Newsgroups: comp.ai.alife
Path: cantaloupe.srv.cs.cmu.edu!nntp.club.cc.cmu.edu!godot.cc.duq.edu!news.duke.edu!MathWorks.Com!europa.eng.gtefsd.com!howland.reston.ans.net!agate!ihnp4.ucsd.edu!usc!crash!snodgras
From: snodgras@crash.cts.com (John Snodgrass)
Subject: Re: The Meaning of Life
Organization: CTS Network Services (CTSNET), San Diego, CA
Date: Sat, 1 Oct 1994 16:27:21 GMT
Message-ID: <snodgras.781028841@crash.cts.com>
References: <34qm2c$ei2@scratchy.reed.edu> <snodgras.779997564@crash.cts.com> <35nvmg$9rj@masala.cc.uh.edu> <snodgras.780521883@crash.cts.com> <36ci34$cjs@masala.cc.uh.edu>
Sender: news@crash.cts.com (news subsystem)
Nntp-Posting-Host: crash.cts.com
Lines: 90

In <36ci34$cjs@masala.cc.uh.edu> clifton@rodin.cs.uh.edu ( Clifton B. Davis ) writes:

>In article <snodgras.780521883@crash.cts.com>,
>John Snodgrass <snodgras@crash.cts.com> wrote:

[del]

>  Well, I am not through failing yet, have bairly started.  Your statement
>of the purpose of computer science is clear and the fact that you think
>something is basicly wrong with the very idea of AI and artificial life
>is also clear.  What is not clear is just what you think is wrong with the
>basic idea or what reasons (other than lack of total success to the present)
>you have for feeling that way.  

      We were talking earlier about the importance of philosophy to program
design, and to what directions one takes as profitable in computer science.
Let's look at the deepest possible level, then, of philosophy pertaining
to computer science. Is nature fundamentally mechanistic? Is the manner in
which theoretical physics depicts nature, as "externally" controlled (by 
physical law) the only scientific way to model nature? Or if that is how
we define "science" (as a mechanistic model) then is it the only profitable
way to model nature? If your answer to this is yes, then it is natural for
you to assume that lifeforms are merely complex machines, that life and
intelligence are emergent from complex systems controlled by physical
law, and therefore that computers have the potential to be truly "alive"
or "intelligent" or even more alive and intelligent than we are (since you
are equating intelligence and life with some property emergent from 
complexity).

      But that is not my own personal assumption. I think the mechanistic
model is one of two possible models. The other is is that nature is 
fundamentally self-organizing. This means that nature can be viewed as 
made up of self-organizing systems, functioning at all size levels, 
and combining into larger and larger systems.  It means that processes 
like evolution, which are anomalous in the mechanistic view, become 
fundamental. It means that origin of life becomes a non-problem. 

      The two views, self-organization (or a systems view) and mechanism, 
are not contradictory, but complimentary -- each is powerful in 
understanding a different facet of nature -- provided we take a more
detached view of modeling -- a view in which we are aware that
we are engaged in purposive modeling, rather than confused about the
distinction between modeling and "knowing" or a model and a "truth". Nature
is best modeled from both perspectives, i.e. using models of both types 
brings us more power and ability to predict nature, and that
is what we seek (here is a yet deeper assumption, of course). A perfect 
synthesis of the two obviously is what we want -- to understand why both 
are needed.

      Strong believers in mechanism do not accept this, and here are motives
for being a strong mechanist: 1) you do mechanistic modeling, and want to
believe this contains all the answers 2) you want to believe in strong AI
or a-life for some other reason. At any rate, it is clear that a machine
does not fit into the category of self-organizing. Someone suggesting a
machine can self-organize, but this is a contradiction in terms. A machine
is something that is designed and deterministic. It has no self-interest
to form a basis for self-organization. It cannot struggle to survive, it
can only be programmed for self-maintenance. IOWs, a machine is by definition 
an example of mechanism, just as a living organism is by definition an
example of self-organization. Mechanistic science describes machines well,
organisms poorly. Self-organizing systems describes organisms well and
machines poorly. Until the synthesis between these two views is achieved,
one is hardly going to accept the mechanist's assertion that machine will
be self-organizing (or conversely that organisms are machines). This
is the view behind AI and a-life as I see it. Mechanists attempting to create
true intelligence or life in machines is like trying to create color 
graphics on a monochrome monitor: you're skipping a step. Show me the
synthesis between the two views, and I'll start to take you seriously. People
like Minsky are still unaware, IMO, that there even are two views. Certainly 
they have plenty of support from the dominant culture of mechanistic 
modelers, but this hardly makes them believable IMO. 

>I am somewhat taken about by your catagorization
>of neural networks as useless.  I see them being used in a number of ways.

      They are used, but let's face it, was the vision their originators
of a cheap, quick and dirty solution to problems better solved other 
ways? Neural nets have fundamental problems you can't eradicate.
A million or a billion trials of a neural net may make it safe to use, 
that's all.

[del]

      Sorry about the above length. Thanks for your patience in reading
through it. Also, the provider I use has had problems, and I regret I did not
get to read many of the posts on this thread. Feel free to e-mail me if you
are interested (pro or con) in the benefits of taking a step back on the
modeling process -- or in the AI vs. augmentation debate.

      JES
