[Date Prev][Date Next][Thread Prev][Thread Next]   [Date Index] [Thread Index] [Author Index

Re: Rosen & Ashby



Hi JohnK,
 
I was thinking along a somewhat different line than you, regarding my blunder. But your point is well taken. I realized that I was committing a fallacious bit of logic as follows:
    If a subset of system X can model system Y, then X is a model of Y.
The conclusion is not a valid one.
 
Regarding simple/Turing-computable/mechanism/etc....
 
"I define a system to be simple if all its models are computable, or simulable." [EL 303]
"We shall say that a natural system N is a mechanism if and only if all of its models are simulable." [LI 203]
 
So, by these two statements, simple systems could be either formal or natural systems meeting certain criteria, whereas mechanisms refers specifically to natural systems meeting the same criteria. However, at several points [LI 186, EL 292, 303, 325, etc.] Rosen also indicates that simple system is a synonym for mechanism, which would seem to imply that simple systems refer only to natural systems. 
 
"A system is called complex if it has a nonsimulable model." [EL 306]
 
As for complex systems, he seems to make no restriction to it being applicable to either only formal or only natural systems. Rosen refers to Number Theory as a purely formal system that is complex [EL 293], and also, of course, to organisms as being complex [EL 269, LI 280]. So, complex seems to be able to be applied to either system type.
 
The conditional criteria for simple, mechanism, and complex is a condition on the models of the system, not on the system itself. Since formal systems as well as natural systems can have models that meet the respective criteria, the condition on the models does not, by itself, limit us to either only formal or only natural systems as being able to be categorized as simple or complex. Mechanism, however, always seems, in his writing, to refer to a natural system only.
 
"We shall further say that a natural system N is a machine if and only if it is a mechanism, such that at least one of its models is a mathematical machine." [LI 203]
 
So, a machine is a mechanism with an additional restriction. This restriction can be alternately worded as: the mechanism can perform the task of simulation [LI 185]. Simulation, in turn, involves the concepts of algorithm, program, hardware and software; it roughly means that an entailment structure of something else (call it the "original system") can be fed as software input into the simulator (which utilizes its own entailment structure as defined by its own "hardware"). In other words, the entailment structure of the original system is not put into correspondence with the entailment structure of the simulator (i.e., the "hardware"), as would happen in a Modeling Relation; instead the entailment structure (material, formal and efficient cause) of the original system is converted entirely into material and formal cause for the simulator hardware.
 
Since machines are defined as a subset of mechanisms, then machine therefore seems to refer to certain kinds of natural systems. However, just to note, Rosen also refers in passing to machines as being systems in either formal or material form. [LI 185, 194]
 
 
So, from all this, my view is:
1) to consider simple and complex to both be able to refer to either formal or natural systems whose models meet the respective criteria.
2) to consider mechanisms as being only those simple systems which are also natural systems.
3) to consider machines as being only those simple systems which are also natural systems and which further meet the criteria of having at least one model that is already a mathematical machine (i.e., it is capable of performing simulation).
 
 
Simulable, or Turing-computable, refers to the condition on the models, not on the original system. (Based on Natural Law, we do not "know" systems in the material world directly, instead we comprehend them via making models of them; therefore, any categorization of what we have comprehended can only be done so by categorizing our resulting models. [LI 203]) It further implies that the models in question are formal system models, since it is only of formal systems that we can sensibly ask if they are Turing-computable or not (that is, we can't sensibly ask if a tree or a rock is Turing-computable; we can only ask this of formal models).
 
[A copy of the original 1936 Turing paper, defining a Turing machine and "computable", can be found here:
I don't pretend to have studied too much of the formal argument in it.]
 
Turing-computability, as a condition, imparts at least two major restrictions on the formal models meeting that condition. First, the models must be entirely syntactic since a Turing machine is only a symbol manipulator - it has no capacity for  handling semantic referents. Second, Turing-computability imposes limits on the type of entailment structures that the model can possess. In short, the formal models must essentially be able to be codified into a program for the Turing-machine.
 
The nature of these restrictions are such that there are close parallels between: 1) the state-based nature of the simulator and the state-based nature of Newtonian mechanics, 2) the recursive nature of algorithm in the simulator and recursive nature of Newtonian mechanics, 3) the requirement of pure syntax in the simulator and the requirement of featureless particles in Newtonian mechanics, 4) the "largest model" being decomposable into (and re-assembleable from) atomic "smallest models" and the reductionism inherent in Newtonian mechanics. The upshot of these kinds of parallels (I think there are more but thats all that comes to mind) is that the range of systems that are describable in Newtonian mechanics formalism are essentially the same as those systems whose models are Turing-computable. This makes Turing-computability an eminently ideal condition to place on models for the purpose of distinguishing simple systems from complex systems. Insofar as these kinds of parallels extend to state-based relativistic mechanics and quantum mechanics, those systems of mechanics are likewise limited to describing mechanisms. Further, insofar as any known computing machine can be represented by a universal Turing-machine means that the criteria of Turing-computability has the broader implication of telling us which systems (via their models) are, in any real sense, computable, and which are not.
 
 
Regards,
Tim
 
 
 
-----Original Message-----
From: ROSEN Forum [mailto:***On Behalf Of John Kineman
Sent: Saturday, November 22, 2003 2:58 PM
To: ***
Subject: Re: Rosen & Ashby

Tim,

thanks. I was writing a reply trying to think about that very point, but ran out of time and deleted it. The gist of it was the same - that we get into a definition problem if we call that case a complete model, as you then have something that can't itself be fully specified, serving as a complete model for something else that otherwise can't be fully specified, and I think that results in a contradiction, or at least an untestable hypothesis. So... I agree with your current statement.
 
A corollary might be that two complex systems cannot precisely duplicate each other's behavior, unless, of course, we restrict them (or they restrict themselves) to simple behaviors.

I'm less comfortable talking about Turing computability. I don't understand its definition that well, so making the translation from that statement to the idea of a simple system or a mechanism is difficult. somehow it seems like just a mathematical way of describing a mechanism, but if you can point out an example of the difference, i.e., a system that is a mechanism but is not Turing computable, or vice versa, I would be appreciative. Same with definition of "simple." I tend to lump simple, mechanism, computable, and machine, and classical all as one; but perhaps that misses some subtleties.

JK

Tim Gwinn wrote:
Johnk,
 
After some more thought, I retract my comments regarding being able to use a more complex system to model another complex system. Such a "more complex" system would have characteristics that make it "more than" a model (excess degrees-of-freedom, etc.), and as such it now seems to me that it would therefore be improper to call it a model. My bad.
 
I also spoke incorrectly in my last reply regarding "no largest model". I should not have said that "no largest model" was necessarily an imprecise form of "no largest Turing-computable model", although that is sometimes the way it is used. Instead, only systems with only Turing-computable models will have a largest model, and that largest model will also accordingly be Turing-computable. So, "no largest model" will imply "no largest Turing-computable model"; so the former phrase is not really simply a shorthand version of the latter, which is what I wrote below. :(
 
Thanks for the discussion!
 
Regards,
Tim
 
 
-----Original Message-----
 -----snip----- 
[JK]
> - complex systems cannot be described completely by any other system,
> formal or realized.

[TG]
This is not necessarily true. It is possible that a complex system can be completely "described" (i.e., modeled) by another system if it is more complex than the system under study.
This is a philosophical escape. If one complex system can fully model another, but by definition no complex system can have a complete "largest" model itself, what is the meaning of the first part of the statement?  
 
[TG]
No, not a philosophical escape. :)
This is one of those cases where context is important. "No largest model" is a commonly used (even occasionally by Rosen), albeit somewhat misleading, shorthand version of the more precise phrase "no largest Turing-computable model". So this phrase does not impact on the ability to model a complex system with another complex system.
 
With this more explicit phrasing it might also now make more sense why the definition of complex system as "A system is complex if it has a nonsimulable model." and "no largest model" (in the more precise form of "no largest Turing-computable model") are closely related. Any number of Turing-computable models can be combined or concatenated into one largest Turing-computable model. Since a complex system will posses at least one non-Turing-computable model, the totality of that system's models cannot be combined or concatenated into one largest Turing-computable model.
 
--snip--

--
© 2003 John J. Kineman
all rights reserved