Life In 19x19

I was rather taken aback by the following opening. White was Fujisawa Rina & Iyama Yuta and Black Ueno Asami & Son Makoto.



It was a Shuffle Pair Go game - a "sub-event" at the 2023 Pro Pair Go Championships (cos play terms seem to crop up quite a bit in modern commentaries!). For that reason, I'm sure the players were playing to the gallery. The starting formation of the sanrensei looks like a nod to amateur tastes, for example. AI disapproves of it.

But the interesting thing was that when the final position of the extract given here was reached, White had a winning ratio of 61%. That is despite the fact that White had made a single-stone ponnuki, traditionally worth 30 points but here hemmed in on the side, whereas Black had made a two-stone ponnuki, the traditional tortoise shell shape worth 60 points, here facing the open centre which we tend to assume is something AI bots love.

Furthermore, the move that saw Black's chances plummet was apparently 25. That seems understandable - two horrible cutting points don't seem like a sensible recipe for success. But after Black 31 could we not assume those cutting points have been usefully dealt with?

But the plot thickens.

If we look at just the NW quadrant and play only those moves on an empty board (8 moves each), KaTrain gives the result as absolutely equal! In order words, a 30-point ponnuki on the blocked side is worth the same as a 60-point ponnuki in the open centre. Jugged hare is just as good as mock turtle soup, but White later ran away with the game.

Can anyone explain these (to me) counter-intuitive evaluations? My best guess is that overconcentration has something to do with it. But while that's something that once cropped up a lot in pro talk on AI, it's barely mentioned nowadays.

My take is:

The evaluation has little to do with one stone vs. two stone ponnukis, and more to do with whole board balance.

-- AI prefers cash to potential.
-- Given what already happened on the right, AI probably thinks Black is staking too much on potential when Black chooses the large outside ponnuki.

I'm not much of a go player these days, but E15 feels slow.

I think it's important to put those numbers in perspective. The AI evaluation of win percentage tends to lean toward absolutes with a very small lead, so a 60% win probability is still very close by human standards. According to my slightly-out-of-date version of KaTrain, it thinks that white is only 1.4 points ahead in the final position. Even at a professional level, I wouldn't consider that a real lead with this much left to play. I'd say this slight difference comes because current AI evaluation tends to value solid territory in the early game, and white has more clear points at this point. But both sides are eminently playable.

John Fairbairn wrote:But the plot thickens.

The flow is good for white in the sequence as played in the the upper left corner. The three space pincer is almost like a tenuki, getting such a great move as part of the sequence is good, it's better than playing the same move in gote. Still the benefit is minuscular, white has to repeat the same success many times to gain a secure lead. Judging from the Computer the inconsistent P18 by black is -0.5 and then white gains 0.4 points more in the upper left then squander 0.4 points with C6.

If it weren't for C6 then white would be playing perfectly. It is really hard not to fall behind when the other player is playing almost perfectly but usually there are some inaccuracies from both sides. I don't think it is unusual to see almost perfect play in the opening and past move 50, so it is to be expected that you can fall behind by a few points in equally matched games. Playing the (computer) correct move is also a style thing, if both players adopt a style of playing sure moves then I'm sure top pros can play perfect games past move 150 almost every game if they wished to, but this will usually not be equally advantageous to both players. As far as I know it's more common for the stronger professional to try to win by playing sure moves than it is in equal matches or for the weaker player to attempt this. It's probably different in pairgo and I think white started fighting on the very next move and took an aim at the tortoise shell, which might have been the intention with C6.

I think the Key stones are O17, K16 and C6。There's no clear way for black to use the upper left group effectively.
If F17 was at F16 or P18 was at O15 it would be clear how to make black's stones work together to create proper influence。 やだやだ, at least I think so . . .

edit: I just realised kavsir also thinks P18 is suspect so . . . That must mean it's, true, right? Haha。Although I see nothing wrong with C6　。。。Maybe it should be a further extension?

Elom0 wrote:edit: I just realised kavsir also thinks P18 is suspect so . . . That must mean it's, true, right? Haha。Although I see nothing wrong with C6　。。。Maybe it should be a further extension?

The difference is not much unless you think openings like sanrensei are out of the question, then you may be fuming - "how can they call themselves pros and play badly"- because this was worse than sanrensei in terms of computer estimates. The move seems reasonable enough to me but on the other hand it is not easy to do worse in this position.

When I analyzed the game the computer preferred the Tennouzan(?) N6.

It interests me how this position is being talked about here - analytically.

My interest stems in part from the fact that I have been re-reading some old stuff as I put together the final section of my Segoe biography. In his latter years, he was concerned (obsessed even) with spreading go internationally, and so he and many go "missionaries" went overseas, and in the typical Japanese fashion they were expected to give a debriefing for newspapers and magazines on their return. Not all were pros. Several Japanese amateurs did world trips, too. Much of the tone was amazement of "dogs walking on their hindlegs" type, directed at westerners not the Chinese or Koreans. And the pros were always quite explicit - the westerners were "weak". But in the more discursive parts of their articles, they would often mention, presumably by way of explanation, that the westerners thought about the game in a different way.

At the same time, quite a lot of westerners, especially those who went to Japan to study go, wrote articles for western go magazines and so on when they returned, and here too there was frequent mention of the Japanese having a quite different way of thinking about the game. I never saw any reference in either case to what these differences amounted to, how they came about or whether they really mattered, but one thing stands out - as it does today - a lot of the western players interested in the game had a strong mathematical bent (their articles or mentions of them were usually preceded by a mini-biog; in many cases I knew them personally).

The classic portrayal of the differences between the Japanese and us at the time, however (this was when the Japanese were first emerging as a huge economic force), was that the Japanese were synthesists and we were analysts. The smug summary here of the effects of this (would it be allowed in today's woke culture?) was that the west therefore produced lots of PhD students whereas the Japanese were good at copying ideas from elsewhere and combining them into a cheap package. The Japanese retort to that, as I have just said, was along the lines of seeing us as dogs walking on hind legs. And I well remember a colleague of mine, a noted financial journalist, remarking how the Japanese bankers on the streets of London and New York were then beginning to "strut".

The two sides know each other rather better now, and I think it's fair to say have also influenced the respective mindsets a little, but if you want to continue to see it as a leap-frogging competition, I suppose Deep Mind has given our side some bragging rights Imagine the PhD quotient in that!

And, in that connection, I now see (I think) the old difference surfacing in study with AI. The Japanese pros (they being the ones I follow most) seem to use AI mostly by just looking at AI game after AI game, in the same way that their traditional advice for improving was just to keep playing over pro game records, to build up one's intuition. People here, however, seem more concerned with the decimal points.

As far as I can tell, the Japanese approach is to find a wood, pick a tree and then walk through the wood, and when they emerge they end up with get a good feeling for how big and/or varied it is. Then they find another wood or even a forest and repeat, and compare the result to the previous woods. The western approach seems to be to find a wood, pick a tree, and then examine its bark, count the leaves, measure the height and span, check for fungus and woodpecker holes and so on, and then claim to be a tree expert.

I can't remember the details and am too lazy to look it up (it will eventually be in the book, I expect) but Segoe was scornful of some western way of studying, claiming that that would never build up your intuition, and you needed (as a pro) to be have great intuition to be able to play your moves within the time limits.

As an example of the way the Japanese were amused by the western obsession with numbers (this was in the context of which was better: the German ranking system of just numbers or the Japanese dan-kyu system), Hashimoto Utaro mentions a European joke that if you could sit at a go board you were 60 kyu. If you could hold a go stone you were 58 kyu. After you'd won your first couple of games you were 55 kyu, and so on.

So, going back to the original position, does an AI bot evaluate it synthetically or analytically? Is it better to be a wood expert or a tree expert?

Of course, no one knows precisely what intuition is. I think the AI players follow the old Japanese advice about developing intuition. When we humans play through pro games we pick up patterns unconsciously. John Fairbairn's old friend, the late T Mark Hall, improved his go strength significantly just from transcribing thousands of games for GoGod. Could the training of AI by playing millions(?) of games develop something like intuition? When Segoe and others were first spreading go in the world people distant from the Asian go centers mostly didn't have access to many game records of pros. Consequently, not having seen thousands of games (trees?), people searched for detailed descriptions so they could recognize something as a tree. As it happens there are botanical taxonomic keys that allow people to identify trees they have never seen before. In go we have used joseki dictionaries in a similar way to how the taxonomic keys might be used. An analogy to language learning might also be a propos. We learn to speak a language by hearing it, trying to use it, and being corrected. If we grow up in an environment of highly educated people we tend to have a higher level of language use. I would say that pros play go synthetically. On TV games the players still "analyze" the game but they don't use a lot of words, they point to an area and use gestures to indicate how they feel about a situation and they may lay out some sequences but there is little discussion.

the Japanese were synthesists and we were analysts

I agree that intuition is extremely valuable. I don't think there's a problem with analyzing - it can lead to better understanding. I don't think there needs to be a dichotomy. However, if you are lacking activities in study that may lead to better intuition, it's certainly good to correct that.

gowan said:

When Segoe and others were first spreading go in the world people distant from the Asian go centers mostly didn't have access to many game records of pros.

A very good point. One reason I prize the (rare) discussion here is that it can bring up points that I had not considered myself, such as this one.

That, in turn, can lead to further stimulating (maybe incorrect, but still stimulating) thought of one's own. In this case, my own further thoughts went something along the following lines.

First, it occurred to me that, while the Japanese would then certainly have had more access to game records than westerners, it was still not easy for them. Very few people would have been able to afford large collections of games such as those of Shusaku. But they did have games in go magazines and newspapers. The two often went together. I have a very large collection of old magazines. Hidden within their pages are lots of clippings of game records from newspapers, made by the original Japanese owners. Even non-Japanese readers would collect these newspaper records. John Power, for example, has a large pile in his attic which he kindly used to provide GoGoD with rare games.

Japanese pros tell lots of anecdotes about travelling round the country to play games with rich amateurs, at whose homes they discover priceless collections of old games. They spend most of their stays in said homes copying the games by hand! There were no photocopiers in those days.

Indeed, it further occurred to me that that could have been a major factor in the live-in system of pupillage. If you start from the premise that playing over lots of pro games is the best way to become strong, you need access to such games. And if you can't afford them yourself, you can be sure to find collections in the homes of existing pro teachers. I have never seen that specifically mentioned as a factor in fostering the live-in system - it is usually explained as an extension of the traditional orei boko apprentice system - but it does tally with records of the Edo families prizing their own records and keeping them secret.

It also tallies with a story that has long intrigued me. Kato Masao was a member of the Kitani school, and in her autobiography, Mrs Kitani described the routine for the children. The day began with rising at 6am, when the children were expected to play over a game from the classic collections. (At 7am there was radio calisthenics, and breakfast was at 7.30.) Kato Masao evidently liked to snooze a bit more before he got up and became adept at finding short games to play over. Implicit in this story, incidentally, is the supposition that they had to learn the game, or at least part of it, by heart. We can deduce from other, similar accounts that they had to expect Kitani to pounce on them and make them replay the game from memory.

Years later, Kato himself recounted how he had stormed his way to title matches as the "Assassin" - killing large enemy groups in seemingly every game. Short games. in other words. But when he got to the actual title match he kept on coming off second best. He was, of course, playing the very best players who could see as far as him tactically, and their groups tended not to die. Long games, in other words. Games at which Kato had little experience, whereas his title-match opponents, who were generally fellow members of the Kitani school who did get up at 6am on the dot and play over longer games, had plenty of such experience.

Kato also recounted how his failures were getting him down, so that he finally asked his co-pupils how to cross this final hurdle. Which he did eventually, but, annoyingly, he never revealed what they told him. And I'd love to know. I have long assumed it was some sort of psychological advice: "Stiffen the sinews" and "Cry God for Harry!" But, today, I'm now wondering whether it was that by playing over only short games he had never learnt how to finish off a won game properly. Once he started playing over longer games, did he improve that extra notch. We know it's not far-fetched because Michael Redmond has told us how he had a similar realisation, late in his career, that his opponents were rather better in the endgame than he was.

As to whether I'm right about the relative scarcity of game records in those days in Japan, I find it easy to believe because I and most others of my generation were once in a similar position with books in general. We had one bookshop in the city and books were very expensive. My own home was initially almost bookless (though we had good public libraries and I built up a nice collection by winning book prizes at school). But in those post-war days we had lots of door-to-door salesmen, and they included people selling encyclopaedias. My parents bought me Arthur Mee's "Children's Encyclopaedia" and I read and re-read almost every word off the page. By a strange coincidence, I found the same set of books in a care home a few months ago, and I could turn to my favourite pages just like yesterday, almost 70 years on!

I can easily imagine the same sort of scene in Japan: a young player suddenly presented with a rare collection in his teacher's home. It does not surprise me if he claims (as told in Kageyama's "Treasure Chest Enigma") that he can remember the old games years later.

It's going off on a different tangent, but this same train of thought led me to realise something about old Chinese games. There is a comment you sometimes see in their commentaries that that particular game exemplifies a good "model" to follow. It's common enough to have registered with me, anyway. It dawned on me that, implicit in such comments, was again advice to play over pro games.

The way I see it, everything stacks up to support the pro advice from auld lang syne to play over pro games as your main mode of study. In your countryside walks through the land of go, be first a wood expert - so you don't get lost. But Kato and Redmond also provide a reminder that you also need to become a tree expert eventually! Learn to follow the stars as well as a compass.

This too been a bit of ramble (which makes me think on "We twa hae run about the braes,/and pou'd the gowans fine;/But we've wander'd mony a weary fit,/sin' auld lang syne.), but my brain enjoyed it. So, thank you, gowan.

And that naturally calls forth a later verse:

And there's a hand, my trusty fiere!
and gie's a hand o' thine!
And we'll tak' a right gude-willie waught,
for auld lang syne.

gowan wrote:Of course, no one knows precisely what intuition is. I think the AI players follow the old Japanese advice about developing intuition.

This reminds of a comment from a chess player back during the Kasparov match vs. Deep Blue. There was a question of whether the computer would ever be able to 'know' a move was going to be bad in the way that a human would--intuition backed by study, heuristics and experience. The question in return was to ask why the computer needed that intuition when it could just go and calculate out the millions of moves to get a firm evaluation.

From what I understand of AI, the first layers of the neural network come up with candidate moves along with a baseline valuation that is refined based on more calculation. I'd said the closest thing to intuition is that first layer. Interestingly, I think if the system plays with only its first layer it's equivalent to a very strong amateur, which tells us just how little we really all know.

I remember a Japanese participant in an old　WAGC saying that Westerners are more analytical。
*rant*
To make glib generalisatiins of groups of cultures for the sake of convenience, European thought about thought is about what can be practically achieved in the physical world, whereas East Asian and pacific thought about thought is about culture。Africa and West Asia and West Asian cultures have both mixed in a myriad of complex ways that would take paragraphs of off-topic text to explain, but I can say that Abrahamism from North Africa and West-Asia does not fit with European culture where Pagan thought will always be more natural。And China and East Europe are odd cases who are as similar to each other to be the same. It's a shame that that Japan outpaced China in the sense that Chinese culture is far easier for westerners to understand without too much time than Japanese culture。

In the west you try understand and gain more power in the physical world and then use that to develop and understand ideals. Japan and not in China and Russia, you start of with an ideal and then。Instead of this, China and Russia have Vyanro and do what the Japanese do indirectly, in Japan vyanro is done indirectly。It's complicated。In any case westerners sound fake talking about ideals and Japanese can sound fake talking about the real world. And they could snob each other, but remember that only the East Asian mentality can take you from being a caveman to some degree of scientific competency, since ironically trying to control reality limits how much you can do that, since it's often creative thinking rather than practical thinking that leads scientific advancements。But when East Asians started sitting around being very impressed with themselves and closing themselves from each from themselves and the rest of the world, the elites thinking only of high culture and looking like the most superior neopotdom to their populations, Europeans took Asian knowledge and advanced far past them。 And history repeats itself since the AlphaGo is exactly thaｔ, before then some pros couldn't think of how else to improve and often just meditated in front of the board。It would be expedient to add that there significant differences betwern the copycat cultures of Japan and China, but this post is already to long. In general Chinese thinking is more European given their proximity to Europe.

The problem is the modern economically based form of globalisation does not lead to an authentic understanding of other culture s, but rather the illusion of it when it's really nothing but main-streamism。None of these apply as we and then people get the illusion they are 'worldly' from from those economically determined pidgeon-hole viewpoints of other cultures. Golly, even subcultures within the same country are not authentically understood。In fact many modern-day so-called 'worldy, digital' people may be surprisee to learn that my mothers Ewe side is prone to a significant cultural misunderstandings of my Father's creole sideーand the differences those historically freed slaves can have between the plethora of ethnic groups that have managed to place a chunk of their territory on the laughably small country of Sierra Leone。 Because of their cringey 120 pixel resolution of Africa which is even false, Northern Africa and Sub-Saharan Africa, while my mind automacally sees five geographic cultural divisions each with many subdivisions, with one of the five divisions seeing East Africa, North East Africa and the West and South of Arabia which includes all of Yemen as all one。

I believe the Exact same thing is going on here。 I am sure Japanese eyes looking at the board would think the same way in dividing the board compared to westerners, that Taiwanese, Korean, Japanese and Chinese pros would think the same of each other about their view of the board。A city person should compare their experience to another city lest they risk drawing the wrong conclusions, 'people in that country are so much more laidback and friendly', and vice versa, Western Amateurs can only really compare themselves to Eastern amateurs。I mean, okay, we have pros, but 。。。Semi-pros, as Stiasny humbly said。I also mildly have contempt for videos like 「differences between the USA and Japan」、 because I firmly believe you must always have at least a third control culture, in the comparison, preferably a fourth as well。And my annoyance isn't because of the number of videos that explain why Japan doesn't tip rather than asking why Americans do . . . Although it did make it worse。Not having VAT included in the price is another reason why for me the best view of America is from a distance！ Do you not care about having no idea how much you have to pay for things? Is everyone there pretending to be rich on the low-key?

In fact I believe that differences of viewpoint of what the game is means in Asia artistic people view it as 'their thing' and in the west scientific people view it as 'their thing' since the different cultural groups differ on what it is they differ on what kinds of people are best at solving it。
*rant*

John Fairbairn wrote:As far as I can tell, the Japanese approach is to find a wood, pick a tree and then walk through the wood, and when they emerge they end up with get a good feeling for how big and/or varied it is. Then they find another wood or even a forest and repeat, and compare the result to the previous woods. The western approach seems to be to find a wood, pick a tree, and then examine its bark, count the leaves, measure the height and span, check for fungus and woodpecker holes and so on, and then claim to be a tree expert.

Well, he would be "(A tree) expert", just not "a (tree expert)".

Regarding AI validating the Western approach... I'm not so sold on that. My impression is that old bots were analytical, while NN-AIs are synthetical.

Take care.