|
James Rose wrote: Finally published words are done to please the audience to a great degree. Working notes are the person's alone and a direct vision of the exploration in all its subtlety, hardship, and beauty. Beautifully written post, Jamie. You talked me into it: Robert Rosen wrote:
"Some Thoughts on Returning from Trip [dated 1/7/94]...
There were two thoughts I remember, which are worthy of
further development.
FIRST was the whole complex of ideas surrounding
the CONDITIONED REFLEX and its relations with modelling relations, and
particularly with encoding/decoding arrows. And hence, the establishment of
relations between language and its referents.
SECOND was in connection with differential equations, and
boundary/initial conditions to determine specific solutions (fix the "constants
of integration" in general solutions) and using these "initial/boundary
conditions" as information in systems driven by information. More specifically,
it had to do with regarding such differential equations as constraints, and
then, as in closed systems, treating the highest derivative (e.g.
"acceleration") as "force" (efficient cause) expressed as a function of state
(material cause or initial state). Remember that this last, making force a
function of phase, expresses a "Newtonian" rather than a "Machian"
view.
In connection with the FIRST, I originally wanted to show
that a modelling diagram constituted a non-fractionable system. Specifically,
fractionating the language away from its referents (or the referents away from
the language) kills the modelling relation irreversibly, and at the same time
creates all kinds of artifacts, in great profusion, and on both
sides.
SO LET ME GO AT THE CONDITIONED REFLEX
FIRST.
Well, you have to start from a reflex. Pavlov started with
stimulus (e.g. food) and a wired-in response (salivation), which he apparently
thought of in causal terms. The stimulus is a sufficient condition to the
response. The response is like a bioassay of the stimulus?!
Conditioning involves generating the response via an
unrelated stimulus (a bell). This was regarded as an instance of learning,
performing an induction on an association of the unconditioned stimulus (food)
with the neutral bell. I WANT TO THINK OF IT, RATHER, AS THE ASSOCIATION OF A
WORD WITH A REFERENT. Indeed, conditioning makes the bell a synonym for the
food, like the Einsteinian "hat check to the hat". Likewise, I can think of the
bell as a PREDICTOR for the food. Anyway, there is a DICTIONARY involved here,
almost like translating one language into another. YES! THAT SEEMS TO BE THE
APPLICABLE METAPHOR.
Indeed, in learning a new language (a new vocabulary, a
new grammar) you learn a new word (here the bell) in the new language to
correspond to the old word (e.g. something about the food, an adjective for the
food) which you knew in the original language. The food itself, the stimulus, is
the referent of both words, and is what creates the synonymy between
them.
I'M NOW LOOKING AT BOTH SIDES OF A MODELLING RELATION AS
AN OLD LANGUAGE AND A NEW ONE, BEING PUT INTO CONGRUENCE (SYNONYMY). THIS
"PUTTING INTO CONGRUENCE" IS THE CONDITIONING. THE ENCODING/DECODING ARROWS
CONSTITUTE THE "CONDITIONING" ITSELF.
So now, the question is: WHAT HAPPENS WHEN I ATTEMPT
TO FRACTIONATE, OR REDUCE, SUCH A SITUATION? What do I irretrievably lose, and
what artifacts do I create?
I said earlier that both sides of a modelling relation
are, separately, adjectives of a larger situation. When I fractionate, I lost
the larger situation. One of the artifacts I create by such a fractionation is:
I create alternative histories. Let me try to explain this further. If I don't
know the history which created a conditioned reflex, I have no way of telling a
conditioned reflex from an "unconditioned" or "real" reflex. I might presume the
salivation response to the bell is exactly the same as the salivation response
to the food. In other words, there is not enough "information" in a reflex
itself to discriminate the one from the other. The "real" history is lost, and
unreal new ones are created. I HAVE TO THINK MORE ABOUT THIS.
For one thing, if the conditioning corresponds to
encoding/decoding arrows, fractionating the two sides of a dictionary (e.g. a
modelling relation) apart precisely lose these arrows. IN SUCH TERMS,
FRACTIONATION SERVES TO "UNCONDITION" THE REFLEX AGAIN. IT SERVES TO WIPE OUT
THE HISTORY. THIS IS PRECISELY THE "INFORMATION" IT LOSES.
Maybe that is precisely why these encoding/decoding arrows
look unentailed; why they look arbitrary, and hence need to be separately
posited. MAYBE THIS IS WHERE WE COME TO NEED REALIZATIONS, WHICH IN EFFECT
RECONSTRUCT ONE SIDE OF A MODELLING DIAGRAM FROM THE OTHER. OR MORE PRECISELY:
WHICH RECONSTRUCT THE REST OF A SYSTEM FROM A FRACTION!!?? AND WHY THIS CAN
GENERALLY BE DONE IN MANY WAYS, IF IT CAN BE DONE AT ALL??!! (NOTE: DOES THIS
LAST REMARK ASSERT THAT REALIZATION PROBLEMS ARE GENERALLY
ILL-POSED???).
So this line of inquiry is beginning to touch on
everything!?
IT MAY BE WORTH RE-EMPHASIZING HERE WHAT I SAID ABOVE,
THAT YOU REQUIRE A "REAL" REFLEX TO EVEN BEGIN TALKING ABOUT CONDITIONING. ID
ON'T EVEN BEGIN TO TALK ABOUT WHERE THAT FIRST ("REAL") REFLEX COMES FROM. NOR
CAN I EVEN BEGIN TO ADDRESS THE "ORIGIN" OF THAT REFLEX IN THE SAME TERMS I TALK
ABOUT THE ORIGIN OF THE CONDITIONED ONE. Likewise, I can't talk about the
"origin" of a first language in the same sense I talk about learning a new
one.
Okay, time to sum up. I have concluded that conditioning
is like learning a new language. I create a synonymy between my owl language and
a new one, by attaching a word in the new language to the same referent already
attached to a word in the old. Each language, by itself, is a separable
fraction, where the fractionation amounts to the erasure of the
encoding/decoding arrows, and indeed, the common referent, which makes words in
the two languages synonymous. NOTICE that by removing referents, I HAVE ONLY
SYNTAX LEFT IN THE FRACTIONATED LANGUAGES. This observation may be crucial in
itself; THE FRACTIONATION LOSES ME THE SEMANTICS.
***
The "evolution of science is from syntax to semantics.From
physics to biology. From facts to meanings. From experiment to
theory.
Hence there is always conflict between an early stage of
science and the later stages of the same science. Between the present and the
future. The "paradigm shifts".
Semantics requires (external) referents. Models and
anticipation?
As science gets t9o be about more and more things,
referents, it gets more and more semantic. More and more surrogacies. At the
same time, fewer and fewer reductions.
What is the bearing of innate vs. acquired? Can we say
innate = syntax? AT THE EARLIEST STAGES OF SCIENCE, EVERYTHING IS INNATE?
SYNTACTIC?
MAYBE TAKE EVERYTHING BACK TO NATURAL
LAW??!!
I must say such things without getting too
technical.
Is e.g. the Hamiltonian mechano-optical analogy like a
conditioned reflex? Just a dictionary? An assertion of synonymies (surrogacies)?
Hmmm.
The Unified Field is reductionistic, syntactic. Like a
Leibnizean "characteristica universalis". IT CAN ONLY WORK IF EVERYTHING IS
INNATE.
Maybe a good example of such evolution in science is
equilibrium vs. steady state. Or unforced vs. forced; or autonomous vs.
non-autonomous. IS THERE ANYTHING INNATE TO A BELL, SAY, WHICH INDICATES IT
MEANS FOOD? ANYTHING IN THE EXTERNAL WORLD WHICH CONVEYS SUCH MEANING? SUCH
SEMANTICS? If you look for such a fractionation, you don't find it, because of
the limits of fractionation!
***
In causal terms, why does an animal eat,
anyway?
The obvious answer is that food relieves hunger. But
"eating" is a verb. A behavior. It should not be an effect. Rather, like any
behavior, it should be an efficient cause of some other
effect.
Indeed, why does an organism follow a gradient, or exhibit
any other anticipatory behavior?
If a behavior is like a mapping, then causing a behavior
is itself a behavior; another mapping, whose "output" is itself a mapping. THIS
CAN BE INNATE, OR ITSELF ACQUIRED. Hmmm. This is getting (M,R)-like, even
here.
Indeed, M is a behavior, a mapping from inputs to outputs.
R is a generator of behavior; another mapping.
The only "acquired" thing in an (M,R)-system comes from
the environmental inputs; what the mappings in M process. As I showed long ago,
these can show up in the repair maps. BUT NOT IN THE REPLICATION MAPS. As I
recall, that is forbidden by a requirement like [mathematical symbol that looks like a small "g" with a cap over
it^, to the power of -1] exist.
***
Wait a minute. In Pavlovian conditioning, the bell only
makes the dog salivate. It doesn't relieve its hunger the way food does. IT IS
THUS NOT A SURROGATE FOR FOOD IN THAT SENSE.
If I want to analogize conditioning to Placebo effect,
feeling better after the doctor gives you a pill is more like satisfying hunger
(a subjectivity; a feeling) than it is like just salivating. SALIVATION IS AN
"OBJECTIVE" SIGN; PLACEBO EFFECT IS JUST TESTIMONY. IT IS
SUBJECTIVE.
Further: the dog has to be hungry (a subjective state)
before it will salivate at the smell of food. It is not conditionable, then, if
it always feels full. Thus, the original unconditioned reflex is a measure of
the subjective state of hunger. "How hungry" a dog is can thus be measured by
how much it salivates at (a) food in the unconditioned dog, and (b) the sound of
the bell in the conditioned one. (Bell now is conditioned to be a surrogate for
food later; not for food now). Maybe satiety is also measured objectively by
"time from preceding feeding".
Eating to relieve hunger is INNATE? Or acquired? Likewise:
taking a pill to relieve pain is acquired; learned. There is also an (innate?)
specificity as to what "food" is. A dog will not salivate at
grass.
I think these things are important. It is important that
conditioning to a bell will not relieve hunger; it will NOT be a surrogate for
food in that sense. IT IS ONLY A PREDICTOR FOR FOOD. The physiological
("objective") salivation is like a tropistic response; like climbing a gradient
of light or heat, or being in an environment where the day is shortening. Moving
in the gradient is only a PREDICTOR for something "objective"; something IN THE
FUTURE.
SOMETHING WHICH CAN GET GENOCOPIED.
***
ISN'T PART OF "OBJECTIVITY" TIED UP WITH THE IDEA THAT YOU
CAN DO A "CONTROL"? HENCE, THE IDEA OF THE CONTROL IS REGARDED AS AN ESSENTIAL
PART OF WHAT IS SCIENTIFIC??
THE IDEA OF A "CONTROL" IS ROUGHLY AS FOLLOWS. IF I HAVE
"CAUSES" x, y, z, ,...OF AN EFFECT E, AND I JUST VARY ONE (say replace x by
x', and keep y, z,... fixed), AND THE EFFECT E DOESN'T HAPPEN,
NO WAIT. IT'S SUBTLER THAN THIS. IF THE EFFECT E ONLY
HAPPENS WITH x', AND NOT WITH x, WHEN EVERYTHING ELSE IS HELD FIXED, THEN x' IS
A CAUSE OF E . THUS, TO DEMONSTRATE THAT A PARTICULAR ANTIBIOTIC, SAY, CURES A
DISEASE, YOU HAVE TO DO CONTROLS OF THIS KIND. KEEP EVERYTHING FIXED, AND JUST
VARY WHETHER YOU GIVE THE THERAPY OR NOT.
DOUBLE-BLINDS ARE ALSO NECESSARY.
SO JUST TAKING A PILL AND GETTING CURED IS NOT
OBJECTIVE.
DO YOU EVEN DO CONTROLS IN
MATHEMATICS???"
***
Well, that's enough for today, I think!
Judith
|