🚀 go-pugleaf

-------------------------------------------------------------------
Date: Thu, 21 Jan 93 19:01:33 PST
From: jrupert@sfu.ca (Jason David Rupert)
Message-Id: <9301220301.AA25651@kits.sfu.ca>
To: erwin@trwacs.fp.trw.com
Subject: Re: Cognitive Networks
Newsgroups: sci.cognitive,comp.ai,sci.anthropology,sci.archaeology
References: <erwin.727662658@trwacs>
Status: R

In comp.ai you write:

>> You seem to discount Dawkins' memes.  I understand that they may
>> not be a particularly effective way to model sweeping cultural
>> change but I think they are extremely relevant in modeling very
>> small, critical changes as well as, perhaps, resistance to certain
>> changes.  The real value, of course, is in modeling specific
>> self-propogating, self replicating, mutating, evolving "ideas, 
>> ideological elements, or idea complexes" - to wit, memes,
>> in a culture, a community, a socio-politico-economic unit.

>The problem with a meme is that it evolves by Lamarckian rather than
>Darwinian evolution. The two have distinctly different dynamics. In
>Darwinian evolution, variation is -->random<--, followed by selection. In
>a Lamarckian system, variation is not random. This leads to distinctly
>different dynamics. (I've demonstrated this in simulation experiments.) In
>particular, the optimal rate of evolution in a Darwinian system is very
>low if not zero. In a Lamarckian system, chaos is not unexpected, leading
>to a need for a non-zero rate of evolution. The evolutionary theory that
>has been developed for Darwinian evolution needs to be recast for
>Lamarckian evolution. In particular, we need to understand innovation in a
>cultural context. I -->think<-- an understanding of cognitive networks is
>needed here.


Something I read a while back by Doug Lenat reminded me of what you are
talking about.  It was in a book titled "Machine Learning", I can't recall
the article's title.  What I liked about Lenat's approach is that he
managed to cast cognitive/ideological *novelty* entirely in terms of
random variation.  This seems to be a necessary constraint unless you
are going to allow organisms/systems to have some sort of prescience
about what ideas/concepts are going to work out.  How Lenat does this
is by allowing there to be a nested hierarchy of evolutionary processes 
where the variation at one level is (partially) controlled by the processes
at the level below.  So long as the bottom level of variation is "random"
then you've avoided the problem of prescience.  New concepts thus can
be the result of true randomness, or they can be directed by "heuristics"
that have evolved at a "lower" (or is it higher?) level.  In the later
case you can argue that nothing really "novel" has been created, just 
the application of some earlier evolved heuristic.

Donald Campbell (who calls his work "Evolutionary Epistemology") also
has similar ideas.

Anyway, I feel that these are a plausible alternative to making a
Lamarckian move... or maybe this is a Lamarckian move just spelled 
out in a way to make it (sound?) compatible with Darwinism.

Comments?


(Also, I can get the proper reference for the Lenat article if you
are interested).


Jason Rupert
Simon Fraser University

--I suspect something of this sort is going on in societal innovation.
--The nested hierarchy of evolutionary events allows a single-point
--mutation to have major effects, thus getting around the problem
--with mutations under normal conditions (they're almost always
--deleterious.)
--I'll have to go off for a weekend to think about this one.
--Harry Erwin
-- 
Harry Erwin
Internet: erwin@trwacs.fp.trw.com