Re: models in the wild (was Re: [Swarm-Modelling] Re: [Swarm-Support] Repast vs. Swarm)

[James Marshall]

Heh heh,
  and some people I've worked with (me included) have called what they
do "simulation modelling". Does that mean modelling a simulation?
  Anyway, various people have already mentioned this, but I think it's
worth emphasising: two issues for ABM vs, for example, DEs is about a
common language, and about reproducibility. DEs have a common language
that is pretty universal... if you've been trained in science or
engineering you can (hopefully!) interpret them. I think ABMs still
haven't reached the same level of consensus on the language, which makes
people suspicious of them... although the concepts are probably much
more universally shared, if people don't call them the same things then
confusion arises. Also reproducibility takes more care with ABM... a DE
model can usually be described much more concisely than its 'equivalent'
ABM, and you won't be able to publish a paper with results/insights from
a DE model without publishing the DEs! Unfortunately it's still quite
possible to publish results/insights from an ABM without even making the
ABM publicly available, and people do (present company excepted of
course). That in itself undermines the scientific nature of ABMs as
commonly used, because they're unreproducible.
  Of course, this is what Swarm set out to address... it's also nice to
see different "toolkit communities" getting together to address the
common language issue across toolkits.
  Another thing that I think adds to people's confusion is the term
"agent"... this has connotations for many people far beyond its intended
meaning in ABM (Glen already noted this). E.g. engineers have asked me
"What is an agent? What do they do that's so special? How's it any
different to, e.g. an Ising model?". I think there's an argument for
sticking to Individual-Based Model or maybe even something new like
(using Glen's suggestion) Component-Based Model or (my favourite, with
tongue firmly in cheek) Bottom-Up Model, rather than ABM to help address
this, but we're probably stuck with current common usage...
  Forgive my ramblings,
    James

> Rick Riolo wrote:
>
> > Just one more note:
> >
> > I call what i do "modeling" as opposed to "simulation"
> > in order to take the emphasis off the criteria of mimicing outputs.
> > Ie, it reminds me that i am simplifying both the
> > mechanisms *and* the outputs.   I like Holland's description of 
> > modeling as
> > being like drawing political cartoons --- we emphasize,
> > even to the point of exageration, certain aspects
> > of a system, de-emphasizing and ignoring other aspects,
> > in order to better understand *one part* of what is "fundemental"
> > about the system under study.
> >
> > For me, "simulation" conotates systems like "flight simulators"
> > where the more details are matched, the better.
> > For the kinds of models i am interested in, that is not the right 
> > criteria.
> >
> > Along those lines, I point my students to the Borges story,
> > "On Exactitude in Science", in Dream Tigers i think.
> >
> > And i guess i do put a fair amount of emphasis on
> > the criteria that the mechanisms in my model
> > should be plausible, if simple, versions of the
> > mechanisms i think are going on in the system being modeled.
> > So its not "no holds barred" to mimic some desired
> > dynamics or structures.
> >  
> > - r


--
Dr James A. R. Marshall
Department of Computer Science
University of Bristol
http://www.cs.bris.ac.uk/home/marshall