Life In 19x19

The general discussion about AlpgaGo's recently released teaching tool is over here, this thread is for discussion of particular moves you find interesting. I was clicking through some old-fashioned joseki (mostly early 80s, featuring Otake and Rin according to waltheri). Here's a summary of AlphaGo's judgments preceding the funky move which caught my eye. AG says black starts at 47.1 win % with a 4-4, 3-4 is only slightly less at 46.7. This is Master version of AG and it did still play 3-4s, unlike AG Zero which rather boringly only plays 4-4s. Not much to say up to 5, small shimari is 45.7 which is surprisingly a little better than big low, but big high is favourite at 46.7, but all small differences. Low approach of 6 tiny bit better than high, but then first notable 2% difference is to approach top left instead of pincer, it seems it doesn't like losing sente after a pincer so prefers mutual destruction: you prevent my shimari so I'll prevent yours. It doesn't like the solid kosumi of 8 but counter pincer at a (and then b-c, treat the approach lightly and split the large area of black potential in front of the shimari and then hurry to enclose the top left.

Click Here To Show Diagram Code

[go]$$B
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 2 . . . . . . . . . . 5 . . . . |
$$ | . . . , . . . . . , . . . . . , 1 . . |
$$ | . . . . . . . . . . . . . . . . c . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , a . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . 7 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . 8 . . . |
$$ | . . . . . . . . . . . . . . . . 6 . . |
$$ | . . . 4 . . . . . , . . . b . , . . . |
$$ | . . . . . . . . . . . . . . . 3 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

The humans continue the classic joseki (it slightly prefers big knight, but within the noise that it chooses moves), and instead of the slide likes to pincer as before, but less difference (I suppose the corner is 2 stones now and heavy so it makes more sense to settle it). By 2% it prefers to approach top left instead of complete joseki at 11, speed to approach 3-4s or make shimaris seems ultra high priority. The 80s humans made the small shiarmi, AG makes big and we only have mainline now. 13, 14 look normal, but 15 surprised me, what is the meaning of this move?! I have sometimes seen such 2nd line attachments against the knight's move of smaller shimari (either 3-4 or 4-4, but this early against a big shimari is surprising as you can be giving your opponent more solid territory in a place that wasn't yet solid. Is it a probe, deciding how to continue lower left?

Click Here To Show Diagram Code

[go]$$Bm9
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . X . . . . |
$$ | . . . , . . . . . , . . . . . , X . . |
$$ | . 8 . . . . . . . . . . . . . . . . . |
$$ | . 7 4 . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . 6 . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 5 . . . . . . . . . . . . O . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . O . . . . . , . . . 1 . , . . . |
$$ | . . . . . 9 . . . 3 . . . . . X . 2 . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Here's the continuation:

Click Here To Show Diagram Code

[go]$$Wm18
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . X . . . . |
$$ | . . . , . . . . . , . . . . . , X . . |
$$ | . O . . . . . . . . . . . . . . . . . |
$$ | . X O . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . 2 8 . . . . . . . . . . . . . . |
$$ | . . X 1 3 . . . . . . . . . . O . . . |
$$ | . . 6 . . . . . . . . . . . . . O . . |
$$ | . . 4 O . 9 0 . . , . . . X . , . . . |
$$ | . . 5 7 . X . . . X . . . . . X . O . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Uberdude wrote:The general discussion about AlpgaGo's recently released teaching tool is over here, this thread is for discussion of particular moves you find interesting.
{snip}

Low approach of 6 tiny bit better than high, but then first notable 2% difference is to approach top left instead of pincer, it seems it doesn't like losing sente after a pincer so prefers mutual destruction: you prevent my shimari so I'll prevent yours.

Unclear what that 2% difference means. It seems to be based upon the win rate of simulations, not upon the judgement of AlphaGo. But simulations of what?

In any event, Master did not like pincers, but Zero pincers more freely. Would Zero pincer here? Quien sabe?

It doesn't like the solid kosumi of 8 but counter pincer at a (and then b-c, treat the approach lightly and split the large area of black potential in front of the shimari and then hurry to enclose the top left.

was popular when I was first learning go, but I never understood it. It eventually fell out of favor, though. I recall an article by Maeda saying he never liked it.

Master pushing through the magic sword table shape joseki was one of the most memorable things from that series, and in that little teaser afterwards DeepMind revealed AlphaGo doesn't like the attachment of a to cut the large knight's move, showing the armpit hit (b) variation and either crawling or pushing and cutting. Well, now we can see just how much it doesn't like it, a whopping 7% worse compared to this bizarre looking 2nd line kosumi! (they show no other choices). Position is from game #35 vs Jiang Weijie. AG then doesn't play in that area for a while. In the Kim Jiseok game with this push through table AG recommends just tenukiing the large knight move (with a 3-3 invasion of course)!

Click Here To Show Diagram Code

[go]$$Wcm16
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . 4 . . . . . . . . . . . . . . |
$$ | . . . X . b . . . . . . . . . . . . . |
$$ | . . . , . . 3 . . , . . . . . X . . . |
$$ | . . . 1 a . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 2 . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . O O X X X X X . , . . . . . O . . . |
$$ | . . . O O O O . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

I suppose the pros of this move are it doesn't contact white so doesn't make white stronger, means white's can't attach at 3-3 because you can cut and capture it, and also gives an option to sneak out on the top side. Alsoif you do play the normal attachment it recommends bump and then ugly empty triangle connect in response to atari and then cut and fight!

This kosumi is perhaps not so weird though, it remindsme of this one now that I've thought about it. Here black is more sealed in so on the defensive and cutting immediately is dangerous, but the corner shape relation with 3-3 attachment for a quick life and potentially looking forward to cutting in the future is similar.

Click Here To Show Diagram Code

[go]$$Wm8 Mok Jinseok vs AnJoyeong 2003: http://ps.waltheri.net/database/game/11932/
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 2 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , 1 . . |
$$ | . . . . . . . . . . . . . . . 4 6 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . 5 . 3 . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . 8 . |
$$ | . . . O . . . . . , . . . . 7 , X . . |
$$ | . . . . . X . . . . X . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Am I the only one who finds AlphaGo's moves (what this thread is for) more interesting than precisely what % means?

Uberdude wrote:Am I the only one who finds AlphaGo's moves (what this thread is for) more interesting than precisely what % means?

You're not the only one. I am also impressed by the moves, as % are too small for my eyes and I've been quickly scrolling through the moves for the past 3 hours, and I'm not even half way through. Wish I had been taking screenshots of all the interesting moves so I could share...

Here's another one which caught my (and Aja's: he shared it with Cho Hyeyeon 9p on facebook) eye. The inside approach is same as outside approach (3-3 invasion of course #1, but by a tiny amount). White pincer 1% off back off with knight move, but that's not a big deal as AlphaGo will sometimes play moves in that range. However, it prefers this 2nd line slide of 7 to the usual hane connect joseki by 43 to 39%.

Click Here To Show Diagram Code

[go]$$Bcm5
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . 7 . . . . . . . . . . . . . . . . . |
$$ | . . . 6 . . . . . . . . . . . . . . . |
$$ | . . 5 O . . . . . , . . . . . O . . . |
$$ | . . 3 4 . 1 . 2 . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Perhaps just as interestingly, it only sees the hane connect as being bad for black (staying at 39%) if white finishes with the 2nd line crawl instead of more normal solid connection at a. If white does the solid connection black goes up 5% to even better than the slide from above: i.e. white only punishes the hane connect mistake if white doesn't solid connect but crawl afterwards. This crawl isn't new to human, but is much much rarer than the solid connection (waltheri gives 36 hits vs >2400). From looking at AG's expected continuations if black adds a move to the slide later white also reinforces, so I think it doesn't want to hane connect and damage the initial approach stone because it is preparing to run out with it and it sees that as a potentially severe attack on the white wall (white can't lean so effectively if black is on the 2nd instead of 3rd line). However, if you make the solid connection the approach stone still has quite a bit of annoying aji so white isn't so good, the downside of crawl is black can cut or the push from the 3rd line jump has more power, but I suppose AG is happy with those continuations. It'll be interesting to see if AG generally prefers this crawl in many positions, maybe it will become the default joseki move in the future.

Click Here To Show Diagram Code

[go]$$Bcm11 (Locally) good for white, but only with 14 not at a
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 5 . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . X O . . . . . , . . . . . O . . . |
$$ | . . X O a X . O . . . . . . . . . . . |
$$ | . . 3 1 2 4 . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Uberdude wrote: It'll be interesting to see if AG generally prefers this crawl in many positions, maybe it will become the default joseki move in the future.

It also does in outside approach in parallel 4-4s. As before it prefers answering approach with knight's move to pincer (as On Sojin 7p said to me in BIBA, pincer a mistake in this position, 3-3 is good), though prefers double approach 2% to 3-3 of the lower right (3-3 at lower left #1 of course!). Again it likes the 2nd line slide, but 11 hane is only punished if you crawl instead of connect, 41 vs 46.5% black win:

Click Here To Show Diagram Code

[go]$$Bcm11
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X 4 . |
$$ | . . . . . . . . . . . . . . . . . 2 . |
$$ | . . . O . . . . . , . . . . O O O 1 . |
$$ | . . . . . . . . . . . . . 5 . X X 3 . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

P.S. At least the Azerbaijani comedy 6 dan wasn't right that p4 extend should be hane as many kyus like to play!

Bill Spight wrote:

It doesn't like the solid kosumi of 8 but counter pincer at a (and then b-c, treat the approach lightly and split the large area of black potential in front of the shimari and then hurry to enclose the top left.

was popular when I was first learning go, but I never understood it. It eventually fell out of favor, though. I recall an article by Maeda saying he never liked it.

That kosumi and the (old/special-purpose) joseki of 3 below are the kind of move where I think the pupil says "that looks kinda slow" but the teacher says "trust me young padawan, when you reach the level of master you too will learn to appreciate the latent power of such strong moves" and pupil may then accept the move into their Go vocabulary on trust. Of course human josekis evolve, opinions change and these fell out of favour, but bots have no such respect for their teacher (although training on human data could give a bit, AGZ ditches that) and can just say "nah, it's slow".

Click Here To Show Diagram Code

[go]$$B
$$ . . . . . . . . . |
$$ . . . . . . 8 . . |
$$ . . . . . . . . . |
$$ . . . . . . . . . |
$$ . . . . . 2 . . . |
$$ . . . 3 . , 6 . . |
$$ . . . . . 1 4 . . |
$$ . . . . 7 . 5 . . |
$$ . . . . . . . . . |
$$-------------------[/go]

Uberdude wrote:

$$B
$$ . . . . . . . . . |
$$ . . . . . . 8 . . |
$$ . . . . . . . . . |
$$ . . . . . . . . . |
$$ . . . . . 2 . . . |
$$ . . . 3 . , 6 . . |
$$ . . . . . 1 4 . . |
$$ . . . . 7 . 5 . . |
$$ . . . . . . . . . |
$$-------------------

Click Here To Show Diagram Code

[go]$$B
$$ . . . . . . . . . |
$$ . . . . . . 8 . . |
$$ . . . . . . . . . |
$$ . . . . . . . . . |
$$ . . . . . 2 . . . |
$$ . . . 3 . , 6 . . |
$$ . . . . . 1 4 . . |
$$ . . . . 7 . 5 . . |
$$ . . . . . . . . . |
$$-------------------[/go]

Does anyone know who invented this joseki? I'm curious whose weird opinion got copied for so many years.

Uberdude wrote:Of course human josekis evolve, opinions change and these fell out of favour, but bots have no such respect for their teacher (although training on human data could give a bit, AGZ ditches that) and can just say "nah, it's slow".

Well, training by self play can yield the same kind of thing. If you are your own teacher, well, you may not find out that some of your assumptions are incorrect, at least, until very much time has passed. In evolutionary terms, when two very similar bots play each other, they each present almost identical fitness landscapes to the other. I think that Kato said that Zen made an error under Japanese rules with 6.5 komi because it had trained under Chinese rules with 7.5 komi. If so, then current bots may be quite brittle. Hopefully not! Curiously, Zero seems to play more like humans than Master, although Master trained on human data and Zero did not. But maybe another Zero -- give us a few days -- would have a rather different style, as well.

We know that AlphaGo favors white at the beginning of the game, but how does that impact play? I think the asymmetry between play for black and white gives us insight into some of AlphaGo's choices.

I like to play this opening for black.:

Click Here To Show Diagram Code

[go]$$B 19x19 diagram
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . a b . . . . |
$$ | . . . 2 . . . . . , . . . c . , 1 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 4 . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . 3 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

AlphaGo prefers the large knight's enclosure at a to the small knight's enclosure at b. (It also gives very high marks to the two space extension at c.) It's not a big difference and it's not entirely uncommon in professional play, but I find it interesting that the more traditional move is not chosen.

However, if we force AlphaGo in a similar position for white:

Click Here To Show Diagram Code

[go]$$B 19x19 diagram
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 2 . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . 1 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . 5 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 4 , . . . . . , . . . . . 3 . . . |
$$ | . . . . a . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

It considers a to be the only move. (Edit: as Uberdude pointed out, this language was too strong. Should have said something like, it considers a to be the correct enclosure.) Now, AlphaGo would never get to this opening - approaching the corner with 5 is too big for black to ignore - but I still find it interesting. It seems to me the large knight's move in the first diagram is a concession to AlphaGo's evaluation that black is behind, so it is playing a little bit riskier in order to fight back from a disadvantage. I think this is important to consider when we look at other joseki / fuseki choices that AlphaGo is making.

jeromie wrote:

$$Bc 19x19 diagram
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 2 . . . . . . . . . c . . e . . |
$$ | . . . , . . . . . , . . . . . 1 . . . |
$$ | . . . d . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . 5 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 4 , . g . . . , . . . . . 3 . . . |
$$ | . . . . a . . . . . . . . b . . f . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bc 19x19 diagram
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 2 . . . . . . . . . c . . e . . |
$$ | . . . , . . . . . , . . . . . 1 . . . |
$$ | . . . d . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . 5 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 4 , . g . . . , . . . . . 3 . . . |
$$ | . . . . a . . . . . . . . b . . f . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

It considers a to be the only move. Now, AlphaGo would never get to this opening - approaching the corner with 5 is too big for black to ignore - but I still find it interesting.

That's overstating it a lot: more accurately a is the only move other than the human approach at b for which DeepMind have reported AlphaGo's evaluation. AG gives a 43.5 to b's 44.1, not much difference and AG sometimes chooses moves 0.6 worse than another. Quite how they choose which moves to report we don't know (I expect they report AG's highest/lowest number for black/white to play plus a few more, reducing as we go down the tree), but I would bet plenty that c-g are all ok for white (not giving black more than e.g. 60% to make it a safe bet) and wouldn't be surprised if some are with a % or two of the reported 2 moves. In fact I wouldn't be so shocked if something like g had a smaller black win % (on at least some 10 minute runs) than the reported a.

You’re right; I should have said that a is the only enclosure it showed. That was a poor choice of words on my part.

One more thing about the extend instead of connect in the 4-4,approach pincer 3-3 joseki with hanetsugi(white has just invaded the 3-3 at the triangle):

Click Here To Show Diagram Code

[go]$$Bc 19x19 diagram
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . 1 . X 2 . |
$$ | . . . . . . . . . . . . . . . . @ O . |
$$ | . . . X . . . . . , . . . . O O O X . |
$$ | . . Q . . . . . . . . . . X . X X X . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

according to AlphaGo,black should immediately take advantage of the position white left with the "wrong" move at the square by jumping to 1, to which white is supposed to reply by cowardly connecting at 2. If this is the right sequence, it's no wonder you would prefer to play at 2 instead of the square immediately, the exchange of 1 for the squared stone seems rather oviously good for black.

Interesting Schachus, it didn't do that exchange in the parallel opening I looked at first, but does in the 2nd if white 3-3 afterwards (they have many choices for white next move, if approach black does s6 instead, what's the relationship?!) Whilst it is nice to make w answer so low, there's also an element of aji keshi that I think means most pros don't want to, keeping s6 r10 etc aji.

<Admin>
Win % discussion moved to viewtopic.php?f=18&t=15312
</Admin>