Author: Albert Silver
Date: 16:03:05 10/11/04
Go up one level in this thread
On October 11, 2004 at 15:49:07, Enrique Irazoqui wrote:
>I have Chrilly's permission to post these remarks he emailed me about Hydra:
>
>"Winning against GMs has to a certain extent already lost its charm. We are
>already in the position were we can only lose. I personally think that this
>expectation overestimates the capabilities of the program. But it is a matter of
>fact that this version played 8 games against top GMs, won 6 and made 2 draws.
>Gives an Elo-rating of >2900."
>
>"Leonxto (Leontxo García, well known chess journalist from Spain) wrote in El
>Pais: "y ademas muestra uns unaudita "comprension" de la estrategia" (and it
>also shows an unheard of "understanding" of the strategy). Actually it has no
>idea at all about this. It has some simple rules like "It is beneficial to
>attack the area around the opponent king, to attack opponent pieces, to control
>the center. It knows a little bit what abut bad/good bishops, when is the knight
>stronger than the bishop". The big surprise for me was, that even this simple
>rules were sufficient to know that the exchange of the light-squared bishop of
>Ponomariov in the first match was not the best idea. The evaluation went up a
>1/4 of a pawn. Generally I saw no GM move which had some flavour of geniality (I
>am speaking only of the Hydra-games). The moves which really surprised me (e.g.
>Qe2 after Nc4 in the last match) were done by Hydra."
Thanks for sharing. As to the remarks on its knowledge, I must say I'm not
terribly surprised. Similar remarks have been noted on Fritz when it was a far
more 'ignorant' program. The depth of the search allows it to see the
consequences of its moves sufficiently to compensate for a lack of knowledge,
and my oh my is Hydra going deeply. Even Ed Schroeder has noted this on Rebel,
and the latest versions have the knowledge parameter at a mere 100 when the
maximum goes all the way to 500. Much as it would be nice to use all the things
it potentially knows, it just hasn't proved useful enough to compensate for the
fewer plies. Still, it would be interesting to see how a real expert in
anti-computer chess fared against it. Much as I respect the GMs who played, I
hardly consider Ponomariov or Karjakin as expert computer opponents, which makes
their suitability for this particular event questionable.
Albert
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.