AlphaGo vs. AlphaGo with Michael Redmond 9p: Game 22
ฝัง
- เผยแพร่เมื่อ 23 ก.ย. 2024
- Michael Redmond 9p, hosted by the AGA E-Journal's Chris Garlock, reviews the 22nd game of the amazing AlphaGo vs. AlphaGo selfplay games. Click here for the game record www.usgo.org/ne...
Absolutely beautiful game. :) Final position if they had played it out to the end more simply is at 59:15. Thank you to everyone involved in making these videos. :)
Wow really wonderful game, thanks Michael and everyone. This was one of the few AG games I've seen that had a nice unpredictable flow of board development, instead of the big sacrifice trades based on 1/2 point calculating but more like large patterns developing, evolving, and redistributing over time. Beautiful to watch it almost settle and then spill all over again, haha.
So glad you guys are still doing these
Beautiful analyzis Michael Redmond sensei.
Thank you for these. Omg they're so exciting. More than pro games, these alphago games make me realise I really don't know how to play this game ~low dan amateur
Yes. More players should be viewing these great vids. Q1: Can Alpha Go get even stronger? Q2: Does Alpha Go have any way of 'explaining it's strategy' of a game to other players? This is part of the "nobility of Go." It's really too bad if the program can't do that. Human players talk about their game afterwards all the time.
Thank you!
Thanks for the content, but could you please add which versions of AlphaGo to the title?
They do - if it is alphago zero, it is in the title - if not, it is the self-play games ;)
@@godave8934 all versions play self-play games and there are three versions of alpha go: lee, master and zero, which have also played against each other. With that amount of combinations, I think that specifying in the title or the description is almost mandatory.
"The 50 game series was published by Deepmind after AlphaGo's victory over world champion Ke Jie 9p in May 2017."
It's in the description of the first video - it is THE self play series :D
deepmind.com/research/alphago/alphago-vs-alphago-self-play-games/
Here is more info about the particular series of selfplayed games, in case you missed it ;)
Every video they get somebody asking this question, I don't understand why... The descriptions are perfectly adequate...Also I wouldn't call "AG Lee vs AG Zero" a self-play. Those are two distinct nets playing against each other.
David Ulbricht The "self-play" title is old and doesn't cover the fact that there are now 3 versions of self-play. Besides, it's annoying to have to check which version is playing every time I watch the video (I'm sorry if I don't have that good a memory), would it be that troublesome to specify the version in the title?
Thank you very much for this high quality content! - What is your thought on the reason behind the exchange of forcing moves? I'd say it is either irrelevant for the AI so random or if the other move is bigger it might even have a higher winning percentage? What happens if they ignore each others forcing moves? Who is winning then? Should be the one ignoring first?
The "AI" has no reason, it is implementing software written by reasoning humans. Once you device a way to score moves the reasoning is over. I'd recommend you read Conway and Berlekamp's "Winning Ways for Your Mathematical Plays". It will teach you what true ingenuity is all about, it is not about efficiently implementing a scoring system (which is all AlphaGo does).
Interesting Elfv1 bot very quickly finds and strongly wants to play the t7 kosumi at the same time AlphaGo did. It doesn't even consider the territory-surrounding move at the top Michael expected. Though like Michael Elf can't understand why white played j4 hane instead of defending the cut, though it would do so with n5 as a lot better than m6 (because if black blocks to make it gote black has to suffer a bad shape empty triangle).
Yay, welcome back!!!!
Hi, another great commentary!!
excellent review!
Thank you for the analysis. :)
These are criminally underviewed. English speakers are too worried about either watching other people hit each other or turn left around a track..
Clap clap. But seriously folks, it is time to let the AI go, literally. The great human advancement made with AlphaGo was not the 9-dan (19-dan??) status of an "AI," it was the proof-in-principle of efficient algorithms for heuristic solutions to computationally hard problems. It is time to unleash all that on severe problems facing humanity, not on an (admittedly cool, but irrelevant) ancient board game.
Second!
First!!!