Author: Christophe Theron
Date: 08:40:51 02/28/99
Go up one level in this thread
On February 28, 1999 at 01:38:46, Peter McKenzie wrote:
>On February 27, 1999 at 21:36:15, Christophe Theron wrote:
>
>>On February 27, 1999 at 14:48:30, Dave Gomboc wrote:
>>
>>>On February 27, 1999 at 11:46:11, Christophe Theron wrote:
>>>
>>>>On February 26, 1999 at 22:09:32, Peter McKenzie wrote:
>>>>
>>>>>On February 26, 1999 at 13:16:01, Christophe Theron wrote:
>>>>>
>>>>>>It is clear to me since several years that being a good chess player is a
>>>>>>serious handicap for anybody trying to write a top level chess program.
>>>>>>
>>>>>>Every rule has its exceptions, so I guess you can find some. But can you list
>>>>>>strong chess players that wrote good chess programs? I'm not even sure Larry
>>>>>>Kaufman can be included in the list, because he does not program.
>>>>>
>>>>>Interesting theory, of course Hans Berliner is the obvious exception to it.
>>>>
>>>>Are you sure? Hitech was a nice program, but it was a long time ago. How would
>>>>Berliner's program compare to current strong programs?
>>>
>>>Pretty well, I'd imagine.
>>>
>>>> Christophe
>>>
>>>Dave Gomboc
>>
>>I'm still not sure. Last time (to my knowledge) Hitech played in the World
>>Championship, it was in May 95 in Honk Hong, on a fast supercomputer.
>>
>>It finished in the 6th place. Ranking of the first places were:
>>
>> 1. Fritz (4/5, 5/6)
>> 2. Star Socrates (4/5, 4/6)
>> 3. Deep Blue, Frenchess, Junior (3.5/5)
>> 6. Rebel, Genius, WChess, Zugzwang, Hitech (3/5)
>>11. Cheiron, Virtua Chess, Schach 3 (2.5/5)
>>...
>>
>>Several microprocessor based programs (on P90 and P120 I think) did better or as
>>well as Hitech which was computing 120000 positions per second. The micros were
>>far from this (except Fritz maybe, but with a much simpler evaluation).
>>
>>I think Hitech could be overplayed by current programs/processor combinations,
>>unless they have improved their hardware.
>
>Is this relevant? Hitech was one of the very best computer chess players of its
>day, its creator was a strong chess player (former world correspondence
>champion), so we have a counter example to your statement of "It is clear to me
>since several years that being a good chess player is a serious handicap for
>anybody trying to write a top level chess program."
As I said every rule has its exceptions.
>The fact that todays Micro programs may or may not be better than Hitech has
>nothing to do with whether Berliner is a strong chess player or not. Techniques
>have advanced since Hitech was developed (around 1985). I think that if
>Berliner had continued to work on Hitech it may well still be one of the best.
That's right, in the case of Hitech it is very difficult to compare...
>Actually the exact meaning of your statement is unclear to me. Do you mean that
>"it has been clear to you for several years that being a good chess player is a
>serious handicap...", or that "it has been true for several years that being a
>good chess player is a serious handicap...". If you mean the second one, what
>has changed in the last several years (ie. since Hitech) ??
What has changed since Hitech is that we are now able to compute much deeper
than what was possible in these days. When you don't compute deep enough, you
need knowledge to avoid simple traps your search cannot see (pins, trapped
bishop, mate threats...).
Being a good chess player may help in this case, but I still have some doubts.
On faster computers, with better search techniques, go can compute so deep that
including some tactical things in your evaluation is not required any more. It
is even counter productive.
So the job is different in this case. You need to include only very simple
knowledge in your evaluation. Your task is to find what is unecessary and remove
it. It's amazing to see that with the right basic knowledge and a good search
your program behaves exactly as if it had high level knowledge.
I see that everyday. Beta testers send me games and mention in their analysis
that Tiger did this or that for whatever (well known) positional reason. They
think Tiger has some specific knowledge, but in fact he has not.
But let's relax. As I said in another post, it's maybe easier to agree that
being a good chess player gives you no special advantage when you write your
chess program.
>Christophe, do you think the more knowledge based approaches to chess programs
>are inferior? I personally think there is more than one way to write a good
>chess program, witness the big speed difference (but similar strength) between
>Hiarcs and Fritz.
The future of chess programs is not to put more knowledge in them. It is first
to remove unecessary knowledge, then to introduce some missing concepts (I think
there are not many of them missing).
I mean that future chess programs may well end up having not more knowledge than
current programs. But they will have more adapted knowledge.
And I think the kind of knowledge you need when you compute 1 million nodes per
second is very far from what you need when you compute 10 positions per second.
That's why no human being so far, good or weak chess player, knows what this
knowledge is.
That's why I think that being good or not at chess does not matter when you
write a chess program.
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.