Newsgroups: comp.ai.alife
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!pipex!sunic!news.chalmers.se!news.gu.se!gd-news!d6250.shv.hb.se!sa209
From: sa209@utb.shv.hb.se (Claes Andersson)
Subject: Re: Evolvable Fitness Formula
Message-ID: <sa209.88@utb.shv.hb.se>
Sender: usenet@gdunix.gd.chalmers.se (USENET News System)
Nntp-Posting-Host: d6250.shv.hb.se
Organization: Department of Scocial Science
References: <3bt1am$q25@srvr1.engin.umich.edu> <3btij6$jph@saturn.haverford.edu>
Date: Thu, 8 Dec 1994 17:38:20 GMT
Lines: 49

>You are missing out something in your model, which I think Claes
>touches on;

>In evolution, we talk about survival of the fittest determining the
>population of system.  However, would you care to define "fittest" in this
>instance?  Well, it actually defines pretty well in retrospect as "those
>that survived".  Given the nature of most systems, it is impossible to
>predict what "the fittest" is _before_ the survival.  But you could say
>that fitness is the ability to survive - which depends on others fitness,
>and the fitness of others depends on you - how can you tell if you are
>less fit than those around you?  You dont survive.  So fitness isnt
>defined until after someone has not survived - or shown to be "unfit". 
>Whatever you define it as, it ends up being a tautology (not a paradox -
>that would be if they survived by not surviving, which is how some people
>try to explain death).  Things survive because they are fit, and they must
>be fit, because they survive. 

>This does not mean that it is not useful, but "survival of the fittest"
>as a central tenet of biology makes certain assumptions that havent been
>resolved - I'm hoping that the mathematics that come out of complexity
>science will do something about this, but I'm not losing sleep over it.
>I have enough things to lose sleep over already.

I agree, the fitness is of course impossible to calculate from the genome. 
Well, it's practically impossible anyway because it would take that one had 
almost all possible knowledge. This is important: Genetic algorithms have 
their way with building-blocks etc. etc. which has with the genome to do. 
One could say that they bypass the lifecycle. If one, by some sort of 
ontogeny form a phenotype of some sort and test it's capabilities and select 
reproducers from the fitness value that every specimen turned out to have, 
the phenotype (which of course is what we seek in an optimisation) will be 
more emphasised. Common sence says that it ought to be better...
   Perhaps if the number of genes is small enough, the genotype and the 
phenotype will be almost the same, or if it is very simple, it is possible 
to map the fitness-landscape by testing each possible genotype.
  The number of genes acting on a specific genotype will be very hard to 
determine when using an evolvable fitness formula. (The zebra's leg and the 
lions tooth again.) And what do we have then? A fitness landscape with an 
unknown, perhaps even not discrete, number of dimensions. 
  The author of the post I'm following up, I'm sorry that I forgot the name 
of the one who wrote it and can't look it up since I'm using a plain DOS 
news-reader.., mentions that it is impossible to know what the fitness is 
until it is clear if it has raised offspring or not. But there can be more 
than fitness and unfitness. The more it either spreads it's own genes, help 
its relatives and children to spread their genes etc. etc., the fitter it 
was. (was, not is)


Claes Andersson. University of Bors. Sweden
