[Inquiry] Re: Theme One Program -- Motivation Notes

[Jon Awbrey]

o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o

TOP.  Motivation Note 2

o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o

One of the most curious things that I learned --
or think that I got some kind of insight into --
during the decade-long process of writing the
Theme One program, is some of the reasons why
empiricists and rationalists seem to have so
much trouble understanding each other, even
when those two styles of thinking inhabit
the very same soul.

The way that this came about was like this.
The stock of code from which the program is
currently assembled initially came from two
different programs, that I worked on roughly
during alternate years, at first only during
the summers, at least, so far as active work,
and not just dreaming or musing or ruminating.

One of these programs was called "Learner", and in it
I deliberately sought to implement a Humean empiricist
style of learning algorithm for the adaptive uptake of
coded sequences of occurrences in the environment, say,
as might be codified in a formal language.  I knew all
the theorems from formal language theory that tell how
limited any such strategy must ultimately be in terms
of its generative capacity, but I wanted to explore
those limits in practical computational terms.

The other program was called "Modeler", and in it
I was seeking to implement a differential extension
of Peirce's graphical syntax for propositional logic
that I had developed in work over the previous decade.

As I mentioned, work on these two projects proceeded in a parallel series
of fits and starts through interwoven summers for several years, and then,
somewhere along the line, I can't remember just when, I realized that the
Learner, one of whose other names happened to be "Index", could be put to
work helping with sundry substitution tasks that needed to be done by the
Modeler, one of whose other names happened to be "Slate", since it evoked
innate ideas of the tabula rasa.

So I began to integrate the Learner and the Modeler,
at first still working on the two component modules
in an alternating intermittent manner, but devoting
a portion of effort toward the task of amalgamating
their principal data structures, bringing them into
convergence with each other, and unifying them over
a common basis.

After another round of seasons and many more changes
of mind and mood and programming style, I arrived at
basically the same graph-theoretical data structure,
strung like a wire through all the far-flung pearls
of my programmed wit.  And yet, as I discovered to
my consternation, the pearls that were polished in
counterposed years had still been glossed to shine
along axes of polarization that were skew in their
grains with regard to each other.  That is to say,
all of the strategies that would seem to be the
smartest tricks to pull from the standpoint of
optimizing the overall program's performance
on the Learning task I would discover to my
considerable dismay the next year were the
dumbest things to pull from the standpoint
of its performance on the Reasoning task.
I eventually came to appreciate this as
a "discovery", but then, what choice
did I have?

Jon Awbrey

o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o