Copy Link
Add to Bookmark
Report
Alife Digest Number 027
ALIFE LIST: Artificial Life Research List Number 27 Friday, June 29th 1990
ARTIFICIAL LIFE RESEARCH ELECTRONIC MAILING LIST
Maintained by the Indiana University Artificial Life Research Group
Contents:
Evolving Networks - New TR
Failing the Turing Test
On the RISKS of Artificial Life
Microbial Growth Simulation
----------------------------------------------------------------------
Date: Tue, 26 Jun 90 05:26:18 PDT
From: rbelew@UCSD.EDU (Rik Belew)
Subject: Evolving Networks - New TR
EVOLVING NETWORKS:
USING THE GENETIC ALGORITHM
WITH CONNECTIONIST LEARNING
Richard K. Belew
John McInerney
Nicolaus Schraudolf
Cognitive Computer Science Research Group
Computer Science & Engr. Dept. (C-014)
Univ. California at San Diego
La Jolla, CA 92093
rik@cs.ucsd.edu
CSE Technical Report #CS90-174
June, 1990
ABSTRACT
It is appealing to consider hybrids of neural-network learning
algorithms with evolutionary search procedures, simply because Nature
has so successfully done so. In fact, computational models of
learning and evolution offer theoretical biology new tools for
addressing questions about Nature that have dogged that field since
Darwin. The concern of this paper, however, is strictly
artificial: Can hybrids of connectionist learning algorithms and
genetic algorithms produce more efficient and effective algorithms
than either technique applied in isolation? The paper begins with a
survey of recent work (by us and others) that combines Holland's
Genetic Algorithm (GA) with connectionist techniques and delineates
some of the basic design problems these hybrids share. This analysis
suggests the dangers of overly literal representations of the network
on the genome (e.g., encoding each weight explicitly). A preliminary
set of experiments that use the GA to find unusual but successful
values for BP parameters (learning rate, momentum) are also reported.
The focus of the report is a series of experiments that use the GA to
explore the space of initial weight values, from which two different
gradient techniques (conjugate gradient and back propagation) are then
allowed to optimize. We find that use of the GA provides much greater
confidence in the face of the stochastic variation that can plague
gradient techniques, and can also allow training times to be reduced
by as much as two orders of magnitude. Computational trade-offs
between BP and the GA are considered, including discussion of a
software facility that exploits the parallelism inherent in GA/BP
hybrids. This evidence leads us to conclude that the GA's GLOBAL
SAMPLING characteristics compliment connectionist LOCAL SEARCH
techniques well, leading to efficient and reliable hybrids.
--------------------------------------------------
If possible, please obtain a postscript version of this technical report
from the pub/neuroprose directory at cheops.cis.ohio-state.edu.
Here are the directions:
/*** Note: This file is not yet in place. Give us a few days, ***/
/*** say until after 4th of July weekend, before you try to get it. ***/
unix> ftp cheops.cis.ohio-state.edu # (or ftp 128.146.8.62)
Name (cheops.cis.ohio-state.edu:): anonymous
Password (cheops.cis.ohio-state.edu:anonymous): neuron
ftp> cd pub/neuroprose
ftp> type binary
ftp> get
(remote-file) evol-net.ps.Z
(local-file) foo.ps.Z
ftp> quit
unix> uncompress foo.ps.Z
unix> lpr -P(your_local_postscript_printer) foo.ps
/*** Note: This file is not yet in place. Give us a few days, ***/
/*** say until after 4th of July weekend, before you try to get it. ***/
If you do not have access to a postscript printer, copies of this
technical report can be obtained by sending requests to:
Kathleen Hutcheson
CSE Department (C-014)
Univ. Calif. -- San Diego
La Jolla, CA 92093
Ask for CSE Technical Report #CS90-174, and enclose $3.00 to cover
the cost of publication and postage.
------------------------------
Date: Tue, 26 Jun 90 13:43:49 BST
From: Robert Davidge <davidge@computing-science.aberdeen.ac.uk>
Subject: Failing the Turing Test
Failing The Turing Test
=======================
The Turing Test requires action at a distance through a given and restricted
medium. No medium exists that both Natural Life & Artificial Life could
exhibit through. Their behaviour or exhibition through the medium
of manifestation is the very indication of Life.
Is this a fundamental difference between intelligence and life? No for AL
sets out to show that life can exhibit through another medium than
earth based carbon chemistry. The Turing Test could only be applied to the
principals of AL if both Natural Life and AL exhibited through the medium
of earth based carbon chemistry or a computer terminal. This is clearly
not possible. Therefore the Turing Test as applied to AI can not be
applied to AL.
Life-as-we-know-it is totally interdependent. Although there are boundaries
of a sort, there is no isolated case of life; could the solar system exist
without the universe? or the earth without the sun? Sea and land are largely
independent at the level of life, but chemical exchange is vital. Within
these great "non-living" systems, the interdependence of life is very marked.
Is this a function of Life? That is, within a system Life acts to optimise
the inter-relation of the constituent parts or energies. What about the
boundaries of these ecosystems? They are special cases in the expansion of
ecosystems and mark a dramatic point in the history of the evolution of life
e.g. methane -> oxygen; sea -> land; earth -> space.
The Turing Test only examines the boundary case, i.e. communication across a
boundary. Life has developed largely to increase this transfer across
boundaries via more and more sophisticated sense streams and response
behaviour. Such sense streams are far in advance of any TTY medium
used to try and demonstrate intelligent behaviour.
Intelligence (response behaviours) then perhaps emerges
from increased utilisation & cross-relation of the sense streams.
Intelligence could be seen as optimisation of the relationships between
boundaries (an organism's boundaries with all other organisms) within
a self-referential (relatively closed) system. Intelligence could then be
seen as an emergent property (behaviour or pattern or method) arising
out of the Life purpose to "optimise relationships within or
between systems". Life is what we call the behaviour within the system
and intelligence what we call the behaviour of one system to another.
If this is so, then we would expect to see markedly different behaviours
within a system relatively devoid of life and a system bursting at the seams
with life. We would also have different measure of the achievement of
Life within a given system. An ecosystem with a huge biomass, but few species
would be in a less advanced state than a species rich ecosystem with less
bioproduction. The quality of the life would also need to be taken into
account. This can only be measured by the range of sense information available
to the Life and the total complexity of its response behaviours to this sense
range.
We could expect to see 3 phases during the development of any life
displaying system:
1. Exploration: The new life moves rapidly to colonise the new system
Species will be few. Intelligence will be low but "physical" prowess
and independence high. These are the pioneers.
2. Exploitation: Once the boundaries are reached the "life wave" will
reflect back on itself producing an enfolding which will appear as
turbulent or chaotic or fractal patterns. Speciation and specialisation
increase causing a great enrichment of the system with specialised
niches developing in the most unexpected of places. Intelligence
and interdependence increases and physical prowess decreases as the
complexity of life increases. These are the specialists.
3. Escape: A point occurs where the exploitation of a smaller and smaller
fractal environment can not absorb the life expansion. Certain members of the
system develop intelligence and/or physical prowess and/or group
relationships beyond that of the system which enables them to penetrate the
boundaries and move into a new and larger system. The process begins again on
a higher turn of the spiral. These are the New Wave.
This cycle or pulse is similar to the Phase idea of macro-mechanisms in
evolution (I believe put forward by Kuhn in 1976). Due to the nature of
time in evolution (which can be thought of as exponential), it is easier
to give examples from sociogenesis than cosmogenesis. An example
from sociogenesis might be the social history of North America from
the Pilgrim fathers to the Space Race. Whether examples from Palaeontology
can be found for this cycle in Biogenesis remains to be seen, but the
change from Dinosaurs (pioneers) to mammals & birds (specialists) to humans
(New Wave) might be an inherent nature of Life rather than due to a great
catastrophe.
This essay has come some way from the inadequacy of the Turing Test
to a hypothesis for the purpose of Life. The point is that intelligence
can be seen as a pattern of behaviours emerging between living systems,
which are themselves expressions of the patterns of behaviour within
a higher, encompassing living system and that Life drives onward from
system to system.
Robert Davidge
Department of Computing Science
Aberdeen University
Aberdeen AB9 2UB
Scotland
davidge@uk.ac.abdn.cs
------------------------------
Date: Thu, 28 Jun 90 14:06:49 MDT
From: cgl%pullet@LANL.GOV (Chris Langton)
Subject: on the RISKS of Artificial Life
Here is an article that appeared on the RISKS digest. I have tacked on
my response to it.
Chris Langton
------------------------------------------------------------------------
Date: Thu, 21 Jun 90 08:34:25 -0400 (EDT)
From: Nathaniel Borenstein <nsb@thumper.bellcore.com>
Subject: "Artificial Life" out of control
The latest issue of the Whole Earth Review has an article ("Perpetual
Novelty") about the growing "artificial life" movement, which works to
create computer simulations of artificial beings, with rather
far-fetched and grandiose long-term goals. I was particularly struck by
the discussion of the idea that some of these people have to release
lots of relatively dumb robots and simply let them evolve. Talking
about one researcher's goals, the article says:
He wants to flood the world (and beyond) with inexpensive, small,
ubiquitous thinking things. He's been making robots that weigh less
than 10 pounds. The six-legged walker weighs only 3.6 pounds. It's
constructed of model-car parts. In three years, he'd like to have a 1mm
(pencil tip-size) robot. He has plans to invade the moon with a fleet
of shoe-box-size robots that can be launched from throw-away rockets.
It's the ant strategy: send an army of dispensable, limited agents
coordinated on a task, and set them loose. Some will die, most will
work, something will get done. In the time it takes to argue about one
big sucker, he can have his invasion built and delivered. The motto:
"Fast, Cheap, and Out of Control."
I think that about says it all. The risks should be obvious, at least
to the people who read RISKS.
Nathaniel S. Borenstein, Member of Technical Staff, Bellcore, Morristown, NJ
----------------------------------------------------------------------------
Here is my reply, posted to RISKS:
----------------------------------------------------------------------------
On the risks inherent in the study of "Artificial Life."
In his note to RISKS 10.11, Nathaniel Borenstein comments on the
risks involved in the pursuit of "Artificial Life".
As the organizer of the two conferences on Artificial Life, I
would like to comment on the risks involved in this field
(there certainly ARE risks!), on how they are being addressed, and on
the way in which this research and those conducting it are often
mis-represented in the public media.
The quick summary is that Artificial Life researchers are NOT poised to
release hordes of self-replicating, evolving robotic-insects or net.organisms
upon an unsuspecting public. Some of the risks are indeed obvious, as
Nathaniel suggests, many are more subtle, but they are certainly NOT being
ignored or run roughshod over by those in the field. In fact, bringing
the risks to light and dealing with them is as important a part of Artificial
Life research as is the pursuit of the "purely" technical and theoretical goals.
Unfortunately, many journalists reporting on Artificial Life have leapt at the
chance to "sound the alarm" about the latter aspect of the research without mentioning that we are equally concerned with the former. I have spent hours
with reporters explaining how seriously we take this aspect of the field, only
to read a week later something to the effect that "...if these young
Frankensteins have their way.....", and I am tired of being slapped in the face with something which we have been very careful to promote: responsibility.
These self-righteous reporters are preaching to the choir, believe me.
For some journalists, this omission is not malicious; the responsibility
issues are just not as "newsworthy" as the possibilities for creating life
in computers, in test-tubes, or using Lego sets, and so get left out of the
discussion in the article. Unfortunately, this leads to the mistaken impression
in the reader that such issues are not of concern to researchers. This is simply
not true. We are VERY concerned, and a good deal of our research and technical
prowess is devoted to addressing such issues.
With all due respect to Nathaniel, there are "risks" in reporting risks perceived
from articles or reports in the public media, and I would hope that those
reading the RISKS notes would take the trouble to find out what's really
going on by contacting the people involved directly before merely repeating
verbatim some reporter's account.
There's probably more to the story....as in this case.
What is Artificial Life (AL)? As the name might lead one to expect, AL is
roughly to the study of "life" what AI is to the study of "intelligence,"
although the approach is certainly different, and I hope that AL will benefit
from the many mistakes and dead-ends encountered in AI research.
Artificial Life is the attempt to study the bewildering variety of processes
involved in the origin, evolution, and dynamics of living things by abstracting
these processes and simulating or even synthesizing them in other media - often
within computers - where they might be studied and experimented with in ways
that would be impossible or unethical to carry out with the "real" living
processes themselves.
Of course, if one takes the view that it is the process itself which is
"alive," not just the material that it informs, then one does NOT escape
the ethical or moral problems by re-creating an identical process in a
computer. Once one has adopted the theoretical and/or philosophical position
that life is a property of the organization of matter, rather than a property
of the matter that is so-organized, then the world - indeed the universe -
becomes a very different place, and once solid ground is solid no more.
The very idea of "Artificial" Life gives rise to an enormous suite
of scientific, technical, philosophical, social, moral, and *-al problems,
questions, dilemmas, and of course risks, and the latter must be addressed
and dealt with before the technology is "deployed," i.e., released into an
environment we will have to share with it.
In the preface to the proceedings of the first workshop on Artificial
Life in 1987, I close with the following:
"Such issues MUST be discussed before we go much further down the
road to creating life artificially. We are once again at a point
where our technical grasp of a problem is far ahead of our moral
understanding of the issues involved, or of the possible
consequences of mastering the technology.
It is, perhaps, not coincidental that the first workshop on
Artificial Life was held at Los Alamos, site of the mastery of
atomic fission and fusion. Both technologies have tremendous
potential for benefiting life on earth, but both also have
tremendous potential for abuse, whether intentional or accidental.
Whereas the "technology of death" was developed in secret, under
government mandate, and with little official attention to social
and moral consequences, this first workshop on the technology
of life was held in the open, by voluntary participation, and we
must see to it that it is pursued with the most careful attention
to social and moral consequences."
At the second Artificial Life workshop, held in February 1990, there was
a good deal of discussion on the risks involved. In a panel session on
computer viruses, Eugene Spafford likened the deliberate release of a computer virus to the deliberate release of a toxic biological virus into the
public drinking water, and this is an absolutely valid analogy. This
is SERIOUS business with serious risks involved. Those of us doing
serious scientific research in Artificial Life appreciate this, and
take their responsibilities in this area VERY seriously.
However, the very existence of computer viruses shows that there are many
people who do NOT understand the possible consequences of their actions,
or if they do, they do not take them seriously. Computer viruses are,
unfortunately, already out there, and it is quite clear that they are only
the tip of the iceberg of possibilities for computer or network based
life-forms.
It is important to note that computer viruses were NOT released by Artificial
Life researchers but by "hackers" (in the perjorative sense of the term),
and that their very existance is one of the reasons for studying and
understanding the general class of such phenomena. However, when one
begins to think about the extent of this "general class of phenomena"
and what it encompasses, one is lead far beyond mere computer viruses
to the space of all possible life, in any of its possible incarnations;
the space of life-as-it-could-be, which includes as a mere subset
life as it has evolved on the planet Earth.
There are obviously many more reasons for studying Artificial Life than because
of the risks posed by computer viruses and their relatives, primary among them
being the fundamental contributions that can be made to theoretical biology
by expanding the empirical data-base upon which our theories of life are built
beyond the single example we have for study. However, when we come
to the point of actually synthesizing, in whatever "hardware," processes
that do many of the same things as living organisms, such as replicating,
adapting, speciating, competeing and so forth, we have to understand that
whether we consider them "alive" or not, we have set in motion autonomous,
physical processes which we may never be able to control, or indeed even
have the "right" to control. It is not nice to mess with Mother Nature, but
it is necessary to understand Her, and it is becoming increasingly obvious
that we will have to broaden our understanding of just what we mean by the
term "Nature" or we may be "messing" with Her without realizing it.
The particular research referred to in Nathaniel's post is being carried out
by Rodney Brooks and his colleagues at MIT. Their motto - "Fast, Cheap, and out
of Control" - does NOT imply that they intend their robots to be out of control
and run amok. Rather, it refers to the fact that the behavior of their robots is
not due to an explicitly programmed central controller, but rather emerges out
of the interactions of a number of semi-autonomous "behavioral agents"
which ARE explicitly programmed. As most of the interesting phenomena in
living systems are "emergent" in just this same sense ("life" emerges from
the interactions of a great many nonliving molecules) it is an appropriate
research topic to understand the way in which complex behaviors can emerge
from the interactions of simple parts. One needs to understand the ways in
which emergent behavior can be employed towards useful ends, but we also need
to understand the ways in which emergent behavior can go bad, so as to
avoid the latter possibility. As we understand very little about the
nature of such "emergent" behavior, we must study it, but study it carefully.
We don't want to be "tickling the tail of the dragon."
In fact, much of what is discussed here in the RISKS forum is as much due
to the emergent, IMPLICIT behavior of complex hardware or software systems
as it is due to EXPLICIT programming or wiring ommissions or commissions.
Complex information processing systems can come to have a "life of their own,"
and exhibit emergent behavioral properties neither intended nor understood
by their designers. It is important that we study such emergent phenomena,
and that we understand that such emergent behaviors can have useful as well
as dangerous consequences.
Anybody involved in the field of Artificial Life - or any other scientific
discipline for that matter - should be required to read "Frankenstein" by
Mary Shelley. Read the original book, don't base your impresions on the
various movie versions you may have seen.
Written in the early 1800's, "Frankenstein" is an eloquent essay on the responsibility of the scientist for the consequences of his research. In
the book, it is clearly Dr. Frankenstein himself who is the monstor, not
his "creation," for releasing something into the world and then washing
his hands of any responsibility for it. This book was written in an
era when the world was finally emerging from the Dark Ages and all of
the horrors visited upon humanity in the name of Religion. At that
point, "Science" was being touted as the saviour of mankind, and anything
scientific was viewed as being above reproach. Shelley was one of the
first to point out that science was equally capable of visiting horrors
upon humanity, and her book certainly qualifies as one of the earliest
RISKS postings!
In summary, it is certainly true that there are risks inherent in the study
of Artificial Life, but I hope that readers of RISKS (at least) will
take the trouble to dig a little deeper than the journalistic reports
on the field and find out what the researchers themselves are up to
and how they intend to manage the risks involved. We as a group are trying
to make sure that we do not "visit horrors" on humanity, be they electronic
mechanical, or some other physical instantiation. Indeed, the early fears
about Genetic Engineering, which were entirely justified, have been
fairly effectively met. The important point is that the risks ARE there
and must be addressed honestly and openly. This should go without
saying in any scientific discipline, but unfortunately this is not
always the case.
We in the field of Artificial Life are attempting to be as conscious as we
can of the possible consequences of our research, both good and bad.
We welcome any and all input, especially criticism, from any source.
We specifically invite people who are critical of the whole enterprise
to participate in the process. Some objections will be silly, of course,
but we need to consider them all.
----------------------------------------------------------------------------
Those interested in further information on Artificial Life can obtain the
proceedings from the first workshop from Addison Wesley. They have a toll
free number for ordering: 1-800-447-2226. The title of the book is:
"Artificial Life: Proceedings of an Interdisciplinary Workshop
on the Synthesis and Simulation of Living Systems"
Christopher G. Langton, Editor.
It is Volume #6 in the series:
Santa Fe Institute Studies in the Sciences of Complexity
The ISBN #'s and prices are:
Paperback ( ~ $25) ISBN 0-201-09356-1
Hardcover ( ~ $45) ISBN 0-201-09346-4
The ISBN numbers are also AW's internal order code.
The proceedings from the second workshop should be out early next year.
There is also an online discussion list. Subscribe by sending your email
address to:
alife-request@iuvax.cs.indiana.edu
I would also be happy to see a discussion of the risks involved carried
out here in comp.risks.
**************************************************************************
Yours in livelier computing!
Chris Langton
Complex Systems Group
MS B213, Theoretical Division Phone: 505-667-9471
Los Alamos National Laboratory Email: cgl@LANL.GOV
Los Alamos, New Mexico, USA
87545
------------------------------
From: Joseph M. Saur <saur@cs.odu.edu>
Date: Fri, 29 Jun 90 08:37:46 EDT
Subject: Microbial Growth Simulation
In a recent AL Journal, you requested information on available AL
programs; well, here's one that I did as a research project last year:
MICROBIAL GROWTH SIMULATION
This is a PC-based program that attempts to reproduce the dynamics
of microbial colony growth through the use of AL techniques. Basically
the user is allowed to set up a population growth curve experiment with
two separate colonies (experimental and control), and to specify for each
the nutritional content of the media. Different types of food particles
are represented by differing colors of pixels on the screen, and the bugs
created grow and divide based on their success at ingesting the food.
Among the choices offered are:
Batch growth (food is created at the beginning only) vs.
Continuous growth (food is added each cycle of the simulation).
Complex media (amino acids, etc.; faster growth) vs.
Simple media (ammonia, potassium, phosphate, sugar; slower rate
of growth).
Glucose or lactate for the sugar.
Aerobic or anaerobic growth if glucose is used.
With simple media, the ability to vary percentages of each component.
Inhibitors (fatal or non) to allow the demonstration of the effect of,
for example, heavy metal ions and/or antibiotics.
The simulation runs on an IBM-compatible PC with 640K of RAM. It can be
set for either EGA or VGA machines (please specify). As for speed, the same
growth curve study that took me 10 hours in the Micro lab will take 3-4 hours
on a 286-based machine, and 15-20 minutes on a 386.
Data from each run is displayed on the screen during the run, again on
the screen after the run in an expanded form (which can be dumped to an
IBM-compatible graphics printer), and is stored in a .CSV file on a floppy in
the A: drive (for further use in spreadsheets, etc.).
If interested, send a blank diskette (5 1/4 or 3 1/2") to:
Joseph M. Saur
Advanced Technology, Inc.
468 Viking Drive
Virginia Beach, VA 23452
Phone: (804) 498-5582
e-mail: saur@cs.odu.edu
Thanx.
------------------------------
End of ALife Digest
********************************
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=---=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
= Artificial Life Distribution List =
= =
= All submissions for distribution to: alife@iuvax.cs.indiana.edu =
= All list subscriber additions, deletions, or administrative details to: =
= alife-request@iuvax.cs.indiana.edu =
= All software, tech reports to Alife depository through =
= anonymous ftp at iuvax.cs.indiana.edu in ~ftp/pub/alife =
= =
= List maintainers: Elisabeth Freeman, Eric Freeman, Marek Lugowski =
= Artificial Life Research Group, Indiana University =
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=---=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
End of Alife Digest
********************************
