Re: MR as ontological, intelligence

["Tim Gwinn" <***>]

Hi John,

 

After thinking some more about this, and re-reading sec. 5L of Life Itself on Category Theory, I am fairly convinced that the encoding/decoding are not identical with the causal relations between the internal model, sensory inputs and control outputs.

 

The encoding/decodings in the MR are covariant/contravariant functors in Category Theory. Functors specify the nature of the relationship, or association, between the objects and morphisms of one category and the objects and morphisms of another category. Rosen speaks of the functorial images as a kind of "metaphor". I think this is why he uses the term "dictionary" when originally describing the Modeling Relation near the beginning of LI: a dictionary specifies the nature of the translation between objects and morphisms (e.g., system elements and entailments) in one system in an MR and those of the other system in an MR.  The encoding/decoding is thus not an entailment between systems, but rather a kind of schema for a particular way of associating systems.

 

In the case of the bird example, that schema or set of dictionaries - the encoding/decoding - would be not the causal relationships per se between the model and sensory inputs and control outputs; but rather, are the particular way in which those inputs and outputs are schematized to and from the internal model. If a sensory input, for example, gets mapped differently to the internal model, either by change in wiring or thresholds or similar alteration,  then there is still a causal entailment between the input and model. However, the encoding (the translation, or, what that same sensory input now "means" to the model) has been altered.

 

 So, in this view, the encoding/decoding are not separate physical structures from the causal entailments nor are they the causal entailments per se, but are instead a matter of the particular manner of the organization of those causal entailments between model, sensory inputs and control outputs.

 

Regards,

Tim

 

 

 

-----Original Message-----
From: ROSEN Forum [mailto:***On Behalf Of John Kineman
Sent: Tuesday, August 26, 2003 11:49 AM
To: ***
Subject: Re: MR as ontological, intelligence

Yes, very interesting. The fact that he cited only two kinds that we know about supports the idea that he did not consider the encodings another kind of entailment. But on the other hand it doesn't rule that out either. My guess is that you are correct that term is used in a standard manner such as "something that causes something else," like a tail follows a dog, hence en-tail-ment; in which case it wouldn't be a very restrictive definition. It seems the context is what distinguishes these things critically, and on that there are three contexts referenced, NS, FS, and abstraction.

It gets deeper the more we dig, eh?

jkineman

Tim Gwinn wrote:

Hi John,

 

I would agree that, in the example of the bird and its internal  (anticipatory) model affecting its behavior, that there are indeed causal entailments between the internal model, the sensory inputs, and the control outputs. 

 

I find this a very curious distinction from Rosen's comments on the "epistemological" MR that I quoted below. I have to think some more about it, as to whether these causal entailments are isomorphic to the encoding/decoding mappings of the embodied MR. I suspect they are not, but I need to ponder some more.....

 

As for a definition of "entailment", I do not know if he gives an exact one anywhere. It refers to a certain kind of relation [LI 46, 56]. I want to say that he uses the word in a fairly standard fashion. In Essays (p. 89), "Indeed, inferential entailment (between propositions) and causal entailment (between external events) are the only two modes of entailment we know about." Hope that helps.

 

Regards,

Tim

 

-----Original Message-----
From: ROSEN Forum [mailto:***]On Behalf Of John Kineman
Sent: Monday, August 25, 2003 9:43 PM
To: ***
Subject: Re: MR as ontological, intelligence

Tim,

Sorry I missed this and commented on later posts without modification of the terminology.

Yes, I do recall reading that now, but it didn't compute in my brain as a clear distinction with the word "entailment," so I suppose I transfered the terms being a bit unorthodox. I think they are more generally called mappings, which is a term general to all the arrows in the diagram. So, I need a definition of "entailment" to know more.

They are indeed not part of either NS or FS, and that distinction is important. They are associated with the necessary "acts of abstraction" which are not represented in the MR itself. This gives them a different character from entailments that seemingly operate without the particular abstraction context of these mappings.

But, and this gets really subtle, we are making this distinction to get the modeling relation to work as an analysis. As such the meanings tend to change with context. A natural system might be, say a bird. Certainly encoding and decoding take place in the bird, which Rosen says builds and responds to internal models - thus the bird is capable of these "acts of abstraction," which would normally appear as arrows between FS and NS. Hence, also, the dynamics of the bird is modified by such models (they are used for feed-forward control). Hence the models have causal effect through these arrows (decoding). So they are legitimately thought of as causal mappings, I would conclude..

In the epistemological use of an MR, say I model a bird, MY model's endoding and decoding relations are not causal to the behavior of the bird (except in larger systems, such as bird management - also it is well known among behaviorists that observers, who are seen, can affect observed behavior, etc.).  But "entailment"  it is inescapable when I consider what goes on inside the bird, where it has an internal modeling relation going on, yet we represent the WHOLE bird generally as a Natural System and say everything that goes on inside the bird is called "causality" which is entailment.

So, my inclination is to think of these distinctions as purely contextual. In the case of encoding and decoding, we don't understand their context. Even when summarized in a FS as "causality" there is an implied context for abstraction that hasn't been accounted for.

Now in my lose writing I called them "functional entailments" whereas I meant "causal mappings from functions to dynamics." But I confess that I don't really know the difference between these two phrases, if there is one other than the context they occur in: "causality" used for NS, "implication" for FS, and potentially causal mappings (as in representing what happens inside organisms as NS) for acts of abstraction that feed back to the dynamics. I'm not sure that "causality" and "implication" are any different than each other, except for their context. Perhaps keeping the convention of distinguising the contex of a mapping with its label -- "causality" for NS, "implication" for FS -- we chould say "influence" for encoding/decoding, or just relationship. But are they really qualitatively different? They all say "A affects B."

Can anyone say more on this????? Is there a clear definition of the term "entailment"??


Cheers,

J. Kineman



Tim Gwinn wrote:

Hi John,

 

I think I now see perhaps the root of some differences in the following (from below):

------------

This surprises and worries me. Is it not accepted that Rosen discussed entailment between natural and formal systems?? I suppose if a strictly epistemological view were taken, that might not be obvious. But in that case, one would be talking only about the philosophy of science, not anything we would otherwise associate with nature. Is that your perspective???
------------

 

The encoding/decodings between systems in an MR are not entailments. In Life Itself (p. 52), he refers to them as "dictionaries", where "encoding" is short for "encoding dictionary, and "decoding" is short for "decoding dictionary".  

 

More directly, from Essays, p. 159 [while describing the pieces of an MR]:

"The crucial ingredients are the arrows 2 and 4, which I call encoding and decoding, respectively.....They do not fit entirely inside either the object system or the model; they do not represent entailments, nor are they themselves entailed. They manifest what Einstein...once called "free creations of the mind", on which he believed science depends....The only condition on them is that they bring the two entailment structures into congruence - that is, that they satisfy the commutativity condition...." [bold added]

 

So, these dictionaries arise from the mind of the person creating the MR, but are not entailments of, or entailed within, the MR. So, an MR requires  [(object system) + (formal system) + (modeler-with-mind)]. When you gave the example of the scientist engaged in an MR with some system, and suggested that the whole (scientist + object-system) could be considered a natural system, I agreed that it could. But that new system is now a much bigger chunk of the world than just the MR that was within the original scenario: it includes the entire mind of the scientist. Without that mind to impute those "free creations", there is no MR.

 

Below, where you use realizations as examples of entailment between F and N systems, I do not think these demonstrate entailment beween F and N in the MR itself, but rather entailments in an application of an MR to a process called "realization".

 

Regards,

Tim