Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 3000 Rated program & GM's

Author: Robert Hyatt

Date: 15:58:40 01/06/99

Go up one level in this thread


On January 06, 1999 at 13:40:19, Mika wrote:

>On January 06, 1999 at 12:23:50, KarinsDad wrote:
>
>>On January 06, 1999 at 10:07:58, Christopher R. Dorr wrote:
>>
>>[snip]
>>
>>>>I think a more interesting opinion question is: "Do you think that a 3000 rated
>>>>chess program available to the general public (including GMs) will drastically
>>>>change the way in which GMs approach the game of chess?"
>>>>
>>>>Cheers,
>>>>
>>>>KarinsDad
>>>
>>>
>>>Now this is a quite interesting question! One which may well be of practical
>>>value within the next few years. What would a 3000 rated program do to GM's, and
>>>how would it change their approach to chess?
>>>
>>>I was actually thinking about this yesterday. I was watching a fantastic speed
>>>game between Ban and Crafty on ICC. One onf the most fascinating games I have
>>>ever seen (human or computer). Crafty had 2 knights and a pawn vs. Ban's 2
>>>bishops. For most of the game, Crafty was being amazingly creative in corralling
>>>the bishops. Using techniques that only a profoundly strong computer could
>>>manage. Now I'm a 2150, and a pretty experienced player, and I suddenly realized
>>>that I simply couldn't understand much of what was going on. I simply wasn't
>>>good enough. maybe with several hours and the help of Crafty, I could, but in
>>>real time, I simply was left to appreciate the beauty of the game without really
>>>understanding how and why things were happening.
>>>
>>>What happens when some program makes Shirov and Kasparov feel like I did
>>>yesterday? It's not far-fetched. I'm rated higher than about 99% of the
>>>tournament players in the country, and I'm already baffled by the computers
>>>sometimes. I've seen speed games where good GM's totally lose understanding of a
>>>position against Crafty. It's very impressive to watch.
>>>
>>>As an opponent, I've virtually given up on being competitive with the top
>>>programs on my M2 266. Crafty, CM6K, Fritz5...they all hammer me mercilessly. My
>>>record against Fritz5 on this machine is something in the neighborhood of 1 win,
>>>20 draws, and about 125 losses.
>>>
>>>Since I can't play against it full strength, I have focused more on letting it
>>>be my teacher. Watching it play out positions where I had difficulty in choosing
>>>a move. I'm guessing that within a few years, it's going to be this way for the
>>>average GM. It is already this way for the average master.
>>>
>>>But a fun thing is that even the GM's won't understand the games that are going
>>>on. Generally, I can understand (in real time) games by human players up to 400
>>>or 500 points better than I am. I'm guessing that's fairly average. I'd like to
>>>find out if weaker GM's watchin Crafty have similar difficulties understanding
>>>what's going on.
>>
>>Chris,
>>
>>This actually hits on a theory of mine regarding chess which is very pertinent
>>to computer chess as well.
>>
>>The theory basically goes that players "accidently" fall into losing positions
>>that suddenly appear within the event horizon.
>>
>>A player (human or computer) makes a move, which from his chess knowledge and
>>current depth lookup, appears to be a strong move. However, a move or two later,
>>this move suddenly no longer appears to be good and effectively loses the game.
>>This is seen easier with computers since their score will go from +1 to -4 in a
>>matter of 3 moves (6 ply) or less.
>>
>>This means that even the superGMs have this problem. However, since they are
>>using a better and more sophisticated set of heuristics than the class B player,
>>the problem is not as obvious and usually not as fatal.
>>
>>Once a 3000 rated program hits the market, the superGMs will be able to use it
>>to discover their (granted minimally) flawed heuristics which can lead in
>>certain position to inferior positions and come up with new heuristics that
>>enable them to improve themselves.
>>
>>One of the main problems that the superGMs have is that there are very few
>>people (zero) in the world with the skill set and the desire to learn from. A
>>class B player can (given the financial resources and the desire) always find a
>>master to learn from. The superGM cannot and must effectively learn on his own
>>(not always the best learning technique). A 3000 rated program (especially if it
>>had a sophisticated analysis mode) could be that "teacher" that the superGMs
>>could use to improve.
>>
>>:)
>>
>>KarinsDad
>>
>>>
>>>Very interesting topic!
>>>
>>>Chris Dorr
>
>I agree that the superGMs would learn from a program rated 3000, if such a thing
>were possible. However, isn't it also true that a chess programmer can only
>encode the knowledge which is available at the current time? Chess programmers
>can't program knowledge that doesn't exist. So, the program that is rated 3000
>achieves its rating because it is simply consistently good all the time, which
>is not something that human beings are good at. Accordingly, a program might
>actually achieve a tremendous rating (perhaps not 3000, however), without
>advancing chess knowledge much at all. The improvement for the superGM would be
>in identifying their own deficiencies, not in developing novel heuristics for
>better game play. In other words, the 3000 computer might just amount to the
>same old tired chess principles projected out over 25 or so ply. That will get
>you a great rating, but it makes the machine only worthly of imitation, not a
>real teacher.
>
>IMHO,
>
>Mika


You are bordering on witchcraft now.  The "a computer can only know as much
as its program" is a cloudy statement.  It won't know anything that the
programmer doesn't know, but when you combine knowledge and search, the sum is
often greater than the individual parts by a huge amount.  IE I don't do
anything about coordinating squares in king and pawn endings, yet crafty can
solve many of these instantly with a deep search (fine #70 was long thought
to be unsolvable by a computer, ever.  It is now solved in a fraction of a
second by most).

So it is more than possible that knowledge plus a deep search produce something
totally new and unexpected...

We see this every day on ICC.



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.