[MUD-Dev] Procedural content generation

[Daniel.Harman at barclayscapital.com]

From: Brian Hook [mailto:bwh at wksoftware.com]

>   "You don't explore procedurally generated universes, you reveal
>   them.  And that's a different experience than exploration."

> This is the "duh" thing to do, but in my first pass I really
> wanted to push the procedural generation as far as possible.  I
> think I've done that, but now I've been convinced that you can't
> "explore" a randomly generated universe -- you're just revealing
> pieces of it.  I think part of the attributes that makes
> exploration appealing is that you are putting together pieces of a
> larger system; with randomly generated content, the larger system
> is, well, random, and things just aren't tied together very
> convincingly.

I think you are making assumptions about the quality of procedurally
generated content, or perhaps your implementation is making
assumptions. If you were to work from first principles, whereby the
first thing you were to generate was a motivation, a raison d'etre,
then there would be no less overriding order behind a creation than
if you let a human steer it.

Now arguably by creating these on an algorthimic level you are in
fact manually planting the seeds, so perhaps some aspects of your
argument still hold. I prefer to think of it in terms of information
theory; in the search space of your randomly generated content, you
have every possible human guided construct. Its just a case of
narrowing the search space, either by direct human interaction, or
higher level algorithms.

Now if you were to ask which I thought was easier to
implement/generate cohesive content, I'd lean towards human
intervention. I just don't believe its necessarily better.

This all remind me of the iterated function system image compression
debacle. The jist of the story is that Barnsley created IFS
compression for images and claimed to be able to compress huge
images into just a few bytes of affine transforms. The problem was
that doing this algorithmically was very hard due to the size of the
search space. It was then found out that some of his initial
demonstrations had been generated by what is known as the 'student
locked in a room algorithm' - he literraly got some grad students to
sit in a room and fiddle with the transforms manually to get a
decent result. It was easier to use humans to refine the search
space than do it algorithmically (the contentious point being he had
lead people to believe it was computer generated).

I'm led to believe that he does now have more viable algorithms that
can produce reasonable results without chained up students, but I've
not heard a lot about them. If you used to read the much missed Byte
magazine, you might remember when this whole debacle occured.

Dan
_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev