World Seeding (was Task Parsing)

[Ling]

On Mon, 5 Jan 1998, JC Lawrence wrote:
> 
> One of the more fascinating aspects of this to me is the world seeding
> you allude to.  Ideally, at the end of the day you as a world designer
> could merely define the physical terrain and its resources, plop in a
> starter population at the locations of choice, hit the "Go!"
> button, and go away to come back later to a teeming populous world
> full of interesting constructions.

This is what I used to aspire to.  Afterwards, I would have sent in the
terraforming crew to add in artifacts, interesting tidbits and shaping
mountain ranges to fit their initials.  I still do in fact.

This is also why I occassionally leave my machine with The Settlers (Serf
City for you Americans) running on demo mode for a day or two (asides
from warming up the room).  It's also why I wanna obtain a Silicon
Graphics thing just to run Polyworld 24/7.

> The problem:  Your AI is of necessity going to be sub-optimal.  As a
> result your system must be heavily weighted toward systemic explosions
> or else it will fail far too often ("The colony, it died Jim!").
> You're going to need that extra weighting just to ensure that even a
> pretty crappy AI can at least hang on to the edges of survival in most
> worlds.

Depends on the game design.  But if it has trillions of factors and
problems to take care of (to be of interest to powermonger players), I
suppose the above will apply.  I remember one AL experiment where
optimally weighted neural nets spawned a population smaller than the ones
evolved over a few weeks because the agents ate everything the saw.

> The problem with this is that once you introduce players, they are
> going to tend to be far more adaptable and inventive than your AI's
> (expert systems to the side), with the result that they'll overrun the
> entire system.  

One reason why the systems I make up sit on stand alone machines and
stagnate.  Then again, I prefer to use the method to initialise the maps
rather than run the mud.  Or perhaps to make the world more dynamic than
it already is.  By having AIs run the world and making sure players never
has the chance to do the same, a mud could be made to be quite dynamic and
realistic.  Any bad decisions could be blamed on a crap king or somfink.

> Its rather like the old game of CRobots: its a good game as long as
> only robots are playing.  Hook one robot up to a human player with a
> joystick and overhead view of the playing field and the game becomes
> RobotSlaughter.

I think this analogy is a bit poo in that the player is not on the same
terms as the 'bots.  Try giving the player the following info instead:

X,Y position, heading, speed, turret facing, camera facing, camera
resolution, detected target (if any).

It'll be a little bit harder without that overhead map...  If what you're
saying is that the AI will have access to less info than the players, then
fairdos.

Final note, ever considered adding an extra tier to your mud?  Let's take
the PBeM approach. :*)  The PBeM players would control a nation and submit
turns in weekly without ever realising it is based in a mud.  They can
advertise for a scout to make a recce into enemy territory and a player
would take up the job, do the biz and write up the report (well, fill in
the blanks).  As usual, there is the problem of synching the two tiers and
certain design problems...

  |    Kilo Lima Lima Oscar dash Niner Four, over and out.        ---o
_O_O_  Electronic and Electrical Division, Loughborough Corp.    <+=+=+>