Author: Pete R.
Date: 15:01:45 05/14/00
Go up one level in this thread
On May 14, 2000 at 13:26:15, Bruce Moreland wrote: >On May 14, 2000 at 10:07:58, Karpov wrote: > >> >>It was not just a lose by comp. The computer played such a stupid game that all >>those who regularly claim that computers are super GMs should not post for a >>week :-) > >It's true that it was terrible. It's also true that any program could play a >game that looks this bad, and if someone wants to dispute that they can be shot >down in short order. > >I don't think that computers are super-GM's, but this evidence doesn't >necessarily shoot down the theory by itself. > >If I may be forgiven a baseball analogy, consider a baseball player who swings >as hard as he can at every pitch. If he misses the ball he falls on his butt >and looks like a fool. Most players don't fall on their butt and look like a >fool that often, but this player does, because he has a different approach. >Whether or not he is going to the hall of fame or is getting a bus ticket back >home is dependent upon how often he hits the ball, because, of course, when he >hits the ball it goes 600 feet. > >Computers aren't people. Their wins won't look like human wins, and their >losses won't look like human losses. Their game has strong points and it has >weak points, and they don't match up with human strengths and weaknesses. They >are capable of making a human look stupid, and they are capable of looking >stupid against a human, we should all know that by now. It's the ratio of these >occurrences that determines relative strength. > >I think that a confident and well-educated strong human player can do unto the >computer more often that it does it unto him. Van Wely certainly qualifies. > >bruce I agree, and forgive my ignorance, but isn't this largely a matter of evaluation tuning? It's impossible to see tactically the consequences of these kingside pawn advances until it's too late, so this has to be compensated by positional evaluation just as a human does. Any human player can look at the late positions and conclude that black has no counterplay as a direct result of white's annexation of kingside space. I'm not saying this would be easy to program, but in theory a perfect evaluation function would have the computer *appearing* to play according to planning, and thwarting the planning of its opponent. No? And isn't pawn structure a major part of eval functions? Such things are beyond tactical evaluation, but modern programs obviously still have a hard time seeing that white can create such a favorable structure within a few moves. Once the tactical realizations sink in it's far too late. Pawn structure patterns are easy for us humans to evaluate, but obviously this is still a computer eval problem.
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.