Author: Fernando Villegas
Date: 20:01:40 04/07/98
After my last game against CTAL -white.exe version-, when I was trying to hold all together -I did not make it- against an untiring attack, I have had something like a revelation about how this program maybe works and the reason of, at the same time, his proficiency against average, experienced, expert an even Fide master class players and at the same time his “weakness” against the very top programs in terms of SSDF ratings. My impression was and is this: that CSTAL not only make good use of a far great knowledge code than other programs, but besides, and maybe more important, it makes use of a search technique very different to the well known Maximin kind of search, where in each node the program pick up the moves that gives the lower rating -or the less high- to his rival. It’s my impression that CSTAL uses a search technique where the purpose is, inside certain limits, to get moves that produces maximal pressure for the adversary and not the minimal score for his best available move. That “maximal pressure” would be equal to get a maximal number of threats and dangers for the rival even at the cost, some times, to let one or more candidate moves that, if picked up, surpass the maximin standard approach. The presupposition that Chris W probably did was a very obvious one: humans are not, inside the tactical horizon of the game, so precise and almost perfect as programs are to select ever the best move to hold back a threat in tactical terms; on the contrary, they get tired, lose concentration, lose nerve and lastly, after doing well under many attacks, commit the last, fatal mistake. My experience with CSTAL is precisely that: he launch an attack after another with the idea that sooner or later, under pressure, you are going to choose one of the moves that maximizes his advantage and not the move that gives you a best chance, even if theroretically exist. In fact, is not that what happens in normal game between humans? How many times a human game goes trough and trough the best moves for each side? For the same reason CSTAL does not so good against computers, that does not get trired and calculate everything inside a normal horizon of let us say 7 to 12 ply. Am I mistaken? If some programmer is there to show the Light.... Fernando
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.