Newsgroups: comp.ai.alife
From: Chris@smithg.demon.co.uk (Chris Gordon-Smith)
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!udel!gatech!swrinde!pipex!peernews.demon.co.uk!smithg.demon.co.uk!Chris
Subject: Re: a theoretical biology for alife...
References: <A.J.Hirst-280395145129@uu-igor-mac.open.ac.uk> <3l9erl$632@gap.cco.caltech.edu> <A.J.Hirst-070495170717@uu-igor-mac.open.ac.uk> <797345616snz@smithg.demon.co.uk> <3m9qte$2la@scapa.cs.ualberta.ca>
Reply-To: Chris@smithg.demon.co.uk
X-Newsreader: Demon Internet Simple News v1.29
Lines: 79
X-Posting-Host: smithg.demon.co.uk
Date: Mon, 10 Apr 1995 21:35:05 +0000
Message-ID: <797549705snz@smithg.demon.co.uk>
Sender: usenet@demon.co.uk

In article <3m9qte$2la@scapa.cs.ualberta.ca>
           vanbelle@cs.ualberta.ca "Terry Van Belle" writes:

> [...]
> 
> >As I understand it, Dawkins' point is that the real battle in evolution is 
> >between genes and their alleles at the same chromosome loci, whereas fitness 
> >has several different meanings, none of which can be easily applied to this 
> >view of evolution (eg because they tend to focus on survival of individual 
> >organisms or of species).
> 
> >Can anyone throw any light on why fitness is used so extensively in EC / GA 
> >even though some biologists are questioning its usefulness?
> 
> If you're talking about the fact that EC/GA fitnesses tend to be
> well defined closed-form equations, then I see that other postings
> have already addressed the issue.

Thanks for this posting and the other responses posted.
> 
> If you're referring to the chapter "An Agony in Five Fits" from TEP,
> then I received a different impression from reading it.  It seems to
> me that Dawkins isn't discarding the traditional notions of fitness as
> much as noting that they are awkward approximations to the "true" notion
> of genetic fitness, which can be defined as the long-term proportion
> of individuals which contain that gene, as opposed to its alleles.
> 
Yes, I am referring to "An Agony in Five Fits". In his conclusion to that 
chapter, Dawkins says:-

        "Before Hamilton's revolution, our world was peopled by individual 
        organism working single mindedly to keep themselves alive and to have 
        children. In those days it was natural to measure success in this 
        undertaking at the level of the individual organism. Hamilton changed 
        all that but unfortunately, instead of following his ideas through to 
        their logical conclusion and sweeping the individual organism from its 
        pedestal as notional agent of maximisation, he exerted his genius in
        devising a means of rescuing the individual. [...] The result - 
        inclusive fitness - was technically correct, but complicated and easy 
        to misunderstand. I shall avoid mentioning fitness again in this 
        book..."

I don't think Dawkins uses the term 'genetic fitness' in the Extended 
Phenotype, but the definition you give looks to me to be broadly compatible 
with the rest of his argument. In my original question I might have asked why 
EC / GA systems use individual fitness rather than genetic fitness (if I had 
heard of genetic fitness when I asked it!). 

> Dawkins' rowing analogy from _The Selfish Gene_ comes in handy here.
> [...]
> 
> This all looks very similar to John Holland's schema theorem.  In this
> theorem, a schema is a template which can match many bitstrings in
> genetic algorithms, because some of its positions can match either 0 or
> 1.  Each schema contains a fitness, which is the average fitness of all
> individuals which contain it.  The schema theorem basically says that
> the rate of increase of any schema (~ Dawkins' genetic fitness) is
> proportional to its fitness divided by the average fitness of all
> individuals in the population.  There's also a correctional factor in
> there to account for a schema being destroyed by crossover or
> mutation.
>

Could genetic/schema fitness be used in some way to determine the way in which 
a GA should evolve, or is it a value that can only sensibly be calculated for a 
particular gene/schema after the evolution has happened?. It seems to me that 
using genetic/schema fitness to drive a GA system would be quite difficult. 
Would the fitness of every gene/schema have to be calculated? This could be 
hugely time consuming, bearing in mind that we are using 'gene' and 'schema' to 
mean any combination of (reasonably close) chromosome loci.

> [...]
> 

-- 
Chris Gordon-Smith
London
UK
Email: chris@smithg.demon.co.uk
