Newsgroups: comp.ai.alife
Path: cantaloupe.srv.cs.cmu.edu!europa.chnt.gtegsc.com!gatech!news.sprintlink.net!simtel!harbinger.cc.monash.edu.au!lugb!latcs1!latcs1.lat.oz.au!agapow
From: agapow@latcs1.lat.oz.au (p-m agapow)
Subject: Re: What's going on?
Message-ID: <agapow.804132037@latcs1.lat.oz.au>
Organization: Calvin Coolidge Home for Dead Biologists
References: <3r6klu$jnk@seralph9.essex.ac.uk> <1995Jun9.112723.17763@zippy.dct.ac.uk> <agapow.803362346@latcs1.lat.oz.au> <3s1nkf$3eg@decaxp.harvard.edu>
Date: Mon, 26 Jun 1995 02:00:37 GMT
Lines: 81

martino@fas.harvard.edu (Carlo Martino) writes:

>>  2) You can get into a large number of arguments over what is/isn't
>>  	emergent. There are a number of ecosystems that demonstrate "emergent"
>>  	behaviour (because they do something their creators didn't expect) but
>>  	other people looking at them can say "Of course that was going to
>>  	happen - it's an inevitable consequence of how your system was
>>  	programmed ..." 

>Wait a moment....  Perhaps I am merely revealing the extent of my
>unfamiliarity with the discussion surrounding alife, but it seems you are
>moving in a direction tangential to the real significance of emergent
>behavior; the point about emergence is not that it is unanticipated, but
>rather that it is undetermined. 

A bit of both. "undetermined" is perhaps not a good word to use, since it
intersects with "unpredictable" and the results of rolling dice aren't
what we'd call emergent. Emergence is behaviour on one level that cannot
be anticipated or reduced to behaviour on a lower level. Not most of the
time we judge emergence "relative to a model", i.e. based on whether our
model of the system shows a causal chain between the lower and higher
level behaviour. And that is often open to opinion.

>I would certainly hope that the designers of any carefully constructed
>alife simulation might anticipate at least to some degree the behavior the
>system eventually exhibits. 

Me too. Be pretty dull otherwise. But how much can you anticipate the
results of an alife model (not simulation) and still maintain the results
are undetermined? i'm sure it's things like this that Peter Cariani was
thinking of when he said that arguments about non-determinism were
"unproductive".

>>  3) And then we get on to semantic and syntactic emergence (being the
>>  	advent of new interactions with the environments and new computations).
>>  	While most Al models have a lot of syntactic emergence, there's little
>>  	or no semantic emergence because that's very difficult to express.

>Again, sorry if my grasp of the lingo is somewhat limited, but how would 
>you characterize semantic, as versus syntactic, emergence?  Both terms 
>are new to me.

No worries - it's hairy stuff. This again is from Cariani (although the
terms are used slightly differently elsewhere). We can think of things
that "happen" concerning an organism as having outputs and inputs and
being symbolic and non-symbolic. You collide with something - that's a
non-symbolic input and output. You sense something - non-symbolic input
and symbolic output. You raise your arm/manible/move - symbolic in and
non-symbolic out. You think - symbolic in and out. Essentially it's cause
and effect.

This i think maybe the biggest problem facing ALife. Most AL models don't
have that non-symbolic end of the equation, they lack the pragmatic
relationship that exists between themselves and their environment. Is this
a problem? Maybe - it certainly makes things different. Physical
interactions and sensing have non-symbolic and non-deterministic
components.

Then there's emergence again. Most Al model eat syntactic emergence in
their sleep (it being the emergence of new computations, new symbol to
symbol functions). But there's little if any semantic emergence - the
advent of new relationships and new ways of sensing and effecting the
environment. It's like a bunch of pre-Cambrian bacteria being unable to
ever evolve into anything with sight or hearing because it's not in their
original specs.

The problem is that AL models have a very rigidly defined repotoire for
interacting with their physical environment. And understandably - i can't
see how you could program in the ability / chance for new interactions to
evolve. It's a bloody hard problem. Does it make any difference. Again i
have to say "maybe".

Hope this helps. Sorry for the length, but as you can see it's a fairly
complex issue.

p-m

paul-michael agapow (agapow@latcs1.oz.au), LaTrobe Uni

  "Teenage sex, nuclear winter, Junior Miss fashion & 
  the mechanics of emotional capitalism ..."
