Life In 19x19

I don't share in the presumed orthodoxy expressed here that bot games are ugly, unnatural, hard to understand etc. In fact I find a lot of them beautifully clear, with a focus on the important areas of the board, great local technique, sharpness, whole board balance etc etc. See for example https://www.lifein19x19.com/viewtopic.p ... 36#p241836 or https://www.lifein19x19.com/viewtopic.p ... 82#p240382 or https://www.lifein19x19.com/viewtopic.p ... 89#p240789 beating up an 8p. Maybe that's because I'm stronger than e.g. Ian and have got quite used to LZ style from playing as it on Fox (and using Lizzie I can interactively ask it questions to understand its play more). Also when LZ can easily beat a Fox 9d it can probably afford to play simpler style, but the games against other people using bots too aren't so different IMO.

P.S. Ninjad by Knotwilg

Just to clarify: for those who say that bot moves are easy or hard to understand, are you talking about bot selfplay games or of bot vs. humans?

What I find hard to follow in bot selfplay games, is the fact that they tenuki quite often. But some pro moves are also difficult to understand, and only become clear (for me) 10 or 20 moves later.

If you talk about aesthetics, or beauty, it's hard to reason why something is so or not so.
Those things come from feeling.

I am re-playing both Invincible and Invisible, so I'm doing the games of Shusaku and the games of AlphaGo (against other humans).
I find the difference in games to be tremendous.

But this probably also comes from a too one-sidedness that arises in the AlphaGo vs. humans matches. It's basically a slaughter, and many games in that book don't go past move 140, 150 or 160.
So perhaps that is also a factor.

But if you go by feeling alone, for me the AI style is "ugly". I'm not saying it's not deadly effective. But remember the last episode of Hikaru No Go, where the old man says: "Go is a two player game".

AI vs human is not a two-player game, and that takes away the beauty.
The most beautiful games are those where both players push each other to their limits.
So perhaps AI vs AI, then?

Yet don't underestimate the psychological aspect. If you know a game comes from 2 computers, you're bound to look differently at it. I'm most likely guilty of that, too. And deep inside you want to find beauty in human-ness, in nature. Not in robots or technology. I think.


It's always dangerous to make a comparison, but take cycling. (for some), it's a beautiful sport to watch. Every race tells a story, like every Go game tells a story.
What would we feel if all cyclists were replaced by robots. Technically they'd ride much better. Yet the beauty of the sport would be hugely impacted.
Sorry for the bad comparison
For some of us, Go tells a story. For some of us, the top Go players bare a peace of their soul on the Goban. What is more beautiful than that? Certainly not an algorithm telling me the 3-3 is technically a better play in that circumstance.

jlt wrote:Just to clarify: for those who say that bot moves are easy or hard to understand, are you talking about bot selfplay games or of bot vs. humans?

What I find hard to follow in bot selfplay games, is the fact that they tenuki quite often. But some pro moves are also difficult to understand, and only become clear (for me) 10 or 20 moves later.

Yes, the self play games look different, and the frequent tenuki give less of a sense of the flow of the stones.

yakcyll wrote: I find AI moves to be quite erratic and lacking said harmony. When humans play, the stones seem to flow, the formations are built; when bots play, the formations emerge as if from vast depths of the goban.

Perhaps when they are playing so that they can play millions of games for trying in a short period of time. Then they mostly rely upon their "intuition" (weights) without reading deeply (for them). By comparison, when bot faces bot with thinking time, perhaps they play more like humans. Case in point. Uberdude has been playing as Lizzie online, with thinking time.

Uberdude wrote:I don't share in the presumed orthodoxy expressed here that bot games are ugly, unnatural, hard to understand etc. In fact I find a lot of them beautifully clear, with a focus on the important areas of the board, great local technique, sharpness, whole board balance etc etc.

----

I have been wondering if the bots have given us a glimpse of what human go may have become without them in a few hundred years. I have recently looked at a couple of ancient, pre-Dosaku, go games with different eyes. The top players' tactical abilities were probably on a par with those of today's top players, but I think that today's top amateurs could give them a run for their money, if not trounce them. The first thing that strikes me is that they (pre-Dosaku) have no feel for overconcentration, which even today's amateur SDKs have. The second thing is that their understanding of the center is not as good as that of today's amateur dan players. (Again, pre-Dosaku.)

In AI games I see some crude and unpleasant looking moves, but mainly I see shocking beauty. You see some incredible sabaki play and some wonderful and sometimes shocking treatment of aji. some of these tenuki plays seem to be so wonderfully calculated in terms of the risk and reward.

Knotwilg wrote:In fact, I have found bot play easier to interpret than games of Go Seigen or Lee Sedol

Absolutely agree with this.
Lee Sedol in particular - I've felt convinced before that even Lee Sedol does not understand the moves of Lee Sedol.

And back on topic:

Marcel Grünauer wrote:Lee Sedol 9p has announced his retirement.

This is really sad.
Maybe it's quite reasonable too, given that what I've loved about looking through some of his games is the sharpness and that may no longer be so special now there are so many sharp young pros.

Lee Sedol via Kirby's rough translation wrote: .. On top of that, more and more, it seems that my matches with Ke Jie 9d haven't had any meaning.

I realise that this is right after the exhibition match, but still, seems odd that he would name Ke Jie personally when talking about his decision to retire.

I remember watching this game between the two players when they played in late 2017. This was after AlphaGo had introduced some interesting ideas that pro players were experimenting with, but before pros had really developed their concepts as I feel they have now.
For me, at the time, something unexpected happened in every corner and there are frequent surprising developments (for an sdk).

MikeKyle wrote:I've felt convinced before that even Lee Sedol does not understand the moves of Lee Sedol.

On the topic of Lee's memorable games, there was this miraculous comeback some years ago against Kong Jie (who was probably world #2 then) following a woopsy death with scant compensation leaving him far behind. But he keeps on fighting and complicating the game and turns it around.



And then this one from the same time just after his 2009-10 hiatus, I call it the 2-space high pincer workshop. Lee is black vs Chang Hao in the BC card cup final.

Kirby wrote: Roughly translated:

This year feels really different. I started playing go when I was 6, and in 1995 I became a pro. It looks like this year is the last year that I'll be active as a competitor. On top of that, more and more, it seems that my matches with Ke Jie 9d haven't had any meaning.

Thanks for the translation!
What does the "it seems that my matches with Ke Jie 9d haven't had any meaning" mean exactly - do you think he is disappointed about losing to Ke Jie more often than winning, or is he talking about something else (such as, they don't meet in important matches anymore? or... does he dislike Ke Jie's AI-style nowadays )

moha wrote:

macelee wrote:Ke Jie's opening was like a ruthless robot (following Fine Art's suggestions pretty closely, including some strange-looking second-line moves)

He was rumoured to have personal access to Fine Art, wasn't he? I guess humans can learn to do many things efficiently, not just playing go but also predicting a certain player's moves.

Is FineArt's playing style different than LeelaZero? I don't think it makes a difference nowadays for a pro to have exclusive access to FineArt - since both FineArt and LeelaZero are stronger than top pros, so Go AI has been "democratized".

What does the "it seems that my matches with Ke Jie 9d haven't had any meaning" mean exactly - do you think he is disappointed about losing to Ke Jie more often than winning, or is he talking about something else (such as, they don't meet in important matches anymore? or... does he dislike Ke Jie's AI-style nowadays )

Sorin: The score between Yi Se-tol and Ke Jie is now 5-13 (5-14 if you count pair go), and for the past couple of years and more they have met usually only by courtesy of specially arranged exhibition games. Add to that the fact that Yi had just experienced his 36th birthday and Ke is only 22. Ke also rubbed salt in the wound (no doubt unintentionally) by telling the press he grew up studying Yi's games.

And all that is within the framework that Yi is an emotional rollercoaster kind of guy. Which is why he didn't actually say he has retired. He said he is planning to make this year his last but he hasn't quite decided whether it will then be a temporary or permanent retirement.

At heart, just a bad case of tempus fugit, I suspect. Been there, done that. Like quite a few on this forum. You're next, uberdude!

sorin wrote:

Kirby wrote: Roughly translated:

This year feels really different. I started playing go when I was 6, and in 1995 I became a pro. It looks like this year is the last year that I'll be active as a competitor. On top of that, more and more, it seems that my matches with Ke Jie 9d haven't had any meaning.

Thanks for the translation!
What does the "it seems that my matches with Ke Jie 9d haven't had any meaning" mean exactly - do you think he is disappointed about losing to Ke Jie more often than winning, or is he talking about something else (such as, they don't meet in important matches anymore? or... does he dislike Ke Jie's AI-style nowadays )

It is difficult for me to say more from the single quote. I can make my own interpretation, but I'd be making some assumptions. I agree with John that Lee Sedol is somewhat temperamental (that's my opinion, not an objective fact) - and it's true that he didn't actually say that he was retired, just that it looks like this will be his last year.

In any case, I think it's somewhat safe to say from the quote that Lee Sedol isn't satisfied with his recent matches against Ke Jie. To say more, I'd be extrapolating.

Another article from a couple of days ago (http://m.tygem.com/news/news_view.html? ... &seq=32539) has another quote:

은퇴를 암시하는 것이냐는 질문에는 “아직 완벽히 정한 것은 아니지만 올해를 마지막으로 휴직이나 은퇴를 고려 중이다. 휴직은 어설픈 느낌이 있지만, 완전 은퇴를 하면 돌아오기 힘들 것이다. 은퇴를 하면 나중에 바둑을 못 두게 될 수 있기 때문에 여러 방면에서 고려 중이다.”라고 말했다.

And my quick translation:

When asked whether he was going to seriously try hard to retire, he replied, "It's still not perfectly decided, but I'm considering whether to take a leave of absence or retire by the end of this year. Taking a leave of absense feels kind of lame, but if I comletely retire, it'll be difficult to return. If I retire, later on it'll get to be where I'm not able to play. Because of that, I'm still considering various angles."

So maybe he's still thinking things over a little bit. Not really sure.

Remember that in 2013, Lee Sedol said he would retire within three years.

forum/viewtopic.php?p=133280#p133280

sorin wrote:

moha wrote:He was rumoured to have personal access to Fine Art, wasn't he? I guess humans can learn to do many things efficiently, not just playing go but also predicting a certain player's moves.

Is FineArt's playing style different than LeelaZero? I don't think it makes a difference nowadays for a pro to have exclusive access to FineArt - since both FineArt and LeelaZero are stronger than top pros, so Go AI has been "democratized".

I think I disagree with this, for pros at least. There are significant strength differences between go bots, and especially as a study tool for a pro I think that half stone could matter. For playing styles I'd guess the differences are even larger.

I have no real idea about Fine Art (apart from its strength), but I think joseki preferences, evaluation balances, opinion on certain lines, even W advantage on an empty board are viewed differently. LZ and Elf also seem to differ, and I'd guess even LZ would differ from itself after a new run from scratch. Selfplay training can reinforce and amplify (initially random) tiny NN preferences. The extent of ladder understanding can also make a huge difference (not for misplaying a ladder move at the top, but for the accuracy of searches where ladders arise several moves deeper).

https://sports.sina.cn/others/qipai/201 ... 142.d.html

Looks like he actually retired today.