AlphaGo vs. The World: Game 51, AlphaGo Master (W) vs. Chou Chun-hsun 9p (B)
Vložit
- čas přidán 12. 02. 2021
- Chou Chun-hsun 9p (周 俊勳): Born in Chiayi, Taiwan in 1980. He became 1p in 1993 and was promoted to 9p in 1998. In 2007, he won the LG Cup to become world champion.
Find the first volume of our book
AlphaGo to Zero: The Complete Games
Vol. 1: AlphaGo vs Lee Sedol 9P
at smartgo books
gobooks.com/
Support my channel at
Patreon
/ michaelredmond
There are some sound quality issues with the AlphaGo series because of the equipment I was using at the time and the internet link.
Produced by Stephen Hu and Chris Garlock
Biographies researched by Stephen Hu
In the first days of 2017, a mysterious online player defeated dozens of top professionals on go sites in Asia. “Master” - the latest version of the AlphaGo AI that had defeated Lee Sedol in March 2016 -- achieved a staggering 60 consecutive wins against many former and current world go champions. In this series, Michael Redmond 9P, hosted by American Go E-Journal Managing Editor Chris Garlock and produced by Stephen Hu, focuses on the key points of each game in brief videos, which will be expanded upon for Volume 2 of AlphaGo to Zero: The Complete Games.
Subscribe for easy access to more Go videos with game analysis, basic and AI josekis, and more!
/ @michaelredmondsgotv
playlist:
Go problems
• Go puzzles
suggested video:
The Direct 3-3 Invasion Keep it Simple!
• The Direct 3-3 Invasio...
Sign up to my Twitter channel!
/ redmondgopro - Hry
another unique video
I remember trying to play manego with the computer with white on an 18x18 board and 7 points komi... the computer made a capturing race in the center which black won by 1 move. :-(
Michael, I hope you'll review some of the UEC Cup games next month. Some of the past heavyweights didn't enter this year (Golaxy, Deep Zen, Fine Art, etc) so my money would have to be on Katago, but we'll see if any of the new names are strong enough to win. :)
Fujisawa Hosai 9-dan was famous for playing mane-go (though not for the entire game). Are there any current practitioners at the top level?
Haven't seen many top-level pros do it; in the amateur world, however, Jonas Welticke (2-time German Champion) is known for frequent mane-go in his matches.
What was he thinking? This mirror-go attempt with black against komi was hopeless.
New to Go. Are there more vids with 9x9 format in the pipeline?
Probably moving on to 13x13 but still with beginners in mind. It is considered a good idea for beginners to switch to larger boards at an early stage.
With all these AI-Zero machines around nowadays, I think it to be a great idea to have some elaborate matches between them, where White has the strict order to play mimic go all the way until it judges it will lose if pursuing this strategy any further. This all, with the sole purpose to figure out what to do with Black about mimic go, from the Zero perspective. To me, mimic go poses some danger to the game with such a big komi, it is up to White to diverge from it at the right time, thereby maintaining the possibility to win without a real effort except the diverging, where it could be too late for Black to do something meaningful about it. In fact, I am surprised that mimic go isn’t played more often, except that it makes you look bad in the go playing community. Somewhere else on the internet someone mentioned that the most famous mimic go player, Fujisawa Hosai, was HATED (his capitals) for doing this so often.
The bots actually can’t handle mirror go very well - like as in humans can win as w playing mirror go against them. Deepmind’s is immune because it plays on a data center. A basic bot on a laptop generally doesn’t do enough playouts to get there and the basic bots didn’t get to train on a million CPU virtual machine. If Deepmind hadn’t decided to run off and play Starcraft this is an interesting question they could investigate.
You say "test the machine" - this was still within the first few days, wasn't it? Had they already figured out / guessed that it was a new algorithm?
This was several months after the AlphaGo-Lee Sedol match and pros were waiting for Deepmind's next move. When Master played 10 games a day against top pros with a perfect win record, I think the players had an idea. I believe that this game was on day 6.
No human ever did anything like this so either there was a new pro better than any previous pro or Deepmind was “trolling” with a hidden bot. Deepmind’s thinking was to see how well the machine would do if the pros didn’t know it was a machine. Of course, had it lost a few games here and there they’d have collected 50 games and we’d still be debating if it was a bot. But since it just kept winning Deepmind eventually admitted it was their bot.