Page 2 of 3

Re: Alpha Zero (redux)

Posted: Sun Feb 04, 2018 4:43 pm
by MartinCarpenter
David Robertson wrote:
Sun Feb 04, 2018 3:40 pm
Jonathan Rogers wrote:
Sun Feb 04, 2018 2:42 pm
maybe it will independently start to share the preferences forged by human experience
Now that would be a breakthrough - computer says humans are right!
Well, human anyway :) I believe that it ended up relying on the Berlin defence vs 1 e4 - Kramnik was a true prophet!

Re: Alpha Zero (redux)

Posted: Sun Feb 04, 2018 5:01 pm
by Jonathan Rogers
Did it say 9...Bd7 in particular was right? 8) :o

But this is my whole point - it could go on further, couldn't it? At what point might we have what it considers its final verdict? (Never?)

Re: Alpha Zero (redux)

Posted: Mon Feb 05, 2018 10:44 am
by MartinCarpenter
Quite likely its converged to a locally optimal point of some sort, and keeping training it won't really improve things that much.

You'd have to introduce a load more 'similar strength but not quite identical' engines and set them playing & training against each other to see.

Might not do much though - presuming that chess is basically drawn in a few thousand different ways in the opening, there's plenty of scope to get good enough to be unbeatable without fixing on one thing :)

Re: Alpha Zero (redux)

Posted: Fri Apr 27, 2018 4:25 pm
by Christopher Kreuzer
And now we have Leela Chess Zero:

https://en.chessbase.com/post/leela-che ... for-the-pc

(AlphaZero for ordinary computers - you can install it on your own computer!)

Re: Alpha Zero (redux)

Posted: Tue Oct 16, 2018 1:53 pm
by Christopher Kreuzer
Interesting article here on AlphaZero and the upcoming World Chess Championship match (between Carlsen and Caruana):

Miguel Illescas previews Carlsen vs. Caruana

Illescas speculates on the impact access to AlphaZero may have on the match, focusing on Caruana being American, and Google being (mostly) a US company. I thought I'd read somewhere that Carlsen had seen some of the AlphaZero games, but I might be misremembering. Whatever the result in the match, there might be something said at the London Chess Classic in this regard, as the initial rounds (semi-finals) are "hosted by pioneering artificial intelligence company DeepMind at Google".

Re: Alpha Zero (redux)

Posted: Sat Nov 17, 2018 10:22 pm
by Angus French
Possibly of interest: Game Changer. (HT Elk and Ruby.)

Re: Alpha Zero (redux)

Posted: Mon Nov 19, 2018 11:01 am
by Mike Gunn
Demis H was interviewed yesterday on the official commentary. He said that the best information on Alpha Zero is on their website but there are more academic papers on it due to appear soon. I guess they will focus on the AI, rather than the chess aspects, though.

Re: Alpha Zero (redux)

Posted: Sat Dec 01, 2018 12:39 pm
by Christopher Kreuzer
A partial transcript of what was said by Demis H (I think he visited the commentary team on 2 days?) on 8 December is here ('A Conversation with Demis Hassabis'):

http://chessforallages.blogspot.com/201 ... sabis.html

Also, the Sadler analysis, forthcoming book and YouTube videos (on the WCC games) are mentioned here ('An AI/NN Blockbuster'):

http://chessforallages.blogspot.com/201 ... uster.html

Re: Alpha Zero (redux)

Posted: Fri Dec 07, 2018 6:47 pm
by Carl Hibbard
10 games (2016 TCEC opening book)


Re: Alpha Zero (redux)

Posted: Fri Dec 07, 2018 6:48 pm
by Carl Hibbard
10 games (no opening book)


Re: Alpha Zero (redux)

Posted: Fri Dec 07, 2018 6:50 pm
by Carl Hibbard
Replay might be a bit slow I had to increase the maximum post size to even paste them in.

20 games with analysis by Stockfish 10:

10 games (2016 TCEC opening book)
10 games (no opening book)

Source was:
https://www.chess.com/news/view/updated ... game-match

Re: Alpha Zero (redux)

Posted: Fri Dec 07, 2018 6:58 pm
by Carl Hibbard
4 selected games with analysis by Lc0:


Re: Alpha Zero (redux)

Posted: Thu Jan 31, 2019 5:04 pm
by Christopher Kreuzer
Interview/article on the Sadler-Regan book Game Changer on AlphaZero:

https://en.chessbase.com/post/interview ... hew-sadler

Re: Alpha Zero (redux)

Posted: Mon Feb 18, 2019 2:31 pm
by Christopher Kreuzer
Another story to add here:

https://en.chessbase.com/post/the-adven ... ing-part-3

Frederic Friedel in the third part of his series on computer chess, reaches the AlphaZero era, with sweeping predictions about the future of artificial intelligence and its impact on society and human civilization.

Re: Alpha Zero (redux)

Posted: Fri Jul 12, 2019 12:06 pm
by Paul McKeown
And another game of high complexity is conquered by an AI: https://techxplore.com/news/2019-07-ai- ... poker.html

The following academic paper may be of interest: https://science.sciencemag.org/content/ ... y2400.full