Life In 19x19
http://www.lifein19x19.com/

Teaching beginners oldschool or new AI joseki?
http://www.lifein19x19.com/viewtopic.php?f=16&t=18185		 Page 3 of 3
Author:  jlt [ Sat May 08, 2021 11:00 am ]
Post subject:  Re: Teaching beginners oldschool or new AI joseki?
What is a teacher good for? I'd say that the teacher is not there to teach fine details, but to tell the student ideas that the student will have to pay attention to when reading books or reviewing games, as in the quote below:

John Fairbairn wrote:
It related an experiment in which two groups were asked to read a long passage. One group were told nothing about the passage and ended up unable to recall more than a handful of sentences. The other group was told that the passage concerned the washing of clothes. "The simple addition of a human goal transformed the gobbledegook into something clear. They remembered twice as much."
Author:  gennan [ Sun May 09, 2021 2:26 am ]
Post subject:  Re: Teaching beginners oldschool or new AI joseki?
I think the role of a teacher varies greatly depending on the student. A random 6 year old, a random adult and a random academic are not the same.

I think there are more academics active on L19 than 6 year olds. And many of them have played and studied go for decades. Our memories of our own path in go are not really representative for an average group of beginners.
Author:  MikeKyle [ Sun May 09, 2021 4:55 am ]
Post subject:  Re: Teaching beginners oldschool or new AI joseki?
If it's not too late to add to the discussion of the specific 3-3 pattern, I have some analysis of what katago thinks is the best move in a range of cases:

At this point:
Click Here To Show Diagram Code
[go]$$
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . 1 3 . . . . . . .
$$ | . . . 2 a b . . . , .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]


Across a sample of 100 mid-to-high-dan KGS games (post AI era), the bot chooses the extension at a 73% of the time and the jump to b 27% of the time (ignoring a couple of cases where katago wanted to tenuki.)
Personally I thought I'd seen the bot choose the jump more often, but this seems consistent with the thread.

At this point:
Click Here To Show Diagram Code
[go]$$
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . 1 3 b d . . . . .
$$ | . . a 2 4 . . . . , .
$$ | . c . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]


Across a sample of 200 mid-to-high-dan KGS games (post AI era), the bot chooses the turn at a 48% of the time, pushing again at b 41% of the time(!), the knights move c 11% of the time and on 2 occasions jumped to d (5 positions where katago wanted to tenuki.)
I'm surprised to see that pushing again is more common than the knights move. This seems like a strategy to take sente to me. The sort of thing that humans still find difficult to accept maybe?

At this point:
Click Here To Show Diagram Code
[go]$$
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . 1 3 a . . . . . .
$$ | . . . 2 4 . . . . , .
$$ | . 5 . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]


Across a sample of 50 mid-to-high-dan KGS games (post AI era), the bot actually played away for 20 of the board positions.
When the bot chose to play locally it played at a 87% of the time with the remaining few board positions each having unique jumps and centre moves as the bot choice.

At this point:
Click Here To Show Diagram Code
[go]$$
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . 1 3 6 . . . . . .
$$ | . . . 2 4 . . . . , .
$$ | . 5 . . . . . . . . .
$$ | . . b a . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]


Across a sample of 50 mid-to-high-dan KGS games (post AI era), the bot chooses the knights move at a 57% of the time, kosumi at b 17% of the time, there were only a handful of board positions where katago chose a pincer on the top side.
From the Waltheri search mentioned below I think this seems pretty consistent with the human choices

Alternatively, at this point:
Click Here To Show Diagram Code
[go]$$
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . 1 3 b . . . . . .
$$ | . . 5 2 4 . . . . , .
$$ | . . z . . . . . . . .
$$ | . . . a . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]


Across a sample of 100 mid-to-high-dan KGS games (post AI era), the bot chooses the old human-move jump at a 89% of the time, and bending round at b 9% of the time(ignoring a few tenuikis.) I really thought I'd seen the block at z come up when reviewing but it didn't come up as top choice in my sample of games at all.
I thought of b as a small mistake previously as black following up around d15 feels too good so I'm surprised to see it as top choice on some board positions.

If anyone is interested I can provide the whole board positions where katago makes each choice.

I'm fairly sure that any of these choices would loose only a fraction of a point in the case that it's not kg's top choice and if played by a beginner, should be described as an excelent choice on almost any board position. I wouldn't want a beginner anyone to think that there is one true way and all others are inferior, but I think looking at examples of good play can be a good platform for discussions.
Author:  Knotwilg [ Sun May 09, 2021 6:15 am ]
Post subject:  Re: Teaching beginners oldschool or new AI joseki?
schrody wrote:
First of all, I'd rather not see teachers reduced to just quoting factual knowledge from books.


I understand you can read into my opinion this way but that's not at all what I was saying.

Let me rephrase your differentiation of knowledge:

- things you know for sure
- things you know about

I'm not advocating limiting teaching to things you know for sure. Exposure to topics can be driven by many factors, the most important of which is what seems to tickle the pupil/novice.

In general I don't like "things you know about" to be presented as "things you know".

Yes, you explain/show a 3 point nakade before a 5 point nakade before the L-group. Yes, you show simple corner patterns before showing variations of 30 moves, traditional or old.

No, you don't show traditional 4-4 joseki while hiding modern joseki because of the idea that the old stuff is somehow more true. No you don't teach "cross cut, then extend" if the data doesn't support it.

Quote:
I sense that the major difference in our views is that you're less willing to teach "wrong" things than I am. I'd be interested to know where you draw the line for yourself and where you'd draw it for professional players.


There's so much to talk about in Go that I don't see why I should narrow that to something I am comfortable with, even though today's insights tell me it's doubtful. Why would I do that?

Quote:
- Teachers should always take the student's level and understanding of the game into account.


Most of all, the student's interest, desire and motivation.

Quote:
- Even professional players will likely teach you some wrong things.


Of course, the things they still don't know. I don't blame any pro for teaching what they genuinely believe is the correct thing to show. I doubt if pros deliberately teach lines of play they know are inferior or not true.


Quote:
- Amateur teachers therefore shouldn't feel that guilty about teaching some wrong things.


Well I don't want to obsess about it. Let anyone do as they please. I'm mostly advocating modesty when teaching.

Quote:
- Teachers should always be open and honest about the limits of their knowledge. I prefer saying "I would play here instead because..." rather than "Your move was wrong. You should play here instead."


Agreed!
Author:  gennan [ Sun May 09, 2021 7:04 am ]
Post subject:  Re: Teaching beginners oldschool or new AI joseki?
MikeKyle wrote:
Click Here To Show Diagram Code
[go]$$
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . 1 3 b . . . . . .
$$ | . . 5 2 4 . . . . , .
$$ | . . z . . . . . . . .
$$ | . . . a . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]

... I really thought I'd seen the block at z come up when reviewing but it didn't come up as top choice in my sample of games at all.

I'm not sure if you know, but z can be a bad exchange / trick move / overplay that may be refuted by clamping:
Click Here To Show Diagram Code
[go]$$W
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . X X . . . . . . .
$$ | . . X O O . . . . , .
$$ | . 4 1 3 . . . . . . .
$$ | . . 2 . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]

I think this explains why :w1: is not a joseki move here and why the bot doesn't recommend it.

But I have seen quite a few SDK games where both players seem to be unaware of this pattern and then they may get this result instead:
Click Here To Show Diagram Code
[go]$$W
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . X X . . . . . . .
$$ | . . X O O . . . . , .
$$ | . 2 1 . . . . . . . .
$$ | . 4 3 . . . . . . . .
$$ | . . 5 . . . . . . . .
$$ | . . . . . . . . . . .[/go]


Edit: I just tested this with KataGo and to my surprise, it wouldn't "punish" with the clamp here. It prefers the safe variation:
Click Here To Show Diagram Code
[go]$$W
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . X X . 6 . . . . .
$$ | . . X O O . . . . , .
$$ | . 2 1 . . . . . . . .
$$ | . 4 3 . . . . . . . .
$$ | . . 5 . . . . . . . .
$$ | . . . . . . . . . . .[/go]

It already sees that as a 0.5 gain for black, compared to:
Click Here To Show Diagram Code
[go]$$W
$$ ----------------------
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . X X . 2 . . . . .
$$ | . . X O O . . . . , .
$$ | . . . . . . . . . . .
$$ | . . . 1 . . . . . . .
$$ | . . . . . . . . . . .
$$ | . . . . . . . . . . .[/go]

It doesn't like the complications of the clamp very much for black and from what I can find in online joseki libararies there may be ladders involved.
So it seems I was a bit too quick in relying on my own supposed "knowledge". It is more incomplete than I thought.
Author:  tekesta [ Mon Aug 09, 2021 8:56 pm ]
Post subject:  Re: Teaching beginners oldschool or new AI joseki?
I wouldn't go crazy over whether to teach new AI joseki or old-time joseki. It depends on what is most used within a group of players. Some like AI style, while others like the 20th Century stuff. Not everyone will understand enough technique to be able to play AI-style (the last frontier). I would sooner have a group of players replay Honinbo Dosaku's games and play freely in their own games, applying what they learned from Dosaku's games. AI style is a very deep rabbit hole, so I wouldn't touch it until I'm 2D or stronger, at least without professional commentary to explain it.

Understanding joseki requires two things. One, a solid understanding of life & death and tesuji to know how the sequences arise in the first place and, two, a working understanding of opening theory to know how the choice of joseki impacts the strategic outlook of a game. Last Saturday I attended a micro-gathering with Christopher Sagner 6D on opening theory and direction of play. He told us that there are situations in which playing out a joseki to completion is actually counterproductive.
Author:  bogiesan02 [ Sat Dec 09, 2023 9:51 am ]
Post subject:  Re: Teaching beginners oldschool or new AI joseki?
I've recently tried to claw my way out of my dusty old cave so I can help new go players. Those who are coming into the game with only the AI openings to observe know (nor care) little about the previous three thousand years of joseki research. To that end, I have acquired K89 and K90 from Kiseido. https://www.kiseidopublishing.com/Other.htm

Are you a fossil? Buy these books.
Author:  Knotwilg [ Sat Dec 09, 2023 3:45 pm ]
Post subject:  Re: Teaching beginners oldschool or new AI joseki?
The two main novelties by AI which players of all strength are adopting these days are the immediate 3-3 and the attachment after approaching. None of these are a deeper rabbit hole than the classic joseki. The ramifications of 3-3 are no more than those of taisha or avalanche.

The complexity an ai follower calls upon himself is the opponent Moyo which arises from consecutive 3-3 invasions. AI is more comfortable dealing with that than the common man. Which is why it might be wise to keep a balance between territory and influence.

Just like in the old days
Page 3 of 3 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/