Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The game of chess can never ever be solved.

Author: Sune Fischer

Date: 07:25:38 11/03/02

Go up one level in this thread


On November 03, 2002 at 10:05:25, Omid David wrote:

>On November 03, 2002 at 09:30:59, Sune Fischer wrote:
>
>>On November 03, 2002 at 09:16:39, Omid David wrote:
>>
>>>On November 03, 2002 at 09:00:59, Uri Blass wrote:
>>>
>>>>On November 03, 2002 at 08:41:28, Omid David wrote:
>>>>
>>>>>On November 03, 2002 at 08:27:42, Joachim Rang wrote:
>>>>>
>>>>>>On November 03, 2002 at 07:37:26, Uri Blass wrote:
>>>>>>
>>>>>>>I believe that chess can be practically solved.
>>>>>>>
>>>>>>>I believe that you do not need to prove the result in order to get a draw in
>>>>>>>every game.
>>>>>>>
>>>>>>>I do not expect it to happen in the near future but I believe that in 2050 every
>>>>>>>comp-comp game between top programs in chess is going to be finished in a draw.
>>>>>>>
>>>>>>>Uri
>>>>>>
>>>>>>well that are concrete bets. Maybe you're right, but I hope not.
>>>>>
>>>>>Percentage of draws between top grandmaters has always been on an increasing
>>>>>course. Nowadays 2 out of 3 games are ended in a draw. It is natural then, to
>>>>>predict that in 50 or 100 years something like 4 out of 5 games will end in a
>>>>>draw, and it could very well happen that in a little over a century almost all
>>>>>the games between top grandmasters end in a draw. But that will not mean that
>>>>>the game is solved, since the draw is the result of strength and knowledge of
>>>>>the two players, not because they *know* what to play to reach a draw.
>>>>>
>>>>>You can call a game "solved", if everyone can learn what to do in a short time,
>>>>>and will then, play the optimal moves forever (like tic-tac-toe). For computers,
>>>>>"solved" will mean that they have a database or heuristic to determine the
>>>>>optimal move at every position. For example, Shaeffer and his research group at
>>>>>the University of Alberta are close to "solving" the game of checkers, in form
>>>>>of having a database of win/lose/draw for every possible position.
>>>>>
>>>>>And according to this definition, the game of chess can NEVER EVER be solved.
>>>>
>>>>If programs always play the best move thanks to search and evaluation then
>>>>The result is the same as the result that they do it thanks to database.
>>>>
>>>
>>>Correct, but the problem is that a program can never play the best move without
>>>such a database!
>>
>>Sure it can, it can find the solution at runtime by search.
>>
>
>Yes, but here we are talking about a problem that cannot be solved by runtime
>search.

Never say never, it cannot be solved by search _now_, but who can possibly
imagine what kind of computational power we have in 5000 years?

>With a branching factor of 4, in order to reach the depth of 40 plies
>alone, you have to search about 10^24 positions. If you have a processor with
>the speed of 100 trillion (100 million million) nodes per second (10^14 NPS), it
>will take 10^10 seconds, or more than 300 years...!

Well, 300 years is a long time, but still infinitely far from "never" :)

-S.


>>Why would it need to be stored in a file on disk for it to be solved?
>>My pocket calculator does not have table of all multiplications and additions
>>hardcoded in ROM, it simply has an algorithm to answer the question at runtime
>>:)
>>
>>-S.
>>
>>>>I am also not sure that the game can never be solved by some database.
>>>>There can be a rule for classes of positions and not for a single position so it
>>>>is possible to have database that may give a move for every position and the
>>>>size of the database may be smaller than the number of the possible positions.
>>>>
>>>
>>>But still the database will be extremely large. Let's say you come up with a
>>>database of _only_ 10^30 needed positions. Where will you store it?!
>>>
>>>
>>>>Uri



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.