Having translated Fujisawa Hideyuki's book on the avalanche I took a special interest in a couple of the comments here and an a related thread. I even made time to write a long article giving chapter and verse for my own claims, but just as I finished I got the infamous Windows blue screen. I haven't the will to do it all again. You'll have to trust me when I say I made my assertions on the basis of five joseki dictionaries, plus commentaries. I'll just give the headlines here.
First, Gomoto said:
Quote:
nuff said
(Go Seigen and Alpha Go agreed on this move, the matter is closed
)
- Click Here To Show Diagram Code
[go]$$B
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . X X . . . |
$$ | . . . , . . . . . , . . . . O O X B . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]
This was meant to prove that a fairly standard human move, the connection at the 3-3 point, in place of the circled sagari, was wrong.
Various problems with this claim. One is that Go never played it, but he did play the 3-3 point (in 1931). The sagari was first played against Go by Fujisawa Hosai in 1958, and very many top pros have played it, too, well before AlphaGo.
Trying to back the claim that there are things that bots are teaching us that we have to "unlearn", Bill took this up later and said:
Quote:
Well, now the bots tell you to play
below, instead.
G
- Click Here To Show Diagram Code
[go]$$Bc
$$ . . . . . . |
$$ . . b a . . |
$$ . . O X . . |
$$ . . O X . . |
$$ . O X . . . |
$$ . . 1 . . . |
$$ . . . . . . |
$$-------------+[/go]
Do you say, Oh, no! I couldn't? Or do you say, Thanks?
Some things wrong with this, too, though not all directly in the quotes above. One claim is that pros say the 3-3 move is joseki, and that's what we have to unlearn. My "chapter and verse" was to prove that pros haven't claimed it was joseki. They use phrases such as "common shape", and what they said specifically was that Black can make this move if he wants to avoid complications such as ladders and kos, and if he does he does not necessarily get a disadvantage. Nothing about equality, It is in fact a fuseki move. There are a couple of josekis that do follow from this starting point. It seems often overlooked that almost all josekis have fuseki moves within them, and not all josekis start on move 1.
But the main mistake is that the bots haven't proved the 3-3 move is either wrong or bad.
It is true that LZ gives the sagari a very slight preference over connection on an otherwise empty board - 45.6% to 44.8% after 18k playouts for the sagari alone, which I've been given to understand in any case lies within the margin of error.
But in the following position, LZ gave the 3-3 connection at A as best at 46.2% (17k playouts) and the sagari didn't get a rating. The second best was the nobi at B and the small avalanche hane at C got a good work-out early on before fading. The play envisaged on the rest of the board seemed to be all in the lower left. Note that White has a stone in the lower right, so the complications of ladders (which Black, in pro opinion, wants to avoid) would generally favour White. LZ seems to be thinking like a pro.
In the following case, however, the sagari A was the top move after 26k playouts (46.0%).
What I found interesting was that the presumed subsequent play was now mostly in the lower right, and the variations shown there looked very ladderacious (may be my imagination as I'm not sure how that ties in with LZ's alleged weakness in ladders).
So what I infer from all that is:
1. The pros have been misrepresented in their opinions, in part, and overall are actually supported by LZ.
2. We don't have to unlearn the 3-3 move yet. We just have to learn to play it for the right reasons (like Go Seigen).
And I can't help but note again that a slew of joseki books and commentaries by humans gave very clear and useful advice using words. What LZ offered could be made sense of, but only by using the words and principles already learnt from the human corpus.
At the very least the matter is not closed.