REPLY Future Development Directions (SD6262)

[System Dynamics Mailing List]

Posted by  j-d <jaideep at optimlator.com>

Kim writes:
"I also like a quote from Jeff Bezos - ex Wall St star-analyst and
founder of Amazon .. "When it comes to the important decisions, data
beats intuition every time"."

This quote reminded me of one of the major reasons I had moved away from
system dynamics (SD). I will try to keep it short because I have mentioned
this in the past, but since we are talking about future directions,
hopefully it is still relevant. I strongly feel that SD got stuck with, and
enamored by, its ability to create unexpected simulation-based insights. For
anyone working with nonlinear systems, this should not have been a surprise,
especially with all the research after Lorenz found the unpredictable
behavior from only three innocuous nonlinear equations. My dose of humility
came after working with the World3 model and realizing that it has hundreds
of relationships (and if I remember correctly 26 nonlinear ODE) that are not
based on econometric studies or strong theory of structure (for example,
economics, flawed as it may be, has axioms and builds a theory/structure of
human economic behavior, but all that is usually made up, along with the
data, as we build SD models). So that is a very slippery slope to start
with. No strong structure (by that I mean equations/relations/tables/graphs
in the SD models are not based on any theory), no data (we say data is not
that important, as the insights can come without data) and we want to
promote SD to hard-nosed people in the corporations!! I am not denying that
there are domain experts and deep thinkers/consultants (many on this list)
who can create great models out of fuzzy data and structures. Personally I
couldn't do it and I suspect many cannot, so they move away from SD, or keep
it as only one of their closet tools.

The way SD is practiced sometimes reminds me of what people say about video
games and its effect on children. Some claim (and I agree) that actually
these games make children smarter. So practice of SD, understanding the
intricacies of simulations, and of different subject areas, of course all
that makes us smarter. But we smart at why we are not recognized as being so
smart (sorry for the pun!!). The problem with depending on video games only
to make one smarter is that everything then is seen through the filter of
these games. So the users are smart in their own make-believe worlds of
distorted realities (coming out of distorted structures, distorted data of
games and hence possibly indulging in distorted behavior - enacting violent
fantasies/acts in the real world?).

Personally I think that the task of creating theories is a lifetime one and
SD wants to compress it into a few modeling sessions or so.

I agree with Dr. Forrester that maybe the time for SD has not come yet. If I
may, I may also suggest that time has not come yet for single and
multi-player dynamic game optimizations (my pet subject), and "optimlations"
(even more pet subject ("optim"ized simu"lations")). This is the the idea
that for complex systems we will never know everything, so simulations are
important to know possible different futures; and because our resources are
limited so optimization is necessary; and because we live in a world of
competing (sometimes randomly behaving) actors over time, so dynamic
(possibly stochastic) games are needed. These "optimlations" build on tested
SD modules.

If I were to look at the future, a hundred years from now, I see this:

There are domain experts (in economics, biotechnology, psychology,
nanotechnology, robotics) who work with SD modelers to build very sound
(structurally and data-wise) tested (i.e., they work in the real-world
unlike in distorted realities of video games) plug-in modules. The only
outward-facing interface is specifications of some parameters and relations.
Like classes in OOP, non-domain experts and non-SD modelers can't mess with
the internals of these modules. These modules must be tested and clean,
otherwise every SD modeler is going to be re-inventing the wheel, as seems
to be the case right now.

Depending on a given problem (precise specs of this are VERY important),
these modules can be selected and defined and connected together (for
example the population, resources, agriculture etc. modules in World3
model).

Depending on the problem, dynamic game (sometimes randomly acting) actors
are chosen (scientists vs. politicians in a climate model, neocons and
deciders vs. everybody else for promoting democracies, etc.) and so on.

Based on these tested modules and specified actors, simulations are done
with different objectives in mind (you are always trying to optimize one or
more objective functions); through these "optimlations" you create mental
models of possible futures that are more reality-based, taking into account
the deep needs of actors, and hopefully then make the best "gut" decisions.

Since any of the above topics can turn your hair gray, it is extremely
important to use the best teaching tools/software for teaching and learning
purposes. The task of software development and model-building can be
simplified much, much more so that instead of patting oneself on the back
that one can build and understand software and models no one can understand,
the focus and value shifts more toward analysis and understanding. Unless in
100 years we have implantable chips in our heads that make us smarter, we
need to make the software and tools smarter.

Until the above happens, we work with what we have and that includes even
Excel spreadsheets, econometrics, whatever - the future of SD/optimlations
for me lies in SDers/optimizers linking with domain experts in different
areas and creating well-tested modules that can form the foundation for work
as I outlined above. And not to forget what Bezos said "When it comes to the
important decisions, data
beats intuition every time".

Thanks for listening to the viewpoints of an outsider - I truly appreciate
how much I have learnt from the brilliant minds on this list.

Best regards

Jaideep

Jaideep Mukherjee, Ph. D.
Posted by  j-d <jaideep at optimlator.com>
posting date  Tue, 13 Feb 2007 11:36:11 -0600