Author: Christophe Theron
Date: 15:32:47 01/01/04
Go up one level in this thread
On January 01, 2004 at 05:54:07, Tord Romstad wrote:
>On December 31, 2003 at 21:23:25, Christophe Theron wrote:
>
>>On December 31, 2003 at 20:58:08, Tord Romstad wrote:
>>
>>>On December 31, 2003 at 13:43:42, Steven Edwards wrote:
>>>
>>>>On December 31, 2003 at 13:31:56, Ed Trice wrote:
>>>>
>>>>>This was the Qh5+!! sacrifice, correct?
>>>>
>>>>Yes; a mate in ten. One variation found by Symbolic's low level search goes:
>>>>
>>>>(1. Qh5+ Nxh5 2. fxe6+ Kg6 3. Bc2+ Kg5 4. Rf5+ Kg6 5. Rf6+ Kg5 6. Rg6+ Kh4 7.
>>>>Re4+ Nf4 8. Rxf4+ Kh5 9. Rg3 Bxe6 10. Bg6#)
>>>
>>>Symbolic is an extremely interesting project, and I hope you succeed in creating
>>>a super strong engine with your unusual approach. However, forced mate
>>>positions
>>>like this one are not a good way to measure progress. Solving them quickly is
>>>easily
>>>achieved by more conventional methods (the above position is solved in 7 plies
>>>and
>>>only a couple of seconds by Gothmog), and does not necessarily imply high
>>>playing
>>>strength in normal games.
>>>
>>>Tord
>>
>>
>>
>>Solving it in approximately the same time with a totally new approach would be
>>extremely encouraging I would say.
>>
>>Current computer chess programs are light years behind human intelligence. Mine
>>included. More "human-like" or "intelligent" approaches are really welcome and a
>>very promising field.
>>
>>I'm fed up with Crafty or Fritz or Chess 4.x clones. I'm fed up with chess
>>calculators. Show me something else now.
>>
>>It is going to be a very hard road. But there is much more merit in it than in
>>writing yet-another-alphabeta-nullmove-hashtables-computer-chess-program.
>>
>>I would *LOVE* to see Chess Tiger torn into pieces by a Symbolic-like approach.
>>The whole computer chess field needs to have his ass kicked by something new,
>>because the current approach is coming to an end: it's very good, but has taught
>>us very little about what intelligence is. It's like the Matrix movies: lots of
>>promises at the begining, only disappointement in the end.
>
>I agree 100% about all of this. But when developing a chess engine using an
>entirely
>new and different approach, I think it is very important to concentrate on real
>games
>from the beginning, rather than mating problems and tactical test positions.
>Otherwise
>you risk to end up with an engine which cannot do anything well except running
>test
>suites. Of course this danger also exists for programs developed with more
>conventional
>techniques, but I think the danger is considerably bigger when you try to do
>something
>new and revolutionary.
>
>Tord
The danger is also in trying to be revolutionary. It should not be your goal.
Take what is good in the current successful approach, and add to it what the
human players do well.
Basically, a grandmaster searches a tree that is several orders of magnitude
smaller than the tree searched by a computer and they both reach a similar
playing strength.
I have always been shocked by this. I have tried to work in that direction. I
have been successful by achieving just a ridiculous fraction of the goal. So
imagine how successful one could be if one could achieve just half of that goal!
There are also possible improvements in the way to evaluate positions, but I
don't think it can be as effective as improvements in the search.
Christophe
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.