Proceedings of the Workshop
Autopoiesis and Perception
held in Dublin City University, August 25th & 26th 1992
edited by Barry McMullin
Autopoiesis and a Biology of Intentionality
Francisco J. Varela
-
As everybody here knows, autopoiesis is a neologism, introduced in 1971
by H. Maturana and myself to designate the organization of a minimal
living system. The term became emblematic of a view of the relation
between an organism and its medium, where its self constituting and
autonomous aspects are put at the center of the stage. From 1971, until
now much has happened to reinforce this perspective. Some of the
developments have to do with the notion of autopoiesis itself in
relation to the cellular organization and the origin of life. Much more
has to do with the autonomy and self-organizing qualities of the
organism in relation with its cognitive activity. Thus in contrast to
the dominant cognitivist, symbol-processing views of the 70's today we
witness in cognitive science a renaissance of the concern for the
embeddedness of the cognitive agent, natural or artificial. This comes
up in various labels as nouvelle-AI (Brooks 1991c), the symbol grounding
problem (Harnad 1991), autonomous agents in artificial life (Varela & Bourgine
1992), or situated functionality (Agree 1988), to cite just a few self-
explanatory labels used recently.
Any of these developments could merit a full talk; obviously I cannot do
that here. My intention rather, profiting from the position of opening
this gathering, is to try to indicate some fundamental or foundational
issues of the relation between autopoiesis and perception. Whence the
title of my talk: a biology of intentionality. Since the crisis of
classical cognitive science has thrown open the issue of intentionality,
in my eyes autopoiesis provides a natural entry into a view of
intentionalty that is seminal in answering the major obstacles that have
been addressed recently. I'll came back to that at the end. Let me begin
at the beginning.
The Causal and Symbolic Explanatory Duality as a Framework for Understanding Vision
Noel Murphy
-
The conventional approach to interpreting biological vision systems and
experimenting with computer vision systems has been overwhelmingly
dominated by a representational view of information. Even more recent
connectionist approaches, though embodying a substantial change in
viewpoint, have only involved a change of the *type* of representation,
to one of a distributed nature. An alternative view is the notion of
information as being constructed and co-dependent rather than
instructional and referential. This is an interpretation based on the
more embracing viewpoint of the complementary causal descriptions and
symbolic descriptions playing clearly defined interrelated and dual
roles, rather than mutually exclusive, or even muddled roles. This paper
examines this radical change in perspective and compares it with a
causality framework and with a position on the nature of perception
which is based on the idea of universals.
Relativistic Ontologies, Self-Organization, Autopoiesis, and Artificial Life: A Progression in the Science of the Autonomous
Part I -- The Philosophical Foundations
David Vernon & Dermot Furlong
-
Autopoiesis is a very powerful way of looking at and dealing with
autonomous systems. It also has some major implications for the
philosophy of science. Unfortunately, it is not clear in what
philosophical context one should go about using autopoiesis. In this
paper, we look at these issues, touching upon the inadequacies of
conventional (positivistic) ontologies and philosophies of science, and
we briefly describe an alternative relativistic ontology. We argue that
self-organization is a necessary condition for autonomous systems and we
highlight the difficulties that this raises for conventional
representational approaches to autonomous systems. We discuss a
methodology for discourse in relativistic ontology (Systematics) and,
based on this, we argue in favour of a spectrum of autonomy. In a sister
as a particular instance of autonomy in this spectrum. We proceed to
describe the progress which has been made towards the development of a
computational simulation of autopoietic organization, beginning with a
formulation in terms of the calculus of indications (incorporating
Varela's extensions to include autonomous forms), and incorporating the
Systematic formulation.
Relativistic Ontologies, Self-Organization, Autopoiesis, and Artificial Life: A Progression in the Science of the Autonomous
Part II -- A Scientific Development
David Vernon & Dermot Furlong
-
In a sister paper, we have looked at the philosophical aspects
of the development of autonomous systems, touching upon the inadequacies
of conventional (positivistic) ontologies and philosophies of science,
and we have described an alternative relativistic ontology. We argued
that self-organization is a necessary condition for autonomous systems
and we highlighted the difficulties that this raises for conventional
representational approaches to autonomous systems. We discussed a
methodology for discourse in relativistic ontology (Systematics) and,
based on this, we argued in favour of a spectrum of autonomy. In this
paper, we try to show how autopoiesis can be interpreted as a particular
instance of autonomy in this spectrum. We now proceed to describe the
progress which has been made towards the development of a computational
simulation of autopoietic organization, beginning with a formulation in
terms of the Calculus of Indications (incorporating Varela's extensions
to include autonomous forms), and incorporating the Systematic
formulation.
Perception, Adaptation and Learning
Alvaro Moreno, Juan Julian Merelo & Arantza Etxeberria
-
We attempt to distinguish, in a biological frame, ontogenetical
adaptation from learning. Ontogenetical adaptation arises as a second
order (sensorimotor) loop on the ground of the operational closure that
provides autonomy and reproductive identity to the living system.
Adaptation ensures, through perception, the functional correlation
between metabolic-motor states and the states of the environment.
Learning brings about a qualitative change in regard to adaptation, the
most generic and simple form of optimization at an individual scale. It
implies the idea of new knowledge, in the sense that the organism links
what formerly appeared as an undistinguished whole. In other words, it
means the capability to change its own codes of meaning. Finally, we
outline some basic ideas for modelling an adaptive sensor embedded in a
(partially) autonomous system, which implies the former distinction
between adaptation and learning.
Artificial Darwinism: The Very Idea!
Barry McMullin
-
The realisation of artificial Darwinian evolution is one
conceivable -- indeed, more or less obvious -- route toward the
realisation of a growth of knowledge (or complexity) in artificial
systems. This paper explores the current state of the art in achieving
Artificial Darwinism, and the prospects for further progress. In
particular, I reassess the seminal work of von Neumann on evolution in
cellular automata von Neumann 1951; 1966a; 1966b). I also review the *Genetic
Algorithm* also review the Genetic Algorithm (Holland 1975),
and the VENUS (Rasmussen et al. 1990) and Tierra
(Ray 1992) systems. I attempt to relate this to the work of Varela, and others, on
the realisation of *autopoiesis* in related (discrete, 2-dimensional,
homogenous) spaces (Varela et al. 1974; Zeleny 1977;
Zelany & Pierre 1976), and I also revisit the Holland
alpha-universes (Holland 1976; McMullin 1992d).
I suggest that while both open-ended heredity
(von Neumann style self-reproduction) and spontaneous autopoiesis
have been separately demonstrated in such systems, the combination of
the two remains a difficult outstanding problem. I conclude by outlining
an avenue for further investigation.
Reality Paradigms, Perception, and Natural Science. The Relevance of Autopoiesis
Dermot Furlong & amp; David Vernon
-
There is an ancient philosophical principle which states that Being is
prior to Knowledge. That statement, I would suggest, holds little
interest for the majority of scientists and technologists, who quite
likely would not see the relevance of the remark to their research
activities or, indeed, to their lives. Such is the entrenchment in our
world view, our perceptual reality-paradigm, that we do not, and indeed
almost cannot, recognise that how we see `the world' is dependent on
what we are, i.e. on our ontological status. To put it bluntly, seeing
the world, for normal science, is the construction of a represention of
an external reality by the scientific spectator, which representation
must be probed to determine its hidden, primary,
mathematically-describable, and fundamentally mechanistic, basis. Such a
scientific `spectator consciousness', and the scientific methodology
associated with it, is the product of a cultural development which, with
its roots in antiquity, found most complete reinforcement in the
successes of the mathematical physics of nineteenth century classical
science. For the greater part this spectator science remains remarkably
unscathed despite the various undermining developments of modern,
twentieth century, physics, and is now finding concrete expression in
the relatively new discipline of cognitive science, as we assail the
question of consciousness -- `just about the last surviving mystery', to
quote Daniel Dennett. Along the way, the terrain underfoot of the
`secure stride of science' has been substituted without any enquiry as
to its suitability, moving from the physical, to the biological, to the
mental, in the quest for absolute certainty, necessity, and
completeness, or at least an acceptable approximation to same! However,
what is suggested here is that normal science is fundamentally flawed
when it is applied to the domains of Life and Mind, and that,
furthermore, very much involved in that flaw is our non-recognition of
the adopted perceptual reality-paradigm. That is to say, our
mis-conception of science has very much to do with our mis-conception of
perception. And what is wrong with our conception of science in its
application to Life and Mind is that the analytic reductionism which
characterises the spectator consciousness stance can never capture the
organisational distinctions which characterise living or cognizing
beings. Scalpels and microscopes may be useful, but not for the
discovery of Life or Mind, for when the analysis is done, that which is
essential is gone.
Constructivist Artificial Life, and Beyond
Alexander Riegler
-
Within this paper I provide an epistemological context for Artificial
Life projects. Later on, the insights which such projects will exhibit
may be used as a general direction for further Artificial Life
implementations. The purpose of such a model is to demonstrate by way of
simulation how higher cognitive structures may emerge from building
invariants by simple sensorimotor beings. By using the bottom-up
methodology of Artificial Life, it is hoped to overcome problems that
arise from dealing with complex systems, such as the phenomenon of
cognition. The research will lead to both epistemological and technical
implications.
The proposed ALife model is intended to point out the usefulness of an
interdisciplinary approach including methodological approaches from
disciplines such as Artificial Intelligence, Cognitive Science,
Theoretical Biology, and Artificial Life. I try to put them in one
single context. The epistemological background which is necessary for
this purpose comes from the ideas developed in both epistemological and
psychological Constructivism.
The model differs from other ALife approaches -- and is somewhat radical
in this sense -- as it tries to start on the lowest possible level, i.e.
avoids several a priori assumptions and anthropocentric ascriptions. Due
to this characterization, the project may be alternatively viewed as
testing the complementary relationship between epistemology and
methodology.
Reconstructing AI
Conor Doherty
-
Symbolic AI is argued to be epistemologically and ontologically
necessary but insufficient for constructing robust AI. Two principles,
embodiment and situatedness, are elaborated which any global theory of
AI must incorporate. These principles require autonomous robotics to
form a basis for AI. Learning is the key to the development of more
autonomous robots. Artificial neural networks are evaluated for their
ability to learn to integrate robust sensory categorisation with motor
control. The future relationship of artificial neural networks to
symbolic AI is speculated on.
|