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!d6170.shv.hb.se!sa209
From: sa209@utb.shv.hb.se (Claes Andersson)
Subject: Re: Evolution Maker
Message-ID: <sa209.80@utb.shv.hb.se>
Sender: usenet@gdunix.gd.chalmers.se (USENET News System)
Nntp-Posting-Host: d6170.shv.hb.se
Organization: Department of Scocial Science
References: <1994Nov25.121045.15938@ucl.ac.uk> <1994Nov2.104653.26949@leeds.ac.uk> <39oc58$gla@news.duke.edu> <1994Nov14.140813.21400@ucl.ac.uk> <3aofme$t38@deadmin.ucsd.edu> <CzuCLK.F5x@freenet.carleton.ca>
Date: Mon, 28 Nov 1994 14:36:15 GMT
Lines: 78

In article <CzuCLK.F5x@freenet.carleton.ca> av574@FreeNet.Carleton.CA (Tim Sallans) writes:


>In a previous article, charris@cs.ucl.ac.uk (Christopher Harris) says:

>>|> >
>>|> >Absolutely! Mutation is a very small factor in the variability of a 
>>|> >population, and is often over-estimated in its power. Crossover and 
>>|> >other operators are far more useful, given a long enough genome and 
>>|> >sufficient population.
>>|> >
>>|> >Chris
>>|> 
>>|> Dear Chris,
>>|> 
>>|> It's not clear whether you are addressing mutation in biota or mutation
>>|> in evolutionary computation.  It would be helpful if you could clarify
>>|> this point for me.
>>|> 
>>|> Regards,
>>|> 
>>|> David
>>|> 
>>
>>I meant in evolutionary computation - but I guess it could apply to biota 
>>as well. That kind of thing is very hard to observe in the real world, due 
>>to the enormous times and populations, but in simulation it's a lot easier 
>>to see. Of course if you have any hard evidence to the contrary I'm willing 
>>to be proved wrong - but I don't mean to discount mutation altogether, I 
>>was making the point so that the other operators would get their
>>"share of the limelight", so to speak.
>>
>>Chris Harris
>>

>It seems to me that normal random mutation may not provide enough
>randomness in genome modification...wouldn't random transcription
>errors contribute more in a biological system?

 Random transcription errors would occur rectangle distributed during the 
entire lifetime. They are just as it sounds: random. If they should have any 
importance, one error would have to be selected for rather than another but 
the selection takes place one time each generation. They act during the 
lifetime and alter a phenotype in a random way. It would be Lamarckism since 
there is no selection pressure involved.

>And, as a corollary, could the probability of events like transcription 
>errors, crossovers, inversions, base pair duplication, etc., be coded 
>in the genome as well, as opposed to externally as a system variable?  
>That way, the genetic search would optimize the search process 
>as well as evolving to score high on the fitness function.  Would this
>accelerate the search over the course of the generations?

 Crossover takes place every time a gamete is produced and it is of course 
gentically encoded (what else?). 

 One thing has to be clear: The genes do not cooperate. They do not act in 
favor of their phenotype, they act in favor of themseleves. A gene is 
something that can reproduce and have influence over its own survival. A 
gene that act in favor for the whole genome will find itself together with 
slightly other genes next time and therefor fail. The genes coevolve to fit 
togehter and much of what they do is apparently altruistic to their 
neighbours.

 A gene that creates a beneficial trait will become more and more numerous 
in the genepool. This is what happens when a GA converges. What would the 
benefit be for the genes that made the other cooperate for finding a good 
phenotype? Selection is nothing mystic, for the genes to predict the future 
by doing unexplainable things is mystic, it is merely the fact that what 
survives, survives.

 BUT when we are talking about AL and especially optimizations, we can of 
course ignore any property of natural evolution we like to as long as we 
improve the result. Optimizations must not be lifelike, they must be good. I 
haven't thought about what you wrote thoroughly but it might very well be 
useful i optimizations.

Claes Andersson. University of Bors. Sweden.
