Author: Steven Edwards
Date: 00:17:00 07/01/05
Symbolic: Light at the end of the tunnel
Consider again BWTC.0031:
[D] 2qrr1n1/3b1kp1/2pBpn1p/1p2PP2/p2P4/1BP5/P3Q1PP/4RRK1 w - - 0 1
It's the well known mate in ten position often referenced in the computer chess
literature. {Incidentially it was Alexander Alekhine playing White in a game
from the mid 1920s.) The MacLisp program Paradise used patterns and planning to
solve the problem with only 109 search nodes using less than an hour back in
1980 on a pdp10. Today, most programs on fast hardware can solve it under
tournament time controls, yet may require many millions of nodes to do so.
The above position has also been one of the freqently used tests for Symbolic.
And, after looking forward for well over a year to having it solved by the
cognitive search, that goal was achieved yesterday 2005.06.30.
Symbolic was able to locate the winning move 1. Qh5+ using forty-one seconds CPU
time (700 MHz PPC750). The search tree had sixty-seven interior nodes and
nineteen leaf nodes for a total of eighty-six nodes. The entire position search
tree was retained in memory and the search was suspended and resumed at
different points multiple times.
Some caveats:
1. Some less than perfect defensive moves were tried, so the the entire
resulting analysis is not perfect. Good enough for the first seven or so ply,
though.
2. While there is nothing specific in Symbolic in regards to the BWTC.0031 test
position, the program's cognitive search understands little more than mate
attack themes.
3. Much of the success is due to the utility of the GA derived mate attack move
suggestion pattern matcher and not a more sophisticated, to-be-written
multilevel pattern knowledge library.
4. There is still a very long way to go.
This page took 0 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.