Jeff Prideaux
[ The following was originally posted to the Artificial Life list by
George Kampis. I liked it so much I asked him to submit a copy here.
Cliff Joslyn ]
Autopoiesis matters
In a recent note, <mcmullinb@vax1.dcu.ie (Barry McMullin)> suggested
new computer simulations for autopoiesis. As a long-time follower of
autopoietic theory, let me be critical with respect to what such a
simulation can bring us, and let me make some remarks about the very
concept of autopoiesis at the same time.
1. Autopoiesis - Its Contribution
=================================
Defined by Maturana and Varela in (1973), autopoiesis means
"organizational closure" (well, yes the *word* itself means
self-production), a state where the preconditions and the results of
existence coincide, in a self-referential way.
Although vague and often misinterpreted, and resting heavily on less
known earlier developments, the notion is an important one,
characterizing life on the basis of a *dynamic* instead of a
*structural* existence, and focusing on a mutuality between the
products and the producers of the living state -- in the small, on the
mutuality between the genes and the proteins, or the knowledge and the
knower.
2. The Computer Model
=====================
Let me be very explicit: taking the above definition seriously, there
can be no such thing as an algorithm for an autopoietic system.
Algorithms are prototyped by machines where there is no feedback from
the outputs towards the defining primitives. No formal system can
depend on axioms that it produces as theorems. An algorithm is, by
definition, an "allopoietic" system, one that produces something else
than itself, whereas an "autopoietic" system should be the way around.
So, for instance, the model described in (BioSystems 1974, vol 5,
187--196) depicts a unidirectional process, which proceeds from a
given input towards a given output. So, this Uribe model is clearly
not an autopoietic system at all. There is no organizational closure
in it. The model does not produce its basic components, only a few
derived ones. Who produces * (the catalyst), for instance (on p.
189-90)? It has to be introduced in the system by hand, and it is not
subject to any production or degradation process. What we see here is
a dependent or controlled process, and not an autonomous process. In
fact any autocatalytic reaction system shows more "autonomy", as there
are at least *cycles* in autocatalysis.
It would be false to look for a tricky outway now, by means of fabrics
that allow for a "better" simulation, where we would have, instead of
the above seen straightforward and crude dependence on what in
chemistry is called a pool substance (one that is never deployed), a
situation where the supporting scaffolding becomes better concealed
behind the clumsiness of an intricate definition and behind the use of
a vast amount of different compounds. The issue is not a matter of
degrees but a matter of principles. There will be no better
simulation, there is no algorithm for self-reference.
3. But Why Autopoiesis?
=======================
Here we come to a different issue: what is it that we want, after all?
A model of autopoiesis at any rate, whatever that might be, or a model
of the living state? Autopoiesis may just not be the right concept to
begin with, if we want *models*. Autopoiesis is not a modeling
methodology at all, it is just a way, and it is a different way, of
looking at things. It is different metaphysically, much as Zen
buddhism is different from science. As it is a wrong question to ask,
what the point in Zen buddhism is, for if you ask that, you know
nothing about Zen buddhism, and, with this start, you will probably
never learn anything, it might be similarly wrong to ask for a model
of autopoiesis. Autopiesis is just a third kind of metaphysics,
besides physicalism and vitalism, containing elements of both but
making allies with neither.
This metaphysics can be best characterized as that of the Kantian
"Ding an sich", the Thing to Itself. This metaphysics involves
self-reference as a third logical primitive besides "truth" and
"falsity", as indeed implied by Varela s own work (see his Principles
of Biological Autonomy, North-Holland, 1979). In such a system, you
have no mathematical decidability, and hence no possibility for
modeling any more. Whereas this system has a high emotional and
intellectual attractiveness for an adventure of thought, it may be a
cul-de-sac with respect to various goals of ordinary science.
So why adopt this system? What do we learn by that for our models?
What should be done differently with it than without it? In its
original form, this approach is just not accessible for such a
treatment. But that does not mean no treatment of life itself (or
perhaps, Life Itself, smile) is possible.
4. Processes, Not Units
=======================
What we may need, then, is a *constructive* instead of *descriptive*
characterization (as in autopoiesis) of the circularity between the
system and its components. For instance, the circle from genes to
proteins and back can be cut off, by making reference to the process
that brings it about. There is clearly no self-reference and hence
nothing "autopoietical" at all in the process in which a given gene
yields a given protein, or a given protein contributes to the
expression of a gene. It is only our perspective (a false perspective
now), a birds-eye view that defines the cell (or the brain, for that
matter) as a global and self-referential unity instead of a web of
local causal processes.
To go for a causal description and to reject the object description is
not reductionism. It can be process thinking (in the tradition of
Heraclitus and Whitehead) instead of the essentialist thinking of
Parmenides and Plato. It is thinking in terms of emergent integrities
instead of in terms of persistent unities. From the causal point of
view, the cell is not a *thing*, it does not *exist*, in this sense.
There are various ways of trying to characterize emergent integrity.
There is connectionism, that tries to transcend the static symbolic
descriptions of the unity called brain, by going to levels below. It
may be an attempt, nothing more. In my view, it is an attempt that
fails, in its essential points. Yet the idea is there. Then there is
dynamical structuralism, with all its oddities, but with its emphasis
on temporal becoming, and on the unfolding of trajectories, instead of
the existence of pre-fixed solutions. How long it can go along that
road is another question. But again, the attempt is there. (Artificial
Life is, at best, something that pertains to dynamic structuralism, in
its present form - maybe it can be called "computational
structuralism").
And there is, somewhere in this list, a suggestion by myself, which
offers to consider the logic of component production on its own.
Developed under the name "component-systems", this theory depicts
productive processes of the kind that may interest the biologist, the
brain theorist and the cognitivce scientist, as complex and
unpredictable dynamic processes, that generate radically new
variables, yet at the same time can be *understood* and in an at least
approximative sense simulated while resting on some ordinary
metaphysics of materialism.
Or there are so many other candidates. Autopoiesis is, in the form
described, an old hat, not necessarily promoted even by its inventors
today - they just dont seem to use this word any more (instead, Varela
uses *enactment*, for instance). But that would be another story.
-- George Kampis
kampis@ludens.elte.hu
Acknowledgment
I thank Barry McMullin for FAXing me p.189 of the cited 1974 paper -on
travel, and deprived from my bureau, this was an important help.