[NKS and epistemology - Symons talk at CNRS] - A New Kind of Science: The NKS Forum

A New Kind of Science: The NKS Forum

Pages:1

NKS and epistemology - Symons talk at CNRS
(Click here to view the original thread with full colors/images)

Posted by: Jason Cawley

John Symons is presenting a paper related to NKS at a CNRS meeting in Paris at the end of this month, at a conference on emergence and philosophy of science.

Symons is a philosophy professor at the University of Texas, editor of Synthese, has written a book on Dennett, and was a student of Hintikka. His homepage is here -

http://www.utep.edu/philos/Faculty/symons/index.htm

See also -

http://www.kluweronline.com/issn/0039-7857
On Dennett - ISBN 053457632X

He sent us an abstract of his upcoming talk, and I provided some comments, included below in the post that follows in this thread. Here is his abstract -

Logically possible mechanisms and the metaphysics of emergence

This talk begins by separating epistemological and metaphysical accounts of emergence in the complex systems literature focusing on Bar-Yam, Holland and Wolfram. Straightforwardly epistemological accounts of emergence can be found in the work of scientists like Bar-Yam and Holland. While Wolfram’s book has been criticized extensively, a generous reading can extract at least one important notion that has been absent from the literature on emergence, namely the idea that there exists a set of logically possible mechanisms that are wider than those we can capture via traditional explanatory strategies, for example via differential equations. Wolfram’s project can be restated in terms of a particular metaphysical construal of emergence. Equations, he argues, can never be the basis for deep insight into nature's workings. Equations are, at best, simply descriptions of an emergent level that is itself the product of simple rules. In place of equations he proposes simple computer programs. The old kind of science struggles with the surface of appearances, the new kind of science is the study of the simple recipes for the surface patterns. Since there's no way to peak below the surface of these patterns we must canvas the logically possible mechanisms that give rise to patterns. One way of doing this is via cellular automata, (CA) which are, after all simply logical representations of possible mechanisms. If one’s goal is to provide an account of all logically possible mechanisms, then CA’s will prove more effective than equations, since there will be CA’s for which no partial differential equation can be given. This talk will briefly relate this approach to emergence with discussions of the metaphysics of emergence in the philosophical community and will discuss its potential drawbacks as an explanatory strategy. In this new kind of emergentist science, physics becomes something like natural history rather than natural philosophy and its method is computer assisted observation, rather than the discovery of mathematical patterns and identities.

Posted by: Jason Cawley

The following are my own comments on Symons' abstract -

I think that Symons' understanding of the specific difference between mathematical equations and simple programs is partially correct and partially optional. He gets that simple programs are a wider case, allowing a greater variety of possible relations. The optional side is he notices they are in a way a more explicit model of underlying mechanism (saying what happens at each step, where an equation specifies a form that satisfies a constraint e.g.), and to him that puts it in a different epistomological box (a model imputed behind an appearance as generating that appearance).

I call this optional because equations can be used in models in that way as well - they needn't be restricted to surface appearances, and one can e.g. have equations that govern model internals, only bits of which rise to the level of observables according to that model - and because one can use a pattern created by a simple rule in the same way equations are usually used, if one wants (just matching a resulting built-up form to a form seen somewhere in nature).

His comment about NKS making physics more like natural history is partially true, one way of seeing the case where one can only watch what the thing does. But it seems to me it might be unnecessarily pessimistic about finding patterns in natural history (on the one hand), and perhaps overlooks how much regularity a simple enough underlying rule - even one with complex overall behavior - could imply, if for example there is one simple rule for fundamental physics. Models can have stable emergent gross properties which fully qualify as traditional natural laws (e.g. a rule may give rise to special relativity as an emergent property). These needn't be in the form of mathematical equations, it is true. But they aren't just contingent empirical history of a single instance, either.

"there exists a set of logically possible mechanisms that are wider than those we can capture via traditional explanatory strategies, for example via differential equations."

Yes. In general, models deal with isomorphisms between model elements and any abstracted set of characters of an imputed or observed system. The simple programs idea is to "scan" for cases stable at the level of some rule repeatedly applied. All models are simplifications based on regularities. Repeated application of the same rule is one kind of regularity, with different characteristics than other kinds. An example of a different characteristic is that rule to resulting behavior can be arbitrarily complex even at the logical level (internal to model, or to system). Which means modeling something differs from predicting it. In general, modeling is easier than prediction, because only some models are themselves predictable - predictable is a proper subset of model-able.

As for potential drawbacks as an explanatory strategy, models whose own behavior is not always predictable clearly have the status of hypotheses or guesses about an underlying mechanism. They purport to capture the real mechanism causing an empirical system to behave in complicated or seemingly unpredictable ways. One older tradition in epistomology has resisted the idea that theories have the status of guesses, trying to ground them in empirical evidence in more direct or constrained ways, rather than putting the "any isomorphism of abstracted characteristics" "guess step" in between the evidence and the model. Such views will have trouble accounting for NKS-esque models.

In addition, though, the tradition in epistomology that freely admits models are guesses has usually stressed the entrance of rigor or control over the variety possibly introduced by guesswork, through deterministic prediction and empirical falsification. (Note, however, that determinism within a model differs from imputed determinism out in the world - they are separable questions about which one can have different opinions. Though certain combinations of opinions about the two may be "more natural" than certain others). NKS does fit into this tradition, but in some cases the systems it deals with - while deterministic by hypothesis - are irreducibly hard to calculate ahead of time.

This leaves the role for falsification pretty much as it was, but it can change the type of observation that counts as a falsification. For example, a model that predicts a system will behave with a simple class 2 behavior is falsified if instead its behavior is clearly class 3. A model that predicts some statistical average of the system's behavior will be stationary is falsified if it isn't. These are different sorts of observations than the apparent position of Venus to 5 decimal places. But their role in the trial and error process of guesswork and revision is basically the same.

Can this sometimes trail off into uncertainty, though, if the model implies the actual system is simply behaving in a complicated (say, class 4) way, where one must simply trace out empirically its actual trajectory? Otherwise put, does NKS imply that some systems can only be "thickly described", even after an underlying rule for them has been found? Yes. NKS not only implies this, it seeks to explain empirical cases in which analysis fails to do more than this, by tracing such cases to repeated application of a complicated enough, but still simple, underlying rule.

Whether this is actually what is going on in any given case of a failure of analysis to yield anything more, NKS regards as (philosophically, not methodologically) an empirical question - or perhaps more properly, as a contingent possibility. We may guess that it is so in case X, and seek a rule that predicts that outcome for the overall behavior of the system. In other words, the predictability of systems is itself one system characteristic that can be checked against models, that can match or fail to match. Failure to match falsifies a guess, successful matching does not "verify" anything, but leaves the guess standing for the time being.

I hope this is useful.

Forum Sponsored by Wolfram Research

© 2004-2009 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