|
Tim, you are missing the essential nugget of meaning here. In the
passage you quoted, you are actually proving my case for me.
Look:
RR wrote: "In either case, there are
never enough purely behavioral
discrimina-
tors that entail identity.We must go to causality for
that kind of identi-
fication?i.e., into how
behaviors are generated."
From page 122 of Essays: "As a prominent example of this kind
of approach [mimesis], we have considered the Turing Test, which says
that if WE cannot discriminate behaviorally between another human being and a
properly programmed machine, then the two must be objectively identified. In
short, we must impute, to anything that behaviorally mimics a mind or
consciousness, its own mind and its own consciousness; we must impute to it a
subjective, inner world just like ours."
He then goes on to say that "there are never enough behavioral
discriminators that entail identity", we have to go to causality (the generating
of behaviors) for that kind of entailment information. In other words, if we
conduct investigations into the causal basis of the behaviors of both the
Turing machine and the human, we will find what entailment relations are
underlying the behaviors. The only way to get at entailment relations is via
causality. Observable behavior alone cannot tell us about entailment. We
have to learn the specific causes of the observable effects, to learn whether
the entailment patterns of the two systems commute.
What he is saying in that paper is a larger issue which I
think is very appropriate for many of the current problems that science is
causing humanity and the rest of the biosphere, today: namely, that
biomimesis has changed from what the original premise was but has lost sight of
that fact, completely. Earlier, on page 122,
Robert Rosen wrote: "In the beginning, the neural net was a
plausible mimic of brain, because it did not depend entirely on a comparison of
the behaviors of each; it rested rather on anatomy. Indeed, as originally
conceived, the behavioral mimicry arose from the underlying mimicry between
biological neurons and switchlike elements, and on a continuity assumption, or
robustness hypothesis, that populations of comparable elements arrayed
comparably would behave in comparable ways. This kind of plausibility has been
entirely lost in the progression from neural net through finite automaton
through Turing Machine, in which comparability devolves entirely on behaviors
themselves, rather than on the way the behaviors are generated. In mimetic
terms, we now have actors (e.g., Turing machines) imitating actors (automata)
imitating other actors (neural nets) imitating brains. What looks at each step
like a gain in generality (i.e., more capable actors) progressively severs every
link of plausibility and throws the entire burden on the actions
alone."
In other words, the Turing machine doesn't even posit that the
causes of the behaviors need to be figured out, via science; all that is
required is to get the behaviors right, and the causes will follow. It's a very
shallow, juvenile approach to things and hardly based on any scientific
principles whatsoever, it seems to me. In mimesis, they apparently dispensed
with the original hypothesis (whether the continuity assumption holds) and kept
the experiment-- but gave it all new relevance and importance as if the original
hypothesis was some law of nature that was applicible to any kind of similarity,
even if only in appearance or behavior.
However, to get back to the original discussion:
Page 84 of Essays:
Robert Rosen wrote: "In one way or another, the concept of
causality dominates our conception of what transpires in the external, public
world. In broadest terms, causality comprehends a system of entailments, which
relate the events and phenomena occurring therein. The concept itself is due to
Aristotle, who associated it with the answers given to the question, "Why".
Indeed, to Aristotle, science itself was the systematic study of "the why of
things," and hence entirely concerned with elucidating such causal
relations."
In his usage, the word "comprehends" means "embodies" and causality
then is the manifestation, in each particular observable situation, of the
entailments which exist, therein. Causality is a step removed from entailment.
For example; it is not only the entailments of the system which are generating
causal behavior, it is also the interaction between the system and a
secondary influence (be it another system or the system interacting with
itself...), constrained by the entailments of the relation through which the
interaction occurs, which constitute the total cause of any given
phenomenon or behavior. However, the underlying reasons for why the
system CAN behave that way, in interaction with whatever
particular secondary influence it may be.... that is the realm of
entailment.
So, the observable behaviors of systems in the
universe are what generate our "why?" questions. However... isn't
it true that we don't only want to know the direct cause, as in
specific "cause and effect". That is not "the why of things"...
that is the "what" of things (as in; "what caused this explosion?"). Aristotle
was looking at underlying causal entailment patterns. As my father has written
many times: All "what" questions eventually become "why" questions, if you
really want to understand any given system or situation. So, rather
than defining what causality IS... perhaps we should ask; Why does
causality happen the way it does? The answer, according to Rosennean
Complexity would be: Because of the underlying entailment
relations.
There is, clearly, a relation between causality and entailment, but
I believe you have gotten their roles reversed in reading my father's work.
Entailment is the word my father gives to the "causal"
foundations of causality.
Judith
----- Original Message -----
Sent: Wednesday, March 23, 2005 10:25
AM
Subject: Re: [ROSEN] Causality vs
Entailment
Judith,
From that page in
Essays:
In either case, there are never
enough purely behavioral discrimina-
tors that entail identity.We must go to causality
for that kind of identi-
fication?i.e., into how
behaviors are generated. One immediate
con-
clusion
is that we cannot generally argue backward from behaviors
alone
to
their causal bases (i.e., analysis and synthesis are not simply
inverse
operations [see
chapters 1 and 6]).Behavioral mimesis is at
best only
evidentiary; it has
been argued that inductions from such evidence,
like
the Turing Test, have no
basis, since at least the time of Hume. [ital. orig., bold
added]
To say that the problem
is that it is ?a linear _expression_ and a linear view? is simply a restatement
of ?analysis and synthesis are not simply inverse operations?, where ?linear?
refers to the synthetic. Since each is not just the inverse of the other, one
does not entail the other.
Regards,
Tim
From: ROSEN
Forum [mailto:*** On
Behalf Of Judith Rosen
Sent: Wednesday, March 23, 2005 8:25
AM
To:
***
Subject: Re: Causality vs
Entailment
Tim Gwinn wrote: one cannot
reliably argue backwards from behavior (effects) to causes.[EL p. 123] The
former does not entail the latter.
Tim you're getting into quicksand, here. My father did
not say this the way you are representing it. All of science argues backwards
from effects to causes. The problem is that such an argument is a linear
_expression_ and a linear view. That is precisely what limits "causality" as a
subject for study. The underlying entailment relations don't follow this
linear, timebound form and are, therefore, more
comprehensive.
----- Original Message -----
From: Tim Gwinn
To: ***
Sent: Tuesday, March 22, 2005 10:29
PM
Subject: Re: [ROSEN] Causality vs Entailment
Hi
Steve,
I think one general point about mimesis Rosen makes is that And
an accretion of behaviors does
generically entail a particular accretion of
causes. Analytic models are not
generically the inverse of synthetic
models. Thus to create something which
mimics the behavior of some original
system does not entail that the mimic
therefore has the same underlying
causal entailment organization - it's
causal basis - as the original
system. So mimetic approaches will
generically result in a lack of synonymy
of entailment structures between
the two systems.
With respect to
simulations, there is again the problem that that there is a
lack of
synonymy of entailment structures between the original system and
a
simulation of it. In a simulation, all the entailment structures
are
collapsed into material cause (i.e., software program) to a simulator
(i.e.,
hardware). Additionally, the simulator does not decode into anything
in the
original system: it is entirely extraneous to the original system
yet
entirely integral on the simulation side.
In both cases, I feel
the term 'entailment' is used consistent with the way
he uses the term
elsewhere.
Regards,
Tim
|