Author: Christophe Theron
Date: 15:14:53 12/12/00
Go up one level in this thread
On December 12, 2000 at 14:51:38, Mogens Larsen wrote:
>On December 12, 2000 at 13:30:32, Christophe Theron wrote:
>
>>This way of explaining things shows that you don't know how programs can plan.
>>It's not about changing weights of the evaluation terms. It's about looking at
>>the position like a human would do and selecting the appropriate "goals", and
>>then trying to achieve them by calculation.
>
>These goals as you call them are guided by weights of the evaluation terms and
>the knowledge of the program, so in reality you're tuning for good execution of
>a certain and limited type of positions/goals. The program will perform more or
>less sane moves until the score inflates and it has a goal. That's randomness.
Randomness???
No you really don't understand what I mean.
>The progress in computer chess (IMO) is that the waiting moves are getting
>better. By this I mean better at aiming for positions where the program excels,
>but not good enough to call it planning. Too many holes and exceptions. I guess
>that speculative scores is a way to bypass that problem. Making it do something
>instead of nothing. I still see that as forced behaviour.
>
>Sometimes the goals are correct and "the plan" works and sometimes not, but it
>contains the randomness of tuned weights and knowledge. The previously mentioned
>carrot on a stick. The main problem is that it's guided by evaluation and when
>that starts to fail or becomes static, planning goes out the window because the
>goal disappears.
I repeat that it is the opposite. Some programs first find a plan, then the
evaluation is changed in order to guide the program to the realization of this
plan.
Not the other way around.
>>What you are saying here is not that program do not plan. You are just saying
>>that in some cases they don't find the right plan, or don't have one. It happens
>>to humans too.
>
>Off course it does. But you're making the mistake of concluding on the basis of
>positions where it finds the plan due to specific tuning or pure calculation.
Of course not. I was specifically speaking about the opposite: planning first,
computation and evaluation after, serving the plan.
>In
>most cases that doesn't involve material gain or exposing the opponents king,
>which makes the program oblivious to planning. Uri's example is quite good in
>that regard.
>
>>It doesn't refute the fact that programs DO plan.
>
>There is no fact to refute. All programs can occasionally find the right moves
>in a position, which would constitute planning in your opinion. However, such
>incidents are indistinguishable from pure calculation without purpose. Either a
>program plans or it doesn't. It plans sometimes doesn't work.
>
>>Chess programs have others ways to plan. Not just bigger carrot and longer
>>stick.
>>
>>That was good in the Chess Challenger era. Programs have evolved, you know.
>
>Yes, they have. But not as much as you think IMO.
Apparently you know better than me what's going on inside top chess programs.
So I let you have the last word and quit this discussion right now.
Christophe
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.