A few years ago, seemingly convincing rumours have been spread saying that, during the opening and middle game, Michael Redmond 9p could make an accurate territorial positional judgement by reading ahead the game using endgame-like sequences of 100 globally optimal plays. The myth was born that using endgame-like sequences was a feasible method for super-strong players for making an accurate territorial positional judgement also during the opening and middle game.

Redmond's own territorial positional judgement during his live commentaries of the 5 games of the match AlphaGo - Lee Sedol debunk the rumour, which was very far from the truth. Besides, he only occasionally showed hints of making non-territorial positional judgement at all.

Whenever Redmond was supposed to make a territorial positional judgement during the opening, middle game, early endgame or middle endgame, he interrupted the unfinished task, only gave very rough counts, said that making a territorial positional judgement was too early or said that it was too difficult [for him] to be done at all at the moment. When making a partial, local territorial positional judgement, he used the core idea of minimal remaining territory after reductions, which is also used by Cho Chikun and me. However, even then Redmond did not apply much beyond the core. Instead, he had difficulties relating, and calculating consistently for, different parts of a position. Although I watched most of his comments, I did not even once witness Redmond doing an accurate territorial positional judgement of the whole position until the micro-endgame. His use of the method of minimal remaining territory after reductions means that, during the mentioned stages of the game, he does not use the rumoured method of very long, global endgame-like sequences.

Nevertheless, he reported how he had studied the endgame: he had taken pro games, removed the last 30, 50, 70 or 100 moves before the final scoring positions, printed out the created positions and then imagined endgames for them. He explained that 100 moves had been too hard, 70 already very hard and 50 significantly less easy than 30. Surely this method of study is worth trying for improving one's late endgame. However, it is not the method he uses for territorial positional judgement during earlier stages of the game.

"100, endgame, counting territory, Redmond" must have been the keywords causing the rumour, and the people creating it might have misunderstood, overestimated his skill and confused positional judgement during the micro-endgame with territorial positional judgement during the opening, middle game, early endgame or middle endgame.

Myth debunked!

How about the other myth of perfect endgame (starting during the early middle game) of Edo players, when games were played with unlimited thinking times?

Speaking in terms of "debunking myths" rings some serious crank bells for me...

I don't think what Redmond can do as a live commentator corresponds very well to what he can do as a player with lots of time on the clock. During the first few games he talked quite a bit about how much harder it is to do analysis whilst talking at the same time.

If he thinks that he needs to talk non-stop, he can count aloud and point to the counted intersections. Could be great teaching. Instead, there was too much hand-waving, interruption, improper excuses even when counting would hsve been easy etc.

If you don't enjoy it, don't watch it. He's making a judgement based on decades of experiences of the right tone and level to set to appeal to the demographic he's targeting himself at. If that's not you, move along, rather than complaining he's not serving your own educational needs.

topazg, my various intentions for watching the streams are meta-discussion. Redmond addressed a wide range of levels. Everybody is interested to know who is ahead, so every commentator should make judgements. It is the commentator's choice whether he just declares who is ahead or demonstrates counting on the board. Redmond often chose to do the latter and spend time on it. So far so good, but then it is also valid to discuss his counting skill. Besides, it is important for everybody to stop desinfomation about unrealistic judgement methods supported by false evidence.

I stopped posting here because a certain person did not know when to shut up. But coming up on deck to try to apply a touch on the tiller is merited whenever especially jagged rocks heave into view.

1. A myth is easy to debunk if you use the straw man tactic of inventing it yourself.

There are many books on positional judgement in the Oriental literature. None of them uses mainly territory counts for assessing the early and early middle game. The usual paradigm is: first, relative strength of groups (or in short - are there any weak groups?); second, shape and efficiency of stones; last, comparison of territories.

By these means a pro like Redmond can look at a position and often say instantly who is ahead, because weaknesses and inefficiencies leap out at them. Even amateurs can use these methods successfully, though obviously, an apparently weak group is not weak at all if you can read out a way for it to live, and, if you are a pro, inefficiencies can be more finely graduated. Territory counting is merely a back-up, and as such doesn't have to be super-accurate.

2.

Unlikely to be a myth. O Meien has shown in great detail how it is done.

On page 126 of his endgame book he says that there is a formula for predicting certainty of victory. What he means by "certainty" is actually a term from Japanese psephology, but will be familiar to students of American elections where a result can be "called" even before the voting has finished. In many respects it is like AlphaGo using winrate rather than precise points. To understand O's theory you need to know how to obtain the "value of a move" by "absolute counting", but that is very easy - easier than de-iri counting, but you can use that instead if you want so long as you make a very simple conversion.

Without explaining the details but just to show how simple his method is, here is how he defines the critcal formula:

Advantage of first move = half the value of a move
Margin of error = half the advantage of first move

To this he adds these riders:

* If it is the opponent’s turn to play, even if you add the advantage of first move and the margin of error to the opponent’s territory, if you are ahead you have “certainty.”

* If there is an outstanding big move for the opponent, assume he can play there then count. Add the advantage of first move to you territory and add the margin of error to the opponent’s territory.

Typically he applies this formula at around move 150, but sometimes around move 100. Rather little reading is required, and certainly within the scope of many amateurs.

Just to show how straightforward the thinking is, here is a section giving his insight into the first position he deals with (this is prefaced by showing the reading required in six parts of the board - between 2 and 6 moves in each case).

First, as per the count on page 124, Black’s territory is 83⅔ points and White’s is 80 points, ok?

Next, it is Black to play, so we calculate the value of his “advantage of first move”, but as mentioned under Diagram 9 the biggest boundary play remaining on the board at this point in time is the block on the lower right, that is the 3-point reverse sente. The “value of the move” for the 3-point reverse sente is 3 points, and so the value of the advantage of first move in this position is 1½ points.

Next, the margin of error. There is a relationship with things like the last play (the final tedomari) and so it is not something we can calculate exactly. However, even if we slip up, there is a limit to this, and what I am saying is that keeping it within a value of half of the value of the advantage of first move is a figure that I have come up with on the basis of my experience so far.

Accordingly, as the advantage of first move in this game at this point in time is 1½ points, the margin of error is half of that again, which means it is ¾ point.

So we add 2¼ points, being the sum of advantage of first move and the margin of error, to the 83⅔ points previously calculated for Black’s territory. The produces a figure of just under 86 points, but if we compare that to the 80 points for White’s territory, we can see that Black does not have enough to give komi.

On the other hand, if we take White’s point of view, he is ahead even if the advantage of first move and margin of error are added to Black’s territory. This situation is what I am calling “certainty.” The fact that there is no doubt about the evaluation has already been proven by the further moves in Diagram 9.

Later examples are more or less as straightforward as this, but add a little extra each time, e.g. showing how kos are handled, and in two examples he shows how Edo players and Yi Ch'ang-ho can safely predict early on that they have won by at least a half a point. Actually the method never changes; he is just adding sidelights.

Now the really interesting point is that O says he did not know about this method when he became a pro. He says: "In Japan, it seems that very few people, of which I am one, are using Absolute Counting, but I have heard that in Korea and China many people use it." He later adds that the method seems to have been "forgotten" in Japan, but implies - from an approving reference to Edo players being so accurate with counting - that it was once used. (O has the advantage, denied to most fellow pros in Japan, of speaking Chinese, but he does not reveal how he learned about it.)

So, until you read this book and all the others by pros on positional judgement, debunking anything seems a little out of order.

I will now retreat back below decks.

This link: http://www.usgo.org/news/2010/06/michae ... ros-train/
may contain the information from which you talk about Michael Redmond's endgame research.

I think a much milder tone should be used when commenting on professionals - here and in other threads - and would actually increase the number of people who give serious thought to your stuff.

Dear John,

Would you kindly mind giving the title, or the ISBN ?

Thomas

_________________
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)

Thank you very much, Marcel.

I own this book in paper, but haven't read it yet.

_________________
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)

This reminds me of Eric Schiller's work: Unorthodox Chess Openings.

_________________
North Lecale

If that's the reason you choose to watch the streams, you are probably in a minority of fairly close to one. I'm sure the points you are making stand you in good stead with any others that may be watching those streams for the purpose of engaging in a lively meta-discussion debate, but I think you might find it hard to get other takers for your goals. Aside from all the rather unsubstantiated assumptions at what Michael may or may not be doing in the official English language stream, using it as evidence to support any claim or counter-claim is just nonsense.

He started the first two games explaining what the point of the board game Go is. Are we to assume then that he's reminding himself how to play the game before embarking on the commentary? Everything he chose to say, show (and, just as importantly, not say or show), is based on providing a valuable stream to those he expects/has been told are likely to be the primary audience. There were a number of situations in the games which showed just a glimpse of variations he'd read in some detail all over the board and chosen not to discuss with the audience until it seemed contextually relevant (such as the L/D status of the black group on the left in game 4 after the "divine" move was played and the White group was escaping to safety).

You simply can't base any conclusions on that stream based on Michael Redmond having omitted to say certain things by assuming that he cannot have seen/read them.

Also, this:

John, territory presumes life, so assessing statuses before territory is, of course, the right order. I have suggested that every relevant aspect of a position must be taken into account incl. development directions, neutral stones, statuses (connection, life and death, stability, LD of imagined invasion groups), options, reductions, invasions, aji, potential, influence, thickness, fights (and their aspects). Among these, strengths of groups can be assessed absolutely or relatively (when e.g. comparing Black and White, we might call it "relative").

Shapes or efficiency have aspects of the past and aspects of the future. Depending on how one makes a positional judgement, either, neither or both (past and future) can be taken into account. E.g., one can use aspects of the past (such as in tewari) as a backup for alternative methods of judgement.

I disagree when you say that territory was merely a backup. I consider it an important aspect because it is directly related to the game aim. - Depending on how and when one makes a positional judgement, a territory count can be accurate or approximate if the margins of error are considered consistently and meaningfully. Near the game end, a super-accurate count is preferable. Earlier during a game, approximations can be tolerated more easily. Then, I prefer accuracy of about 1/2 point because such enables strategic decisions between different, but almost the same values and the related next moves. With approximation, some decisions cannot be made meaningfully but can only be made with, as you might call it, "certainty". However, such approximate certainty involves arbitrary decisions when sometimes correct decisions can be distinguished from wrong decisions when relying on (more) accurate values.

Since we agree that there are various aspects worth taken into account when making positional judgements, it is meaningful to expect such also from Redmond.

Thank you for outlining absolute counting. However, I wonder whether small fractions are everybody's preference. Surely, it must be an alternative to update counts of the whole position from time to time so that one does not lose oneself in iterated smaller and smaller fractions. Fractions are good for deciding among next move candidates. Elsewhere, you have expressed your disliking of fractions; here you mention certainty and a tolerance of a margin of error; so I think you would agree that we need not overemphasise tiny fractions during the opening and middle game.

The book you mention and the method of absolute counting have nothing to do with Redmond's counting during the live commentaries and the debunking of the rumour that he used endgame-like sequences 100 moves long during the middle game. As interesting as absolute counting as a topic is, introducing it does nothing to my claim in the thread title.

mhlepore, you suggest a milder tone. This is meta-discussion. I do not say "Maybe his judgement was not the best part of his performance" when, elsewhere, even a newbie has been clearly aware that Redmond's territory counting has been weak. Improvement comes from avoiding mistakes, and in particular from avoiding to learn from weak aspects of a (live) commentary. If we bury the mistakes, unnecessarily many will accidentally learn from them. Redmond has made predictions of the winners not supported by continued game developments or eventual winners, predictions sometimes disagreeing with Kim's judgements. Everywhere (not only in this forum), I see doubts about Redmonds judgements or bad advice for the AlphaGo team or other programmers as a consequence of believing his judgements naively. Clearly pointing out weak positional judgements helps those wishing to learn, or letting their programs learn, faster. BTW, for that task, accurate judgements are more helpful than margins of errors or uncertainties.

Let me explain it once more: I have had various reasons to watch the streams.



At other times, he went into tiny fractions of local endgame counts. (His commentaries were meant for a wide range of strengths in the audience.)

Marcel, I have not made my claim on the basis of the AGA webpage. Nor have I suggested that very long judgement sequences must be shown in a live stream.

Sure, but you haven't said what the basis for this "myth" is, so people are forced to guess. Why do you think this myth exists to be debunked in the first place ?

It might be worth noting that Redmond tried to improve his ability to read out long endgame sequences as a form of training. He claims this training actually improved his endgame technique. But he also said that this depth of reading is not necessary in order to play a perfect endgame. Reading out the largest moves and getting their sequence right is sufficient, without carrying through the analysis to the much smaller moves, whose sequence can be deferred until later. These statemants may seem slightly contradictory, but (some) humans are sometimes comfortable living with contradictions.

topazg, the basis of the myth was go forum messages (maybe on godiscussions) suggesting what I mentioned about endgame-like 100 moves sequences during the middle game in the OP. Maybe you do not recall it, but I do because a) I was impressed about Redmond's allegedly reported skill at that time, b) impressed about the method of positional judgement because it presumes extraordinarily correct and deep reading with following counting and c) annoyed because the suggestion had been that other methods of territorial positional judgement (such as mine) would be less skillful (and in fact they would be, because, you know, perfect play reading 100 moves deep during the middle game would indeed be much more skillful than estimating the count +- 1 point 100 moves later).

I was unaware of those rumors, which do not seem convincing to me.

I did hear something similar about Lee Chang-ho, which did not convince me, either.

Anyway, it sounds like a better title for this thread would be something like,

Debunking the myth that Michael Redmond judges opening and middle game positions by reading 100 moves ahead

It sounds like you are questioning Redmond's positional judgement, not internet rumors about superhuman reading ability.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Bill, thread titles cannot be as long as we wish. If you reread my OP, I also criticise Redmond for his weak positional judgement during opening and middle game.

I think a better thread title that would not be beating around the bush would be "Another topic where the words Positional Judgment are used so that I can sell more books". That might be a little bit too long though..