Author: Ratko V Tomic
Date: 09:09:10 10/24/99
Go up one level in this thread
> i think around 30 ply or so with a average (current)eval and perhaps > 40 to 50 ply with only material . Even 30 plies for exact minimaxing (with over 10^23 nodes examined with optimim move order) is far beyond any conceivable technology. Selectively, yes, they even go today 30 plies thay way. As for the 50 plies, with over 10^38 positions, that would be a "galactic" sized computer, it won't happen. There are also limits on the minimum size of computer circuits and the customary doubling of speed every couple years won't go on for much longer. We may see programs examining perhaps 10^12 positions, equivalent to 15-16 plies of exact minimaxing. Not a big deal for a well motivated and well prepared GM to subdue. The present style programs, essentially alpha-beta searchers with refinements, are a dead end. Their efficiency (the proportion of non-junk nodes to all nodes examined) drops toward zero with increasing depth, which is why even otherwise seemingly unsound selectivity (be it the null-move or gradual ad hoc forward pruning increasing with depth) appears to work better at higher depths -- even a random discarding of nodes will end up discarding proportionately more junk at higher depths, thus it will seem more accurate (and less wasteful of computing resources). But the programs which will achieve the unquestionable superiority over the top human players (and, of course, over the present style programs) will be of a smarter kind, with several layers of control and search trees, where alpha-beta is only the lowest layer guided between the nodes of the sparse next higher level tree by the reasoning components (similar to what humans do when they use computer for in depth analysis).
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.