Newsgroups: comp.ai.alife
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!swrinde!pipex!sunic!news.uni-c.dk!imada!news
From: breese@imada.ou.dk (Bjorn Reese)
Subject: Re: Genetic Game
Message-ID: <1994Nov22.181501.10042@imada.ou.dk>
Sender: news@imada.ou.dk (USENET News System)
Nntp-Posting-Host: rubens
Reply-To: breese@imada.ou.dk
Organization: Dept. of Math. & Computer Science, Odense University, Denmark
References: <3areie$phu@cville-srv.wam.umd.edu>
Date: Tue, 22 Nov 1994 18:15:01 GMT
Lines: 39

[I look away for a moment and suddenly the newsgroups are flooded with
 topics that I find incredible interesting. Because I don't have too
 much time I'll just list my thoughts roughly as they came to me.]

I've been considering similar task as has been discussed here. I've even
considered using GAs for Tetris as a supplement to the usual way, ie.
increasing the speed, to increase the level of difficulty. I've never
attempted this in practice, as I was being too concerned with the subtle
act of balance that is required to make it successful; on one side it may
turn out to converge too slowly, and on the other side it may converge
prematurely.

You may want to consider coding visual and audio effects into the genes.
Things like colours, shapes, and movement patterns may have an unconscious
effect on the human player.

A way to make the enemies more realistic is to use probabilities instead of
binary genes - assign a probability to a given gene rather than a binary
value. This would also make them more immune to noise (mistakes and
misinterpretations by the human player). In this case you might be better
off using Evolutionary Programming rather than GA.

Usually you start with a random genepool. This may not be a very good idea
here; a "lucky" initiation _may_ cause one or more very good enemies to
emerge in the beginning of the game. This is undesirable as these types
of games usually starts with weak enemies, which gets tougher and tougher
as the game progresses. Instead you could breed some below-average enemies
which you could choose from at the beginning. Care must be taken to avoid
premature convergence though.

A fact that has always bothered me is that the majority of (arcade-) games
promote mindless shooting and violence. I've been wanting for a long time
to design a game that promotes social behaviour, especially cooperation;
a game where your progress is slow if you opt for the mindless shooting,
but fast if you choose to cooperate with others. I could well imagine this
progress being based on the Prisoner's Dilemma.

--
Bjorn Reese
