Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Is LC0 the future of computer chess?
- - By billyraybar (***) [us] Date 2018-11-18 00:42
Is the LC0 model the future of computer chess? Is alpha beta safe for the foreseeable future? Is AI where it's at?  I'm asking for a knowledgeable individual to sum up where computer chess is and where it is going. Thanks in advance.
Parent - - By bob (Gold) [us] Date 2018-11-18 02:58
Jury is out on the LC0 question.  And will likely remain out due to the lack of available information.  Alpha/Beta is already more than enough to defeat the best humans in the world, and they will continue to get stronger.  Whether something will surpass alpha/beta or not is unknown, and really not very important from a pure chess perspective anyway.  The death of alpha/beta has been reported many times in the past.  Each time proved to be an error.
Parent - - By Carl Bicknell (*****) [gb] Date 2018-11-18 10:10
The main problem as I see it isn't alpha beta or neural networks, but chess itself. We are running out of head room for future progress. Perfect chess may be less than 500 elo above what we have now (SF on biggest machine available).
Parent - - By Sesse (****) [gb] Date 2018-11-18 10:38
Depends on your definition of “perfect chess”. Is 1. e4 better than 1. h3? Both are probably draws.
Parent - By Carl Bicknell (*****) [gb] Date 2018-11-18 13:37

> Depends on your definition of “perfect chess”. Is 1. e4 better than 1. h3? Both are probably draws.


At the very least (and this is a poor definition, but probably a likely one for CC) - an engine which will no longer lose a game. I can imagine in 10 years SF would be unbeatable.
Parent - By Labyrinth (*****) [us] Date 2018-11-18 22:15
I think what they need to do is make a large enough neural net to account for special cases, and then have it train on chess960 and shuffle chess, along with the tablebases/tactics/puzzles. That way it should have seen enough to make it quite strong. The problem, among others, is that this takes a long time. Doing this, and perhaps using a hybrid approach with conventional chess engines might help reach a new level.

The really crazy thing is that while there is a plateau happening, it's not happening nearly to the extent that I thought it would. I really thought we would have hit a wall already, yet progress, while slowing a bit, is still fairly stable. It's astonishing how deep chess really is.
Up Topic The Rybka Lounge / Computer Chess / Is LC0 the future of computer chess?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill