The Teaching tool contains AlphaGo Master’s evaluation of openings found in strong human amateur games up to 2017.
Of the first 15 moves in this example, 6 are suboptimal human moves according to AlphaGo Master. From move 17, this opening starts to diverge from human play. So only from move 16 until move 35, the sequence represents AlphaGo Master’s 1st choice moves.
And then there is the matter that AlphaGo Master was trained on human games, which may give it a bias in its evaluations. It is probably weaker than the Zero style AI that came later (AlphaGo Zero, Leela Zero, KataGo).
About the lower left corner enclosure (D4 & C6), playing a knight move from a 4-4 stone is quite common for all strong AI. They tend to play simple moves in the opening while focusing on taking territory in the corners.
This particular enclosure has also become one of the most common opening moves in strong human games in recent years. Surely you are aware of that?
The black stones looked low Chinese, so I placed seven white stones in a conventional pre-AlphaGo way, identified the upper right as a high-approach-and-tenuki joseki, and arranged the rest around the other black stones. I was lucky my presuppositions were true.
Sorry for delaying you all. I’ve just been lurking here, but haven’t had the courage to actually play yet. Maybe in the next one. Please don’t feel the need to wait for me though.
These are all 141 white stones in the final position of a korean pro game. 139 black stones are missing from the board. In total 327 moves were played, and all dame have been filled. There is no seki. The scoring breaks down like this: (korean rules are very similar to japanese)
Black
White
Territory
51
45
Dead stones
9
6
Prisoners
22
25
Komi
6.5
Total score
82
82.5
Note that dead stones and prisoners refer to the opponents stones, i. e. there are 9 dead white stones on the board.
Hopefully I didn’t mess up any of the counting.
Scoring
1 point / correctly placed black stone
BUT
if you have any incorrectly placed black stones, you get 0 points.
Thus the strategy should be to only place stones you are 100% sure about. Placing all 139 stones is definitely impossible without being super lucky, but guaranteeing 100 points should be pretty easy. The question is, who will push the furthest for those extra last few points?
Even though it is pretty easy to figure out which areas are territories, it’s probably not obvious to everyone. So I’ll let each person figure it out for themselves. I think if they are obvious it’s pretty quick to place them anyways.
Realized I missed one stone White P1 was missing, fixed now. I’m going to do another count of the stones to double check everything is correct now.
Edit: Yep, everything adds up correctly now.
Edit 2: Ok, one more embarrassing mistake, I wrote that the game was japanese when it was in fact korean. I got confused because an earlier candidate game I looked at was japanese. Shouldn’t make a difference to anyone’s guess, but I’ve corrected the post.
I’ll decide that this would constitute cheating. People have different access to databases and different knowledge on how to search them, so it seems a bit unfair.
(More importantly, searching a database doesn’t seem as fun to me. But if there are some interesting challenges connected to it, maybe we can make a later round for something like that!)
Human pros tried various openings, and I did see the discussion about the first 15 moves might be human, so I didn’t know what to expect from a mixed AI and human moves game. And some strange uncommon enclosures popped up in recent years, and some of them did came from AIs and pros studying released AlphaGo games.
Also, I need to admit my fuseki is pretty weak, I didn’t study so much fuseki theories as much, I tend to stick to a few of those I am familiar with. My favor is the mid-game fighting and contact fighting, revolved around semeai.
I kind of feel the same as well, except I didn’t try to make guesses; I just concluded in advance that it’d be too hard
Maybe there’s some tricks one is supposed to figure out to help, like large areas of connected empty space probably have Black stones in them (around the border) - at least if the game got to counting and was a close enough result.
White P1 was missing, fixed now. I’m going to do another count of the stones to double check everything is correct now.
