SUO: RE: An article on the pitfalls of metadata
All,
Thanks for your insights on this issue. The very
diversity of your opinions buttresses my belief
that the best way to construct a useful ontology
is to stick with the well debated concepts in
WordNet, which are the closest to peer-reviewed
general concepts we presently have available.
But of course, the axioms aren't there. It seems
that the axioms are far stickier to build correctly
than the concepts though.
Every point of view expressed on this topic has
been unique and very idiosyncratic. We each see the
elephant in a different way because of our attention
focus, our values, our instantaneous goals, and our
drives.
That leads me to believe that personality (whatever
that is) drives beliefs, which become tentative
axioms in individuals. So a bottom-up algorithm that
derives the intermediate to top level axioms from
terminal level data requires a corresponding
simulation of personality.
Some have believed that a program must be situated
in the real world to develop a workable intelligent
system. Striving, failing, learning, and evaluating
actions have to be repeatedly performed. So a process
improvement ontology is the key initial upper ontology
for a bottom-up learning algorithm.
Our human inability to agree on much of anything
seems to confirm that speculation.
So the real requirement for an intelligent system is
the fact that an ontology must be represented, modulated
and measured. Two instances of that algorithm, given
different data to learn from, would produce two different
personalities based on the growth of that simple initial
top level ontology of process improvement.
Is that something a few of us can agree on?
Further discussion is welcomed.
Thanks,
Rich