Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: chess and AI.

Author: Bruce Moreland

Date: 14:21:34 06/27/01

Go up one level in this thread


On June 27, 2001 at 04:10:07, Pekka Karjalainen wrote:

>On June 27, 2001 at 00:33:26, derrick gatewood wrote:
>
>>ok,  I made some general statements about AI and chess and how they have been
>>closely related...  this is the response I get back.  I was wondering what you
>>guys think about this.  If you can rip its reasoning apart,  please do.  I would
>>like a little support when I do it  =)
>>
>
>  It looks like he is arguing that the approach used for chess will not work in
>other domains.  And it will not produce "real AI", whatever that is.
>
>  I think he is, broadly speaking, correct.  Whatever that game Starcraft is, it
>probably is not amenable to the same techniques that computer chess is.  There
>are a whole bunch of games that are much harder for computers than chess is.
>Nothing special about that.

Since this pertains, I will describe Starcraft.  This is not intended to be a
complete description.

A Starcraft creature has a certain amount of durability, it is able to move
along the ground or through the air at a certain rate, and it may have special
abilities such as the ability to repair structures or other creatures, the
ability to build structures, the ability to exploit resources, or the ability to
blow things up.  The game is real-time, so if you tell a creature to go over
there, you can watch him move, while you tell another guy to go somewhere else.
If your guy sees an enemy guy, he'll shoot at him if he can.

A Starcraft structure has a certain amount of durabilty, may have some mobility,
can be used to convert resources into creatures, or can be used to research more
advanced structures or creatures.

The basic idea is that have some creatures exploit resources and build
structures, in order that you can make more creatures which go blow the hell out
of the opponents.  If you lose all of your stuff you lose.

There are hundreds of games of this genre by now, I think.  I play them
sometimes, and my eldest son has been playing them competently since he was
three years old.

>  So, he is saying that Starcraft is really complex in a computational sense.
>All that means that we cannot use the same approach of AB-searching with
>enhancements to play it like chess.  Some other approach might work better,
>though.

It's easy in a computational sense.  The strategy is to make a clump of
buildings, make creatures that collect resources, make a bunch of offensive guys
and run them at the opponent.  These guys keep shooting until they destroy all
of the opponent's stuff or are themselves destroyed.  Repeat until you win or
lose.

The problem is that it's hard to find a real "best move".  It's more about
creating guys at a high rate and in a proper mix, and doing damage to your
opponent without losing more than you destroy.  The AI in most of these games is
pretty bad, but I'm not convinced that it has to be bad.

These games are much less structured than turn-based games like chess or go, so
you have to adopt a very generalized heuristic approach.

I doubt that the AI's in any of these games are really computationally expensive
the way they are implemented now.

bruce




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.