Correct, and perhaps a better way of saying that same thing. I have a bundle of natural system derived metrics for use in comparing the behavior of models with individual instances of physical systems. There's still a gap in making the system features I can extract from nature connect with what statistical modelers will want to plug into their models, however. The gap seems to represent a significant real disconnect between the designs of individual physical systems and the designs of computer models of them.
One of the features I think would help the most to make computer models more similar to individual instances of physical systems is that every sub-system act as an individual, and be given the behavior of 'exploring' it's domain. System 'exploration' is a property that I think could be given mathematical definition, but has not yet as far as I know. It basically means 'variation at the fringe' where the number of experiments in the region of successful experiments is variously self-controlled. The problem for modeling that I see is the question 'fringe of what??'. Because a computer model is iterative there are loops of effects. In nature, loops of effects like that bundle into individuals that act as wholes. I know how to identify them and a little about how to explore them, but not how to write programs to emulate them. Perhaps that's because I see their every feature to be essential, and maybe that's not quite necessary for building somewhat useful models that carry more of their authentic structures? Phil Henshaw ¸¸¸¸.·´ ¯ `·.¸¸¸¸ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 680 Ft. Washington Ave NY NY 10040 tel: 212-795-4844 e-mail: [EMAIL PROTECTED] explorations: www.synapse9.com > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Marcus G. Daniels > Sent: Wednesday, April 04, 2007 6:04 AM > To: The Friday Morning Applied Complexity Coffee Group > Subject: Re: [FRIAM] One of my projects > > > Phil Henshaw wrote: > > It's missing the scientific reality component though, the step of > > identifying how what we think is *different from* reality. > If it's possible to identify how a model is thought be not > representative of reality, then new mechanisms can be added. Then, > differences in model behavior can be quantified both relative to the > simpler model and relative to some set of data metrics that > characterize > known `reality'. > > Still, don't you need some sort of method of a) validation > of results > e.g. Doug mentioned the 1918 pandemic flu: > http://www.mail-archive.com/[email protected]/msg01646.html and b) finding patterns in the discrepancy in the results found? Suppose one posited that temperature had some role in the distribution of a pathogen in some environment. Data could be collected from actual cool and warm environments (by experiment or from a historical account). Meanwhile a model could implement the hypothesis about the role of temperature. If the distribution of the pathogen in the model doesn't match the data, then the model can be elaborated or changed. Or one can consider collecting more precise data if the model suggests finer distinctions in outcomes than could otherwise be witnessed. Rinse and repeat until the model acts like reality. Now try predicting distributions from new datasets. No earthshaking changes to scientific method here. It's just that the predictions can be high dimensional if needed and model mechanisms aren't constrained to be analytically tractable. ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org
