Show Diagram Code
[go]$$Wcm6 :w6: and alternatives
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 1 . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . O . . . . . a . . . . . , . . . |
$$ | . . . . . c . . . b . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

By comparison with :w6:, Elf's top choice, a loses 3½%, b loses 3%, and c loses 2½%. If the side extensions are playable, what's to explain?

That does not mean that we should not listen to his explanation, as it reflects his high level knowledge and understanding of go. But it may also represent assumptions that will not pan out. OC, it is more likely that b is an error than c, but not by much. The difference between them as regards a Black invasion at C-03 are real, but may be inconsequential from the perspective of the whole board.

This is one reason why the margin of error is important. What differences matter, and when? Which require explanation, and which don't?

It's one thing to say, for instance, that you should generally avoid making sanrensei early in the opening, because sometimes it will be an error, and rarely will it be best play, and getting down to brass tacks about it. This is a new challenge for go players, because for the first time our best players are deaf and dumb. It's like we are human animal psychologists trying to explain the behavior of gorillas, except that they are the humans and we are the gorillas. ;)

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #16 Posted: Tue Jun 23, 2020 2:50 pm 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Bill, of the various bots around, I think Elf is more side-oriented than most, in that other bots like LZ will dislike the side extension more (in relative terms, maybe not in percentage points as Elf has big percentage swings). I noticed that Elf quite likes the 3 space high pincer to 4-4 approaches whereas LZ did not ~1+ year ago. LZ has become more partial to them, I suspect in part due to some Elf games in the training, and in part naturally discovering that it is sometimes a good move and moreso than it previously thought.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #17 Posted: Tue Jun 23, 2020 3:00 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Uberdude wrote:
Bill, of the various bots around, I think Elf is more side-oriented than most, in that other bots like LZ will dislike the side extension more (in relative terms, maybe not in percentage points as Elf has big percentage swings). I noticed that Elf quite likes the 3 space high pincer to 4-4 approaches whereas LZ did not ~1+ year ago. LZ has become more partial to them, I suspect in part due to some Elf games in the training, and in part naturally discovering that it is sometimes a good move and moreso than it previously thought.


Thanks much. :) As Keynes pointed out in his Treatise on Probability many years ago, generalization requires differences in conditions, which, OC, was hardly a new thought at the time. ;) A major reason for my Opening problems for AI series is to get differing opinions of different bots. I think that they are more likely to agree on human mistakes than on which plays are best. :)

As for Elf liking the sides more than other bots, that's interesting. Elf regards some of Takagawa's side extensions as losing more than 10%. :shock: What must LZ think?

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #18 Posted: Wed Jun 24, 2020 5:00 am 
Lives in sente
User avatar

Posts: 914
Liked others: 391
Was liked: 162
Rank: German 2 dan
It may be more of a tangent, or a continuation to the next move, but how does Go Seigen's “White 8 is bad” fit in here?

_________________
A good system naturally covers all corner cases without further effort.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #19 Posted: Wed Jun 24, 2020 6:16 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Harleqin wrote:
It may be more of a tangent, or a continuation to the next move, but how does Go Seigen's “White 8 is bad” fit in here?


Or maybe this :w8:?

Click Here To Show Diagram Code
[go]$$Wcm8 Side extension
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . O . . . . . |
$$ | . . . O . . . . . 1 . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . b . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . a . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

In Go Seigen: 21st Century Go, vol. 10, p. 126, Go Seigen says that he does not care for it, and gives a long sequence starting with Black b that he likes for Black. For :w8: Go Seigen prefers a. Elf also prefers a, but only by ½%. ;) Maybe other bots have a stronger preference.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: More Shibano AI magic
Post #20 Posted: Wed Jun 24, 2020 7:03 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
It may be more of a tangent, or a continuation to the next move, but how does Go Seigen's “White 8 is bad” fit in here?


Many a true word is spoken in jest. I have done a lot to propagate the joke, but it was T Mark Hall who first made the observation. Over the years we discovered many examples, and in fact one of the highlights of our weekly meetings in a Korean restaurant was sharing the latest sightings.

With Mark's death in 2013, the hunt came to an end. But in 2016 (just before AI rocked the boat) Ishida Yoshio published a book that wasn't a million miles from the same idea. It was a book about how changing the kakari changes the game and (I'm quoting from memory here) the most decisive point in the game was therefore move 5. His examples ranged beyond move 5, as it happens, but it was always a move very, very early in the game.

Again from memory, Ishida wasn't saying one move was better than an another. He was rather trying to show the reader how to think about the opening. The fundamental point, I think, was that a move very early in the game had a proportionally higher effect than a move later on. Thus, on move 5 (ratio 3:2) the effect of the extra Black stone was much greater than on, say, move 25 (13:12). There were, however, other books around the same time that played the same sort of numbers game even quite late in the fuseki. I think it was Mizokami who demonstrated a method of choosing how to play (aggressively, neutrally or cautiously) by counting the proportions of opposing stones in various areas of the board.

In Ishida's case, however, having a large effect so early in the game is tantamount to having control of the game. It may be, therefore, that we are seeing something similar in AI play. The Direct 3-3 on move 5 may not be inherently any better than the same move on move 25, but what is better is simply having some degree of control of the game. I hasten to add that control doesn't have to mean bossing the opponent round. It may, for example, mean making probes - if you've got better maps, you can get control of the high ground later on.

I have a hunch that a lot of AI play can be explained in such simplistic terms. After all, as Michael Redmond wisely said, go is really a rather simple game. It's rather like when the military strategicians turn to Sun Zi's Art of War. In Sun Zi's time, warfare was relatively simple - no cyber this, no bio that. No numbers! At any rate, it's easier to spot controlling factors with a simplified layout.

Going back to why White 8 was so often bad in reality as in jest, I think it might be fair to say it was because White was trying to wrest control of the game far too son: when the proportions were still 4:4 followed by 5:4, as opposed 20:20 followed by 21:20.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 29 posts ]  Go to page 1, 2  Next

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