🚀 go-pugleaf

draper@ais.org (Patrick Draper) writes:
>So, do you know how the enemy algorithms work? I've always imagined that
>they were some sort of weighting system, with high risk/threat grid areas
>getting the lions share of troop allocation.

From considerable empirical evidence, I hypothesize (but it's ONLY a
guess) that WGB's Interstel Empire's computer opponents are
reductionistic "peephole" automata that react to their immediate
surroundings according to hard-coded response tables that are written to
maximize local "combat results" (in terms of city-turns of production
(CTP) killed versus CTP risked), with only the most rudimentary levels
of higher-order behavior, such as "look-ahead", communication,
co-operation, co-ordination, etc.  For instance, an enemy army adjacent
to a city ALWAYS attacks the city.  (OK, I guess; so do I.)  Enemy
fighters always attack any adjacent enemy units; if multiple enemy units
are adjacent, they seem to attack transports, armies, and other ships,
in that order.  Destroyers always attack any adjacent ships, favoring
transports; transports always run from anything.  Basically, it appears
that there's no real "computer opponent", but rather a set of tables,
one for each type of unit, that lists for every adjacent enemy unit the
appropriate response.  When no enemy units are adjacent, the computer's
units mill about with only the most general guidelines; armies tend to
move to the shore, or in the direction of your latest beachhead;
fighters swarm in the general direction of the front line, going on
search-and-destroy missions that usually end up costing the computer
2-for-1 in CTP losses; destroyers wander around, looking for your
transports; the computer's carriers have absolutely no idea what's going
on.  The foremost reason Interstel Empire lost all attraction for me was
that I could not for the life of me detect any hint of a single
"computer opponent" mastermind that was marshalling all of its forces in
a coherent struggle against me; instead, it was like thousands of
virtually independent battles against individual ants.  Even when I gave
the computer opponent 2:1 production and 3:2 combat advantages, my
advantage in cooperation between my forces proved to be insurmountable,
and I still steamrolled the computer flat from the word go (very slowly,
granted, but absolutely relentlessly).

Still dreaming about an open-architecture Empire that allows us to write
our own plug-and-play computer opponents ...

Eric Wang
wang@cs.uiuc.edu