Copy Link
Add to Bookmark
Report
Alife Digest Number 042
ALIFE LIST: Artificial Life Research List Number 42 Saturday, October 6th 1990
ARTIFICIAL LIFE RESEARCH ELECTRONIC MAILING LIST
Maintained by the Indiana University Artificial Life Research Group
Contents:
SAB conference
----------------------------------------------------------------------
Date: Sat, 6 Oct 90 20:50:45 EST
From: Liane Gabora <liane@cogsci.indiana.edu>
Subject: SAB conference
Here is a summary of the Simulation of Adaptive Behavior conference that
took place last week in Paris, that I sent to the Indiana University
Cognitive Science mailing list. It was suggested that I post it to the
Alife list as well.
I thought some of you might be interested in a summary of the
"Simulation of Adaptive Behavior" conference:
The purpose of the conference, as stated by the organizers, was "to
bring together researchers in ethology, ecology, cybernetics,
artificial intelligence, robotics, and related fields so as to further
our understanding of the behaviors and underlying mechanisms that
allow animals, and potentially robots, to adapt and survive in
uncertain environments". There were sessions on (1) perception and
motor control, (2) cognitive maps and internal world models, (3)
motivation and emotion, (4) action selection and behavioral sequences,
(5) ontogeny and learning, (5) collective behaviors, (6) evolution of
behavior, and (7) architectures (i.e. neural networks, genetic
algorithms, classifier systems), organizational principles, and
functional approaches. It was definitely multidisciplinary. Informal
discussion topics ranged from benefits/difficulties of six-legged
versus four-legged robots to how can we build consciousness into
artificial creatures.
Many nice demos with fancy graphics. One that I found particularly
intriguing was a very realistic-looking simulation of the flight of a
flock of birds. There was no "central executive" in the program that
determined the behaviour of the flock; each individual was simply
following a few simple rules, such as "avoid obstacles", "aim at
keeping within a certain distance from the centre of the flock",
"don't get closer than a certain distance to other individuals", etc.
When the flock approached an obstacle, it split in two and some went
to the left of the obstacle while others went to the right. The flock
then united on the other side of the obstacle. This kind of
phenomenon, wherein complex and interesting behaviour emerges out of
the interactions of individual units following simple rules, came up
again and again. (Emergence was definitely a buzz word at the
conference, as elsewhere -- see recent comp.ai.philosophy discussion.)
In this case, the units were individual "birds"; in other programs the
units represented various drives (such as hunger) in a single
individual that competed to be fulfilled; in neural networks the units
were neuron-like processors. There were also robots that could do
various tricks such as display aggressive-looking threats when you
stomped the floor, but then run away when you chased them! For those
of you who were at Alife II, the imaginative and charismatic
zoosystemicien Louis Bec was back with his "Machinations Zoologiques"
and "Sulphosomes" -- biologically plausible (at least somewhat)
creatures designed down to the finest physiological detail. This was
the only nonserious presentation. (Since two people have already
asked, I might as well add that Jonathan Post was not there.)
What follows is a brief summary of a few of the papers.
The proceedings will be coming out as a book by MIT press in 2
months. I have a booklet containing the abstracts, which people are
welcome to borrow. Of course if anyone would like a copy of my paper,
titled "A model of the mechanisms underlying exploratory behaviour",
I'd be happy to send it to them.
Michael Arbib and Alberto Cobas presented Rana Computatrix, a set of
computer models of visuomotor coordination in the frog and toad. In
their model of prey-catching, the extent to which a "perceptual
schema" is active reflects the system's confidence that a certain
object is present, and the extent to which a "motor schema" is present
reflects the system's readiness to perform a certain action.
Behaviour arises from the parallel activation of multiple interacting
schemas.
Rod Brooks talked about efforts at the MIT AI lab to build robots that
are self-sufficient for extended periods of time and can perform
useful tasks in environments that aren't specially built for them.
One of their robots, for instance, walks around the lab looking for
pop cans, shaking them to see if they're empty, and if so, throwing
them in the trash.
David Cliff (or "Visual Cliff", as Michael Arbib referred to him) gave
a lively talk in which he presented a program that simulates the
visually-guided flight of the hoverfly. His program is a closed
system; input to the "retina" feeds into the "brain" which determines
the appropriate motor action. This action is then performed thereby
changing the location of the fly, which in turn changes the input to
the retina, etc. He calls the approach of embedding cognitive
apparatus within a sensorimotor system: "computational neuroethology."
John Koza discussed a technique for genetically breeding populations
of lisp programs to solve problems that range from figuring out
complex sequences to enabling an artificial ant to follow an irregular
trail. In this paradigm, individuals are programs (which he displays
visually as trees) and crossover amounts to swapping s-expression
(branches) from one program to another.
David McFarland discussed the general notion of adaptation. He
distinguished between evolutionary adaptation, which involves changes
in the genome and has a long-term effect, physiological adaptation,
sensory adaptation, and adaptation by learning. He discussed what
would be necessary in order for a robot to be adaptive.
Richard Sutton presented an architecture that uses trial and error
reinforcement learning to create an internal model of the world. It
then intermixes conventional trial and error learning with
*hypothetical* trial and error learning to reach optimal behaviour
more quickly. This paper might be of interest to people here working
on counterfactual thinking.
I hope this account gives some impression of what went on at the
conference.
One last comment: I've noticed that the list of faculty members
associated with the Cognitive Science program at IU does not contain
anyone from the Biology Dept. The pamphlet I received on the program
stresses the goal of understanding the mental abilities of human
minds: it mentions knowledge and intelligence, teaching and learning,
etc. I think that the study of human or machine intelligence is of
limited value unless it is embedded within a fairly large framework
that addresses issues such as (1) What conditions bring about the
existance of autonomous systems that are able to survive and
replicate? (2) Under what conditions do such systems evolve complex
adaptions such as the ability to learn? (3) Consciousness,
motivation, and emotions are as much a part of the human psyche as
intelligence. Why might they exist? How do they "work"? Perhaps
more interaction with people in biology -- particularly evolutionary
biology, population genetics, and ethology -- would enrich our
understanding of cognition by providing us with perspectives that help
place it within an appropriate conceptual framework.
------------------------------
End of ALife Digest
********************************
