This is a brief history of our discussion:
Luis wrote:
The structural coupling, or situation, of a classifying,
self-organizing, agent in its environment is the source of meaning.
Alexei wrote:
It seems that the distinction of external vs. internal (i.e., self vs.
environment) is one of major assumptions in your theory. But I think that
this distinction is rather artificial. My computer is more "internal" for me
than my hair. Does the nest of a bird belong to environment? It seems
that there is no need to distinguish Umwelt from Innenwelt. I understand
semantic closure of Pattee as SELF-CLASSIFICATION, and environment as a
part of "self".
You oppose syntax and semantics following Morris. But there are
several levels of syntax! You give an example, that the xerox copy
is syntactical. But what if somebody, who knows the alphabet but
does not read, will copy the text character by character. It looks
like this is pure syntax, but it involves recognition of characters.
We may go further and assume that the text is copied word by word.
Each word is recognized, but the text is not understood. Each word
has a meaning and this kind of copying is difficult to consider
as syntax only. My understanding is that the distinction between
syntax, semantics, and pragmatics exists in hierarchical systems:
syntax corresponds to low-level interpretations, semantics - to
medium-level interpretations, and pragmatics - to high level
of interpretation. But there is no distinct boundary between them.
In simple systems with just 1 hierarchical level, syntax = semantics =
= pragmatics.
I agree that syntactic/semantic dualism is beneficial
for living organisms, but it is not essential for meaning of signs.
Primitive signs have no syntax (e.g., hieroglyphic). Complimentary
duplication of polymer sequence resulted from a long natural selection
of more primitive organisms that had no code et al.
Luis Rocha wrote:
Actually, I do not use the term "self" except in referring to theories
that already exist as such (e.g. self-organization). See, my point in
this paper is precisely to show that the notion of autonomy (or self,
and internal versus external) does not exist in self-organization or
situated action. In section 2.2 I say that a consequence of my argument
"is that the concept of autonomy alone is not enough to characterize
living organisms..." This is a very subversive idea in a session
dedicated to the autonomy of intelligent systems! People in this field
typically, as you well point out, make that assumption.
My message is instead that self-organization offers only personal,
dissociated, dynamics (self-reference as Salthe says), but situated
action shows that this self-organization can only come up with a
semantics that is completely intertwined with the pragmatics of a
coupled system/environment interaction. In this sense, the semantics of
a self-organizing system is as much "self" as it is "other". Indeed the
only true autonomy that living, evolving, systems possess is the
syntactic autonomy of the genetic code which implements Von Neumann's
scheme. This is an autonomous code that exists in each cell and not in
situation with an environment. This syntactic autonomy is THE defining
characteristic of all life on earth and the provider of the ability for
open-ended evolution. It is furthermore defined by symbols in the strict
sense of the word (discrete, short number inert tokens) and has a
definite syntax.
This is my attempt to make a serious attack at all theories that rely
too much on selforganization alone such as autopoiesis and others, at
least as an explanation of the living organization. Autopoietic or
self-organizing systems, when in situation (structurally coupled) with
an environment have no autonomy at all, only when some sort of genetic
code is introduced (or Von Neumann scheme) do we have some sort of
autonomy: syntactic autonomy.
My point is that synactic/semantic dualism is essential for open-ended living
organisms. Signs have no syntax, but they are also not as powerful in
communication as symbols are (thing of formal languages). Pure syntax
allows the displacement of observations and actions in time and space.
Signs only do half a job. In any case, I make no claims about human
language, perhaps we do have some level of intertwined syntax/semantics
based more on signs and less on symbols. Perhaps we have both but who
knows how they are organized. The genetic system is easier ;-)
Alexei wrote:
This is probably the most important point of our discussion. You think
that semantics is the relationship between the system and its
environment. For example, our idea of a tree corresponds to some exteranl
objects in the environment. This is the logic of Morris and Von Neuman
and many others. In contrast, the idea of Pattee was that semantics is
self-measurement. Thus, the initial sign is "self" which is not opposed
to any environment. Imagine a proto-organism that is able to capture
resources, grow, and then divide. If such an organism collides with
the resource unit, then this resource becomes the part of "self". But
those resources that are far away simply do not exist for this proto-
organism. What this system is measuring is only self, and nothing else.
The idea of environment has been developed much later when organisms
became able to use distant sensoring (e.g., vision). Then they could
differentiated between "near self" and "far self"; and we use the
word "environment" for the "far self". Environment is actually a part
of self which is "Umwelt" according to Uexkull. Learning about
environment is expanding our "self". This does not mean that other
"selves" do not exist, but they are viewed as parts of my own
universe.
Von Neuman wrote that evolution of self-reproducing systems is
possible if they are able either to measure themselves and
then to create a copy according to these measurements, or to
code the process of construction. Coding is definitely a more
advanced mechanism of self-reproduction, but self-measurement
is the basic mechanism. Coding is just a particular case of
self-measurement which is focused on a sub-system (e.g., DNA/RNA).
As I can remember, Von Neuman thought about self-measurement in
anthropomorphic way, i.e., an organism measures all its dimensions,
writes them down on a blue print, and then strats building a copy.
In other words, measurements are first coded and then used for
construction. But in nature, self-measuring may go without
intermediate coding. An autocatalytic system produces offspring in
the area of convergence to the same attractor. This self-measurement
is not a code, its a hieroglyphic; it has no syntax.
Pattee's semantic closure is self-measurement that does not necessary
require a code. Codes evolved later. This is the point where Pattee
moved further compared to Von Neuman!
I mean that in primitive self-reproducing systems it is impossible
to separate syntax and semantics. I experimented with
cellular automata with additional "separation rule" and found
evolutionary processes similar to that in nature. See
http://www.gypsymoth.ento.vt.edu/~sharov/biosem/txt/isas98.html
However, there was no code in these systems. Their structure
can be viewed as syntax; but it is also semantics because it
is self-interpreting (semantic closure).
The difference between us is that I try to understand how complex
systems develop from simple systems, whereas you try to understand
how complex systems function. I agree with you that for a given
function of a complex system you can find an appropriate level of
description where semantics will be well separated from syntax. I
just argue that at the origin of signs there was no distinction
between syntax and semantics. They split at the point when the
hierarchy appeared.
Luis wrote:
Pattee did not prescribe semantic closure as a definition of autonomy.
Semantic closure points out that any system with semantics works as a
closed loop. But this loop typically includes the environment. Semantic
closure does not necessarily imply that the closure is contained in a
organism. Indeed, Pattee gives many examples where that does not happen
in his original discussion of semantic closure. The only place where it
seems to happen for sure is with the genetic code (and Von Neumann
scheme). Lately, when Pattee discusses semantic closure he talks a lot
more about this semantic closure with an autonomous self-reproducing
code. He insists quite strongly on the necessity of a symbolic code. My
notion of syntactic autonomy has been designed to make that point more
clear. In other words, syntactic autonomy refers to those systems that
observe an autonomous semiotic closure capable of open-ended evolution.
This is quite different from systems that are capable of semantic
closure with their environments without any kind of semiotic autonomy.
I do not understand semantics without pragmatic environmental
selection. I disagree that Pattee's idea is that of measurement
dissociated from an environment, in fact I believe it is quite the
opposite, but only him can clarify this. I see semantics as situated
action in an environment, this by definition cannot be called an
autonomous process.
I talk about that in http://www.c3.lanl.gov/~rocha/sr.html ,
the point is that the template-type reproduction of autocatalytic sets
is VERY restricted in comparison to the coded version which is
open-ended. Also, notice that in Von Neumann's scheme, the organism does
not measure itself, it simply constructs itself from an inherited
blueprint which may even be mutated and therefore affect construction
(leading to open-ended evolution). This was Von Neumann's great
observation: that self-reproduction by symbolic description is much more
effective than by self-inspection or what you call self-measurement.
Pattee also speaks of measurement as fundamental for living organisms,
but as far as I read him, not to imply self-measurement in the coded
construction. I believe he sees measurement as a result of interaction
between organism and environment (at least I do). The only measurement
that affects genetic blue-prints is established phylogenetically from
natural selection.
I stress that it is when syntax becomes
separated from semantics that life, in the sense of open-ended
evolution, begins.
Alexei wrote:
There is no qualitative boundary between template-type reproduction
(I use your terms) and coded reproduction. This is a continuum.
Besides coded DNA organisms use a lot of non-coded inheritance
(e.g., membranes, left-right stereoisomery, environment). Did not
you think that environment is inherited too? Not all kinds of
template-type reproduction are so limited in their evolution.
I observed emergence of new species and r-selection in my model
of CA with a separation rule. There is no algorithm for determining
if the evolution is open-ended in the same way as there is no
algorithm that will tell if the computer program will stop after
finite time. May be life on Earth in not open-ended? May be we are
doomed because of some intrinsic feature of our biosphere that we
don't know yet? Natural selection is the only measuring mechanism of
evolutionary potential, but it never gives the final verdict.
In order to explain how coded information appear you first need to
understand how uncoded information develop and function.
I have a model of evolving self-reproducing systems without
anything that resembles the genetic code (CA with a separation rule):
http://www.gypsymoth.ento.vt.edu/~sharov/biosem/txt/isas98.html
The same is with prions. You cannot say that there is no syntax in
these systems because they have specific structure with relations
between elements. The only problem is that this structure can not
be copied faster than interpreted, i.e., it should reproduce
itself in order to make a copy.
Imagine a self-reproducing polymer which grows by attaching
monomers on its end. Reproduction is completed when the polymer
becomes broken due to some disturbance. There is a semiotic
process here: the polymer recognizes the monomer in the process
of its attachment. If recognition is not specific, then a
wrong monomer may be attached that will stop molecule growth.
There may be evolution towards better recognition.
Luis wrote:
I disagree (that there is no qualitative boundary between template-type
reproduction and coded reproduction.)
In section 1.4 of my paper the second-but the last
paragraph talks about that. If proteins were reproduced by template or
self-inspection, you would only be able to re-produce those proteins
that either happen to have the ability to copy themselves as they are,
or those that can construct themselves from available parts in their
environment. But this last kind (the only reasonable to postulate),
would need to be unfold at will so that its aminoacid chain could be
copied, and the refold to their original form. Well, with the genetic
code, or description-type replication, ANY protein can be produced, not
from itself, but from a description. In other words, such a system can
produce ANY protein (open-endedness), not simply those that can fold and
unfold to be copied by self-inspection of template-based reproduction.
This is not a continuum, it is a big step.
***************************************************************
Now we come to the current point of our discussion. This is
my reply to Luis:
It looks like our discussion goes around the notion of open-ended
evolution. I view evolution as a branching process; each branch is a
new species (or quasispecies, following Eigen and Schuster). The
evolution is open-ended if the number of branches has no finite
limit when time -> infinity. My point is that there is no effective
method to determine if the evolution is open-ended.
>If proteins were reproduced by template or
>self-inspection, you would only be able to re-produce those proteins
>that either happen to have the ability to copy themselves as they are,
>or those that can construct themselves from available parts in their
>environment. But this last kind (the only reasonable to postulate),
>would need to be unfold at will so that its aminoacid chain could be
>copied, and the refold to their original form. Well, with the genetic
>code, or description-type replication, ANY protein can be produced, not
>from itself, but from a description. In other words, such a system can
>produce ANY protein (open-endedness), not simply those that can fold and
>unfold to be copied by self-inspection of template-based reproduction.
>This is not a continuum, it is a big step.
RNA are able both to fold and to copy themselves. Protein is not
a good example. You argue that the variability is limited to those
molecules that can fold and unfold. But this is the same
self-organization constraint that you seem to recognize in the
next paragraph:
>Sure, but those are the self-organizing (both enabling and restraining)
>constraints that I acknowledge in reproduction. I have argued for the
>importance of this embodiment elsewhere.
You view open-ended evolution as the ability to produce ANY protein.
But some proteins may be deadly harmful and kill the organism.
This is tha same as molecules that can not fold and unfold.
These molecules will be just less successful in natural selection
than those that can fold and unfold. In both casees, natural
selection reduces variability. No difference!
>But without descriptions, they (self-reproducing systems) cannot
>be open-ended.
Why not? They have both heredity and mutations. When a system
produces offspring in the region of convergence to the same
attractor, then offspring inherits this attractor from the
parent. If some offsprings eventually converge to another
self-reproducing attractor it will be a mutation. There may be
infinite number of self-reproducing attractors, and the
evolution will go infinitely.
But I agree that coded self-reproduction makes evolution faster
and more efficient.
>Well, I don't know what you mean by uncoded information. I don't
>consider the embodiment constraints we talk about above to be
>information, they are constraints on dynamics. I find it important to
>distinguish, perhaps a bit too conservativly, between information and
>dynamics.
If you put a boundary between information and dynamics, then
you will not be able to understand the origin of information.
Some kinds of dynamics can be viewed as information processes.
This is the most fundamental idea of complementary descriptions
developed by Pattee.
>> But I have a model of evolving self-reproducing systems without
>> anything that resembles the genetic code
>I will take a look at that. But there are several self-reproducing
>schemes, such as Langton loops, that self-reproduce greatly and without
>syntax. I agree with that. the point is that they are not open-ended,
>but very restricted as I describe above.
Langton's loops were never tested in the presence of noise.
May be they can evolve too. And even if they can not evolve, it
is not a heavy argument because what I am saying is that some
systems EXIST that are able to evolve without coding, rather
than ALL systems without coding can evolve.
>The point is
>that until the code for descriptions is introduced, this evolution is
>limited. Von Neumann himself discussed this with his notions of trivial
>and non-trivial self-reproduction.
What do you mean by "evolution is limited"? I don't argue that
coded information enhances evolution. I argue that systems with
non-coded information can evolve infinitely. Actually, I belive that
these "organisms" with no coding eventually will develop a coding
system. Invention of coding is a metasystem transition which
increases evolutionary potential in the same way as any other
metasystem transition.
-Alexei
-------------------------------------------------
Alexei Sharov Research Scientist
Dept. of Entomology, Virginia Tech, Blacksburg, VA 24061
Tel. (540) 231-7316; FAX (540) 231-9131; e-mail sharov@vt.edu
Home page: http://www.gypsymoth.ento.vt.edu/~sharov/alexei.html