Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Position Evaluation vs Selective Search

Author: Bob Durrett

Date: 14:26:26 11/20/00

Go up one level in this thread


On November 20, 2000 at 14:30:04, Daniel Kang wrote:

>On November 20, 2000 at 13:13:21, Bo Persson wrote:
>
>>On November 20, 2000 at 12:14:57, Bob Durrett wrote:
>>
>>>On November 20, 2000 at 11:55:10, Bo Persson wrote:
>>>
>>>>On November 20, 2000 at 11:06:46, Bob Durrett wrote:
>>>>
>>>>>Would it be possible to evaluate a position so well that only one next move
>>>>>would need to be considered?  In that case, "selective search" would not involve
>>>>>any selection at all [i.e. nothing to chose between].  The "selection" would be
>>>>>done during the position evaluation.  In the limiting case, only ONE line would
>>>>>need to be evaluated, except in the cases when two or more moves were found
>>>>>[during the position evaluation] to be of equal value.
>>>>
>>>>If it was possible, you would have solved chess!  :-)
>>>>
>>>>Just let the program run from the start position, and see if it comes up with
>>>>e2-e4 or d2-d4 as the optimum opning. Repeat for 40 plies and you end up in a
>>>>check mate for white (or a draw??).
>>>
>>>You are assuming that one or the other of 1.e4 or 1.d4 is the better move.
>>>Maybe not a valid assumption!
>>
>>Of course we don't know for sure, as the evaluator isn't finished yet. I would
>>be *very* surprised if it was 1.a3 instead.
>>
>>>Incidentally, the position evaluation software may have to declare two or more
>>>moves "equal" if they appear to be reasonably close.  How "reasonably" would be
>>>defined in this case would be up to the programmer.
>>>>
>>
>>My point was supposed to be that the current programs already *do* this
>>evaluation, by doing a lot of calculations for each potential move from the
>>position. The calculations are called "search".
>
>I'd also like to add that this characteristic (looking at all moves rather than
>blindingly following their intuition to limit the search to a couple of good
>moves) is probably a strength of computer programs rather than a weakness. The
>reason why humans rely on pattern matching rather than searching for move
>selection is because their searching capacity is severely limited, not because
>searching is an inferior method. In some sense, a human brain is a massively
>parallel super-duper fault-tolerant computer specialized for pattern
>recognition, yet to be matched in power by silicon-based number crunchers. I
>just don't think it's a sound idea to imitate the human process, when current
>methods have already yielded programs that outplay almost all humans, with
>machines far less sophisticated than brains of even the dumbest humans.
>
>Dan.

A tree can get extremely large very fast if you look at all possible moves from
a position.  Even if you were to look only at two moves for each position, think
about the size of 2**n where n is the number of half-moves.  If that is OK with
you, then think of what happens to 3**n, 4**n, etc.  As the number of moves gets
larger, the tree explodes.  Hence people's efforts to prune and selectively
search.  The problem is real!



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.