It’s no secret that I’m a fan of the Clossi approach to move selection when playing Baduk. That being said, I’ve had some doubts about the actual procedure and have given concrete examples in specific positions of my doubts.
Today, for some odd reason, the concept of logging levels in a computer program appeared to be a good way to move from qualitative use of the Clossi approach into something more quantitative.
In other words, the Clossi approach simply asks “do I have a weakness”? But it does not assess the level of weakness numerically. I think it is safe to say that there are numerous levels of weakness in a Go position. Let’s how if the logging levels available to a computer program can be useful in quantifying them:
So what in the heck does this have to do with assessing a Go position?
Let’s get concrete by looking at a position from this game:
OK. Now as Black in this position, kyu players tend to freak out about their stone that might get put in atari and forget about everything else. But we are going to “log” all the weaknesses in terms of computer logging levels. We are going to assess the weaknesses as critical, error, or warning. So:
First at ‘1’ we see a weakness that is not life threatening. So it merely serves as a warning (value 30).
At ‘2’ we see a weakness that can lead to cutting apart groups. And cutting and connecting is a critical aspect of baduk. But even after the cut, there are 2 running directions for the group, so this is an error (value 40) – it is more urgent than 1 – so why in the HECK did I play ‘1’????!!!! I.e. if you want to play a defensive move, then play the more “critical” defensive move.
Now, at 3 and 4 we see *CRITICAL* weaknesses – White can lose stones and base quickly if White does not respond. Each critical weakness carries a value of 50.
So now let us do “weakness algebra” to find who truly has a weakness.
Black has a warning (30) and an error (40), totaling 70 points of weakness.
In contrast, White has 2 critical weaknesses adding up to 50 + 50 or 100.
Now, looking at the AI assessment we see red and yellow on the weaknesses of Black, meaning that Black should not play there, because those are not critical. And we see green at White’s vital point, meaning that it critical
In my twitch channel, someone summed up this AI analysis quite succinctly:
and the colors are showing us where the weakness is!
Even though question 1 in the Clossi approach is a yes-or-no question implying a singular weakness, the answer is NOT yes or no based on a single weakness!
The best answer is relative to the process of adding up your weaknesses (plural) and opponent weaknesses (plural). So in this case if Clossi asks “do I have a weakness” the answer is: “yes I have 2 or more weaknesses, but opponent is weaker, so I really dont have any critical weaknesses. so let’s move on the question 2”
Now, we ask: does the opponent have weaknesses and the answer is yes.
So now we move to the 4 ways to milk an opponent weakness
Milking an opponent weakness is a good term because we only want profit out of the series of moves. Yes, it would be nice to kill, but we really just want to opponent to feel some heat, and let us collect profit in the meantime.
So the Clossi approach lists 4 ways to milk your opponents weakness. Let’s work through them and see what move to move:
- remove base, where “base” is connected 3rd line stones? So, there is no base to remove.
- surround? yes, but the surround move should have “quiver factor” to use a Bruce Wilcox term: the opponent should feel some “heat” from your move. Therefore O11 is growing your weak group but not scaring the opponent. P11 is making a statement to White: “you are about to be surrounded and killed… what do you want to do about it”?
- reduce eye space – i would say that reduce eye space and surround are the same thing, but the distance from the opponent stones makes the difference. I.e. S8 “surrounded” white but because it was a close encounter it is more of a reducing eye space surround. Another way to see it is that surround is about eliminating running directions and reducing eye space is forcing the opponent to shield off his area. No reducing eye space moves available
- play the vital point – here was the AI move – i guess because it had analyzed the ability of White to run and knew that White would not get away…
Now, supposedly surround has higher priority than playing the vital point for obvious reasons – what is the use in playing a killing move while opponent can run away?
but in this case the AI is suggesting play the vital point, perhaps because White absolutely must defend at 3 or lose an entire group with the follow-up at 4? Well let’s not wonder let’s see what the AI was thinking:
Ah, so the AI likes the “vital” point because it can connect his groups efficiently. When white fixes with R11, it persists in going after the weakness forcing White to respond again at R12.
THEN white plays the surround and then uses a good series of moves to enclose White without getting cut to pieces as White desperately flails away at the position attempting to find a second eye or way out of the sea of hungry black stones.
The Clossi approach to move selection has many predecessors, which Shawn Ray (Clossius) lists at his official page. I felt it necessary to question the approach in the past and also in this game. Out of pure grace, the concept of computer logging levels got into my brain and led me to try to quantify my quandry in this quizzical position. Evolution proceeds slowly step and step and we always look back at the first attempts and see how much further we have gotten along.
Perhaps these few steps towards removing emotional irrational move selection and replacing it with objective numerical assessment will be of assistance to others. Only time will tell!
There was an earlier blog post exploring this idea but I think that this one position allows a tremendous amount of insight into weighing a baduk position and choosing a move.
The point values assigned to various “logging levels” is not accurate. For instance, if there had been only one critical weak point, it still could outweight the “warning” and “error” weak points.
Chess has long-established point values for pieces:
- Rook: 5
- Pawn: 1
- Bishop: 3
- Queen: 9
so we may be able to take the fact that one piece is 3x more valuable and equate it to Go … something like: creating a connection is 3x more valuable than making gote territory.
KaTrain color levels?
KaTrain makes it very simple to see how good a move is: it uses 3 colors, red, green and yellow to indicate whether you should play somewhere. So perhaps I should “color my weaknesses” and “color opponent weaknesses” and then follow kelieon’s advice:
always play in the weakest place