Newsgroups: comp.ai.alife,comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!pipex!uunet!hearst.acc.Virginia.EDU!murdoch!uvacs.cs.Virginia.EDU!mjd4c
From: mjd4c@uvacs.cs.Virginia.EDU (Michael J. Daniel)
Subject: Re: Genetic Game
Message-ID: <CzE1u0.666@murdoch.acc.Virginia.EDU>
Sender: usenet@murdoch.acc.Virginia.EDU
Organization: University of Virginia Computer Science Department
References: <christian.02rw@darkin.demon.co.uk> <Cz7y6p.EoK@murdoch.acc.Virginia.EDU> <3aavs4$8hp@bones.intellicorp.com>
Date: Thu, 17 Nov 1994 01:37:12 GMT
Lines: 33
Xref: glinda.oz.cs.cmu.edu comp.ai.alife:1317 comp.ai.genetic:4314

In article <3aavs4$8hp@bones.intellicorp.com>, treitel@bones (Richard Treitel) writes:
> 
> In article <Cz7y6p.EoK@murdoch.acc.Virginia.EDU>, mjd4c@uvacs.cs.Virginia.EDU (Michael J. Daniel) writes:
> |> I think the second child to play this game will kill the hardest guys
> |> first, and easily beat the game.
> 
> Tweak the fitness function to include some measure of how close the
> invader gets to the bottom of the screen (or wherever it is going)
> and how quickly it gets there.
> 
> Alternatively, challenge the human players to think more
> strategically, by *inverting* the fitness function, so that the first
> invaders they kill are the ones most likely to re-appear.


My whole point is these things are not going to work the way we
think they will.  The niches that probabilities settle into
under slightly different conditions are really hard to 
predict.  The execution of a genetic algorithm is an
exercise in dynamic equilibrium.

You really need to build and test it.


Michael


PS: Reality has a way of bitting our logic, chewing it up, and
    spitting it out.

PPS: Or its better to first do the experiment and then
     pretent there was a logically explanation for it
     all the time.
