More on Joseki and Approximating Equality

[RobertJasiek]

Where in your analysis did we consider that high 5 move? [...] every judgement in go is suspect.

To make judgements more meaningful, first create (more) stable / quiet positions, then evaluate.

It's too large a state space.

With a quiet position, the follow-up variations space is "small".

This is what your method, what no method, can substitute: the processing power of many minds over many years.

A known method overrides unknown mind processes, which are inaccessible for everybody else.

So when we say joseki, we mean 'battle tested'.

So what? Unless all those battles are recorded and analysed WRT to a particular joseki, they provide no or only little information, e.g., because other moves of the game can have greater impact on the "battles".

That's a stronger claim then any amount of analysis done over the course of a move in a game of go.

No. For something to be a stronger claim / evidence, it must be described and it must be compared to the analysis.

the combined work of all the players exploring the consequences of joseki.

Where does this combined work consider even the basics of stone difference and influence properly? ALA even the combined work fails on explaining well the basics, it is by far not worth as much as you claim.

he presents an "algorithm" which takes as input the positional judgement of a player and outputs a positional judgement.

Not "of a player". The evaluation is essentially independent of the player doing it (if only it is done meaningfully according to the method).

(Iirc his method requires evaluation of "thickness" and "stability"

No. It requires stone difference, territory count and influence stone difference. Application is easier for a quiet and stable position. Thickness is not required, but it can be one of the other, possibly significant aspects to be considered optionally.

will RJ's method significantly improve upon the quality of the positional judgement?

1) My methods for whole board positional judgement.

2) My "joseki" evaluation method.

(1) is more generally applicable and more frequently relevant. So if you seek improvement in judgement, I suggest you start with it.

For (1) and (2), the quality of my positional judgements in my played, kibitzed or studied games / positions and strategic decision making due to them have improved greatly.

the algorithm takes as input the judgements of a 3k

The method (2) does not take player judgements as input.

[Polama]

Where in your analysis did we consider that high 5 move? [...] every judgement in go is suspect.

To make judgements more meaningful, first create (more) stable / quiet positions, then evaluate.

It's too large a state space.

With a quiet position, the follow-up variations space is "small".

Ah, I think here is where we disagree. Let's imagine I read ahead as white, analyzing many stable variations, and find a path to equal results that suits me. I play 4, intending to reach those. Let's imagine my opponent instead reads only 5 to great depth and finds that a complicated fight leads to an advantage for him, and that I must take a small loss to keep things simple. I am now behind. In this particular example it's unlikely my opponent could analyze that exhaustively, but in others he could, or he may have a prepared trick play.

Even though I prefer stable, quiet variations, to get to those variations I must accept that my opponent has the option of diverging into complexity. I can only control one player in a game. I can play moves that minimize my opponents aggressive options, but at the same time I must often pay a cost to force us into simplicity.

If, though, I know that move 4 has been analyzed for many hundreds of years and nobody has found a refutation, I can more safely ignore 5. If my opponent forces us off onto a complicated fight I had not read yet, I can be more confident that strong players have read this path, and at the very least some obvious continuation is even or good for me. If, on the other hand, I'm in uncharted territory, I must be more cautious of fighting paths for my opponent, because one of them might lead to an advantage fairly quickly, at a depth my opponent can read.

I can only consider a small selection of positions even spending 20 minutes a move. If a move is established as even, I can use my time to look deeper and analyze deeper stable positions to select a path. If a move is novel, I should spend some of my time making sure the complicated attacks my opponent might launch can at least be sidestepped, if not fought to a draw or win. Thus, knowing whether a move has stood up to examination of many eyes dictates whether I should spend my time looking past it, or spend my time making sure I'm not missing a strong counter by my opponent.

If somebody shows me a sequence and says it's joseki, I want to know whether they mean the end position is even, or if they mean (in the very limited sense possible) that all the variations for each player all the way to this point have been examined, and found to be safe. Because that dictates how I can spend my time in an actual game. Right now, somebody can say 'this is joseki, but better for black', or 'this is even.' and I know both how they evaluate the position and if the surrounding state space has been well explored. If, instead, we start saying 'this is joseki' to mean 'this is even', I know longer have the other information. We've made communication harder. Whether a position is well vetted is interesting to me, so it should have its own word for efficient communication, and in fact it _does_ have its own word: joseki.

[RobertJasiek]

Also among the traditional so-called josekis there are lots of those resulting in fights. However, in the literature, they are shown up to moments of partial stability and straightforward fighting options for the involved instable groups. Although the positions are by far too complex to be read out, they are also sufficiently partially stable to allow evaluation. Therefore, the difference between our views is not as great as you might have thought.

[Polama]

Also among the traditional so-called josekis there are lots of those resulting in fights. However, in the literature, they are shown up to moments of partial stability and straightforward fighting options for the involved instable groups. Although the positions are by far too complex to be read out, they are also sufficiently partially stable to allow evaluation. Therefore, the difference between our views is not as great as you might have thought.

I suspect that's true. I think we're very much in agreement about the general ideas in discussion. I approve heartily of the attempt to formalize evaluation, and agree that for actual play it's a more important skill than knowledge of established sequences. In fact, I think the similarity has made the discussion a little trickier, as we've tended to overshoot each others positions =)

My argument has been communication based: that to many of us, it's interesting to have a distinction between a long established sequence and a newly introduced sequence, however that sequence is derived. That word has traditionally been joseki, so to move it to a related but different meaning would work against efficient communication.

[RobertJasiek]

The Ishida speaks of "virtually equivalent to joseki". ("Equivalent" is a bit too strong though. "Approximately equal" is more correct. But maybe "virtually" should better have been translated to "approximately"?:) )

[leichtloeslich]

The evaluation is essentially independent of the player doing it (if only it is done meaningfully according to the method).

Those are some nice weasel words you got there.

Does that mean that a computer program that knows nothing but the rules of the game can use your method and gain evaluations of local positions equal in quality to that of a high dan amateur?

If so, I think you may have just found the holy grail of computer go: a computationally non-expensive evaluation function that produces useful results.

If not, it would appear that playing strength does play a major role in your evaluation method.

[RobertJasiek]

It is not quite as you suggest; some programming effort and working out of algorithmic details are still needed. However, I think that it can be done by an applied expert program using the method, implied assumptions and the related principles for territorial positional judgement. More precisely, the basics of the method. When it comes to the "other aspects", the programming effort becomes huge. Possible maybe, but several months of coding at least.