From newshub.ccs.yorku.ca!torn!utcsri!rpi!batcomputer!cornell!uw-beaver!pauld Wed Aug 12 16:52:14 EDT 1992
Article 6553 of comp.ai.philosophy:
Newsgroups: comp.ai.philosophy
Path: newshub.ccs.yorku.ca!torn!utcsri!rpi!batcomputer!cornell!uw-beaver!pauld
>From: pauld@cs.washington.edu (Paul Barton-Davis)
Subject: Re: Memory and store/retrieve.
Message-ID: <1992Aug3.220654.20920@beaver.cs.washington.edu>
Sender: news@beaver.cs.washington.edu (USENET News System)
Organization: Computer Science & Engineering, U. of Washington, Seattle
References: <1992Jul31.160209.26718@mp.cs.niu.edu> <1992Aug3.151610.21034@puma.ATL.GE.COM> <1992Aug3.200351.3632@mp.cs.niu.edu>
Date: Mon, 3 Aug 92 22:06:54 GMT

In article <1992Aug3.200351.3632@mp.cs.niu.edu> rickert@mp.cs.niu.edu (Neil Rickert) writes:
>  I guess I'm troubled by its apparent need for an overall top down
>design.  Somebody has to say "lets start evolving features so as to
>prepare a future linked list arrangement."  But I see evolution as
>primarily bottom up.  Features are enhanced as they provide benefits.
>This being said, it is possible that something could evolve in a bottom-up
>design, yet the resulting effect be interpretable as an apparent linked
>list arrangement.

I'm a little troubled by some of the appeals to evolution I've seen
appearing in this thread. Two salient points, IMHO:

	1) there still is no good consensus on the neutrality of
	   most mutations. I have been quite suprised to find 
           a dearth of simulations for this; although there are
           plenty of theoretical position papers, my searches so
           far have not shown up anyone who has actually demonsrated
           that for mutations to persist, they must be beneficial.

	   For my part, I actually wrote a small test program to
           try this out. Its appalling simple, but it confirmed my
           own intuition - mutations that affect reproductive fitness
           have to have a *very* dramatic effect in order to affect
           their own (rather then a system of which they later become
           a part) survival, in the presence of environmental 
           noise (weather, predators, disease, etc.)


	2) Sort of a corollary to the above. "Features are enhanced
           as they provide benefits". Take a look at Tom Ray's 
           work with Tierra, and note the evolution of an "unrolling
           the loop" optimization. Too bad we don't have a good
           phylogeny tree for this, but here's a nice CS-related
           evolutionary path in which there would appear to be
           no middle ground, nor benefit from anything prior to the
           emergence of the entire unrolled loop.

-- 
"Fun" is a word invented by advertising executives to sell soft drinks.


