What does "natural" mean in go?

[gowan]

Hikaru Nakamura, one of the best chess players in the world, in an interview on the chess.com site had some provocative remarks about the effect on chess of the development of the very strong computer programs. (https://www.chess.com/article/view/hika ... chess-boom) Here is an interesting quote from that interview:

What’s your opinion on computers? Have they been good or bad for chess?

"It’s weird, because with technology, I think that much of the natural feel for the game has almost been lost. So I think the weaker GMs and IMs are a lot worse than those of 20, 25 years ago. They understand how to use a computer, but their actual intuition, in terms of the moves they find over the board, is not as good. Basically computers have meant that the top players have gotten phenomenally good, while everyone in the middle has gotten worse."

This suggested the question of whether the arrival of the super strong AI go playing software has had a negative effect on human go playing. Before the development of the AI go players many many human players considered pros as close to perfect players and there was (still is) a tendency to imitate pros' play without actually understanding it. Now I see many amateurs imitating AI moves again without understanding them. Pros also are adapting AI plays though perhaps with more understanding. As a result go looks a lot different. When I look at an AI vs. AI game it often looks very strange. The moves don't look natural to me. So, I ask the question in the title. Nakamura in the quotation above suggests that chess players who have taken on the style or moves played by AI programs have lost a natural sense for the game. Can the same effect be happening in go? And what does it mean when we say a move or style of play is natural?

[Bill Spight]

The chess engines of the past generation have relied upon deep tactical reading. As a result, they often find plays that humans have not been able to develop a feel for. The advent of neural network engines, such as Leela Chess Zero, may change that.

As for modern neural net go bots, they also play differently from humans, but sometimes the humans are actually better at deep tactical reading. I am optimistic that humans will be able to learn a lot from them in a relatively short time, and not just those at the top.

As for what is "natural" in go, I think it means what you are used to.

[RobertJasiek]

If we do not presume it to mean nothing or everything unspecific, natural (or flow like water) can mean to always play the most valuable endgame-like move throughout the game. We would presume that we can always identify it. Of course, we can't. We can, however, strive to achieve it while avoiding too complex, unpredictable fighting moves.

[jlt]

Natural means "which can be found without too much reading". Perhaps with enough training the top choice of a raw neural net (i.e. LeelaZero with 1 visit) can become natural for a strong human player, but that's not obvious. The file of the network LZ157 is something like 200MB. Even if the brain can find ways to compress the information, we may be approaching or exceeding the capacity of the brain.

[Kirby]

To me, natural means to be aligned with common human reasoning about the game. E.g. playing 3-3 early in the opening still doesn't seem natural to me, even if it's a strong way to play. I can't really come up with reasoning as to why it's good that I really believe - other than simply because the computer plays that way.

So early 3-3 invasion is still "unnatural" to me.

[Bill Spight]

To me, natural means to be aligned with common human reasoning about the game. E.g. playing 3-3 early in the opening still doesn't seem natural to me, even if it's a strong way to play. I can't really come up with reasoning as to why it's good that I really believe - other than simply because the computer plays that way.

So early 3-3 invasion is still "unnatural" to me.

But the early 3-3 invasion is natural to DDKs.

[Aidoneus]

This makes me think of this article: https://explorebaduk.com/2019/01/27/tak ... territory/

Takemiya says that "This is go the natural way."

[Uberdude]

The file of the network LZ157 is something like 200MB. Even if the brain can find ways to compress the information, we may be approaching or exceeding the capacity of the brain.

gzipped 157 is about 40MB, or 20MB if using the half precision optimisation. That's tiny compared to most estimates of human brain memory capacity, which are on Terrabytes scale. Of course the way we store things is different to a computer, so it's hard for us to remember a sequence of 1s and 0s which would be just a few bytes for a computer, but easy to remember a face which might be several megabytes as a computer image (if stored as a bitmap, but maybe we store it more like a smaller vector graphic).

[jlt]

It's hard to say if go knowledge is more "textlike" or "imagelike". It's imagelike because we use visual memory to remember shapes, josekis, etc. but it's textlike because we need to pay attention to every detail. So for the moment we don't know if all of LZ157's knowledge can fit into a human brain, maybe, maybe not.

[jaca]

To me, natural means to be aligned with common human reasoning about the game. E.g. playing 3-3 early in the opening still doesn't seem natural to me

it's human nature to believe - whether learned from self-teaching, or from teaching by others.

Beliefs, once formed, are hard to unform. That's the nature of synapses, because the more a synapse is tickled, the firmer it becomes, regardless of whether the tickling arises from reward or punishment.

So the only chance you have of getting somone (including yourself!) to not do a behaviour that you don't want them to do, is to not react to it, so the synapses that create that behaviour do not become reinforced. https://www.psychologistworld.com/behav ... nditioning

Beliefs are the axioms of (subjective) reasoning - hence the phenomenon of post-hoc rationalisation.

So it's hard to reason someone out of their beliefs - and often impossible, especially when the belief is highly connected to a strong emotion.

Here (starting at 4m02s)

is a shining example of a shiny belief that most believers like Patrick Moore will probably still find impossible to unbelieve, for it's just so unbelievable that it could have been faked by the wizardry of Stanley Kubrick!

i would guess that many will still refuse to believe that what they held to be true for so many years (as i did myself until very recently) is actually false - but i find it hard to believe that anyone could still not unbelieve in the light of the further evidence that is offerred in the punchline starting at 6m52s.

Not doing early 3-3 is a folk-wisdom passed down the generations. Even Michael Redmond 9-pro says he can't get himself to play it, even though he acknowledges it's Zero's favourite, and he acknowledges that Zero is quite good...

In the last few days, i've been listening to Laizy (leela-zero) and despite feeling somewhat uncomfortable about doing an early 3-3 myself, and even laughing out loud when i see her doing yet another one yet again, i am starting to think she may have a point....

[John Fairbairn]

To me, natural means to be aligned with common human reasoning about the game.

I think Kirby's view is the best approach to an answer. If I may try to put it another way, it's to do with expectations - the results of our reasoning. That can be misleading, of course. I find it "unnatural" in a sense that a board-games player with a name like Hikaru Nakamura can be a chess player and not a go player....

And another possibly misleading track we (and Nakamura) are following is that it's all really to do with AI, or even computers. At any rate, for me this chimed with the section where Cho Hun-hyeon is scathing about the new generation of go pros in his book "Go with the Flow":

The creative potential of go doesn't seem to be able to keep up with its popularity. Go styles have become so homogeneous that a player with a distinctive approach is hard to find. I have to admit that nowadays rookies have a very strong foundation in go and they play very well. But wait a minute. Their styles ring a bell. I have seen someone else play like them before. I could not help but feel that they were emulating veteran players, or were following the pattern others had created. At around mid-game, when the momentum is there and the spectators have been waiting long enough for a jaw-dropping move that could either make or break the game, the anticipation is quickly disappointed by a normal, obvious move. Go fans have started to complain that the game has become unexciting. What happened?

He goes on to berate the way go is taught nowadays in Korea:

The teacher spoon-feeds the rules and the formulas and the students are expected to learn them by heart. The goal is to make them play the game and to please their parents with quick results. ... ... They are expected to place stones according to the formula they have memorised , which turns their match into a test to see who digested the most information, instead of a battle of ideas. ... ... Students may be trained for multiple-choice questions, but never for essay questions. Learning to merely reproduce knowledge and skills is akin to moulding machines programmed to produce individuals who think alike, at the expense of nurturing individuality. I find it very disturbing that most go programmes in Korea for young players are skewed to short-term goals.

There is much else in similar vein. The reference to multiple-choice questions seems a key point. This approach pervaded education in very many disciplines a few decades ago, where it likewise got quick results but with shaky side effects (which took a long time to emerge, of course). There were similar quick-fix approaches, of course, and one in chess was the popularity of computer-like "algorithms" for making you stronger. Count +2 for having control of a central square, +1 for having 2 bishops, -3 for being unable to castle - that sort of thing. Initially it was all a numbers game but was later dressed up in a more discursive way that was still fundamentally computer-like. At the time computers were not all that strong in chess, so we couldn't say that people were imitating computer moves. They were trying to imitate computer thinking. I believe that what Nakamura is referring to in chess started out like that and so now has a long history behind it. It has simply been accentuated by the super-strong chess computers and people (even grandmasters) are now gulled more and more into copying moves - trying to play like a computer instead of trying to think like a computer instead of trying to think like a human...

We might assume that go players have jumped straight on to the high end of this paradigm, but I think Cho is showing us that the "rot" set in much earlier, well before AI, with more general education trends being applied to go.

His remarks chimed with me for another reason. It just so happens I have been editing the final draft of my book about Genjo & Chitoku. As most people here will know, they had starkly contrasting styles. This (as Cho implies) is one reason for their popularity with fans. There is also the fact that they played each other so many times over a long period that we can see their styles emerging. But because the way I do these books is to collect as many different commentaries on each game as possible, a further layer of fascination is added because we see cases where Genjo or Chitoku chose one move and five or six old or modern pros chip in with stylistic preferences of their own, and of course Elf and Lizzie add their onw takes. It is the complete opposite of formulaic. But at the end of it all, it is Genjo and Chitoku that the fans and pros are admiring. Humans admiring fellow humans as humans, irrespective of the actual moves.

That seems to be part of what both Nakamuara and Cho are talking about. As to whether they are right, though, I suppose the jury's still out and will remain so for a long time. At least until we all get our computer chip brain implants so that human limitations become irrelevant.

[Knotwilg]

For a beginner, an early attachment is natural.
For an experienced player, an early attachment is not natural.
When I told a beginner an early attachment is not so good, a high dan player (Bill) said he was not so sure.
Lately I learnt LZ likes an attachment as move 6 against the Chinese opening.

[Kirby]

Maybe better understanding what is NOT natural to you is part of getting stronger at go.

[Bill Spight]

For a beginner, an early attachment is natural.
For an experienced player, an early attachment is not natural.
When I told a beginner an early attachment is not so good, a high dan player (Bill) said he was not so sure.
Lately I learnt LZ likes an attachment as move 6 against the Chinese opening.

In the 1990s on rec.games.go I suggested the attachment against the 3-4 in the Chinese opening and got told off in no uncertain terms by a stronger player.

Edit: I was a US 5 dan then, for those who may not know.

[gowan]

I've seen and heard the word natural used in different situations. Takemiya characterizes his own style as natural. A pro teacher tells a student that the student's move in a certain position was natural and then says that some other move is better. In the first example, Takemiya's style is rather difficult to imitate and, after a period of imitation by other pros, was more or less abandoned by other pros. In the second example the implication seems to be that the student's move could be seen as following standard shape driven practice but the pro's move did not but was more effective.