Experimentation vs. Theory in Soft/Wetware

Michael Cheney cheney at ucla.edu
Mon Jul 10 14:42:31 EST 1995


In article <3thvel$l6l at nntp3.u.washington.edu>,
meir at zoology.washington.edu (Eli Meir) wrote:

| No computer program is going
| to tell us very much more than we already know.  Models are useful for
| synthesizing data, and for making predictions that we can then test,
| but, except in physics, the model will never tell us whether we figured
| things out correctly.  For that we have to ask the beast in question. 

Why do you make the exception for physics?

I am new to this field, but here is my understanding of this subject:
Running experiments (numerically or otherwise) on a model can never tell
you whether or not you have the "correct" solution.  What it can tell you
is whether your current theory can account for observed phenomena.  If it
can, then you can ask questions of the model that would be harder to ask
of the original system, and then try to verify your models answers with
"wet" experiments.

The problem is complexity.  When you do wet experiments you use your
accumulated knowledge to interpret your experimental results.  But there
are so many factors involved that one cannot intuitivly "see" what is
going on.  But a model can tell you what your knowledge implies should
happen.

So in my view, models can't REPLACE wet experiments.  You need both.

Like I said, I'm new to this, so if I'm wrong please let me know!
 -mike

-- 
Michael Cheney                         cheney at ucla.edu

http://www.seas.ucla.edu/~cheney



More information about the Neur-sci mailing list