Author: Pete R.
Date: 09:28:06 08/17/99
Go up one level in this thread
On August 16, 1999 at 12:59:10, Ed Schröder wrote: >>Posted by Will Singleton on August 16, 1999 at 02:58:46: >> >>>GM's are by far superior when the topic is understanding the game of chess >>>and that will remain for a very long time and maybe even after 100 years. >>> >> >>Ed, >> >>It's great to see these games on ICC; I really like your initiative. I must >>disagree with your statement above, however. 100 years? Did you perhaps mean >>10 years? > >No typo :-) > >>10 years ago, you were using perhaps a 68020 at 25 mhz. You are now >>running at 600 mhz, effectively about 30x faster (considering cache etc). In 10 years we >>might have 20 ghz machines, which means that a 3 minute think today will >>take 6 seconds then. This is not even considering advances in chess algorithms, >>which will certainly occur. > >This is all very true. > >>So, I must take issue with your concept of "chess understanding," since >>that can only be measured by wins and losses. If a computer beats a GM, by >>definition it has better understood the game. > >I disagree. > >If a computer beats a GM the computer is simply stronger. I consider stronger as >something else than understanding chess. > >>In 10 years, it's clear that no human will be able to contend with commercial >>programs running on off-the-shelf hardware. Since you have a propensity for >>gambling, would you like to make a wager on that? > >I also believe that in 10-15 years a PC is able to beat the strongest human on >earth. > >Doing so this doesn't automatic mean the computer does understand chess >better than the human. > >This is what I mean with my 100 year statement: Look at any GM-COMP game >and you will see that the human ALWAYS gets the initiative. The computer always >has to defend. Why? Because the GM knows how to play chess. The GM >understands chess, the computer can not match the feeling for (long-term) >strategy, the feeling for certain positional long-term sacrifices etc. etc. > >What usually happens is that GM's throw away the gained advantage by >overlooking smart, deep and sometimes very surprising defences of the >computer. > >IMO the current struggle in GM-COMP can also be redefined as: > >"Strong understanding of chess + a not optimal search algorithm (many leaks)" > > versus > >"Average understanding of chess + a perfect search algorithm (hardly any leak)" > >IMO a search-depth of 11-13 plies plus decent human-alike or anti-human chess >knowledge is sufficient to beat GM's to 2600. > >Maybe 14-16 plies is enough to beat GM's to 2700 and maybe 17-20 plies is >enough to catch Kasparov. Who knows... > >But in the end the human knows how to play chess. A computer will never >find moves like Ra1 (Fischer) or Rxd4 (Kasparov) You mean the Topalov game? Fritz found that one in a few hours, and I did an overnight run with Crafty 16.5 or so (whatever was out at the time) and that found it too. :) I understand what you're saying of course, but who is to say that no one can program a computer to *plan* in chess. To look at the opponent's moves and try to deduce his plan, and then see which plan can be implemented? It's tough but it could be done. It may be harder to do than regular improvements using the current methods, but I think that's the main difference in how humans and computers handle chess. A plan makes it easy to be highly selective in your move choices and what you analyze, and if this could be implemented in software even to a moderate degree it would make the programs much stronger. If it were done successfully it would be increasingly hard to say who has a greater "understanding" then.
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.