[MUD-Dev] Geometric content generation

Adam Martin ya_hoo_com at yahoo.com
Thu Sep 20 09:27:25 New Zealand Standard Time 2001

----- Original Message -----
From: "Koster, Raph" <rkoster at verant.com>

>> A game system like that (that I use because I'm familiar with it)
>> is Achaea's PvP combat system, at least to some extent. It's got
>> a huge number of possibilities in it, though they do not, by any
>> means, all intersect with each other. As I regularly add more of
>> them, it just increases to get more complex, though these days I
>> rarely add elements that connect to a huge portion of the other
>> elements. The downside (if anyone that's a gamer can call it that
>> without feeling guilty) is that the disparity between the great
>> fighters and the newbies increases, which can make things
>> discouraging for newbies.

> Complexity? Or is each addition simple, and the aggregate
> emergently complex because of the interactions? For example, the
> movement of each piece in chess is simple, but the game has
> emergent complexity because of the amount of movement types.  

When I was doing research into Genetic Programming, and
experimenting with suitable "problems" to attack in the form of
optimisation tactics for computer games, it became clear it was very
easy to find / invent a game with exponentially increasing numbers
of possible games derived from a small number of unique possible
movements/interactions. However, the difficult bit was finding one
where there was no obvious perfect tactic, which usually was
manifest by the goal of the game being sufficiently removed from the
base actions performable that the chance of finding a robust
strategy through pure random play would tend quickly to 0.

Obviously, GP was not limited by the random-attack not working, but
it did also require that emergent tactics *could* bring benefit (in
terms of score, i.e. some measurable in-game quantity) even when
they were far from refined - this being essentially how GP
progresses in the early stages of evolution (random discovery of
tactics, which then theoretically have a much better chance of
remaining in the gene pool to be reused and later refined into
advanced tactics).

I found it interesting that both these criteria for selecting a game
for GP-work also seemed effective (I would say crucial, but I didn't
carry this part of the research any further than merely noticing the
correlation) at selecting games that were historically known to be
highly popular.

At about that time I started noticing other characteristics common
to the highly successful games (across all genres), but these were
the ones that were most important to GP.

In summary, I'd agree with Raph's suggestion that the complexity of
games like chess comes about as an emergent behaviour - as is the
complexity of a fractal derived from a simple equation - but IME
there are big nasty pitfalls when trying to design a game to do
this, and to simultaneously ensure it is actually fun. I would guess
this is similar to the problems with algorithmically generated
content, where it seems that in general people find it very hard to
create content this way which is as deep as they could create from
scratch by hand - when you are creating a game from simple rules
only, it is very hard to produce an emergent behaviour that fits
with your original vision of the gameplay (which is really an
expression of what you want the emergent behaviours to look like).

Adam M
MUD-Dev mailing list
MUD-Dev at kanga.nu

More information about the MUD-Dev mailing list