More on Joseki and Approximating Equality

[Polama]

Who cares? It is, as described, straightforward to relate evaluation of positions to sequences.

The minmax algorithm? Straightforward to describe, not even remotely straightforward to implement. Even if we're conservative and say we'll only evaluate 4 possible moves per position, 8 moves is already over 65,000 positions to evaluate. Any sort of fighting at all, and 8 moves is far too short. And although we may start with a peaceful exchange, we need to make sure that the most aggressive fighting moves for the players don't work.

This is why the standard sequence definition is used. Even with thousands of years of go study, the opinions on joseki are still fluid. New exchanges in any of the branches of the minmax tree can ripple through and change long accepted sequences.

In a game we evaluate a tiny fraction of the moves and hope for the best, but success in even many games doesn't promise you haven't been making a series of terrible moves.

[Boidhre]

Why not just drop the word joseki altogether and say you're doing theory on common corner sequences? Then you avoid the argument about what joseki means and get to do exactly the same research no?

[skydyr]

Why not just drop the word joseki altogether and say you're doing theory on common corner sequences? Then you avoid the argument about what joseki means and get to do exactly the same research no?

Or even corner positions, to eliminate the sequence and get at the end result.

[Boidhre]

Why not just drop the word joseki altogether and say you're doing theory on common corner sequences? Then you avoid the argument about what joseki means and get to do exactly the same research no?

Or even corner positions, to eliminate the sequence and get at the end result.

Exactly. Neutral jargon free language allows precise definition of what you want to study without any potential confusion amongst your readers. I fail to see the down-side of this.

[RobertJasiek]

Polama: Not pure min-max is applied by a go player, but something reducing the calculation complexity by considering only the relevant / interesting variations. Therefore I wrote "must be applied to the relevant sequences". (It is possible to err and overlook part of the relevant sequences, but anyway it is like decision-making can be made.)

Boidhre, skydyr: We can speak of "corner josekis, corner joseki-like variations and other corner variations and the resulting positions of such sequences, for which approximated equality is studied". "Joseki" is just a shorthand for that. ("Joseki" is also just a shorthand for "standard sequences", because it is always open to discussion which are / were / will be those standard sequences.)

[Boidhre]

Boidhre, skydyr: We can speak of "corner josekis, corner joseki-like variations and other corner variations and the resulting positions of such sequences, for which approximated equality is studied". "Joseki" is just a shorthand for that. ("Joseki" is also just a shorthand for "standard sequences", because it is always open to discussion which are / were / will be those standard sequences.)

And here is the core problem for you. That's your definition or understanding of the term and not everyone else's. If you use a word like Joseki you need to be extremely careful that you conform to the normal usage or people will get very confused by what you're saying. Get rid of the potential debate of the meaning of a single word, it just makes it more difficult for people to actually approach your research because every time they see joseki being misused (in their eyes) it'll be jarring and distracting. You're generating heat not light with this approach.

[Polama]

Polama: Not pure min-max is applied by a go player, but something reducing the calculation complexity by considering only the relevant / interesting variations. Therefore I wrote "must be applied to the relevant sequences". (It is possible to err and overlook part of the relevant sequences, but anyway it is like decision-making can be made.)

Yes, we have lots of heuristics to avoid exhaustive minmax. But my point about 4^8 was that that's already absurdly pruned. One the primary traits of the opening is how unconstrained your choices really are.

I agree that something like minmax is very useful during play, branching out into a handful of interesting paths that seem like the best choices for each player, evaluating the resulting information, and making a decision based on that. And having a principled approach to that evaluation is a powerful thing.

But it's not only possible you've erred and missed relevant sequences, it's certain. When somebody says an exchange is joseki, it communicates to me that this has stood up to extensive analysis by many professional level players.

With a joseki, I can start at the beginning and jump 15 moves to the future and say "this is a path we could take that ends like this". I can be confident that (assuming I'm careful about surrounding stones), I'm not going to be in trouble if my opponent changes direction (other than into another joseki inappropriate for this position).

If, on the other hand, somebody tells me "here's a corner sequence I played that gave me good results", that's interesting information I want to examine, a good path for studying if either player made a mistake, something to emulate in my games even. But I wouldn't start at move 3 and say, 'ok, I can take this path for the next 10 moves'. I'd cautiously proceed down that path, re-evaluating every step, not necessarily expecting to end up where the speaker ended up.

And that's my objection to the use of the term joseki for any good seeming corner exchange we come up with. The distinction between "professionals have been playing this for 50 years, and nobody has found a way for either side to get an advantage", and "I looked at the continuations I could think of and they seem alright" is an important practical one.

[RobertJasiek]

The distinction between "professionals have been playing this for 50 years, and nobody has found a way for either side to get an advantage", and "I looked at the continuations I could think of and they seem alright" is an important practical one.

The former: this describes that (after the first few moves) alternative better moves have not been found, but it does not evaluate the resulting position.

The latter: Yes, but "seem alright" is not what I suggest. I suggest an evaluation method.

[Polama]

The distinction between "professionals have been playing this for 50 years, and nobody has found a way for either side to get an advantage", and "I looked at the continuations I could think of and they seem alright" is an important practical one.

The former: this describes that (after the first few moves) alternative better moves have not been found, but it does not evaluate the resulting position.

The latter: Yes, but "seem alright" is not what I suggest. I suggest an evaluation method.

I didn't mean "seems alright" in terms of your positional evaluation, I mean seems alright in terms of 'this position is actually reachable with excellent play by both sides". To a first approximation, this is a reasonable position to arrive at, presumably through this sequence of moves. Is there something extremely subtle that gives a player a chance to sidestep this branch into a better one for them? It doesn't appear so, but we can't know with any confidence yet.

My point is that the former and the latter are orthogonal. Your approach does something traditional joseki don't (a very precise measure of quality), but it doesn't do the main thing traditional joseki's do (extensive confirmation that much better moves don't exist for either player). Thus to call the positional evaluation joseki creates confusion: your approach isn't doing _more_ than the traditional joseki, it's doing something different.

[RobertJasiek]

Is there something extremely subtle that gives a player a chance to sidestep this branch into a better one for them?

I am not worried so much by the subtle, but by possible better alternatives of a) tenuki or b) an EARLY different move choice.

it doesn't do the main thing traditional joseki's do (extensive confirmation that much better moves don't exist for either player).

My evaluation does not exclude such considerations.

your approach isn't doing _more_ than the traditional joseki, it's doing something different.

Here you compare method with sequence:) So I am not sure which thing you are referring to that allegedly was not being done more.

[Polama]

it doesn't do the main thing traditional joseki's do (extensive confirmation that much better moves don't exist for either player).

My evaluation does not exclude such considerations.

your approach isn't doing _more_ than the traditional joseki, it's doing something different.

Here you compare method with sequence:) So I am not sure which thing you are referring to that allegedly was not being done more.

This discussion arose from your use of phrases like "recognize joseki yourself", "determine joseki on your own". While your evaluation of a position (or your approximate min-max for evaluating a sequence) does not exclude such considerations (the absence of superior paths at earlier points in a sequence), neither does it provide them, or provide a tractable method for determining them. For the traditional meaning of joseki, 'determine joseki on your own' just doesn't make sense: the reason being that traditionally joseki are less about obtaining equal results and more about definitely attaining a minimum quality of result.

Evaluating the balance of a position is not equivalent to evaluating that all the moves leading to it were right. That you could theoretically recursively evaluate all options does not mean that you could do so in practice: supported by the evidence that old joseki do get thrown out, despite hundreds or thousands of years of use.

This is the linguistic mismatch I'm trying to point out. A process for determining joseki would be a process that looks at a sequence and determines that neither player had another option that was superior. At the moment, relying on millenia of professional play is the best we've got for that. You're suggesting a method for evaluating a position. I can buy that it's superior to the alternatives for evaluating a position. It does not follow that it can be applied to determine joseki at anywhere remotely approaching the alternative of relying on collected experience.

[Bill Spight]

This discussion arose from your {Robert Jasiek's} use of phrases like "recognize joseki yourself", "determine joseki on your own". While your evaluation of a position (or your approximate min-max for evaluating a sequence) does not exclude such considerations (the absence of superior paths at earlier points in a sequence), neither does it provide them, or provide a tractable method for determining them. For the traditional meaning of joseki, 'determine joseki on your own' just doesn't make sense: the reason being that traditionally joseki are less about obtaining equal results and more about definitely attaining a minimum quality of result.

That puts me in mind of what an 8 dan pro lecturer said in Kyoto years ago: "If I play it, it's joseki."

[RobertJasiek]

For the traditional meaning of joseki, 'determine joseki on your own' just doesn't make sense: the reason being that traditionally joseki are less about obtaining equal results and more about definitely attaining a minimum quality of result.

I rather think that also the traditional thinking strives very hard towards approximating equality as well as possible.

Evaluating the balance of a position is not equivalent to evaluating that all the moves leading to it were right.

But... I do not suggest to abandon knowing representative / standard variations, so that one need not evaluate "all" the moves.

A process for determining joseki would be a process that looks at a sequence and determines that neither player had another option that was superior.

Again, this overlooks a) possible tenukis and b) possible EARLY different move choices.

It does not follow that it can be applied to determine joseki at anywhere remotely approaching the alternative of relying on collected experience.

Yes, because my method is more powerful: it determines joseki-equivalent results of given positions even long before collected experience for them is available.

What do you think are players doing when they use a traditional approach described as "during their games, they go beyond established patterns and develop very good, new patterns on their own"? Somehow, those players must have a sense of approaching equality without any longer relying on collected experience. I think that such players do subconsciously something similar to what I suggest doing with my evaluation method.

Not the collected experience is so important, but an understanding of when one approaches equality.

The collected experience serves mainly as a profound test sample with which to calibrate my method or such players' advanced subconscious understanding.

Even a proverb has got such insight: do not abide by josekis blindly.

[Polama]

Here's the original diagram for this whole discussion.

$$B
$$ ----------------
$$ | . . . . . . . .
$$ | . . . . . . . .
$$ | . . 4 6 7 . 3 .
$$ | . 8 1 5 2 0 . .
$$ | . . . . 9 . . .
$$ | . . . . . . . .

Click Here To Show Diagram Code

[go]$$B
$$ ----------------
$$ | . . . . . . . .
$$ | . . . . . . . .
$$ | . . 4 6 7 . 3 .
$$ | . 8 1 5 2 0 . .
$$ | . . . . 9 . . .
$$ | . . . . . . . .[/go]

Let's look at move 4. Is move 4 definitely good for white? Do any of these marked points, or a tenuki, lead to a black advantage?

$$B
$$ ----------------
$$ | . . . . . . . .
$$ | . . f e . . . .
$$ | . . 4 b c d 3 .
$$ | . . 1 a 2 . . .
$$ | . . h j . . g .
$$ | . . . . i . . .

Click Here To Show Diagram Code

[go]$$B
$$ ----------------
$$ | . . . . . . . .
$$ | . . f e . . . .
$$ | . . 4 b c d 3 .
$$ | . . 1 a 2 . . .
$$ | . . h j . . g .
$$ | . . . . i . . .[/go]

Let's look at f, which is among the less likely continuations.

$$B
$$ ----------------
$$ | . d . . . . . .
$$ | . a 5 . e . . .
$$ | . 7 4 6 . . . .
$$ | . b 1 . 2 . g .
$$ | . c . f . . . .
$$ | . . . . . . . .

Click Here To Show Diagram Code

[go]$$B
$$ ----------------
$$ | . d . . . . . .
$$ | . a 5 . e . . .
$$ | . 7 4 6 . . . .
$$ | . b 1 . 2 . g .
$$ | . c . f . . . .
$$ | . . . . . . . .[/go]

Here's just one possible continuation. Do any of the marked moves lead to, say, an unexpected sacrifice of the corner where black gets better than expected thickness? Or a chance for black to live small, but with an attack on white? Are we willing to categorically state no good can possibly come from this for black with good play?

$$B
$$ ----------------
$$ | . . . . . . . .
$$ | . . . . . . . .
$$ | . . 4 . . . 3 .
$$ | . . 1 . 2 . . .
$$ | . . . . . . . .
$$ | . . . . . 5 . .

Click Here To Show Diagram Code

[go]$$B
$$ ----------------
$$ | . . . . . . . .
$$ | . . . . . . . .
$$ | . . 4 . . . 3 .
$$ | . . 1 . 2 . . .
$$ | . . . . . . . .
$$ | . . . . . 5 . .[/go]

Here's an unusual response by black. That 1-5 gap looks ugly. But are we positive this isn't the start to complicated fighting that gives black an edge?

Where in your analysis did we consider that high 5 move? If it's good for black, we should call that diagram not-joseki, and stop playing 4 here.

In a game, whether intuitively or through your method or any other we explore a tiny fraction of the possible boards. Our opponents explore a similar sized space, so we get comparable result. If the example 5 above unexpectedly leads to an advantage, I would never expect my opponent to stumble onto it in a game. If they play it, I would never expect the two of us to plumb any meaningful fraction of the continuations in that one game.

Except for yose and life and death, every judgement in go is suspect. It's too large a state space. Statistics tells us that if one player sits down to a board and says one thing, and 200 years of experience by 10,000 professionals suggest something different, the 10,000 are probably right. This is what your method, what no method, can substitute: the processing power of many minds over many years.

So when we say joseki, we mean 'battle tested'. We say, it's been offered up as even, and it's yet to be conclusively refuted. That's a stronger claim then any amount of analysis done over the course of a move in a game of go.

None of which is an attack on your process. It's an explanation why those of us who object, object to your calling it joseki. Of course players have to find moves on their own during a game. Of course they should train and learn and make the best moves they can, and of course they can't call 10,000 professionals over to do it for them. But we don't call those moves joseki, because the thought given in the space of one move, no matter how strong the player, can't fairly be called equivalent to the combined work of all the players exploring the consequences of joseki. It's a useful distinction to have between "I played this in a game because I thought it was even and it turned out even" and "This has been played over and over by many people, enough of whom found it turned out even".

[leichtloeslich]

Except for yose and life and death, every judgement in go is suspect.

I basically agree with most of what you say, and I think what you're getting at has something to do with RJ's method of evaluation being somewhat circular:
he presents an "algorithm" which takes as input the positional judgement of a player and outputs a positional judgement.
(Iirc his method requires evaluation of "thickness" and "stability", both of which heavily rely on playing strength/experience.)

So that's how that search space gets pruned down to manageable proportions: we use our judgement while looking at candidate moves.

An obvious question is: will RJ's method significantly improve upon the quality of the positional judgement?

Like, the algorithm takes as input the judgements of a 3k and outputs the judgement of a 2d. I'm highly suspicious that it does.