[NKS and a typology of systems theory distinctions] - A New Kind of Science: The NKS Forum

A New Kind of Science: The NKS Forum

Pages:1

NKS and a typology of systems theory distinctions
(Click here to view the original thread with full colors/images)

Posted by: Jason Cawley

After a recent conference several of us attended in Champaign, Stuart Umpleby ( http://www.gwu.edu/~umpleby/ ) asked an interesting question about NKS. He has studied the history of various forms of systems theory. One of his students, Eric Dent, came up with a set of concepts that he thought marked off systems theory approaches from earlier science. The idea was to trace the variety within systems theory to different subsets of these common conceptual distinctions. What Umpleby asked is, in effect, where NKS would fit in this scheme, which of these concepts does it use? I thought some here might be interested in what I came up with in response.

The eight concepts Dent identified were holism rather than analysis or reductionism, relationships rather than entities, interaction with environment as opposed to closed or isolated systems, indeterminism as opposed to predictability, circular casual loops rather than linear casuality, self organization, observation as part of or directly related to systems, and reflexivity or interaction between a system and what is known about it. Some of these are common to essentially all systems theory traditions (e.g. holism), while others are only found in a few (e.g. observations incorporated aka "constructivism"). These categories stem from Umpleby and Dent, from a 1998 paper. What follows is my own assessment of NKS in these terms.

Holism/Reductionism - NKS avoids classic reductionism because the systems it studies typically depend on complex interaction of the system's parts. It does try to trace holistic behaviors to relations among formalized and simplified abstract parts, but sees the patterns of interrelation as the essential thing. It finds principles that apply to these patterns that are indifferent to the underlying elements, and still more general principles that apply without regard to any specific set of rules for relations. The limited commitment to reductionism that remains is just to trace these phenomena to concrete patterns of relations. The same is found in Bertalanffy (explaining away "vitalism"), or in Ashby's desire to find mechanisms. So, NKS shares the systems change, here.

Relationships - At the base level, NKS definitely focuses on relations among formal elements rather than on those elements themselves, which are typically taken as completely simple. NKS privileges the base level of analysis of any given formal system in this respect, though. It does not generally apply the relationship rather than entity idea up multiple hierarchical levels (as e.g. a thoroughly "relational" Hegelian system might). NKS speaks regularly of a given pattern of primitive relations according to some rule as an entity and studies the characteristics of that
entity. Above that level it certainly classifies and looks for common principles, but e.g. it treats rule 30 as a "critter" in its own right, not something delineated by its relations to other elementary CAs. Yes at the bottom level, no as to a whole hierarchy view.

Environment - NKS does not typically focus on interaction of a system with its environment on an ongoing basis. Typically the environment is thought of as providing a set of initial conditions out of some possibility space. While some NKS work looks at things like perturbations or interventions, that is not the main subject. The system is set up and just goes. Continual interaction with the environment is treated as obscuring the inherent, internal behavior of the system. Basically no, NKS does not share this systems view.

Indeterminism - NKS seeks to account for phenomenal indeterminism, but without positing underlying randomness or primarily considering ensembles or statistical properties. The behavior of irreducibly complex systems is unpredictable, by methods other than direct emulation and formal experiment. It is quite common that one does not know what a typical NKS system will do. But not because any randomness has been injected at the bottom or posited from outside. It does not use monte carlo methods or random walks. Insofar as NKS shares the interest in phenomenal indeterminism and regards it as real and ineradicable when the system is sufficiently complex, I'd say NKS shares the systems view here. It just adds a particular take on the cause of observed indeterminism, trying to trace it to computational irreducibility.

Causality - NKS uses circular causalities only in the sense of repeated applications of simple rules, recursion at a formally base level, mappings onto a pre-existing set or a close analogue. Beyond this it considers a uni-directional flow of steps, essentially a linear casual train, but of many variables or elements rather than only a few. As a special case it does consider some exceptions - multi-way systems e.g. It maps the entire evolution of a system rather than retaining a single state and updating only that copy, or in other words NKS formalisms generally leave a trail of the system's previous states. Yes (base recursion) and no (definite dynamic trajectory).

Self Organization - NKS is focused on dynamics of its systems rather than primarily on equilibria or other limit behaviors. But it definitely sees the main characteristics of its systems as emergent properties of complex interactions of elements and subsystems. In addition to explicitly considering self-organization from random initials, more generally it classifies systems largely by typical behaviors they tend toward, not dependent on a few outside causes or initial conditions. So, other than the minor variation that the full dynamics are the focus, not just limit
behaviors, yes NKS does share this systems view.

Observation - NKS examines typical methods of analysis as data processing systems similar to the rules it studies. The Principle of Computational Equivalence relates perceived complexity to system behaviors as sophisticated as those available to analyse them. It views reducibility as a kind of race between the computational resources of the observor and those inherent in the system. Other than this, however, it does not incorporate the observor into models. The behaviors of NKS systems are fully specified,math-like entities, completely reproducible. NKS does consider issues like behaviors appearing differently in different representations, but expects its main principles to apply across all available representations. E.g. a universal system will look complex to any observor. NKS does not really share the constructionist view here. But not because it does not consider the matter. Rather, it sees possible principles beyond constructivist variety.

Reflexivity - NKS does not consider this seriously. It does not treat matters arising from self-reference as essential. Rather, it sees universality as the critical issue. Reflexivity does arise in universality in the form of a similar space of possible emulations, including self emulations. But that is all. NKS does not consider seriously any effect of experimenting on a system or observing it, on the behavior of that system. Its systems are formally specified. Rule 110 can't do something different because you are looking at it. The limits on our ability to say what it will do have to do with its universality and the inherent complexity of its internal relations, not on its interaction with an observor.

So, NKS incorporates holism, indeterminism (as emergent), formal relations and recursion at the lowest level, self organization, and crucial computational similarities between observors and observed - but not continually embedding in an environment or reflexive alteration by being observed. The above applies to NKS as it stands to date. One can imagine incorporating other bits of previous forms of systems thinking, though in some cases it would miss the point of NKS (e.g. indeterminism as emergent).

I hope some of this is interesting.

Posted by: Stuart Umpleby

Thanks very much to Jason for describing NKS in terms of Eric Dent's dimensions of systems science. I did not fully understand all of the language and references, since I am new to NKS, but on the whole his reply is very interesting and useful. There is one issue that I would ask Jason to think about further and that is reflexivity. It seems to me that NKS functions at two levels, maybe more. The first level is what I would call programming or cellular automata. Here NKS uses a different mathematics than, say, calculus. There is a second level, however, and that is the discourse about a "new kind of science." This thread is part of that discourse. This discourse has at least two purposes -- to reflect on possible social and scientific implications of NKS and to generate some new lines of inquiry in NKS.
In recent years a new type of literature about science has emerged. I would refer you to the books, "The Rhetoric of Economics," and "Writing Biology: Texts in the Social Construction of Scientific Knowledge." This is a literature created originally by English majors who chose to analyse not merely novels and poetry but also the writings of scientists. They have gone beyond analyzing plots and character development to examining the rhetorical strategies used by scientists to persuade their peers, the public, and funding agencies. (Each audience requires a different rhetorical strategy.)
It may be that one needs to shift the type of analysis when one engages in a different mode of reflection, but I think the different modes of reflection are there. Will awareness of this fact alter the way that NKS is done?

Posted by: Jason Cawley

A bit meta, so much so that my first inclination is to ask for another sentence. Instead I will give you my general thoughts on the two cultures boundary and what I think NKS has to say there. That may be a wider subject than you intend. If so bear with me - you can always narrow my focus again with a follow up question.

In general I prefer to get my philosophy straight, rather than roundabout filtered through general theory and literary criticism. Rather than the english major quoting Peirce, I'll read Peirce. Rather than the nth version of applying Foucault to X, I'll read Foucault and the things he himself read. If I want to consider multivoicings in a text, I'll read Bakhtin and Plato, not a sociologist reading high school biology textbooks. Of course those applying it may notice various things peculiar to their own subject, and add value that way. But it doesn't add all that much to be told isomorphism is like metaphor.

Then there is the occasional tendency to willfully miss the point, where advanced theory applied turns into a Strunk and White analysis of Shakespeare, or a political talk show shouting match about electron orbitals.

My general attitude toward the subject is that a little philosophy is frequently not very helpful, and the solution is to increase the dosage. By that I mean, it is not infrequent for someone exposed to any single basic philosophic idea - or to some small connected set of them - to develop the notion that they are critical to everything, and to think most other people don't get it, and to project imaginary significance to, well, unlikely places.

By increasing the dosage I mean learning every aspect of a score of philosophies, and the arguments and historical transitions between them. Then when an idea comes along, it is often not entirely unfamiliar and at least its genus is understood. A man standing between half a dozen opinions and judging them all is free, in a way a man who can't see outside of one is not.

The crossover ideas - both ways - between the "two cultures", are sometimes frankly pathological. I've seen the serious book length analysis that declares all things self similar are good, all things simple are bad, and society must become self-similar tomorrow, lest the world perish. The same for "equilibrium", in more than one form. Some going the other way, socially constructed this and that, reminds me of nothing so much as Hannah Arendt commenting on "German Physics in four volumes". The fates spare us that strand of shallow relativism.

Now, are there constructive things I think NKS has to say about the zone of crossover? Certainly. I think some on the social science and humanities side have been dealing with systems that are in fact complex for a long time, and some of them have developed ways of dealing with systems that do not easily reduce. And with it, a justified resistance to some forms of reductionism.

I think NKS manages to classify system behaviors, at a purely formal level, that map to different practical methods, not merely within the sciences but farther. Sometimes you have to watch the system and see what it does; nothing else will work. Sometimes thick description, historical narrative, is necessary to track a critical chain of events. Looking for simple explanations in such cases is a misplaced faith in reducibility.

When real historians objected to a half dozen overly simplistic philosophies of history (by philosophers or historical amateurs), they were basically just saying it is often class 4 not class 1 or 2 behavior. I recall the college line about Anthropology, that it is where short pithy stories get to count as science. About some things, thick description is all one is going to get. One might ascertain certain substructures and a few of their rules of interaction. But what actually comes from those interactions may remain a long empirical story, irreducibly. History is often like that. If you read the extensive historical notes in the NKS book, that is often what you will see.

Or take apparent randomness. NKS shows one should expect it in many cases where multiple elements interact in non-additive ways. One should not expect it to always yield to analysis, to e.g. always be composed of multiple cycles or stacked self-similar forms. Those can give rise to apparent randomness that is analyzable. But do not exhaust randomness, much of which is irreducible. Practically speaking, we resort to statistics and probabilities in such cases, describing only gross features as the only stable ones.

It may be a question whether, e.g. some economic series involves analyzable apparent randomness (e.g. overlapping rule 90s from complicated initial conditions, or multiple cycles with various lags) or real randomness (e.g. rule 30 like intrinsic randomness generation, or a random walk). But neither thesis is absurd; it is an empirical question. Views that regard everything as a priori reducible are just wrong. Views that think any apparent randomness implies no underlying regularities, or still stronger no simple rule generating that randomness, are just as wrong.

Are some of these things changes, at least from the self understanding of certain branches of science c. 1950 or 1900? Sure. If one looks back one usually finds some sensible people saying so, though often not fully understanding the reasons. They may understand it as a methodological divide, or as a realm of history rather than of nature, or get a bit more sophisticated and think in terms of necessary preconditions for reduction or analysis. NKS can inform that sort of two cultures boundary discussion by making clear what sorts of behaviors can and typically do arise even in simple formal systems. Making the concrete question in some given case, what kind of system do I have in front of me, what class of typical behavior?

I don't think this amounts to anything like reflexivity. Some previous system theory thought that phenomena clustered around self-reference were critical to various interesting problems. This was part of a general atmosphere, in philosophy and in various sciences, in say the first two thirds of the last century.

I doubt I need to recount all the examples. We had paradoxes of set theory, Popperian openness founded on the difficulties of self-prediction, self-duplication as a proposed definition of life, feedback based control theory, etc. The philosophic foundations of that thinking go back to Fichte and Hegel, who also saw reflection as a kind of microcosmic infinity (making a virtue of what had in ancient times been considered an infinite regress, etc). And all of this implicitly referred back to a still older frame (as, to be sure, a critical innovation in some ways stepping beyond the old view) in which the important distinctions were seen as inside vs. outside, or mental vs. physical, or subjective vs. objective.

Now, NKS simply does not think those issues are fundamental. Yes there are possible paradoxes one can construct involving universal systems, to show various things about them - some of which go back to the above period. And emulations or encodings are used to prove things. But to NKS, irreducibility is the primary phenomenon, not self reference or chains of emulation. The claim is that beyond a very low threshold, the behavior even of simple formal systems can get as complicated as anything in our universe.

That happens even in formal systems, "inside". It is observed and claimed to happen for the same reasons in natural ones, "outside". The epistemological problem has been rotated. It no longer runs observer to system, but system rule to system behavior. The cause of e.g. difficulties of prediction, arises because the behavior of a system even given its fully deterministic rule is not trivially knowable from that rule. "But that is just logic"; sure, but enough logic, and involved enough logic, is not trivial it is hard. This rule to behavior gap is seen as arising in purely formal systems or in natural ones in precisely the same way. The epistemological problem happens at the level of software.

And that simply means the fashion of privileging the observer -observed dyad is no longer all the rage. Nor witty ways of tying it into knots the real source of complexity. Self reference is not the problem, complexity is. Self reference in the presence of complexity is certainly a problem (we knew that, that is data), but it is so because the complexity is a problem already, even without anything else (self reference, inside-outside gaps, etc).

I hope at least some of that is comprehensible - helpful would be gravy.

Forum Sponsored by Wolfram Research

© 2004-2008 Wolfram Research, Inc. | Powered by vBulletin 2.3.0 © 2000-2002 Jelsoft Enterprises, Ltd. | Disclaimer
vB Easy Archive Final - Created by Xenon and modified/released by SkuZZy from the Job Openings