Major contribution by Takao Shinji to AI debate

John Fairbairn · Post by **John Fairbairn** » Sat Dec 19, 2020 7:59 am

Takao Shinji is a multiple major-title holder. He has been described as "inheriting the best" of his teacher, Fujisawa Hideyuki. He is Japan's national team coach, tasked with preparing the young cadres to face international competition in the AI age. In other words, he knows more than a bit about go, he is still at the cutting edge, and he has to find a way to convey new knowledge to young pros.

He has lifted the curtain a little in some new material entitled "Understanding honte." Arrestingly, his first example is entitled "Modern honte." Modern refers to what has been learnt from AI. As far as I'm aware, honte and AI have not been discussed together before. If they have, it has escaped my attention. Hane Naoki has written a book contrasting real honte and fake honte, and that covers much of what Takao says, but Hane did not have AI in mind.

Consider also what honte means for the typical amateur. With only a little exaggeration, I'd say it means the sort of slow, prophylactic move that is only rarely made, and when it works it produces a lovely warm glow. It's a "once a year allowing yourself two helpings of Christmas pud and a fine port afterwards in the snuggest of armchairs" kind of rare but heavy feeling.

But what Takao is telling us now is that honte lite is in, and these zippy new moves are no longer once-a-year occasions. You no longer get these weapons from the quartermaster's store for special occasions. You have to wear them on your body at all times. Takao shows many examples from his training games with the national squad. There can be multiple examples in each game. And of course this is what Takao has chosen to reveal first. That all seems highly significant, not least because this concept relates to the middle game. Most AI inferences so far have related to josekis and fusekis.

Incidentally Takao and Hane are old sparring partners from the Nagoya branch of the Nihon Ki-in. Both are fond of thick play and obviously share insights. But Takao's style is usually described as "thick but bold," which possibly gives him a slight edge in appreciating the new kind of AI honte play.

Just show everyone understand what a honte is, look at Takao's first example. Takao is White against Kato Chie, and it is his move.

The White group in the upper right needs a honte. A honte in the first instance is a safe repair move, fixing the roof before the rain arrives - a prophylactic move. But it should ideally do more than that. It should store strength that can be used later to pressurise the opponent. A characteristic of honte by the traditional definition, as Takao points out, is that it is easy to see the damage that can be done if it is omitted. It is quite hard to see what the pluses are. In the diagram above, if White omits his honte, Black will peep at A and follow up soon after with another vital-point move at B, forcing White to squirm into an ugly, eyeless, heavy shape. So the usual recommended honte move here is White A. Which is a simple shape move, no messy calculations involved. It looks pretty but it is not easy to forecast what the extra benefits of this move might be.

For that reason, pros feel A is actually slack. But its hidden benefits are so important that Takao still recommends it for amateurs.

The "modern" honte, however, the sort of move that young pros now have to learn, goes above and beyond. It is White A in the case of the diagram below.

Variations shown by Takao are embedded here. I'll leave you to draw most of the necessary conclusions, but the one to stress is that the modern honte is not just the traditional 守ってから狙う - a prophylactic move that aims at a follow-up. Rather it is a prophylactic move that aims at an immediate follow-up, or one which puts pressure on immediately.

A further difference, which you can only get from the other examples not given here, is that the more dynamic kind of prophylactic move now required brings with it the need to consider much more than simple shape. You now have to factor in things like direction of play and momentum (choshi).

As to why honte are now to be considered necessary more often, it's not just that they are more dynamic and more aggressive. One of Takao's examples was a game was with Sumire. He noted that her recent improvement was based on spotting minute weaknesses and taking advantage immediately to build up an early lead. This is typical of AI play. If you turn that round to view it in a different way, you can say that players now are, or are becoming, better at spotting weaknesses. There are therefore more weaknesses than you might think. And so more hontes are necessary than you might think.

salerno · Post by **salerno** » Sun Dec 20, 2020 7:16 am

This is very interesting, John, thank you for sharing.

I'm curious about this new material, "Understanding honte." Is it a book, a series of articles, or something else?

Thanks a lot

John Fairbairn · Post by **John Fairbairn** » Sun Dec 20, 2020 8:50 am

It's a new Go World magazine series (in Japanese of course).

gowan · Post by **gowan** » Sun Dec 20, 2020 9:15 am

It seems to me that there has been an increase in importance of speed in choice of move. Go Seigen was famous for speedy development and that grew more important as the Chinese and Korean "schools" of go emerged. The shoulder hit John illustrated defends against a Black play at A and gets out faster. The recommended shoulder hit might be thought to strengthen the black stones in the middle of the upper side but, depending, they might also become heavy.

Uberdude · Post by **Uberdude** » Mon Dec 21, 2020 12:55 am

This links in with one of the major things I've picked up from bots: they really hate it if the opponent can make a beneficial exchange in sente. Of course humans didn't like it too before AI, but the degree bots don't like it and the changes it leads to in their play to avoid it are new. A common example is a peep getting answered with submissive connection, such as in this (now largely obsolete) 3-3 joseki.

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

That this exchange benefits white (and there's a plausible game plan where White can reasonably get it following a split of the side), coupled with the not making 2nd line hane connect as turns wall to thickness, is pretty much responsible for the revolution of playing 3-3 earlier and then not playing the hane but extend or jump for move 5 of the sequence. And honte moves are often about removing the opponent's opportunities to make such beneficial exchanges; in Takao's example they would be the shape point peep getting a local answer to defend the cut, but with the honte black can no longer get that as white can reasonably resist and answer in a more efficient way.

Ember · Post by **Ember** » Mon Dec 21, 2020 1:27 am

Thank you for sharing, John!

It's wonderful to get a glimpse of the latest interesting information. But it also saddens me that so little ofall the AI discoveries is actually translated and available in the west, too..

John Fairbairn · Post by **John Fairbairn** » Mon Dec 21, 2020 4:34 am

Here is another example from Takao to ram some points home. It was the one that surprised me most, incidentally, because I would not have described the honte move as a honte (although I might well have come up with the old Chinese equivalent laocheng - another case of old Chinese go presaging AI play).

Takao, to play, is Black against Ueno Asami. Part of the skill required is find in the minute weakness in Black's position, but I don't think that's much of a spoiler as the main point is the formulation of Black's response.

Solution:

Uberdude · Post by **Uberdude** » Mon Dec 21, 2020 4:56 am

John Fairbairn · Post by **John Fairbairn** » Mon Dec 21, 2020 6:19 am

Purely by chance, I just sat down to do some more editing and came across an example which shows yet again, and in a different facet from those we have all mentioned before, how much Go Seigen was ahead of the curve.

Black (Go, against Kitani) to play and make the "modern" honte and explain why (i.e. where's the hidden weakness?). You may wish to note Go spent 23 minutes on it.

Solution:

gennan · Post by **gennan** » Mon Dec 21, 2020 9:50 am

gowan wrote:It seems to me that there has been an increase in importance of speed in choice of move. Go Seigen was famous for speedy development and that grew more important as the Chinese and Korean "schools" of go emerged. The shoulder hit John illustrated defends against a Black play at A and gets out faster. The recommended shoulder hit might be thought to strengthen the black stones in the middle of the upper side but, depending, they might also become heavy.

A while back, InSeong Hwang 8d gave a video lecture about the "history of go concepts".

One of those "go concepts" was "speed", which he puts into his history as a 20th century development, started by the Shin Fuseki. In this lecture he mentions Go Seigen and AlphaGo as examples of "speedy" players and in other lectures he often also mentions Cho Hunhyun as a "speedy" player.

The section about "speedy development" starts here: https://www.youtube.com/watch?t=779&v=OkkXooEDwKg, but the whole video is well worth watching IMO.

EdLee · Post by **EdLee** » Tue Dec 22, 2020 8:36 pm

10

stones, 11

stones. 1 missing

on the board, it seems.

Life In 19x19

Major contribution by Takao Shinji to AI debate

Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate

Re: Major contribution by Takao Shinji to AI debate