Ohashi Hirofumi has published excellent commentaries in
Go World on several of the games played by AlphaGo as Master in the 60-game series at the turn of the year - and promises more to come. The commentaries are decent sized, easy to understand and above all insightful as regards the ways AlphaGo is making pros rethink the game. We are lucky to have had already some similar Chinese and Korean commentaries, but this is the first substantial set I have seen in Japanese. A notable feature of Ohashi's version is that he compares games with similar openings very well.
One position, below, especially caught my eye. It is from Game 23. AlphaGo is Black, playing Kim Cheong-hyeon.
- Click Here To Show Diagram Code
[go]$$W
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O O . . . . . . . . . . . . . . . |
$$ | . X O X . O O . O . . . . . . . X . . |
$$ | . X X O . . X . . , X . . X . X . . . |
$$ | . X O O . . . . . . . . . . . O . . . |
$$ | . . X X . . . . . . . . . O O . O . . |
$$ | . . . . . . . . . . . . . . X . . . . |
$$ | . . . . . . . . . . . . . . . X . O . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . Q . . . . . . . . . . . . . . . . |
$$ | . X Q . . . . . . . . . . . . . . . . |
$$ | . X Q . . . . . . . . . . . . . . . . |
$$ | . X Q . a . . . . . . . . . . . . . . |
$$ | . . X Q . . . 2 . 1 . . . X . , X . . |
$$ | . . X Q . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]
My fascination with this position can be traced in previous threads here. We were discussing a position where I raised a query about a shape that everyone else was treating as thick. I noted that I had recently seen several commentaries where a shape I (and evidently just about everyone here) would consider to be thickness was regarded by pros as "not really thickness." I'm not sure I convinced anyone, and it seems only a very few even paused for thought, but I have since seen further examples and now we have a case where the goddess AlphaGo apparently thinks something is "not really thickness."
Here is the relevant portion of Ohashi's commentary:
Quote:
White extended to 1 only to see Black invade nevertheless at 2. Since overall Black is ahead on profit, it is common sense for White to want to extend as far as possible to make his triangled stoneswork, but faced now with the Black peep at 'a', it's impossible for him to know which groups are the strong ones. Black is asserting that, "The white triangled stones are NOT thickness!" Even so, it would have been hard to make a narrower extension than 1 and so end up with over-concentrated shape. This sort of intuition (!?) about thickness is, I believe, the secret of Master's strength.
Needless to say there are very similar insights about thickness elsewhere in his commentaries, and likewise in comments by other pros about AlphaGo. Since, as I have said, I have seen an increasing number of comments of this tenor recently, even outside the AlphaGo ambience, it may be that pros had already discovered truths about the phenomenon of a doorstep sandwich of thickness dissolving into panada even before AlphaGo, and they are welcoming AI's confirmation of their suspicions.
Incidentally, the human in Game 23 has previously been given in this forum as Kim Chi-seok. Ohashi is a pro well plugged into the internet scene and he also writes regularly for
Go World on AI in go. I'd be surprised if both he and
Go World got the name wrong. In addition, the handles used in the two games so far attributed here to Kim Chi-seok were quite different. But that just stacks the odds in favour of Cheong-hyeon - it's not proof. Does anyone have
hard evidence in favour of Chi-seok?