21.118 pedagogical value of simulations

               Humanist Discussion Group, Vol. 21, No. 118.
       Centre for Computing in the Humanities, King's College London
  www.kcl.ac.uk/schools/humanities/cch/research/publications/humanist.html
                        www.princeton.edu/humanist/
                     Submit to: humanist_at_princeton.edu

   [1] From: sramsay_at_unlserve.unl.edu (30)
         Subject: Re: 21.112 pedagogical value of simulations

   [2] From: "John G. Keating" <john.keating_at_nuim.ie> (130)
         Subject: Re: 21.112 pedagogical value of simulations

--[1]------------------------------------------------------------------
         Date: Fri, 22 Jun 2007 07:11:18 +0100
         From: sramsay_at_unlserve.unl.edu
         Subject: Re: 21.112 pedagogical value of simulations

On Thu, Jun 21, 2007 at 10:38:19PM +0100, Humanist Discussion Group
(by way of Willard McCarty <willard.mccarty_at_kcl.ac.uk>) wrote:
> Where do we sort in all this? Is it fair to say that since the
> phenomena we study are also not directly observable, our simulacra
> play a similar role? If we make a distinction between modelling
> something we can get to otherwise, e.g. by reading or looking, and
> simulating that which we cannot get to except after the fact, such as
> possible patterns of influence, then could we draw a parallel between
> computational physics and, say, a computational literary studies? Are
> statistical studies of literature an example?

I am reminded of a visit I made a few years ago to a robotics lab at a
university CS department, where they were studying stereoscopic
vision. Their main robot was a large, lumbering beast -- in essence,
a full-size PC motherboard mounted atop a six-wheeled chassis -- that
navigated the room by judging distances between objects. It was the
sort of thing that required lots of people to keep it up and running.

Being a software guy, I naturally asked the question: "Why build all
this in hardware? Why not just create a simulation that can prove
that your idea is sound?"

I'll never forget the director's answer. He looked at me and said,
"Because simulations are doomed to success."

Steve

-- 
Stephen Ramsay
Assistant Professor
Department of English
Center for Digital Research in the Humanities
University of Nebraska at Lincoln
PGP Public Key ID: 0xA38D7B11
http://lenz.unl.edu/
--[2]------------------------------------------------------------------
         Date: Fri, 22 Jun 2007 07:10:38 +0100
         From: "John G. Keating" <john.keating_at_nuim.ie>
         Subject: Re: 21.112 pedagogical value of simulations
Dear Willard,
 >Thanks to my colleague John Lavagnino, I have come across a letter to
 >the editor of the American Physical Society News 16.6 (June 2007),
 >"Can Simulations Really Teach Physics?" by Robert Shafer (Los
 >Alamos). He is responding to the assertion that since real events
 >happen too fast to be observed in the laboratory, it's better to
 >watch simulations of them in slow motion on the computer. He makes
 >the case for "doing the real thing, even if the equipment has to be
 >improvised", rather than watching it being done.
Having trained as an experimental physicist (at postgraduate level)
and studies both experimental and theoretical physics at
undergraduate level I was terribly excited by this letter, and the
rekindling of internal debates I believed were settled!
As a student, I was one of those kind that spent as much time writing
programs to assist with the experiments, as I did conducting the
experiments -- quiet often receiving no marks, and sometimes
receiving less marks for my programs. Despite my inclinations towards
simulation, I found that there was no substitute for "doing the real
thing" ... until I encountered Quantum Mechanics. I then realised
that the Heisenberg Uncertainty Principle (which informs that the
position and momentum of a particle cannot be simultaneously measured
with arbitrarily high precision, i.e. there is a minimum for the
product of the uncertainties of both measurements) informed that even
with perfect instruments and techniques the uncertainty inherent in
nature meant that physical experiments would provide limited
understanding of physical phenomena. I spent a distressing final year
searching for a new experiment that would verify any quantum
mechanical effect, with no success. In the end, I adopted a "quantum
like approach" -- all approaches (experimental, theoretical and
computational) existed equally, until I was forced to fix on one or
another (at which point I would be completely unsure about the
appropriateness of the other two).
 >Recently, in conversation with a physicist at UCLA, I asked about
 >computing in his discipline, specifically whether simulations of
 >otherwise unobservable realities -- let's say, just to have an
 >example, subatomic events at the core of an imploding star -- produce
 >anything anyone can be certain of. His answer was that now there are
 >in essence three kinds of physics -- theoretical, experimental,
 >computational -- and that in computational physics "they do things
 >differently there" (to quote L. P. Hartley's novel). If the
 >simulation is plausible, matching everything else one can know, then
 >it takes on the status of something one can learn from.
This is interesting -- I believe simulations can tell you nothing,
everything and something in between. I remember speaking to a physics
professor, about the value of simulations, at a conference about
simulating the dynamics of the Earth's mesosphere. He was presenting
results from a semi- empirical model based on observed data and an
established general circulation model. My concerns rested in the
generation of the computer programs, and whether one could trust the
programs? I believed that I could write a program that gave the same
answers as his model -- my program would not encapsulate any physical
theories nor use observed data. He challenged me to do so, and I
wrote a program that interpolated the output of his model. My program
was only a couple of hundred lines long and was just a simulation of
his simulation. I cheated a little, but I had made my point. I
believe that computational physicists certainly do things
differently! In order to know what their programs are doing, we need
to be involved in their generation, or have access to the program
models, design, code and test suites.
In response to your question, I believe that there is evidence from
particle physics studies where simulations of unobserved realities
(new particles) aid later discoveries. The greatest value of
simulations, I believe, is that they can show you where to look.
 >Where do we sort in all this? Is it fair to say that since the
 >phenomena we study are also not directly observable, our simulacra
 >play a similar role? If we make a distinction between modelling
 >something we can get to otherwise, e.g. by reading or looking, and
 >simulating that which we cannot get to except after the fact, such as
 >possible patterns of influence, then could we draw a parallel between
 >computational physics and, say, a computational literary studies?
  From a theoretical perspective, I think you would have to (i)
establish if there are analogous concrete natural uncertainty laws
that apply to your area of study, (ii) unambiguously determine the
uncertainty model, and (iii) prove (or at least establish)
relationships between observables, i.e. minimum for the product  of
the uncertainties of both measurements (or observables). Where,
exactly, are the uncertainties that lead to phenomena not being
directly observable? Are they emotional reactions, etc. arising from
reading or observing something? I  wold have reservations about
selecting emotional reaction, for example, as an "uncertain
observable," as I do not think is as invariant or absolute an
observable as say  (physical) position. Furthermore, if one
subscribes to Baroness Susan Greenfield's theory that Tomorrow's
people will be influenced (physically and emotionally) by today's
technologies, then there is even less reason to be concerned about
observables -- we'll just make people the way we want them (to be and
to behave). Tomorrow's peoples' patterns of influence may, in fact,
be programmable.
 >Are statistical studies of literature an example?
I know very little about statistical studies of literature, Willard,
but I'd like to tell you something of my ongoing arguments with
humanities researchers about "statistical studies" of historical
records. If I was mathematically, or physically minded, I would pay
careful attention to source sampling. For example, if I wanted to
take hundreds of manuscripts, digitise them, and optimise storage, I
would perform sound techniques like principal component analysis to
ensure that I could rebuild the source from the sampled data.
However, I find that researchers who are dealing with textual records
are often happy to only extract and make data sets from records of
interest to them, for example, building a database of Irish prisoners
in 1800 from prison records. Later someone may, using the same
source, build a database of French prisoners for the same period,
using the same records. It would appear, in such cases, that the
observer is more important than the observation; something that would
concern any physicist!
I believe that there a problem with the latter approach; and it
relates to the title of your post "pedagogical value of simulations".
Students working with the digital (albeit image based) copy of the
manuscript can essentially have a similar experiences to working with
the original. Selective data set generation, however, is not a good
sampling approach as it may not be possible to rebuild the source
from the samples. This means that simulations using these data have
poor pedagogical value as the  simulations are really only valid for
the original observations and not the source. Students, however, may
draw conclusions about the source rather than the observation. It is
a lot like my simulation of a simulation (described earlier).
I believe that it is always a good idea to utilise the best possible
digitisation schemes available to "capture" a source. There is little
pedagogical value in learning with sources derived from poor
sampling, especially in the absence of a complete theoretical model.
Best wishes, John.
Dr. John G. Keating
Associate Director
An Foras Feasa: The Institute for Research in Irish Historical and
Cultural Traditions
National University of Ireland, Maynooth
Maynooth, Co. Kildare, IRELAND
Email:  john.keating_at_nuim.ie
Tel:            +353 1 708 3854
FAX:    +353 1 708 3848