Life In 19x19

Uberdude wrote:Ke Jie 9p lost yesterday to Huang Yunsong 5p (http://www.go4go.net/go/games/sgfview/56666) and Aja Huang made the following interesting comment on facebook with a link to an article on the game http://new.eweiqi.com/portal.php?mod=view&aid=27246.

Aja Huang wrote:Ke Jie 9p might receive wrong information about AlphaGo's opening.

文章中提到"卫冕冠军柯洁对黄云嵩一局，执黑的柯洁序盘就不走寻常路，第11手不挡右上角对手的点角，脱先挂左上，据说这是网上流传的AlphaGO自我对弈棋谱视频的着法。"

AlphaGo在自戰譜中並沒有脫先掛左上。AlphaGo總是先拐擋33那一子完成定式後，再去掛角。在我們第2局解說的視頻中，樊麾老師也沒有顯示棋譜的手數。柯傑九段應該是接受到錯誤的訊息了。

From the garbled machine translations I get the impression Ke Jie's approach of 11 at the top left was a bad move (and white 12 hanging connection a good one) and gave him a bad game, but that he may have played this move based on rumours that AlphaGo likes this move, but then Aja says this rumour was incorrect information. Can anyone fluent in Chinese shed some light on this please?

$$Bm11 Ke Jie (black) vs Huang Yunsong
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O . 2 . |
$$ | . . . . . . . . . . . . . O . X O . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 1 . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+

Click Here To Show Diagram Code

[go]$$Bm11 Ke Jie (black) vs Huang Yunsong
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O . 2 . |
$$ | . . . . . . . . . . . . . O . X O . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 1 . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

And the following conversation between Jennie Shen 2p and Aja Huang in the same post...

facebook comments wrote: Jennie Shen 这种错误信息其实很多，有时候是记者断章取义，有时候这种信息都不知道哪里来的。

Aja Huang 如果網上有流傳這種AlphaGo自戰譜的錯誤信息，導致棋手們的誤解，實在非我們所樂見。不過無論如何，再不到幾週自戰譜加上解說就會公布了。

Jennie Shen Aja Huang 很多时候棋手只是开个玩笑，就被记者当成故事写了。脱先这手棋以前也见过的，我觉得所谓“布局败”不过是句玩笑话。

Aja Huang Jennie Shen 謝謝你，我明白。我只是不希望在棋譜解說公布之前，大家接受到錯誤的訊息而被誤導了。

It sounded to me like the reporter was quoting out of context some joke response or word play?

But I too am only relying on machine translation.

Here is an argument that is not so good.

Click Here To Show Diagram Code

[go]$$Wcm8 Different sequence
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . 1 . 5 . |
$$ | . . . . . . . . . . . . . O . 4 3 . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 2 . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Suppose that had been the slide in the top right instead of (I suppose) the 3-3 invasion, and that had approached the top left corner. So far, so good. Now suppose that had played on the 3-3. In this position is clearly a bad play, gaining almost nothing while strengthening White. Even if is inferior to, say, responding in the top left, it's not that much inferior, is it?

I agree with Bill's analysis that looks a bit weird with in place. Given the different order of moves, I was going to ask, why would you ever leave as it is? After all, continuing can be sente, and black can still play the approach on the left:

Click Here To Show Diagram Code

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

After thinking about it more, I'm thinking that the reasoning at the time must have been a probe. For example, black doesn't have to block that way. He could do this way:

Click Here To Show Diagram Code

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

Which is quite common, too... So my thinking is that Ke Jie must have been intending to probe the top left. After the opponent responded, he could decide which way to continue in the top right.

But of course, white's move in the game prevented either of these options:

Click Here To Show Diagram Code

[go]$$B
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . O . O . |
$$ | . . . . . . . . . . . . . O . X O . . |
$$ | . . . O . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

I like the idea of probing, but since white has this response, perhaps it's bad timing?

Kirby wrote: So my thinking is that Ke Jie must have been intending to probe the top left. After the opponent responded, he could decide which way to continue in the top right.

I like the idea of probing, but since white has this response, perhaps it's bad timing?

Indeed the approach looks like a probe: if white answers with the knight's move then black may block the right to take sente instead of separating as the top side is now less valuable. I found one pro game with this approach and that's what happened: http://ps.waltheri.net/database/game/61578/.

Kirby wrote: Now, if AlphaGo is correct, it very well may be that this board is even for both sides. But what about the bottom right corner?

$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . X O . |
$$ | . . . . . . . . . . . . . . . . X O . |
$$ | . . . , . . . . . , . . X . X X O C C |
$$ | . . . . . . . . . . . . . X O O O C C |
$$ | . . . . . . . . . . . . . . . . C C C |
$$ | . . . . . . . . . . . . . . . . C C C |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . X O . |
$$ | . . . . . . . . . . . . . . . . X O . |
$$ | . . . , . . . . . , . . X . X X O C C |
$$ | . . . . . . . . . . . . . X O O O C C |
$$ | . . . . . . . . . . . . . . . . C C C |
$$ | . . . . . . . . . . . . . . . . C C C |
$$ ---------------------------------------[/go]

White has something like 10 points of territory, but I would totally bet that black's influence is worth more than that.

So what has happened here?

AlphaGo sees its ideal way of how the fuseki will play out, and is willing to play locally inferior plays in order to get the globally ideal result.

I'd assume that any evaluation of this position would want to include the fact that B has played an extra stone? If we consider 10 points to be the price of sente, then one would hope that B's influence is worth at least 20 points.

In fact, willingness to take a locally inferior result in order to get a better global result is implied in any tenuki until you get to completely settled positions late in the game. I'd consider this a perfectly human way to play...I don't happen to have a joseki book handy, but I wouldn't be surprised if this position was straight out of one.

Sure, Mef. My point is that AlphaGo's analysis suggested that LSD's joseki move was wrong. I personally feel that some people have taken this to mean that the joseki was wrong.

I argue that AlphaGo's analysis is only meaningful taking into account the entire fuseki, and the analysis doesn't extend to the same shape in other board positions.

Much like it's important to choose the correct joseki based on global situations, AlphaGo's analysis of what the best move is is highly dependent on that particular board position.

Kirby wrote:Sure, Mef. My point is that AlphaGo's analysis suggested that LSD's joseki move was wrong. I personally feel that some people have taken this to mean that the joseki was wrong.

I argue that AlphaGo's analysis is only meaningful taking into account the entire fuseki, and the analysis doesn't extend to the same shape in other board positions.

Much like it's important to choose the correct joseki based on global situations, AlphaGo's analysis of what the best move is is highly dependent on that particular board position.

Personally I would simply say "AlphaGo would make a different joseki choice than LSD did". Whether or not it's better or worse we can't really know right now.

I think that the go world is just now encountering some issues that the chess world has had plenty of time to deal with...Namely that we have strong computers and that they can evaluate arbitrary positions to an arbitrary level, but that the analysis may have a bit of a false precision attached to it.

If we'll allow a chess analogue, I just let the shredder app on my phone do some soul-searching about how it felt 1) e4 should be handled. After a good bit of consideration (a depth of 15 ply!) it has decided that 1) ...e5 is a full tenth of a pawn better than 1) ... c5. While I am fairly confident that this app plays at or above a grandmaster level (especially on the timescales I allowed it to work), I'm not quite ready to declare the Sicilian defense dead.

Similarly, until we can know how AlphaGo evaluates it's preferred position vs. Lee's, it's hard to get much of a takeaway other than "Oh, this other joseki is also playable". If AlphaGo thinks that the tenuki variant wins 49.9% of the time, while Lee's "standard" follow-up variation wins only 49.6% of the time, I think most of us mere mortals would consider them both reasonable and move on. If the analysis said that Lee's move was only winning 30% of the time, then we might want to take a second look and find out if it had truly discovered something.

Mef wrote: Personally I would simply say "AlphaGo would make a different joseki choice than LSD did". Whether or not it's better or worse we can't really know right now.
...
Similarly, until we can know how AlphaGo evaluates it's preferred position vs. Lee's, it's hard to get much of a takeaway other than "Oh, this other joseki is also playable". If AlphaGo thinks that the tenuki variant wins 49.9% of the time, while Lee's "standard" follow-up variation wins only 49.6% of the time, I think most of us mere mortals would consider them both reasonable and move on. If the analysis said that Lee's move was only winning 30% of the time, then we might want to take a second look and find out if it had truly discovered something.

I kind of agree with you, except that I don't consider AlphaGo's analysis "joseki", per say, because I don't feel that AlphaGo thinks in terms of joseki. Joseki are these common patterns that we've come to know, which simplify the game for us. We can use them or modify them, because we've done some sort of local analysis and think that the result is close to even.

AlphaGo has its own method of evaluating the position, and doesn't care at all about joseki.

From this, we can learn that "non-joseki moves" like the ones that AlphaGo plays are possible. But I don't feel we can treat these new moves in the same way as we treat "joseki", since AlphaGo's intention was not to produce a locally even result. We already know that AlphaGo plays locally point-losing moves sometimes to somehow increase its confidence in winning. We can consider these moves as new ideas, but shouldn't directly apply them unless we identify for ourselves that we truly think they are appropriate.

Kirby wrote:

Mef wrote: Personally I would simply say "AlphaGo would make a different joseki choice than LSD did". Whether or not it's better or worse we can't really know right now.
...
Similarly, until we can know how AlphaGo evaluates it's preferred position vs. Lee's, it's hard to get much of a takeaway other than "Oh, this other joseki is also playable". If AlphaGo thinks that the tenuki variant wins 49.9% of the time, while Lee's "standard" follow-up variation wins only 49.6% of the time, I think most of us mere mortals would consider them both reasonable and move on. If the analysis said that Lee's move was only winning 30% of the time, then we might want to take a second look and find out if it had truly discovered something.

I kind of agree with you, except that I don't consider AlphaGo's analysis "joseki", per say, because I don't feel that AlphaGo thinks in terms of joseki. Joseki are these common patterns that we've come to know, which simplify the game for us. We can use them or modify them, because we've done some sort of local analysis and think that the result is close to even.

AlphaGo has its own method of evaluating the position, and doesn't care at all about joseki.

From this, we can learn that "non-joseki moves" like the ones that AlphaGo plays are possible. But I don't feel we can treat these new moves in the same way as we treat "joseki", since AlphaGo's intention was not to produce a locally even result. We already know that AlphaGo plays locally point-losing moves sometimes to somehow increase its confidence in winning. We can consider these moves as new ideas, but shouldn't directly apply them unless we identify for ourselves that we truly think they are appropriate.

I think the same would be said about pros. Instead of playing the joseki, they always look for the best move. Taking a look at my (somewhat dated) version of GoGod, the shape made if B tenukis after drawing back appears over 100 times. Given that a joseki sequence was being played, AlphaGo didn't agree with a particular joseki move and was ultimately expecting a different joseki result...I would have a very hard time classifying this as something other than two strong players disagreeing on a joseki choice.

Mef wrote: I think the same would be said about pros. Instead of playing the joseki, they always look for the best move.

Pros think of joseki, research opening patterns, and come up with new moves. They are always looking for the best move, and can deviate from joseki, but the idea of joseki is a basis for their research. This is different than the way AlphaGo thinks.

Anyway, we know from previous discussions that when we disagree, neither of us changes our mind, so I'm not particularly interested in going around in circles about this with you.

I'll leave it at this: I disagree, and it's OK with me if you have a different opinion.