Life In 19x19

I have a question about counting: There are books such as "Positional Judgement and High Speed Game Analyses" which claim that a professional could tell in a split second how many points white and black could expect in a given position in a Go game.
When I watch the live coverage of the match between Lee Sedol and Alpha Go with 9 Dan professional Michael Redmond, I can't believe it. It seems to me that even for a world-class player it is not always clear who is better and how many points will be scored at the end of the match. The co-commentator is Dan Player himself, and to him it seems to me it's even more unclear.
How is it now in the practical Go of amateurs below the Dan level? Do you have to know how many points are there around exactly, or is it more a matter of feeling?
Why I ask such a strange question? I come from chess, I have Elo 2260 there, and I can say at 90% of the positions at once, what is the estimate.
Go is more of a mystery to me

When I watch the live coverage of the match between Lee Sedol and Alpha Go with 9 Dan professional Michael Redmond, I can't believe it.

That's probably because "it's not counting as we know it, jum."

They are not normally counting up each point of sure territory, and then doing a raft of hypothetical boundary-play sequences around more open areas, and counting those 1, 2, 3, 4.... They can do that, and may, at least once or twice, actually do it. But in fast games there are more important things to think about in the brief moments available.

So they use various alternative techniques.

A common one is to count just the mistakes. They start with the premise that the game starts even and (between pros) almost all moves are NOT mistakes. So when they spot the few mistakes, it is easy to make a mental note of them. Mistake in this context tends to mean things like overconcentration or some similar inefficiency. For example, making a ponnuki capture with five stones instead of four. Pros furthermore have a very good intuitive sense, based on long experience, of what each type of mistake is worth.

There are other similar heuristic techniques, some of which are beyond the ken even of normal pros. Yi Ch'ang-ho was famous for "counting", which too many people seem to think is the 1, 2, 3, 4 type. It was rather his ability to sense how many points he would get out of fuzzy shapes in the centre, which he could superimpose on the other heuristic techniques, that made him the envy of other pros.

There are also non-heuristic techniques, of course. The main one is sheer memorisation. A typical pro will know by heart the de-iri or miai count (ways of counting the value of a move rather than a position) of maybe 1,000 common positions. The very first book on boundary-play counting, Genan Inseki's Igo Shukairoku of 1844, was of this type.

And add to all that the fact that many pros do actually have the ability to make a pretty accurate 1, 2, 3, 4 type of count at high speed. They may have to sacrifice some precious time, but that's often easy to do. For example, in 30 seconds a move, if the opponent plays an atari that forces you to connect, don't do it at once but spend your 30 seconds counting. Plus you can count in your opponent's time.

Having said all of that, pros DO miscount, and - believe it or not - even sometimes make mistakes in counting up the game at the every end when the game is over and time is no longer a problem (though maybe fatigue is).

Many thanks for the detailed answer! Wow, then the whole thing is even more difficult than I had thought. Now I would like to know how many Kyu player games end with a surprising result for both players... But I will see for myself.
I had thought that maybe 1 year of playing practice should be enough to have a pretty reliable estimation of the own game.

jumapari wrote:I have a question about counting: There are books such as "Positional Judgement and High Speed Game Analyses" which claim that a professional could tell in a split second how many points white and black could expect in a given position in a Go game.

That's a lie.

Well, when the book came out pros did learn heuristics to guess the approximate territorial count early in the game. But, AFAICT, they were not necessarily those given in the that book. And in some cases traditional estimates were found to be wrong.

How is it now in the practical Go of amateurs below the Dan level? Do you have to know how many points are there around exactly, or is it more a matter of feeling?
Why I ask such a strange question? I come from chess, I have Elo 2260 there, and I can say at 90% of the positions at once, what is the estimate.
Go is more of a mystery to me

(Emphasis mine.)

Join the club.

One thing, which is mostly what the book shows -- it never impressed me -- is that you can make a conservative estimate of more or less the minimum territory you can be fairly certain of for each player. Then you can try to see how each player might get more points. In one of his books aimed at dan players, Sakata said that amateurs can be satisfied, at the beginning of the endgame, with a territorial estimate of the current position that is correct within 5 points or so. As the game progresses territory gets easier and easier to estimate. BTW, even Sakata's book did not provide an estimate for the value of having the move.

Nowadays you can get a good estimate of the final score using KataGo. Some other bots provide that, as well, but I don't know enough about them to be confident of those estimates. OC, KataGo's estimate takes into account who has the move.

I would imagine that, if you play around with KataGo and try to guess its estimate for different positions, you can quickly develop the ability to come up with guesses that are consistently within 5 pts. of its estimate.

I had thought that maybe 1 year of playing practice should be enough to have a pretty reliable estimation of the own game.

Oh, I think that after a year you can get with 3 pts. of KataGo's estimate for the majority of positions.

As I already wrote, I am a absolut beginner. I don't even have a Kyu rating. Because of Corona I play exclusively against my computer program "Many Faces of Go" - with 1 - 2 hours time to think about it on the 19x19 board and with 18kyu. After 150 moves I usually check if I like to win or lose. In the beginning I was always on loss, and now I'm always on win. The day before yesterday I even won against 12 kyu
Actually, I'm slowly developing a feeling for how to estimate the position. But it's not about points at all. More about a feeling. Probably this is the normal way, too.

As a 5k to 1k, during the middle game I counted "safe" territory of territory regions / moyos with more or less roughly outlined boundaries according to Cho's book with some error, which came down to globally ca. 2 points as a 1k.

As a 5d, during the middle game I count "current territory" of territory regions and open moyos according to my two positional judgement books with an error of globally 1/2 point, assess influence, neutral stones, positional fighting aspects etc.

For the counting of territory, this presumes a "quiet" position. If hot fights are active, read until the position is quiet to derive the current territory. It can be counted by imagining peaceful sente reductions settling the territory boundaries. The sente criterion is my invention enabling a global sequence reducing all regions of the opponent in alternating play. Detailed assumptions describe how to construct such reductions reasonably and fairly etc. All this can be fast enough because they are assumed to be peaceful, only incremental updates are needed and there are convenient shorthands for actually counting intersections and their dead stones. A fresh incremental count (as needed after a major exchange) takes between 10s and 3min but an incremental update after every ordinary new move takes only 1s or a few.

Assessing influence comprises life / connection status and counting. Usually, I count numbers of stones and calculate the influence stone difference (B minus W influence stones) of an open region. This is also my invention and incredibly fast.

Similarly, I assess the neutral stone difference. My invention again and equally fast.

So the judgement of a position is determination of statuses, fighting options and the mentioned values. If somebody is ahead on territory by a certain amount, the opponent must have favourable other options and values to overcome that amount. With unfavouable aspects, he is behind.

This describes that positional judgement can be done at all, fast enough and accurate enough.

Now, most professionals do not study, let alone apply, my theory. Instead, you see some having difficulty to judge well and fast, or others judging subconsciously and fuzzy. Of couse, there is also the emergency method of reading deeply and counting only territory of an imagined mostly settled position. Such is less secure because long sequences overlook many branches of alternative fights.

Imagining some fights can be necessary but short quiescence reading and then judging all aspects is better. Imaging fights can be too difficult for kyus but otherwise my methods can be applied by SDK or dans if one learns a few methodical details.

Wow, I know of no other forum where you can get such competent and profound answers. Thanks a lot!

jumapari wrote:... a professional could tell in a split second how many points white and black could expect in a given position in a Go game...

Here's my actual experience with a go professional:

My rating was ~5k AGA, and I was playing in a simul against a pro at the AGA Go Congress. I think it was Cathy Li. Somewhere in what I would consider late middle game, I saw a cool move which could threaten one of the pro's groups, forcing her to defend and allowing me to cut off and kill several stones. I did this, and the pro smiled at me and said "you win!". I was almost speechless, as it was not at all clear to me that I was even ahead. As I said, it seemed like late middle game to me.

I asked her "what do you mean?", and she said that by capturing those stones, I was ahead by 4 points and would win. I wanted to play out the game anyway (which wasn't the polite thing to do, and I don't think she really liked it, but I couldn't help myself), and sure enough, at the end of the game I was ahead by exactly the 4 points that she predicted.

Remember, my game was just one out of 5 or so simul games, with several other games against players stronger than I am. She wasn't paying particularly much attention to my game. Nevertheless, not only did she know the score, but I came to realize that she had set up the position as a test for me. Set it up, without my even being aware of it, so that if I saw the correct move I would win, and if not I would lose.

Sure, professional go players are human and will not always be able to do this. But they are sometimes capable of amazing things.

After her declaration, she must have made lots of intentional mistakes to compensate yours and keep the count constant at 4.

RobertJasiek wrote:For the counting of territory, this presumes a "quiet" position. If hot fights are active, read until the position is quiet to derive the current territory. It can be counted by imagining peaceful sente reductions settling the territory boundaries. The sente criterion is my invention enabling a global sequence reducing all regions of the opponent in alternating play.

Cher Robert, you may have discovered the idea of sente reductions independently, but it is hardly a secret. See https://senseis.xmp.net/?SenteGainsNothing

EricBackus wrote:My rating was ~5k AGA, and I was playing in a simul against a pro at the AGA Go Congress.
{snip}

I asked her "what do you mean?", and she said that by capturing those stones, I was ahead by 4 points and would win. I wanted to play out the game anyway (which wasn't the polite thing to do, and I don't think she really liked it, but I couldn't help myself), and sure enough, at the end of the game I was ahead by exactly the 4 points that she predicted.

I am no pro, but if I am behind by 4 pts. in the late middle game, or even the early endgame, playing against a 5 dan, much less a 5 kyu, I expect to win.

But in a simul pros are playing to teach, as a rule. In my lifetime I have heard of only 2 or 3 pros who played for blood against amateurs. The point was not to be competitive, but to find the best play, regardless of who their opponent was.

If you played the rest of the gamem perfectly, more power to you. But the pro probably had to match your level of play.

RobertJasiek wrote:After her declaration, she must have made lots of intentional mistakes to compensate yours and keep the count constant at 4.

Bill Spight wrote:If you played the rest of the game perfectly, more power to you. But the pro probably had to match your level of play.

I'm sure you are both correct. Nevertheless, it was an amazing experience for me.

Jumapari, here's an example of me (a mid dan amateur) counting: viewtopic.php?p=256245#p256245. I guess this might take me 30 seconds to a minute in s game.

Bill also did a series of exercises where a strong bot said one player was leading an opening position and we tried to figure out who. Some people just look at the board and use their judgement without counting, some counted, I used the technique of imagining the sequence of play and summing the mistakes so far. viewtopic.php?f=15&t=17122