Is this opening really so bad?

[bugcat]

This is a sanrensei with an old school joseki result, an opening has been played (against various left side positions) a few times professionally.

The free OGS bot gives a score of White + 5.7. I know that bot can be very unreliable, so could anybody double check this evaluation please? I'm sceptical that Black really loses his entire first move advantage within thirty moves.

I recorded it as a screencap because I can't remember how to make diagrams on L19.

[John Fairbairn]

The free OGS bot gives W+5.7.

I'd be interested to know what you and anyone else think W+5.7 means.

I find it hard to believe that anyone thinks it means that, all other thongs (and things) being equal - or whatever caveat you care to impose - White will win by 5 or 6 points.

In any event, my impression (and I think pros may view this in a similar way) is that the bot is not measuring territory at all, but rather is measuring the initiative.

I base my view of what pros believe on the basis of the sort of remarks they make. They normally never talk about points. In a no-komi, for example, they may say, "Black has lost the effect of first move." In other words, White now has the initiative. They do not say Black has lost 6.5 points. Or they may say, actual territory is equal but White has the thicker game (i.e. White has the initiative). Pros seem inclined to talk this way also in chess, although the meaning is a little different. I don't think anyone in chess ever talks about being 2.8 pawns ahead, unless there's a bot nearby.

It's difficult but useful to think what the differences in degrees of initiative are, e.g. slight, strong, very strong. At the very least it makes you think what factors might go into it, and also what sort of things you might do with the initiative.

In the present case, I'd say White has a close-to-strong initiative as it's his move and he can decide where the next "special operation" will start, and he has a lot of choice. But it's not a very strong initiative because Black has large choice of plays in the lower centre, and so can disrupt quite a lot of White plans on the left side.

[Kirby]

I would say that it depends on you: if you can use this position to win, keep using it. If you start losing a lot, try to figure out why.

[jlt]

My outdated Katago says W+4.8

[dfan]

I'd be interested to know what you and anyone else think W+5.7 means.

There are a few things this can mean for an AI, but it generally means something like "if the AI played both sides from here on, the average result would be White winning by 5.7 points." It can also mean something like "If the komi were adjusted by 5.7 points, both sides would have an equal chance to win", which is similar but not quite the same.

In any event, my impression (and I think pros may view this in a similar way) is that the bot is not measuring territory at all, but rather is measuring the initiative.

Well, it's measuring what it thinks the result of the game will be, that's all. It doesn't separate that into territory, influence, initiative, safety etc. That may sound like nitpicking but I think it's important to keep in mind what it's really saying. The good news about the fact that it's a raw number that needs interpretation is that it leaves something for the humans to do!

Anyway, I already said the following to bugcat elsewhere, but to have it on the record here:

Modern AI is way less intimidated by moyos than pros used to be. The AI as White says "I have so much cash at the bottom and I will find a way to erase your right side, no problem". It's not unusual at all to find pre-AI pro games that have a territory-vs-influence fuseki where the AI thinks the influence side is just toast early on.

Another good example can be found at https://lifein19x19.com/viewtopic.php?f=15&t=17051 (and note that that game was eventually won by the player that AI thinks is way behind).

[dfan]

I don't think anyone in chess ever talks about being 2.8 pawns ahead, unless there's a bot nearby.

I meant to respond to this too. Top-level grandmasters do use engine-eval-speak a lot, and I hate it. I'm sure it comes from sitting at the computer all day. They'll totally say things like "Yeah, this position looks like -0.4 or something but I thought I had chances" or "White's got to be +2 in this ending." Yuck.

[bugcat]

I don't think the opening looks particularly good for Black, for two human reasons.

1. The N5 stone hasn't been netted
2. There's a soft spot at Q6


Here's the information I work with.

1. Is this opening comprised only of joseki and known structures? Yes, but with an old joseki.
2. Has it been played professionally? Yes, in a few games, mainly from 1959-62 but with one outlier in 2011 (using Waltheri; GoGoD might have more).
3. Do I like the look of it? Not really (see first section).
4. What's the bot eval? And is it a good bot?

I evaluate this data, and if it's contradictory I progress to 5. Consult stronger players.

In this case it's 2. that doesn't quite fit with 1., 3. and 4. But if we regard the 2011 game as an isolated throwback, and guess that the opening was thought of as refuted at the professional level after 1962, we can lever the 2. data into line.

On the matter of how meaningful bot evaluation is, I'd say that an opening being in the region of five points lost is a "cause for concern". It's a flag that indicates to the player that they should check the formation carefully for defects and consider examining it more thoroughly (eg. with some variations, group strength labels and counting). Of course, the evaluation itself means nothing on its own.

[bugcat]

On that note, it could be nice to have a bot mod that does away with point margins in the UI and replaces them with vaguer comments.

0-4 points: "Even game"
5-9 points: "B/W has an edge"
10-19 points: "B/W is comfortable"
20+ points: "B/W is winning"

These margins could be calibrated by player level, like how mistakes are on AI Sensei.

[kvasir]

White at R7 (a) threatens to connect to the corner by ko. This aji means black's moyo is reduced. There is also the negative that black O6 (x) missed a chance to play K2 (b), the bots often want to play K3 (c) as white. Kind of just guessing but the aji could be worth 5-6 pts., missing K2 (b) is 3-4 pts. then there could be ~4 pts. extra for white if white didn't need K3 (c), that is white doesn't have sente according to the bots. YMMV

===Edit too many typos in the moves, who knows if I got it right now

I'll just make a diagram right now

$$c
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . a . . |
$$ | . . . . . . . . . . . . . Z . . . . . |
$$ | . . . . . . . . . . . . O X . . . . . |
$$ | . . . O . . . . . , X X X O X X X . . |
$$ | . . . . . . . . . c . O O O X O O . . |
$$ | . . . . . . . . . b O . X X O . O . . |
$$ | . . . . . . . . . . . . . . . O . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$c
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . a . . |
$$ | . . . . . . . . . . . . . Z . . . . . |
$$ | . . . . . . . . . . . . O X . . . . . |
$$ | . . . O . . . . . , X X X O X X X . . |
$$ | . . . . . . . . . c . O O O X O O . . |
$$ | . . . . . . . . . b O . X X O . O . . |
$$ | . . . . . . . . . . . . . . . O . . . |
$$ ---------------------------------------[/go]

[kvasir]

I'd be interested to know what you and anyone else think W+5.7 means.

I have the same view. It is how many moves or half-moves you have lost/gained. One half-move is about 6 pts. or just komi, of course a simplification but it works. That is it works even if you don't have a computer, it can be much easier to notice if either player played something slack or not than it is to count more ephemeral points. It doesn't work if you make many mistakes in your games, but strong enough players can obviously know how the game is going without counting a lot.

I learned this in Pandanet title match kibitz long before computer could play any good. I'd misquote and misattribute exactly what was explained then. I guess what one ends up learning is one's own theory, so for me it is about half-moves (or tempi as chess people call it).

[Knotwilg]

With hindsight everything is easy but

White has komi
White has sente - and a pretty evident spot to use it for, reinforcing the upper left corner
White has territory on the lower side, be it without much potential to expand

Black has a framework on the right side
Black has influence towards the center, though there's some aji in White's cutting stone.

It's not to hard to see why AI finds this to be to White's advantage.

Before AI times, I may have called this even though or even prefer Black. Nowadays I would be more aware of the potential overconcentration in Black's side star point and the influential stones, if White reduces from both sides. Alternatively, I would have more confidence today in finding defects in Black's influence and turning it into a target for attack, should Black try to make his stones operate on too grand a scale.

BTW bugcat, diagrams on L19 have the same syntax as on SL, but you need to embed them in the "go" keyword.

[John Fairbairn]

Black has a framework on the right side

Small point, but potentially important.

From my perhaps peculiar standpoint of analysing go terms closely, it seems almost universal of amateurs to talk like this, even high dans. It seems too loose.

Pros are normally rather different, especially in commentaries. They will qualify the moyo in some way (as here, I'm certain). For example, they will refer to a budding/incipient/fledgling/partial etc moyo, reserving just moyo for one that has actually been made or is about to be. (And if the opponent neglects it, and the moyo tightens up, it will usually be called a jimoyo, but we'll leave that for another time.)

It's not often easy to say when a moyo is complete, but the distinction seems important because action against a moyo is very much about timing (as in e.g. "invade one move before the moyo is complete"). If you don't make the proper distinctions, your timing will likely be all off.

I don't think the present case is yet a proper moyo. I infer that you agree, because you talk about White's likely next move being in the upper left.

Maybe tightening up on the language could usefully tighten up on the evaluation and timing.

Since many people mention AI's insouciance in face of a growing moyo, I would recommend looking at Kitani's games in his strongly territorial phase. He holds back before taking action against the opponent's moyo just like the bots do. But then he could spot and analyse and evaluate a yose-ko 40 moves ahead. Just like the bots do

[lightvector]

I'd be interested to know what you and anyone else think W+5.7 means.

I find it hard to believe that anyone thinks it means that, all other thongs (and things) being equal - or whatever caveat you care to impose - White will win by 5 or 6 points.

Even though the following isn't accurate to how the value is computed, I think the best way for players to conceptualize might be something like:
"If you gave black an additional 5-6 points, the bot would be no longer sure which side it preferred".

This phrasing makes it clear that:
* It's not a literal prediction of the final score (because it isn't).
* It's a preference by the bot, not some sort of objective truth about how *you* should value the position yourself - because maybe given your skill level, playing style, or understanding of your own strengths and weaknesses, you might have a different preference.

Similarly, I like the phrasing "the bot prefers white by 5-6 points" a lot more than "white is 5-6 points ahead". The former again makes it clear that it's a preference by the bot, rather than an objectively accurate calculation of the value of a position. Because, of course, because Go isn't solved, and people have also observed bots making mistakes and misevaulations, just like everyone else does.

A lot of people seem to have an instinctive negative reaction at quantifying things with a number, and I think for many people this is in part because in common parlance, giving a number like this connotes more precision and certainty than there is, especially this early in the game.

But I think numbers are useful because they give you a compact way to indicate the degree of something, if you can mentally shed any preconceptions that the number is suppose to connote some sort of precision or certainty. I think this phrasing even helps address that a little. If you forget about AI, and instead took a friendly pro willing to patiently humor whatever questions you had for them, and you asked them which side they would prefer if we gave black an extra 2 points, 4 points, 7 points, etc... at some point they would probably switch from preferring white, to being uncertain, to preferring black. You could pin a number on that crossover point, and with the context that this is how the number was produced, I think mostly everyone would understand what that number means and what it doesn't. This of course isn't the best way to use the time of a pro willing to teach you, there are many far better things you can do if you have a real human teacher, but with AI tools you work with what you have.

And I also think some people go too far in the other direction, considering such numbers to be almost meaningless. I think they are meaningful! If a strong pro judges "I personally prefer White here, but if were a no-komi game I would not be sure", depending on the position, an amateur might rightfully dismiss that value as having relevance to their own play because pros are at a totally different level of play. But it's still a meaningful thing to know about the position, and maybe in other positions it could materially weigh one's judgment. As long as you conceptualize them as described above, as explanationless preferences by a player with a particular known style (and even some known flaws), but still a very strong player's preferences - I think that's a good starting point for the right intuition on how to weigh them.

[RobertJasiek]

"Go isn't solved": For those 19x19 go positions that are solved by mathematical theorems, AI programs can still make mistakes because they do not apply the solutions.

[Kirby]

I like lightvector's clear explanation, particularly the part here:

* It's a preference by the bot, not some sort of objective truth about how *you* should value the position yourself - because maybe given your skill level, playing style, or understanding of your own strengths and weaknesses, you might have a different preference.

Bots are great in that we can get their opinions about a given board position. And those opinions may be sourced from an entity having stronger go ability than our own.

However, the way to play that gives you, the player, the best chances of winning could be quite different than what the AI recommends.

It's for this reason that I find value in gaining a lot of game experience and just trying out positions you might be curious about. It's in that manner that you can acquire your personal intuition regarding what works for you and what does not.

In the long term, it can be valuable to look at areas where your own intuition differs from that of the AI... But it's still important to have your own intuition. and perspective - though it may sometimes be totally wrong!