AlphaGo vs. AlphaGo with Michael Redmond 9p: Game 30
Vložit
- čas přidán 4. 04. 2019
- Michael Redmond 9p, hosted by the AGA E-Journal’s Chris Garlock, reviews the 30th game of the amazing AlphaGo vs. AlphaGo selfplay games. Sgf file posted here: www.usgo.org/news/2019/04/alph...
The 50 game series was published by Deepmind after AlphaGo's victory over world champion Ke Jie 9p in May 2017.
Produced by Michael Wanek & Andrew Jackson
Thumbnail image of Rock 'em Sock 'em Robots by Lorie Shaull - Own work, CC BY-SA 4.0, commons.wikimedia.org/w/index... - Hry
Michael: 'The black player would be upset, white would be happy, and they would both be wrong.'
Chris: 'That's the headline of this game.'
One of the best phrasings yet in this series. I would give more than one thumbs up if I could. ^^
+
Thanks to Michael for following his most plausible endgame to the end. It's helpful to see what AlphaGo is looking at.
Always love when he does that. :)
Great video! Thank you both! Love the history along the way.
Thanks to both of you for this once more exiting game analysis.
On page 100 of Invincible. The games of Shusaku, it says he was faced with that taisha move for the first time
Invincible is a wonderful book.
I am so ready!
1:03:18 Final position if they had it played it out instead of resigning. Incredible game and commentary. Mindblowing stuff! :)
Yey, back again :)
Great job thank you very much! I would love to see some analisis of some top profesional games too (from Iyama Yuta, Ke Jie, Park Junghwan, etc) .
At 6:30, when White play B7, Black can play C8, and then cut at D3 as alphago played with KeJie. Therefore B7 is not a very good move. I always love Michael's game reviews, but bit surprised he forget this variation.
+
The invasion by Black on the left side of the board has a strange feel to it, almost like it's exactly what a losing AlphaGo should have been doing to turn things around, instead of the way it always plays moves that will only work if the opponent misses some easy response. Basically, I wonder if adding some extra logic to the algorithms has been tried, so that a losing AlphaGo would try to complicate the game so much that it could win against a lesser opponent. Unfortunately in this particular game, as Michael showed, there seems to be a straightforward refutation that rather peacefully avoids all the complexity that followed. Too bad. Beacuse although it's fun seeing two heavyweights trade equally powerful punches for most of the match, I can't help but think that a bit more intelligence from AlphaGo when the game is starting to go downhill would improve the late game experience for those of us reviewing the games. After all, who doesn't love to cheer for the underdog?
That reminds me of another thought I had: it would be interesting to see if a "teaching Alphago" could be made that would play evenly against any level of opponent, and challenge them to overcome some late game comeback, yet always keep things within a few points for either player.
Ladder is good for black no?
After black B2, AlphaGo Teach doesn't even consider C2. The only moves it gives for white are the game move D2 (56.3 % for white) and C5 (54.9 % for white).
After white D2, its only variation is fascinating (it ends with 57.5 % for white): online-go.com/demo/view/385281
Please give Michael an instance of Leela. Zen sucks in comparison