[Note: I'm sorry to say that it is no longer possible to provide copies
of these tapes, internally (at Apple) or externally, as I have none of the
time, the physical resources, or the support personnel to make it happen.
- larryy 4/16/96]
To: Interested parties
From: Larry Yaeger
Date: 7 Oct. 1987
Re: Notes on the Artificial Life Conference at Los Alamos, NM,
21-25 Sept. 1987
These notes are intended primarily as an accompaniment to a set of edited
videotapes of the conference available through the Vivarium office in LA
(xxx-xxx-xxxx). Copies should currently be available in both Valley Green
3 and Mariani (I may be reached at either the Vivarium office number or
at xxx-xxx-xxxx in VG3 to find out who has these tapes). The tapes consist
of 12 hours of material extracted from approximately 29 hours of raw conference
footage. The original video was shot in 8mm on a Sony CCD-V5 AutoHandyCam
using (too thin) 120-length tapes (15 of them). These were edited onto 3/4"
U-matic tapes. The currently available edited tape sets have been dubbed
from these 3/4" tapes to the VHS format (6 T-120's at the SP speed).
Following are brief descriptions of the contents of these tapes, plus an
even briefer description of the items left off of these tapes.
A brief caveat: the quality of these tapes is adequate for documentation
of the event and even for the study of the technical content of the talks,
however they are not remotely related to "broadcast quality".
Most of the presenters' slides and transparencies should be readable, but
there are certainly some that are not (they weren't readable in person,
either, for the most part, though a few marginal cases are readable in the
original 8mm and not in the final dubs - I believe John Holland's transparencies
fall into this category, I'm afraid). And there are tape glitches, drop-outs,
off-camera voices, bad edits, etc. Sigh. The fundamentally interesting nature
of the source material lends a value to these tapes above and beyond their
video quality, however.
The criteria for and fact of inclusion/exclusion of materials for these
tapes was admittedly personal and subjective. Basically, presentations were
judged for inclusion based upon the essentially orthogonal (though frequently
correlated) measures of speaker presentational skills and technical merit
(or my perception thereof). Where technical merit greatly outshone presentational
skills I have tried to edit the material to the essential thesis of the
presenter. In some cases, I have edited or completely excluded materials
that may have been of legitimate technical merit due to some redundancy
in the material, either with respect to other presenters at this conference
or to other, more widely attended, conferences (such as Peter Oppenheimer
and others' graftal plants work). Inclusion in the tapes does not imply
a personal endorsement of the ideas, just as exclusion from the tapes does
not imply their condemnation (for the most part). I have probably erred
to the side of retaining too much of the material, rather than too little,
as I found so much of it to be so very interesting. The brief descriptions
below (also personal and subjective) may permit you to pick and choose the
talks you wish to hear if you don't feel like committing 12 hours to ingesting
these presentations. Oh yeah, you can also skip the first 6 minutes of the
first tape - the ambient sound only, sloppy hand-held scenery is there primarily
for me to help recall the environment surrounding the event.
First of all, I guess a few comments on the nature and purpose of the Artifical
Life (AL) conference are in order. Chris Langton, of the Robert J. Oppenheimer
Center for Non-Linear Studies at Los Alamos National Laboratories, organized
the event to bring together researchers in a variety of disciplines, all
of whom are working on some aspect of the study of the nature of "life"
(artificial or natural). I believe Chris had a number of purposes in mind
for the conference (probably including some political ones given the human
genome mapping project going on at Los Alamos, and the stressing of the
appropriateness of a national laboratory as the site for management of significant
cross-disciplinary research such as AL). Top of the heap, though, was, I
think, a desire to provide the framework for extracting some fundamental
"laws of nature" governing the emergence of life-like behavior
from arbitrary systems (chemical - organic or not, computational, whatever).
Additionally, I think there was a genuine altruistic desire to cross-fertilize
amongst the various disciplines represented at the conference.
The participants in the conference showed their understandable biases towards
their current fields of endeavor... Origin of Life studies centered around
enzyme catalysts, RNA strands, and clays... computer models of ecological
systems with various methods of modeling the indigenous organisms' behaviors
and environmental interactions, with and without evolutionary selection
operating on these behaviors... cellular automata... neural networks...
genetic algorithms... nanotechnology... ethics... and other esoterica. The
behavior modeling, ecological models, neural networks, and genetic algorithms
topics relate most directly to the Apple Vivarium work, though almost every
presentation yielded up some useful insight or metaphor for some aspect
of the Vivarium work. All in all it was certainly one of the most interesting
conferences I've ever attended.
One of the more fundamental, and thus more general understandings to develop
out of the conference was the relation between what we call life and a sustained
negative entropy (positive information) condition [Miller, Pol]. As usual,
of course, this is nothing new... Schrodinger's "What Is Life?"
put forward this proposition some years ago (I am currently looking for
a copy of this - anybody know of a source off hand?). This may have implications
for defining a better set of neural network training rules, or for rigorously
defining the presence of "life". The applicability of classical
energy maximization principles to the modeling of biological phenomena was
also a powerful observation [Kauffman, Goel]. And formal analysis of the
statistical properties and implications of recombination (otherwise known
as sexual reproduction) was truly enlightening, especially as contrasted
to simple mutation [Holland].
And so now to the video and the speakers represented therein (in the order
in which they appear on tape - which is almost entirely chronologically
consistent with their original presentations). A few of these have sufficiently
self-explanatory titles that few comments are needed, though by and large,
the relation between the titles and the contents ranges from ephemeral to
inexplicable.
Pete Miller, LANL Director?, 'Welcome' (7 min)
Chris Langton, LANL, 'Definition & Overview' (53 min)
Conference organizer. Chris spends a lot of time on his cellular automata
work as well.
Stuart Kauffman, U. of Penn., "Is the Logic of Life Self-Organized?"
(58 min)
Postulates closed-loop catalytic cycles as a logical precursor to 'life
as
we know it'. Applies chemistry, statistical mechanics, and graph theory
to carry out a formal analysis of the likelihood of the development of
self-catalyzing sets of amino-acid monomer/polymer reactions. Basically
concludes that such closed monomer/polymer catalytic loops should be
common, hence long protein chains should be common, hence life pretty
much had to be! Heavy Origin of Life (OL) orientation.
A. Graham Cairns-Smith, U. of Glasgow, "Materials for Artificial Organisms"
(50 min)
Demonstrates the similarities between biological functions and the
imperfect, self-replicating behaviors of some clays. Suggests that these
clays would have been an appropriate vehicle for imposing organization
and compartmentalization on early pre-biotic materials. OL related.
Otto Rossler, U. of Tubingen, W. Germany, "Explicit Observers in an
Artificial World" (47 min)
An application of classical physical mechanics to chaotic systems.
Seems to predict something akin to the Heisenberg Uncertainty Principle
from first principles with classical mechanics and some phase-space
trickery. I'd be interested in hearing what some stronger mathematicians
than I have to say about this work.
Steen Rasmussen, Tech. U. of Denmark, "A Quantitative Theory of the
Origin of Life" [Excerpt] (17 min)
Offers one method of analyzing the statistical probability of the origin
of
life. Concludes a grad student could do it in a year or two with an olympic
sized swimming pool.
Louis Maggiulli & Ernest Tihanyi, "A Braitenberg Vehicles Simulation
System" [Excerpt] (3 min)
A software implementation of about levels 1 and 2 of Braitenberg's
Vehicles (if you haven't read this wonderful MIT Press, Bradford series
book by Valentino Braitenberg, I highly recommend it!). Unfortunately,
this is the level at which most people who attempt this sort of thing end
up stopping. Still, perhaps someone will use these efforts as a spring-
board for the development of higher-order vehicles someday. Basically,
Braitenberg says that analysis of something as complicated as the brain
is just too hard, so let's try synthesizing something instead, and see what
we get... and then in a series of gedanken experiments he designs vehicles
1 through 14, adding slight additional complications at each stage (which
complications are all fairly soundly based on biophysical principles -
though this is not revealed until a 'biological notes' appendix). The end
result is a design for a vehicle that might very well be expected to
exhibit "intelligent" behavior. "Seductive" is a word
Alan Kay has aptly
used to describe Vehicles.
John Wharton & Bruce Koball, "A Test Vehicle for Braitenberg Control
Algorithms" [Excerpt] (5 min)
A really interesting hybrid hardware/software, programmable Braitenberg
vehicle. This is a real, 3-D self-controlled 'car'. It has a fairly flexible
control system, and is programmable via MacBrain (a fairly hot neural
net simulation software package for the Mac). Certainly one of the
classier implementations of Braitenberg's ideas.
William Bricken & Eric Gullichsen, "A Self-Organizing Distinction
Network" [Excerpt] (2 min)
Based on Spencer Brown's Boundary Mathematics (as discussed in Laws of
Form), they have implemented an interesting sort of graphic theorem
prover (though they don't call it that).
David Jefferson, UCLA, "RAM: A System for Simulation of Complex Biological
Systems" [Excerpt] (25 min)
A Lisp-based ecological system model. They actually presented several
example applications, but I have included just a "foxes, rabbits, and
cabbages" sample application discussion. Each animal's behavior is
a few
pages of Lisp. The system supports natural selection (evolution) of the
programmer-assigned gene alleles. This is simultaneously one of the
systems greatest strengths and its greatest weaknesses. The ability to
include evolutionary changes in the indigenous population is an
interesting and valuable computational research aid. Unfortunately,
however, as with all of these types of systems, the complete complement
of available genes and their manners of expression must be fully assigned
in advance by the programmer-as-god. Hence the system is incapable of
evolving any previously unspecified behaviors, much less unspecified
organisms. Still, this is interesting work. And no one at the conference
presented, or was even able to make suggestions for a truly "open-ended"
approach that could conceivably overcome this inherent appriori-
specification limitation, though the need for such an approach was voiced
repeatedly.
Hyman Hartman, Berkeley, "Cellular Automata, Reaction-Diffusion Systems,
and the Origin of Life" [Excerpt] (21 min)
This talk didn't really offer any particular answers to anything. And
mostly the presentation consisted of watching cellular automata do their
thing on a special accellerator board in an IBM PC. However, a
central question was raised regarding our inability to analyze or express
the extremely complicated behaviors that can and do arise from repeated
application of extremely simple rules to even simpler initial conditions.
Using an extremely simple cellular automata system as a very appropriate
foil, Hartman makes the point that the "information" present in
the
highly complex emergent behaviors of such systems must be intrinsically
present in the completely deterministic rules and initial conditions that
precipitated them. There is a question here as to what is being
conserved? And where is the mathematics (the Calculus of Chaos) that
will allow us to analyze such emergent properties as straight-forwardly
as the way the Calculus allows us to determine the area under a curve?
Narendra Goel, SUNY, "Microcomputer Modeling of Biological Systems"
[Excerpt] (24 min)
Goel uses a variant of cellular automata to model the components of a
bacteriophage. He is then able to show that based purely on mechanical,
"free energy" principles his components are able to assemble themselves
into the bacteriophage form, and exhibit a bacteriophage's most prominent
behavior of punching a hole in a "cell membrane" and inserting
its own
"DNA" into the cell's body. The models that he uses are gross
over-
simplifications of the molecular structures they are meant to mimic,
however, the fact that he is able to coax such behaviors out of his model,
using only energy extremitization techniques is certainly suggestive that
complex biological phenomena can be explained through classical
mechanics. Of course this has been accepted in principle for some time
with respect to chemical reactions and the function of neurotransmitter
receptors and immune system receptors.
Mitchell Resnick, MIT, "kid's views of the brain" excerpt only
from "Lego, Logo, and (Artificial) Life" [Excerpt] (2 min)
Just a delightful collection of children's renditions of their versions
of
what goes on in the brain. I figure most Apple folks know enough about
Lego/Logo, or can find out, so did not include the rest of the talk.
John Pearson, SRI-David Sarnoff Research Center, "Competitive/Cooperative
Behavior of Neuronal Groups in Brain Function" (61 min)
Probably my pick for favorite talk, though that may be largely due to its
relevence to my current area of work. Pearson is able to computationally
demonstrate self-organization of somatosensory maps in a simulated
cortex with simulated input from the fingers of a hand. He compares the
self-organizing behavior and resultant maps with those of the known
structure of a monkey's brain. The self-organization of the maps, their
plasticity, and the dependence of their relative scale on the amount of
sensory input all compare well with known behaviors of biological brains.
James G. Miller, UCLA & U. of the World, "A General Theory of Living
Systems and its Applications" (76 min)
This talk is actually a bit rambling (note its length!) and almost entirely
avoids getting to those applications mentioned in the title. However,
Miller is a pretty good speaker, has quite a few interesting anecdotes,
and
makes an important point regarding the nature of "life" as being
negentropic. Unfortunately, he spent so much time on his personal
anecdotes and an impassioned, but unnecessarily extended plea (he was
"preaching to the choir") for more generalist and systems-oriented
thinking in scientific research and education, that he had to flip rapidly
through his slides showing the meat and potatoes (theory and
applications) of his premise in a few minutes right at the end of his talk.
John C. Higgins, BYU, "An Ethical Calculus for Artificial Life"
(19 min)
Higgins delves into Turing tests for artificial intelligence and Benthem's
calculus of the law (a +1/-1, pleasure/pain encapsulation of decision
making by the summation of relative desirability of all potential results
of a given action). From these he decides that it is necessary and possible
to develop an algorithm for the detection/provability of "life"
in an
artificial system. Well... I liked his talk anyway. Seems to me though,
that at least Turing didn't feel that such an algorithm was necessary (who
knows whether he thought it was possible)... as he relied exclusively on
the completely subjective response of human interrogators for his
determination of life-hood.
Richard Dawkins, U. of Oxford, England, "Biomorph Evolution" /
"The Evolution of Evolvability" (53 min)
Watch out Carl Sagan. Dawkins is a terrific speaker and also a pretty
mean thinker. He is author of The Blind Watchmaker and is a dyed-in-the-
wool Neo-Darwinist. He demonstrates his Biomorph system, which
permits various forms of highly interactive evolution amongst 16
different gene types, on a Macintosh. His software is supposed to
become available with the second printing of Watchmaker. As he himself
points out, however, he has pretty much exhausted the approach he has
taken to date, and, as has everyone, he has hit this brick wall of not
knowing how to implement an open-ended system that is not wholey
specified by the programmer-as-god at the outset when he/she defines
the available genes and the genotype to phenotype mappings.
John Holland, U. of Mich., "Genetic Algorithms" (63 min)
Probably my second favorite talk. Some top notch formal analysis of the
statistical behavior of natural selection, especially contrasting pure
mutation with recombination (aka sex). Interestingly, mutation is
essentially relegated to the role of protection against dead-ends
resulting from the more dominant recombinatorial selection process,
at
least for some statistical distributions of desirability within the
genome space. I would be very interested in seeing the same comparison
between mutation and recombination for a more well-behaved desirability
distribution. I suspect this was Ted Kaehler's favorite talk.
K. Eric Drexler, AI Lab-Stanford, "Nanotechnology and Biological Systems"
(45 min)
Well, guru Drexler was up to his same schtick again (still?). There
is something about his manner of presentation that makes me
cringe a little bit (perhaps a spill-over from my reaction to his
admittedly popular press version of all this - Engines of Creation),
however, the underlying ideas he is presenting are dangerously close
to unassailable. Basically he has done some fairly clear and far-sighted
thinking about the implications of our development of the ability to
manipulate and build with molecular building blocks. He points out that
there are a variety of paths already being explored that can (and he
believes will) lead to this capability - biotechnology/genetic-
engineering, tunneling electron microscopy, and others (there is even the
new light-bottle atomic containment and so-called light-tweezer
technology that he doesn't mention). Well, assuming that the techniques
will come to hand, he then carries out some basic engineering style
estimates of the capabilities of such molecular machinery. The
projections are truly astounding, and some of his suggested applications
of the technology would indeed fundamentally alter our relationship to
physical reality. Fascinating, albeit somewhat hyped-up stuff!
Hans Moravec, CMU, "Human Culture - A Genetic Takeover Underway"
(44 min)
After a distinctly nervous, but quite humorous (though the audience
failed to appreciate such gems as symbiotic flint arrowheads and an
allusion to Oolon Caluphad's Some More of God's Greatest Mistakes)
beginning discussion of cultural evolution and some robotic experiments
at CMU, he moves into a very intriguing analysis of when human level
intelligence may be expected to emerge in computers. Bottom line is
20 to 50 years depending on whether you want to do it in a supercomputer
or a Mac VII+e/*.
Doyne Farmer, LANL, 'Summary of Workshop' (23 min)
A surprisingly accurate and eloquent wrap-up of the ideas presented at
the conference.
Chris Langton, LANL, 'Thank You and Good-bye' (3 min)
Peter Broadwell, "Plasm - A Fish Sample" (18 min)
A demonstration of a simulated fishtank inspired by Ann Marion, Alan Kay,
and the Vivarium project.
Bill Coderre, MIT, "Petworld - Simulated Animal Behavior" [Excerpt]
(9 min)
A non-evolving ecological environment implemented in association with
the Vivarium Project.
Mike Travers, MIT, "Neural Nets for Artificial Animals" [Excerpt]
(9 min)
I ran out of room or I would have included all of Mike's talk; it was
delightfully succinct and to the point. I was able to keep the majority
of
his demonstration of BrainWorks, a Braitenberg Vehicles inspired wiring
kit and vehicles environment. This work was carried out as part of the
Vivarium effort at MIT.
And then the presenters that I did not include in the edited tapes, but
who are mostly present on the original 8mm tapes:
Gerald F. Joyce, Salk Inst., "Non-Enzymatic Template Synthesis of Informational
Macromolecules"
Theories related to RNA strands as the Origin of Life. Moderately
successful experimental attempts to manipulate genetic polymers.
Hans Sieburg, Salk Inst., "Artificial Immune Systems: Theory, Implementation,
Application"
Some good ideas that were unfortunately communicated too poorly to
warrant inclusion. Cellular automata were used to model blood transport
of antigens. Antigen concentrations were then communicated to a
separate cellular automata model of antigen binding to body tissue. I think.
Craig Reynolds, Symbolics, "Flocks, Herds, and Schools - A Distributed
Behavioral Model"
Same material Craig presented at SIGGRAPH.
Mark Clarkson, "Core Wares - A Challenge"
A fun diversion based on A. K. Dewdney's Scientific American column
"Computer Recreations". On a simulated processor, programs compete
by
trying to "stay alive" while "killing" the other program.
William R. Buckley, "Of Worms, Gliders, and Core War"
More on core wars.
John Park, USAF, 'Animat'
I forget.
Joao P. Leao, Boston U. Artificial Physics Lab, "Artificial Physics
come to (A)life"
Ditto.
Pauline Hogeweg, U. of Utrecht, The Netherlands, "MIRROR beyond MIRROR:
Puddles of Life"
A pseudo-ecology based on a simple, algebraic "dominance" function,
overlayed with too much anthropomorphizing and philosophizing.
Stewart Wilson, Rowland Inst. for Science, Cambridge, Mass., "Genetic
Algorithms and Biological Development"
An okay talk, but that's all it was. Nothing he described had been
implemented, and he based everything (all natural selection) on an
appriori-defined fitness function. Accordingly, the hypothetical system
should be able to evolve some explicitly specifiable behavior, but can
never even demonstrate any emergent properties, much less evolve.
Aristid Lindenmayer, U. of Utrecht, The Netherlands, "Discrete Models
for Morphogenesis and Gene Expression in Multicellular Organisms"
Used production rules to generate artificial flowering plants. I probably
should have included this due to his role as the father of the field (they're
called "L-Sytems" after "L"indenmayer), but the presentation
was awfully
muddled, perhaps due to language problems, and as a result, he made the
subject appear more difficult than it is or needs to be.
Dr. P. Prusinkiewicz, "Realistic Plant Modelling and Growth Simulation
Using L-Systems"
Graftal pseudo-plant growing techniques. Very pretty stuff.
Charles E. Taylor, "A System For Simulation of Complex Biological Systems"
I think this was another simulated ecology.
Norman Packard, "Evolution of Bugs in an Artificial Ecology"
Another artificial ecology. A pretty simple one it seemed, though this
was just a demo, and it was hard to tell.
William B. Dress, Jr., "A Synthetic Insect Showing Genetic Improvement"
This was a neural net model of an insect, if I remember correctly. It
demonstrated some interesting behaviors and had its brain running on
the Novix plug-in board for the Mac, I think. He never really communicated
the underlying models very effectively, however, during the talk (though
in private communication I got a better idea of what he was up to).
Alexander Pol, "Neuron-Like Behavior of Synthetic, Excitable Proto-Cells
Made from Amino Acids" and "Energy Information Relationships Within
Living Systems"
Had some interesting ideas on information-positive, negentropic nature
of "life", but his difficulty with English made this pretty tough
to digest.
Milan Zeleny, Fordham U. at Lincoln Center, "Recent Experiments in
Synthetic Biology: Living Forms from Inanimate Matter"
Autopoietic crystals.
Peter Oppenheimer, NYIT, "Would Be Plants: Computer Simulation and
Extrapolation"
Same graftal plant stuff he has shown at SIGGRAPH.
H. H. Pattee, U. Center of Binghamton, "Life Beyond Logic, or What's
Wrong With Cellular Automata"
One of those it can't be done because I don't know how to do it talks.
Spent most of his time trying to make sense out of his premise that life
evolves its environment as well as itself. If you look at his main premise
as a rejection of appriori-specified fitness functions in favor of a systemic,
interactively determined fitness, then I agree wholeheartedly, but if this
is
what he meant, he never got it across to me.
Michael Conrad, Wayne State U., "Computer Test Beds for Evolutionary
Theory"
Another computer genes = phenotypical attribute, put 'em down in an
environment and watch the numbers change approach (a simulated
ecology). It was actually fairly well implemented and fairly well
presented, and I probably should have included this on the edited tapes.
I really only left it out because the same basic ideas had appeared in
enough other places that I felt it would be a bit redundant, and I didn't
want the tape set to end up as long as it did!
Richard Laing, Logic of Computers Group, "Artificial Organisms - Future
Directions"
Spent most of his time reviewing Von Neumann's self-replicating
universal computing systems. Suggested some possibilities for inclusion
in new artificial organisms: self-replication through self-inspection (a
limited form of Drexler's disassemblers), Lamarckian style inherited
learning, ... Discussed self-replicating factories and a study of same
prepared for NASA, briefly.
