Rosen and von Neumann - Part 2

[Tim Gwinn <***>]

Howard,

 

(continued from Part 1)

 

2) Another topic on which Rosen makes claims with regard to von Neumann's ideas is the inequivalence of computation and construction. Your remark:

> In Life Itself, p. 234, he says it
> "parenthetically," and with no references, as if it were an
> established fact that von Neumann confounded simulation with construction.

The full paragraph and the following one, which appears near the end of section 9E of the chapter on "Relational Theory of machines" in Life Itself:

    "Indeed, under this interpretation, [9E.5] describes a simulation of f by Phi, rather than an entailment of f by Phi. Nothing could more starkly illustrate the anomolous character of simulation and the mischief that can arise through the confusion of causal categories. I remark parenthetically that the confounding of simulation (computation) with construction, which lies at the heart of, e.g., von Neumann's well-known discussion of "self-reproducing automata," arises precisely here and rests entirely on the equivocal and inconsistent hardware/software distinctions to which I have just called attention.

    The distinction between hardware and software, then, is a completely meaningful one, but it must be maintained consistently when decoded back to a natural system. What I have concluded, then, is that it cannot be maintained consistently between the two diagrams [9E.4] and figure 9E.1. Hence, if the former is a mathematical machine, the other is not, though it is a perfectly good mechanism." [p. 234]

 

As Judith has already pointed out, there is no skullduggery here. He mentions it as an issue which is tangential to the main goal of the book (as was 'evolution'). Thus it warrants little more attention than it receives there.

 

Regarding whether or not "von Neumann confounded simulation with construction", the issue here as I see it is that, according to Burk, there were two models of "self-reproducing automata" proposed by von Neumann: the kinematic automaton and the cellular automaton. I believe it is in regard to the latter model, the one which was developed into more detail (e.g., the 29-state machine)by von Neumann,  that Rosen is addressing his remarks.

 

Kampis makes very similar remarks about the "mischief" of simulation, in Self-Modifying Systems in Biology and Cognitive Systems, and states it quite well:

    "It is truly genial to simulate the very machine on which the process is running, and to include it into the reproduction, as von Neumann did. It is genial but impossible. For by doing so, we give rise to another machine, a new one which is not again reproduced, and so on, ad infinitum. We may define new and new simulations but will never transcend the limits of the system. It is the same as with self-modification. An algorithm (a system) cannot modify or reproduce itself; it is only another system that can modify or reproduce it. Self-transcendence, which would be required in both cases, is within the formal domain impossible.

    What is misleading and is responsible for the unfounded hopes for self-reproducing automata is that the ultimate Machine, remains hidden during the simulation. It has no transition function (within the given way of description), or, better to say, its transition function does not interfere in a visible way with the computed transition function. It is, however, essential that there is such an embedding automaton (non-reproduced) in which the others are reproduced; otherwise the process could not proceed.

    But how could we be misled at all? It must be more than clear by the definition, that in a state-determined system, and consequently in any computational system, it is only the states of the system that can change and nothing else. Consequently, only states can be 'reproduced', and thus reproduction reduces to state occurrence, or state propogation, in other words, to the domino type of reproduction - whatever tricks we use. If we look at cellular automata, no matter how complicated they are, we only see state patterns, just as in the most primitive systems (like the Wolfram automata). And what is more important: the cellular automaton itself sees only these state patterns.

    The notion of self-reproducing automata are results of an intricate interpretation of these configurations by an external interpretation theory (the recursive function theory of automata). But the very automata interpret them as states, they just do not know about our notions. Only a few people realized this before. In a remarkably frank passage, M.A. Arbib, otherwise an advocate of computational concepts at any rate, comments in his 1969 book: "Did [we] really construct machines or [have we] only given copying routines that are somehow different than construction?" W. Stahl (1966) said this: "In it [i.e., in von Neumann's automaton] self-reproduction is not of an automaton but merely of a configuration of automata states...". Both authors then went on, forgetting their own remarks, to construct automaton systems. Lofgren (1972) was explicit to speak out that automaton reproduction is possible only in a suitable automaton (or automaton -like supporting environment). R. Rosen (1959, 1986b) made similar remarks.

 

Thus, the notion of "self-reproducing cellular automata" is rather deflated by this account, and its relevance to the physical-realm question of self-reproduction is questionable at best. This does not indict von Neumann the person - it only indicts the relevance of that class of models.

 

 

3) This discussion also precedes, and is distinct from, the entire question of realization - the flip side of "computation and construction". This involves its own set of problems. As you state it previously, "he [von Neumann] simply assumed that his abstract concept could be materially implemented." That is fine. But that in itself does not preclude the examination by others of what problems - both logical and practical - might occur in attempting a realization. Some of those problems are examined by Kampis. Some also by Rosen, such as in AS [p. 419]. Rosen discusses the inequivalence of causal categories. Namely, that the inequivalence of these categories means that manipulation of efficient cause in a computer cannot necessarily be translated into manipulation of material cause in a physical system. Again, these are not attacks on von Neumann himself. 

 

 

4) Another claim by Rosen is that his and von Neumann's definition of 'complexity' differed significantly.  I think this has been already discussed sufficiently that I need not go into it. 

 

 

As I see it, these are the four main topics where Rosen discusses and/or criticizes von Neumann's models or ideas. I see nowhere where Rosen "falsely accused" von Neumann in any of these areas, nor do I see that Rosen committed any ad hominem attacks by these remarks. These topics relate to several different issues with regard to different aspects of

 

If von Neumann did himself raise concerns about some of these same issues (such as "...at which point the analogy ceases to be valid"), then to me that only reinforces the notion that the claims and criticisms Rosen (and others) raised are about entirely legitimate concerns.

 

Tim

 

> -----Original Message-----
> From: ROSEN Forum [mailto:***]On Behalf Of Howard
> Pattee
> Sent: Tuesday, March 30, 2004 9:34 PM
> To: ***
> Subject: Re: Howard's challenge #1
>
>
> Judith,
>
> I apologize for calling Bob's friends (is that OK) by other
> names. I'm generally in agreement with most of your long
> responses, but I think you have lost the issue that I first raised.
>
> Judith: The biggest problem my father had with von Neumann's
> theory was not the secondary stuff, which as you mentioned, seems
> to agree quite a bit with
> "Rosennean" findings-- it's the primary concern that made my father reject
> the whole shebang. Von Neumann's basic premise was what he disagreed with
> entirely and wanted to distance himself from.
>
> HP: I understand all that, but it is not the issue I raised. All
> I suggested is that it would be best for Bob's reputation if his
> friends, colleagues, relatives, (what should I call them?), would
> not continue to defend Bob's specific no-longer defensible
> argument with von Neumann over self-replication. This has nothing
> to do with whether von Neumann's had a good philosophy or even a
> good model of replication. It has nothing to do with the
> technical meanings of "equivocation" or "confounding" or
> "invalidation" or any other words that Bob has used to discredit
> von Neumann's model.
>
> The issue, to put it bluntly, is whether Bob falsely accused von
> Neumann. In other words, the science of the matter is not the
> issue. It is the ethics of Bob's refusal to engage his critics on
> the issue and his unresponsive, uncritical repetition of
> essentially the same charge for over 40 years.
>
> If this is the case, then I think you would understand why it is
> not good for Bob's reputation to continue to justify it by a
> diversion, explaining that the biggest problem Bob had was that
> he just disagreed entirely with von Neumann's basic premise
> (whatever that is?) and that led him to reject the whole shebang.
> That is not considered a good scientific or philosophical
> argument, although it may be a good psychological explanation.
>
> I have given brief quotations from von Neumann showing that Bob's
> assumptions (in his paper, Bull. Math. Biophysics, 1959) were
> mistaken. There are a lot more. I can also find Bob's
> restatements of this misinterpretation in many papers. I think it
> had become a habit. In Life Itself, p. 234, he says it
> "parenthetically," and with no references, as if it were an
> established fact that von Neumann confounded simulation with construction.
>
> I would like go on to discuss why I think some of von Neumann's
> ideas were consistent with Bob's, specifically his Theory of
> Games. I think that iterative game theory might fall in the
> category of Bob's impredicative models.
>
> Howard