Re: A whiff of reality...

[Darren Schreiber]

I found this kind of entertaining (since I wasn't the one receiving the
rejection) because I have heard the same comments made about tools such as
statistics packages.  For instance, "I never use canned stats packages
because I always want to know what goes on under the hood."  Of course,
there is some logic to this, especially in statistics.  There are times
when not knowing how a regression package treats missing cases or where/how
it deals with the last digits of a calculation.

However, tossing out all statistics packages would be fall under the baby &
bath water cliche.  Reinventing the wheel, or Swarm functionality, is a
horribly wasteful enterprise in general.  First, it is a waste of
researchers time in recoding things that probably have been done better
elsewhere.  Second, most researchers do not have a comparative advantage in
software development which is costly in time and accuracy.  Third, this can
preclude the accumulation of knowledge and experience that benefits a
scientific endeavor proceeding from trial and error.  Fourth, since most
people don't code in a terribly transparent way, it makes it especially
hard to validate a model and its associated program.

Now, the development of alternative agent-based environments is of course
useful when the interface, processing, etc are consciously tailored to
fulfill other needs.  But, this is really the provenance of qualified
software developers, not your everyday researcher.

Imagine trying to teach undergraduates how to do regressions by using
X-Lisp to perform matrix multiplication.  I learned how to do this as a
graduate student and it was a useful exercise, but I get far more
accomplished when I use visual analysis tools like Data Desk (one of my all
time favorite programs) or stats packages like SPSS and Stata.

Some day, I really hope to teach agent-based modeling to undergrads and
graduate students.  My hope is that there will be a nice drag and drop
interface that can vary the topology of the interactions, the utility
functions and nature of actors, etc.  This would expedite and reduce the
cost of model development for professional researchers as well.

I remember in college having a long running argument with a DOS using
friend who kept arguing that Macs and GUI interfaces were a terrible idea
and that people needed to learn to use computers.  Or, lawyers I've known
who have insisted on using obfuscating legalese in contracts, briefs, and
arguments.  These are examples of an insecure elitism.

If we have concerns about canned packages, they are obviously best dealt
with through open-sourced and well supported development environments like
Swarm.  I am not willing to push Marcus' suggestions to its logical extreme
and start learning the abacus to avoid errors in chip design and
manufacture.

To maintain the integrity of the agent-based modeling enterprise, we need
to be vigilent about reviewing our own and other's code.  But, this author
has just lost perspective of the enterprise and the conference will suffer
a loss from it.

        Darren


>We just had the experience of having a conference symposium paper
>rejected as the result of vitriolic comments by someone very offended by
>our suggestions that (a) individual-based models (fish, in our case) are
>more mechanistic and general if important behaviors are emergent instead
>of forced; and (b) computer implementation issues are important and
>modelers should take advantage of software professionals and tools like
>Swarm and our Swarm-based fish modeling package. An exerpt:
>
>"I very strongly DO NOT believe that the development of software
>packages is the answer to the lack of use (of agent-based models) in
>management. In my opinion, modelling software packages ... have a great
>potential for abuse, especially in a management situation, where the
>person using software that they are not REALLY familiar with try to
>apply it to an inappropriate situation. It is too easy for assumptions
>to go unacknowledged, and results to be applied too generally and
>uncritically in these situations...It is also possible to too easily
>publish studies based on prepackaged models where the modeller lacks
>understanding of what the model is really doing.
>
>Many IBMs are rather simple conceptually, and ... basic programming
>techniques can be perfectly adequate. The modeller, by programming their
>own IBM, gains in understanding of all the assumptions and formulations
>included in the model."
>
>(This was especially interesting because another point we made was the
>importance of fully specifying a model on paper.)
>
>Initially I was just going to send this to Glen R. as a warning in
>preparing the talk on software issues in agent-based ecology that we
>asked him to give this summer. But maybe others have had similar
>experiences and suggestions for dealing with it? (We chose not to waste
>any more of our time on this particular symposium.)
>
>Steve
>--
>address@hidden
>Lang, Railsback & Assoc.
>250 California Ave., Arcata CA 95521
>707-822-0453; Fax 822-1868
>
>
>                  ==================================
>   Swarm-Modelling is for discussion of Simulation and Modelling techniques
>   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
>   please send a message to <address@hidden> with "help" in the
>   body of the message.
>                  ==================================


_____________________________________________

                 Darren Schreiber
                  Attorney at Law
                 Graduate Student
             Political Science, UCLA
                address@hidden
        http://www.bol.ucla.edu/~dschreib


                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================