Author: Bob Durrett
Date: 12:11:57 02/22/04
Go up one level in this thread
On February 22, 2004 at 13:59:23, Christophe Theron wrote: >On February 22, 2004 at 12:14:55, Bob Durrett wrote: > >> >>With absolute correct play on both sides, is the outcome draw or win? That is >>the question we cannot answer now, with absolute certainty. >> >>Here is an experiment which could be done. The participants could be >>all-silicon or a mix of silocon and human. >> >>Write a computer program [!!!] to do take-backs automatically. >> >>In the beginning, start with a game which looks reasonably free of obvious >>mistakes. This game might be taken from human-human, human-silicon, or >>silicon-silicon praxis. >> >>Then just let the computers run, using very high search depths, a very strong >>engine or engines, run on an extremely fast computer, and give it a lot of time. >> >>After each game is completed, have the machine(s) programmed to go into >>post-mortem analysis mode and let it/them find a candidate improvement. Then >>make the indicated "take-back" and let it run again. >> >>This whole thing should be fully automated. [I'm not sure how to automate the >>humans.] >> >>Ideally, many such computers should be running in parallel but with different >>initial games. >> >>There is a requirement to achieve stability so that the thing doesn't go into >>some sort of loop revisiting the same lines over and over again. >> >>Give it a few years. >> >>Repeat this experiment a few thousand times [preferably run in parallel] to gain >>confidence in the findings. >> >>Conceptually easy, except for the stability part. Might cost a few $$$$. >> >>Bob D. > > > >Conceptually, an alpha-beta search is doing several thousands takebacks per >second. Your experiment looks like a very long alpha-beta search with a very >selective algorithm. > >It would give the same result: a very deep search full of holes, proving >nothing. > > > > Christophe It would "prove nothing" in the sense of deductive logic but would provide additional data to improve estimates and confidence in a statistical sense. What we are talking about is hypothesis testing. Statistics is well suited to that purpose. The alpha-beta algorithm is a brainless process. The process discussed offers plenty of opportunity for human inputs. Not brainless at all. I do not see how you can be sure the product would resemble swiss cheese. It would be difficult to prove that, too. Bob D.
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.