The first newspaper involvement in was a Next Move (Tsugi no Itte) column in Meiji times in Japan. As a result its title has become the equivalent of a technical term, appearing in go-term dictionaries. At some point, but long after World War II, when knowledge of English became more widespread in Japan, a new term came into being. It was besu poji. It was from English. You may like to guess what the English was before reading on.

It was a double abbreviation of besuto pojishon or 'best position'. Apart from the odd formation process, it was peculiar because it's a phrase we English-speakers would never use in that context. In fact, we would probably not even understand it properly even in the right context. The formation process is odd, and is one the Japanese are rather addicted to, but we have a similar thing in English - smog is from SMoke and fOG, for example. But the Japanese use it even with foreign words and sometimes mix up the languages. The most famous example perhaps is bakku sha-n, from English 'back' and German 'schön'. It was used to refer to a person who was great to look at from the back but you revised your opinion when she turned round.

So we must be prepared to find a Japanese nuance in these neologisms. But I lazily never bothered to think about it in the case of the besu poji problems, mainly because I never bother doing such problems.

But my attention was grabbed today by an article in the latest Go World which had the extremely odd headline of three straplines, all of which said essentially the same thing. The top line used the old phrase Tsugi no Itte. The second main line urged the reader to find the besu poji (besu poji wo sagase) and the third line referred to the 'right move in actual games' (jissen no itte).

That triplication was enough to get me to look at the problems (or, rather, their solutions, to see whether something special was intended. Not entirely to my surprise, I think I detected a sense in which all three lines were the same but different. Given the size of the besu poji strapline that was evidently the main nuance, and I think the solutions did convey something different. And that could be useful for us amateurs.

Most amateurs look at problems with an aggressive mindset that one Japanese summed up in a book he called something like the Five Envies (I'm too lazy to go and look it up properly). Basically, he described amateurs as typically playing in a way that showed they just wanted to deprive the opponent of ANYTHING - life, territory, thickness. Or, if he nasde an extension you felt you had to make a bigger one. I have always felt that that is afair description of most amateur play. After al, how often we do hear the questions "How do I punish....?"

To be fair, most Next Move problems can reasonably be said to fall under that category, too. But the Besu Poji problems are rather different. Indeed, the answer is sometimes a choice of moves, not just the 'one' implied in 'itte'. It turns out that the "correct" rendering of besu poji in Eng;ish is not 'best position' but 'best positional play'.

The summary of the solutions given in the cases I looked at were: (1) White settles his shape and so had no need to worry about being attacked; (2) Black can live in good shape; (3) The result is fair for both Black's ladder breaker gets compensation for giving White territory; (4) White gets a high position (kurai) and can survive any attack with 3 or 'a' next.

It's a subtle change in mindset that, I believe, can pay off.

But it became more interesting because the four positions given first were positions where the best positional play was decided from a human perspective (albeit from a pro rather than an amateur perspective, perhaps!). But the next four problems required the reader to find the best positional play from an AI perspective. The answers then were: (1) It makes it hard for Black to use his thickness; (2) White hits the vital point with a sacrifice and so gets the potential to deprive Black of a base; (3) White expands into the uncultivated area in the lower left while forcing Black towards his stronger side; (4) White 1 is a distant probe that gives him momentum to defend his territory on the left.

As you can probably tell from those summaries, the AI play was more positionally complex. It was typically a two-punch combination rather than an attempt at a single knockout punch. Furthermore, the AI's first punch was a surprising move. The follow-up was ordinary, but more telling because of that first jab.

Maybe there are too few examples to make a sound judgement, but I also got the impression that virtually no reading is required. The first jab by the AI limits the range of responses by the opponent, but if the opponent does try to vary from the line shown, it is always possible (and necessary?) to achieve a similar result with only a slightly different second punch - an uppercut to the chin instead of a body punch, but still not trying to switch to a roundhouse KO punch. There are other episodes in this series I might go back and take a look at.

A different article in the same issue gave another perspective on that, I thought. Shibano Toramiru is running a series on new perceptions brought to go by famous players of the recent past - rather in the way that AI has changed perceptions. In this issue he is looking at Otake Hodeo and he makes an interesting summary of Otake's much admired style. It was that he studiously avoided making thickness and then using that to create a melee of contact plays. Instead, he prioritised "yoritsuki over fighting' and played quietly while making sure he never fell behind in the overall play. I thought was an ideal way to describe AI play. Yoritsuki is where you approach but do not come into contact, or specifically in go terms where you play assertively to keep pressure on while thus also keeping control of things overall.

White vaguely on the topic of AI, I'll also mention that Nakane Naoyuki WON his 3-stone game (Game 4) in his 12-game uchikomi match with Katago (20 seconds a move). Katago resigned on move 270 when almost 20 points behind. I repeat: Nakane WON.

Maybe this is his youtube channel https://www.youtube.com/@nakaneigo-doujyo/featured

There are a few of those Japanese pros that have published matches like that on youtube. One thing that surprised me when watching such videos is that they are usually using clocks. That to me is masochism. But apparently playing with clocks is such second nature for Japanese pros that they don't think this is an unnecessary inconvenience, or maybe they would almost always win those handicap games if they weren't relatively fast.

My tentative take on the short time limits is that it is to equalise things. I think that top pros believe they have put in so many years of playing over games that their intuition (the memory databank) effectively matches the memory databank of the bots. There was a brief period when the human pros had some surprises, such as the early 3-3s and the shoulder hits, which were not in their own memory databanks, or at least not on the front shelves thereof, simply because they had rarely seen them. But now they have seen these moves a lot in AI play and have caught up. The result is that the list of policy (candidate) moves the best humans come up with can correspond with AI well over 90% of the time. I have seen a figure of over 95% for Sin Chin-seo.

Short-term reading (but in the sense I mentioned elsewhere on L19, that is like reading chunks just as we sentences or even paragraphs instantly) is also a skill that the top pros have, and they can more or less match the bots at that (for the same reason that is part of their intuition). Longer time limits would just favour the bots' ability to do something more like ultra-deep brute-force analysis. The game Nakane game I referenced above is a good example of how the pros can cope in the sort-term. The game revolved around a position that had a complex double/triple ko which was going on at the same time as a couple of other kos were available, and Nakane coped with that better than katago. Also, when behind, katago tried some outrageous but (according to Nakane) very deep rip-off tricks. Nakane saw through them.

I am still inclined to believe that the bots "know" no more than the best humans (at fast speeds), but when because of an accumulation of small mistakes by the humans, whereas the bots make only very rare mistakes. The human mistakes tend to be less because of a lack of skill and more because of non-go factors such as tiredness, overconfidence of lack of confidence, fear, etc.

Game 6 of Nakane Naoyuki's 12-game uchikomi match against Katago has been published. Katago won with Nakane taking three stones for the second time. The score is now 5-1 to Katago and Nakane (who won once with 3 stones) will now have to take 4 stones.

In the latest game, Nakane decided to take a crude territory-grabbing approach. He found what we all find, that such a strategy tends to have two outcomes. One is that your territory (in this case mostly on the first and second lines) tends not to be as big as you think it will be and just gives the opponent lots of forcing moves. The other, as result of all those forcing moves, the opponent makes a lot of territory in the supposedly infertile centre. He further found it impossible to adopt the usual shinogi tactic of a one-weak-group strategy (i.e. living small in the centre). This was largely due to lack of time (20 seconds a move), and for the same reason he finally succumbed to an oversight (missing a tesuji by Katago) and a blunder (not connecting his own group). Nakane believed he would have been just ahead if he had avoided his final blunder (on move 184), but of course that overlooks the point that he had already frittered away his three-stone handicap.

There were no moves by Katago that looked especially AI-ish. It was just a case of Nakane digging his own grave with an ill-conceived strategy.

Looks like this playlist? https://www.youtube.com/watch?v=Ljv9Dtt ... 1nn_dhdZzC I've spent a few minutes watching with captions and auto-translate turned on. The translation is bad but I could get used to it ;-)

It is a bit hard to follow when Go terminology is translated literally. But he has some good advice, for example solving tsumego with the heart

Nakane Naoyuki 9-dan lost Game 7 of his 12-game uchikomi match against Katago - on 4 stones, resigning after 257 moves. He did at least find a ko threat that Katago had to answer.

I think he could lose 6 stone games if he doesn't change his approach. The thing is that he always makes the game interesting for himself (and the fans). When the game is interesting that means white has a chance. The game doesn't need to be interesting when there is so much handicap and white shouldn't have any chance. I'm sure he wouldn't lose often at 3 stones if he made sure the games were rather unexciting.

You can compare to how the Shibanos do it

In his 12-game uchikomi match against Katago, Nakane Naoyuki won Game 9 at 6 stones and Game 10 at 5 stones so that he has now won three games out of 10 (the third win was with 3 stones).

It is 20 seconds a move. My own feeling is that extra time would benefit Nakane more than Katago, but Game 10 was incredibly complicated from the very beginning, every move being plaved in a veritable minefield, yet Nakane navigated his way through safely. And, although it took him 303 moves, he won by 18 points. However, he did say he made sure he made a special effort this time, which presumably indicates he was showboating a little in previous games.

Whether or not the short time benefits the computer or the human is a tricky question..

If you have two computers and one is 3 times faster than the other, then using the faster computer and 20 seconds per move is equivalent to playing against the slow one where it has 60 seconds per move (if you weren't able to think during its time)

And you would still say 'player x won with 3 stones against katago' even if it was the slow computer..

But sure, if you use short time settings and an adequately poor PC, then there's a better chance for mistakes from the AI

A more precise statement: suppose that a given human and a given computer play together, and handicap is adjusted so that both have 50% winning chances. Is it true that if we keep the same handicap, then the human's winning chances increase if thinking time is shortened?

The question I implicitly posed was the opposite of the one being answered. I said "My own feeling is that extra time would benefit Nakane more than Katago." In other words, should he have increased the time limits in his match against Katago?

If we look at human faults in quick go (including Nakane's mistakes in this match), it seems clear that mistakes in reading through lack of time or lack of concentration are the biggest bugbear. If extra time eliminates such gross mistakes, I'd expect the overall percentage benefit of extra time to be very large. But it seems that AI bots come up with good policy moves instantaneously and rarely make mistakes of the type that humans can exploit. So the benefit of extra time for a bot may be real but tiny in practical play.

I think we see the same thing in amateur go. Stronger players who just care about results will almost always insist on short time limits in handicap games. Those who care about improving in even games will usually ask for more time. Admittedly, there is a point at which, for weaker players, where extra time is a sheer waste, but once you know enough about go to know how complex it can be, but (because you are an amateur) your intuition is still largely unformed, extra time seems valuable.

I am fairly sure of that. What I am less sure about is whether a 9-dan pro's intuition is so vast that extra time for him might be less valuable than I am assuming.

How many stones did professionals say that they would need to take from God, I can't remember if it was 3 or 4.
Have we been able to check if KataGo has an input stream from God?

_________________
North Lecale

It's been either, whenever I've seen it, and it's always been from a truly elite player (e.g. Go Seigen, Rin Kaiho). I'm not sure if it was ever meant to be taken seriously or whether it was just a polite answer to the stupid kind of questions some journalists are prone to ask (e.g. how did you feel when the plane fell on top of you?).

But while there have been answers to the opposite extreme ("I only understand 1% of go"), I've never seen a higher figure than 3 or 4 stones.

Nakane did win a game at 3 stones, and I reckon he could repeat that with a bit more care and effort but whether he'd score 50% seems doubtful to me. And, of course, he's weaker than Go or Rin were - never won a title, for example.

We also know that Go Seigen could give four stones, in quick play, to Yasunaga, who was a 4-dan.

It looks like the 5 stone game was uploaded today on youtube or is it a new game?

It is difficult to estimate at which handicap he'd can win half the time from only a few games. It is still fun to make predictions.

Based on the level of skill exhibited in these games I'd guess around 2 stones, maybe between 2 and 3 since the time is really fast. Especially since only the human player appears liable to a collapse. His approach so far is to play very interesting and long games and that makes it more difficult to get back to low handicaps. If you want to play on two stones then maybe you can't avoid an interesting game anyway, however what I see lacking is the ability, or rather the will, to clinch the game. Since he can gradually adjust his approach and that there maybe aren't many opportunities to clinch a win in such a 2 stone game I'll predict that he can get to 2-3 stones with a little bit of practice.

More time would be a huge plus. It is harder to avoid going down the rabbit hole with KataGo if you don't have time to consider number of alternatives.

It's a new game. There are now two 5-stone games in the series: Nakane resigned game 8, but won game 10 by 18 points, looking well in control after some carnage at the top right.