Author: Jeroen van Dorp
Date: 07:44:04 01/16/02
Go up one level in this thread
>Unfortunately, I find your approach excessively materialistic. Even with a >massive material advantage, you can still lose. Technically *and* theoretically not. If the evaluation function is as correct as can be, you can't lose based on perfect play. That is an assessment of the engine, within it's horizon limits. We know it's flawed. But the statistical approach of win/lose chance has this same flaw, as you can always beat the odds. 1% winning chance is still a wining chance. > >To use an analogy, it's like choosing a car on the basis of its speed alone. >What an unhappy chap you'll be when the car is delivered, and you find that you >cannot legally drive it on the public roads! But the engine *has* the specs, it's up to you to do with it what you want. The specs only tell you: the car is able to drive 200 mph, nothing more. It doesn't tell you *if* you will do it, just that it's possible. Maybe the makers were (a bit) wrong, depending on their knowledge of automotive technique and all dependable variables surrounding top speed. Like chess programmers. Your solution to the board analysis will give you the specs of the engine: it can win because it has x advantage. It doesn't look at your possible performance, because that's unknown. Will I be able to use the power I have on the chess board to speed up to the maximum given by the engine? What you suggest is actually nothing more than a rating system. Applying a rating-like system to the engine evaluation is: "your chances to win are x". We know that this is the same as: your calculated rating is x, mine is x+y, and based on y difference you have a win chance of r (where 0<r<100%). It doesn't however tell you what *according to the engine* the most objective possibility of the game at any given moment against a perfect opponent is. It also can give you an indication of win/lose chances after several games, and not the actual game. Remember that I assume that the evaluation of the engine is 100% correct. I don't count in horizon limits, flaws and bugs in the engine, and well known computer play flaws. It's just a theoretical comparison between a (theoretical) perfect engine giving either statistical win chances (rating system against a normal opponent) or material based differences (win potential against an opponent that's perfect as well). J.
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.