Xref: utzoo comp.ai:7717 comp.ai.philosophy:8
Path: utzoo!attcan!uunet!cs.utexas.edu!helios!tamuts!n025fc
From: n025fc@tamuts.tamu.edu (Kevin Weller)
Newsgroups: comp.ai,comp.ai.philosophy
Subject: Re: Emergent properties (was: What AI is exactly)
Summary: still trying to move this thread to comp.ai.philosophy exclusively
Message-ID: <8629@helios.TAMU.EDU>
Date: 28 Sep 90 22:14:16 GMT
References: <15132@venera.isi.edu> <84118@tut.cis.ohio-state.edu> <1990Sep27.185805.21493@ncsuvx.ncsu.edu>
Sender: usenet@helios.TAMU.EDU
Followup-To: comp.ai.philosophy
Organization: Texas A&M University
Lines: 221

In article <1990Sep27.185805.21493@ncsuvx.ncsu.edu> fostel@eos.ncsu.edu (Gary Fostel) writes:
>
>  Some of the recent posts have criticised the term "emergent property"
>  as a euphamism for "we don't understand" and some have defended the
>  term by examples of the application of the term and some have tried
>  to justify the term as a valid one more abstractly.
>
>  If I assemble a device from wheels, pedals, metal tubes and such,
>  and it happens to become the most efficient transportation device
>  around, is that an emergent property of the parts?  I doubt that
>  defenders of the term would like it to be so.  But why not?  Probably
>  because the transportation property was a goal of the design process
>  that controlled the assembly.

Your example is indeed in harmony with my definition of emergence
since, for instance, a metal tube can hardly be said to have an
"efficient transportation device" property in miniature!  The property
is of the whole system, not of the parts.  This is not to say that the
properties of the tube at its own description-level (or at lower
levels still) don't play a part; that would be the opposite one-sided
view.  Whether the transportation property was a design goal or not is
simply beside the point, just as the ultimate goal of AI researchers
has no destructive effect on the intelligence in any systems they
design/build (other factors can have destructive effects, of course).
Emergent properties are simply properties, independent of the
designer's (or designers') plans.  Your device is probably not the
BEST example of emergence because the "transportation device"
description-level is not that much higher (more complexity-abstract)
than the component level(s), but it does serve as a simpler analogue
of more complex systems.

>                                 Now suppose a Venusian engineer viewed
>  this as a process of putting wheels, metals tubes, pedals, and an
>  engineer, and a few tools, in a room.  These "parts" may well have
>  an emergent property, from the perspective of the Venusian, since
>  they had no expectation of a transportation function arising from
>  the collection of parts.  (Yes the engineer is a part in this view.)
>  I identify this as an emergent property because I believe it would
>  satisfy most of the ad hoc definitions I have heard.  

You are correct again regarding the emergence in this example, only
from another level of description (that of the Venusian).

>  I do not expect most supporters of the "emergent property" term to
>  like this use of the term.  They will not like it (I am guessing of
>  course) becuase they will feel that they can identify the source
>  of the property which has emerged ... but of course the Venusian,
>  having utmost contempt for the large water based carbon compound in
>  with the metal parts, will not be able to identify the source and
>  the "emergence" is viewed from that creatures perspective.

Your guess (concerning my "not liking" this usage) is wrong.  Your
illustration is perfectly compatible with my understanding of
emergence, although as I said before, it is not the most useful
application of the term.  However, it does bring up one aspect of
emergent properties that I failed to address in my original posting
(sorry).

There are usually many different ways to describe an organized system,
these different ways corresponding to different levels of abstraction,
called description-levels.  For example, the collective behavior of an
ant colony is considerably more complex and purposive than that of the
individual ants, so at one description-level, we have a bunch of ants
which are each behaving on the relatively simple level of programmed
automata, while at the colonial level of description the whole is
acting for a larger purpose, and often doing a fantastic job of it.
There may be other examples more effective for you in getting the
point across, but to keep this posting from getting TOO long, I'll put
references at the end of the article.  The key points are: physical
systems in the universe vary in their complexity and organization, the
simpler systems requiring much fewer (and lower) description-levels
than the more ordered ones; and emergence is always RELATIVE to
whatever description-level is being considered, so that the more
removed a property is from those of the system's constituent parts,
the more useful the term is in describing the synergistic effect of
their ordered combination.  THERE IS NO CONFLICT between the hardware
and the software description-levels.  They are complementary.  I
suspect that if we ever manage to build an artificial intelligence,
the construction will involve some combination of knowledge from
several levels of description (physics, chemistry, neurophysiology,
cognitive science, etc.).

>  A second example: if you put large pine forrests, rabbits and foxes
>  together in northern Canada, you will get a 10 year cycle of boom
>  and bust in the populations of rabbits, foxes and young pine trees.
>  (Rabbits LOVE to eat pine needles, far more than carrots.)  Is this
>  pattern an emergent property?  From the perspective of a naive and
>  innumerate individual, the answer is certainly yes.  The cycle is
>  there, it was not predicable (by them) and it is not easy to 
>  identify the source in myopic analysis of rabbits, foxes, or pines.
>  From the perspective of an ecologist or someone versed in simple
>  dynamical systems theory, it is not an emergent property.  It can
>  be predicted, modeled, and well explained, based on properties of
>  the constituent elements, e.g. kilocalories needed, supplied, 
>  gestation period, etc.
>
>  Again, I would expect advocates of the "emergent property" term to
>  be somewhat bothered by this situation, but I believe it is because
>  they will tend to automatically associate with the technically
>  astute view of the dynamics.  Two hundred years ago, ecologists
>  knew it happenned but did not understand at all why.
>
>  But, it was not called "an emergent property of rabbits, foxes and
>  pines".  It was simply an unexplained experimental observation.  That
>  does not have nearly the same gloss, but it is more accurate.  By
>  saying that X is an emergent property of {A,B,C...} and by providing
>  some sort of definition for the term "emergent" an unsubstantiated
>  conclusion has been reached.  A few people seem to make this part
>  explicit in the use of the term, directly or indirectly saying the
>  explanation WILL NOT come from reductionist methods, not simply that
>  it HAS NOT come from that source.  Certainly, in the case of neural
>  systems (real or synthetic) it is not known that a suitable means
>  of reductionist explanation will not be found.  Just that it has not
>  been found.

On the contrary, I am not bothered by it at all.  Your example is
simply not one of emergence.  I don't claim that ALL such phenomena
are emergent.

What I am trying to say is that SOME phenomena CAN NOT be explained
SOLELY on the basis of component properties.  If this weren't so, we
would be capable of explaining every phenomenon using microphysics
alone if given a total knowledge of all physical laws (or perhaps the
much-sought-after Grand Unified Theory).  But you say, of course,
since all the other sciences derive from the most fundamental (i.e.,
REDUCED) law(s) of nature!  Then, would you be willing to say that we
should abandon all studies in chemistry, biology, meteorology, and so
on in favor of physics alone?  It would explain everything, wouldn't
it?  ONLY AT ITS OWN LEVEL OF DESCRIPTION!  Think about it.

Individual memories are not perfectly localized in the brain, but
rather, they are stored as "tendencies" of the overall network to
approximate patterns functioning for memories (a distributed memory
system).  My having a memory is represented on the hardware
description-level by a pattern of electro-chemical processes in my
brain.  You can scrutinize the individual operating neurons to your
heart's content and never see the full significance of the event.
It's the *pattern* that counts on this description-level, and you must
step back and look at the whole system to see the operation.

Note that I am not advocating any kind of dualism here.  Different
levels of abstraction can all be referring to the same stuff, so if
anything, abstraction-level is more of a monistic concept.  It is
philosophically classified as a form of materialism, but, of course,
any good physicist knows the difficulty in defining "matter" and
distinguishing materialism from "spiritism" of a sort.

>  (So-called "chaotic systems" are an interesting counter point, since
>  there is some analytic evididence that there are classes of systems
>  for which it is not possible to predict specific behavior, based on
>  ANY measurement of the system.  But in these cases, it IS often
>  possible to characterize the sorts of behavior that the system is
>  capable of.  I take it that emergent properties deal in the currency
>  of behavior characterizations, not specific predictions, so the
>  behavior of a chaotic system is not an "emergent property".)

I'm not clear on what you mean by "currency of behavior
characterizations," but I can say that chaotic systems are composed of
components that obey physical laws just like components of ordered
systems.  However, the only true high-level properties of such a
system to count would probably be its very chaos and its range of
possible behaviors.

>  People studying intelligent systems seem to operate as if they already
>  know what the suitable "atoms" of the systems are.  Since they
>  are unable to explain the observations based on properties deduced
>  from these atoms, they reach for terms such as "emergent properties"
>  rather than doing good science and looking to reformulate the basic
>  hypothosis in new ways.  Hiding behind a pseudo-science of "emergent
>  properties" will probably delay the real struggle: to find more
>  suitable analytic tacts and more suitable atoms to form the 
>  foundation of a "proper" scientific explanation.
>
>  I doubt advocates of "emergent properties" will like that either.

No, we don't really know enough about the "atoms" of intelligent
systems to build one yet.  As I said, low-level and high-level
properties are complementary and inextricably bound together in highly
organized systems, and neglecting either kind will probably result in
failure.

Your narrow definition of "good science" is slowly falling into
disfavor simply because it is no longer considered adequate for fully
explaining certain natural phenomena such as quantum events, weather,
life, intelligence, etc.  Science should still be largely
reductionistic, but there are some modern scientific problems that
need a more open-minded approach if we ever hope to make any headway
with them (scientifically).  The experimental method and
quantification are not rejected.  The essence of science is untouched.
Consult the references below if you want to see more scientific
support for my position.

>----gary----

-- Kevin

PS: Thank you for responding with such a well-thought-out article.  It
forces me to clarify my position not only for you, but for myself as
well.  One thing: I'm cross-posting this message to comp.ai.philosophy
since ours the kind of discussion that really belongs there.  If you
intend to post a new followup article, would you be so kind as to post
it in comp.ai.philosophy only?  I'm sure that everyone else in comp.ai
would appreciate that.  Thanks!


Further reading:

Bohm, David.  _Wholeness and the Implicate Order_.  Routledge & Keegan
	Paul, 1980.
	David Bohm is a physicist by profession.  This work really
drives home the holistic nature of quantum mechanics and related
disciplines.

Hofstadter, Douglas R.  _Godel, Escher, Bach_.  Basic, 1979.
	On the pitfalls of singleminded reductionism.

Peat, F. David.  _Artificial Intelligence: How Machines Think_.  Baen,
	1985.
	The final chapter discusses the explications and implications
coming out of the modern quest for a specific definition of
intelligence.