Re: [Swarm-Modelling] Announce: metaABM 1.0.0

[Miles T. Parker]

On Nov 25, 2007, at 11:24 AM, Marcus G. Daniels wrote:

> Miles T. Parker wrote:
> > > Using a general purpose programming language, there's the  
> > > possibility of inferring a DSL by having a computer find or refine  
> > > minimal programs that explain some pattern of input and output data.
> >
> > If you find that palatable, then surely one that is crafted by  
> > humans is not such a bad thing? :)
> To me the palatability of it is just a matter of how much data is  
> available.   If there is a lot of data, and it is complex, then a  
> computer is a tool to find order in it.  If there is not a lot of  
> data, then sometimes perhaps all that can be done is invent stories  
> to think about ways a system might work, and how one might obtain  
> data.   But unless there is the possibility some data, however  
> qualitative that can ground the theory, the activity is not  
> empirical science.
...
> Coefficients for formulas are just a special case of functional  
> programs.  Both can be fit to data, and compressed.
> The story doesn't need to be invented by a human to be rationalized.  
> _______________________________________________


But are you claiming that that is privileged over the age-old process  
of the observing nature intuitively, noticing regularities and then  
verifying that through measurement? Or that a "limited" DSL driven  
model that matches "reality" as well as some automated reduced model  
is somehow less preferable?

> > > Ultimately what should matter are rules of predictive utility, not  
> > > pretty stories..
> >
> > Can we *predict* what direction a flock of birds will fly?
> It seems conceivable that by using high speed 3d digital cameras,  
> data on weather conditions, info about the time of year and  
> geographic location, and some health/performance stats on the birds  
> themselves, radio collar info of past movements,  it might be  
> possible to find regularities after many observations.  Not 100%  
> prediction of course but maybe better than random.   And if  
> regularities were found, it seems conceivable that some of those  
> regularities might have to do with group size, the physics of  
> drafting, recent food intake, etc., and maybe even give novel  
> insights to an ornithologist.

Well sure, but I'm simply claiming that you can *also* make novel  
insights w/o all that work. My intuition is that understanding of  
complex systems at this point is so limited that going to all that  
trouble would not be the most efficient way to improve knowledge. And  
in fact, that is exactly what Boids did, work that is now being  
confirmed with more quantitative methods. You can't make very  
effective decisions about what avenues of research will be most  
fruitful without relying on informal intuition. And I guess I am not  
one that sees prediction per se as the only goal of science -- I like  
explanation. I'd say more strongly that the POV of science as only  
being valid when it can predict a *particular* outcome is too narrow.  
Perhaps that isn't a bad description of the difference between  
Simulation and Modeling, and I think why I resist the lumping together  
of "Agent-Based Modeling *and* Simulation" with no distinction between  
the two.