Computer Chess Club Archives


Search

Terms

Messages

Subject: Centaur Impressions and a bit more -or less...-

Author: Fernando Villegas

Date: 15:48:15 01/05/00


With his usual kindness -mixed with a drop or two of subtile humor, after all
the man knows me pretty well- Dgeordge Vidanovic has demanded my impressions
about these Russian programs - Centaur, Mirage- as if I was knowledgeable guy of
the sort capable of saying something objective about anything. I am not. The
other day Christophe shot me some broadside with plenty of sarcasms precisely
because my obsolete cheers for Sargon V and I must say he was damn right.  I
have told this many times: I NEVER play programs Vs programs, except in the rare
cases when I have been asked to do so.
But then, precisely because of that maybe, maybe -it is a very great MAYBE- I
could say something not entirely irrelevant when the program under observation
has been fitted to play humans, no more.
In fact, Mirage, Centaur and CSTAL, -and perhaps also Century in a degree- are
just the class of programs I love to play, or better said, with which I do not
care to lose games. Humanlike in chess computer is still a debatable thing and
to some here more a lot of hot air that anything else. But it is not.  It is not
because is not only an impression without other reason that the program
launching wild attacks, but there is something in the search techniques and
evaluation altogether different in them.   At least in the case of CSTAL -I
dared to say it many post ago- probably the algorithm is not just a variation of
the current maximin technique. If you look carefully the PV’s you can see that
many lines are generated not looking for the best answer to a presumed best
answer in each ply, but looking at which move generates more possible bad
answers for the opponent. That is, which line is more tricky. The idea is that
in a real game the pressure, even if not entirely sound, is more important that
teorical precision. CSTAL see chess like a fight, not like a puzzle to solve.
And in a real fight you lose because sooner or later, if you are pressed, you
will miss the right continuation. And if you win, you will do because the
opponent made a mistake you saw and knew how to profit from it.
In the case of Centaur what is human is the selection of moves. He does not
seems to be interested too much in every legal move, but in a little set of
relevant moves. Relevant to what? To what is happening in the board. So,
mistakes are possible, but the game get a sense of direction. Besides, the
probable errors that a technique like that can produce give a chance to the
human side. The kick of these programs is precisely that you never feel that the
computer is not prone to tactical mistakes and that every line he has seen is
something like the Voice of God. Of course you can still win with programs
machines-like, but only if you are so positionally skilled you can steer the
bloody thing to a lost game. But for a middle age and middle level player that
is not possible, of course. And which is the pleasure to test your strenght
againts a mechanical lever? Or to win because you previously disabled half of
his strenght? I prefer to lose to Fritz in full gear than to win him because I
setted it to play like a 1600  player.
Now probably you ask yourself: where I can get Centaur...?
But in the meantime you solve that puzzle, this is what author say about his
technique:
"This window shows the Centaur thinking process. To choose the next move Centaur
first overlooks all possible moves and obtains the distributions of probable
gain or loss for each of them, then the move having the maximal more probable
estimation gains a vote. Centaur makes attempts to disprove this move. During
this procedure the move can gain more vote or be rejected, this causes selection
of another move and so forth.

This algorithm is more intimate with a human thinking then a tedious speed
exhaust of the others' chess programs.

When any move obtains first vote, there appears the move string and the color
bar showing the probability of gain obtained by this move. The gains or losses
(if negative) are shown in the 'pawn' units, that is gain the number of pawns.

The blue slide bar shows the range in which the gain (or loss) lay with the
probability of 99% at the current state of approximations. The red bar contains
the gain with the probability of 90%. The vertical line marks the maximum of
probability.

When analyzing the moves, the spread in possible gains is getting less, and
Centaur approaches a decision. After all, Centaur makes the move obtaining more
votes. The bar height shows the comparative number of votes.

Naturally Centaur continues thinking during your turn, and you can use its
analyses as a hint.

The Thinking window takes the same place together with the Open book window
because of they never used simultaneously. When they both are open, the Thinking
window is visible only if the library is exhausted..."



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.