jwes wrote:Isn't it just as likely that the program will find something better than a draw rather than something worse than a draw if it chooses the delayed draw?
Quite possibly.
It's just the horizion problem.
You can't search forever, so you have to decide when to stop. That's basically decided by how much time you have left, how far down in material you are, and whatever heuristics you've programmed in.
There's never a guarantee, of course. But if you are down a queen or so, tthen a draw looks pretty darn good. And it's probably in your best interest to take the quickest one you can find.
Delaying it, hoping for your oponent to make a mistake (Hyatt's "Meat makes mistakes") is a little foolish unless your program is really strong and your opponent is under time pressure.
Actually, I think the problem is the exact opposite: When there are two draws, how do you get the program to choose the more distant one? With PVS and iterative deepening, once you find the first draw, it will become the PV, and the second will just be another fail low. You want the deeper draw because the shorter is a draw, while the deeper is a draw plus the chance that one of the positions that evaluated as worse than a draw will evaluate as better than a draw when it it is searched deeper, e.g. this position.
[d]1r5k/pnnbp1p1/q1p3Rp/3pB1p1/3Pp1P1/3r4/P6P/3Q1R1K w - - 0 1
You need to search pretty deeply (or have good extensions) to realize that this position is better than a draw.
If you prefer the distant draw then you may discover later that the distant draw is no draw because the opponent has a move that you did not search deep enough so I do not think prefering the distant draw is better when you play against equal opponent.
When you play against weaker opponent prefering distant draw seems to be better and when playing against stronger opponent prefering short draw is better but the difference is very small and I think that it is not a good idea to spend time on prefering short draws.
In tablebases position you may want different scores for draws as Ed Trice suggested but I do not think that his way his optimal because there may be cases when there is one drawing move that is easy to find and having the number of drawing moves and the number of losing moves is not optimal to decide about the practical probability for a mistake.
I guess that combination of this knowledge and some static evaluation without tablebases may be better.
Uri Blass wrote:If you prefer the distant draw then you may discover later that the distant draw is no draw because the opponent has a move that you did not search deep enough so I do not think prefering the distant draw is better when you play against equal opponent.
Uri
This whole discussion makes no sense if the draws are not forced.