Re: Feature Engineeriing, Self-Inconsistency, etc.
Folks,
These discussions keep running into a common theme:
there is no such thing as a one-size-fits-all suit
of clothing, ontology, computer, database, etc.
We do need standards, but as everybody has pointed out,
we keep getting more "standards" than we want. The best
we can hope for is something like clothing standards:
1. A rough classification based on distinctions such as
Male-Female, Adult-Child, Large-Medium-Small, and
graded sizes for better fits.
2. Variations based on a wide range of purposes: weather,
formality, activities, fashion, ethnic customs, and
the inevitable innovations in technology.
3. The product of various options in points #1 and #2.
But there will also be many people (dwarf, paraplegic, etc.)
and activities (astronaut, firefighter) that fall outside the
normal range. And there will also be people who can afford
to pay enough for "bespoke tailoring" for a better fit.
Some comments on previous comments:
JA> Some mix of SUO (defining a number of non-controversial
> lowest common denominator things) and special-purpose
> ontologies seems like the obvious evolutionary course.
That's basically the idea that buying an ontology is
similar to buying off-the-rack clothing, provided that
the option of hand tailoring is available.
RF> ...the cheapest way to keep all the ways you can
> structure them (all the meaning) will be to hold on to
> the examples and "project out" the structure you want
> when you want it.
Not always. Off-the-rack clothing is cheaper than bespoke
tailoring. If you have applications that fall into a well
recognized category, you can take advantage of mass produced
ontologies. But if you don't provide the right options,
people will have to fall back on hand tailoring.
RF> If only because (given the "curvature" of knowledge =
> inconsistent orderings) the maps they make are only going
> to remain accurate within a very narrow domain.
That's true of clothing. But a good selection of standard
sizes is sufficient for many people in many activities.
RM> In the "Challenge of Knowledge Soup", you cite Kant,
> Wittgenstein, and Waisman on the "Limits of Definability".
> Should Heisenberg be mentioned here, as well (or perhaps
> under "Limits of Measurement")?
Yes, that's one more problem to consider. But the mismatch
between a continuous world and discrete language created
most of the really nasty problems long before measurements
became precise enough to notice quantum effects.
RM> You also cite Whitehead on the "Limits of Logic". Should
> Godel be mentioned here?
You could. But I would put Gödel's incompleteness in the
same category as Heisenberg's complementarity: it's irrelevant
for almost all practical reasoning problems. Mathematics and
physics developed to a very high level of sophistication before
Gödel published his papers, and even today, most working
mathematicians couldn't care less about the foundations of logic.
It just doesn't affect most working mathematicians, and it has
had even less effect on nonmathematicians.
John Sowa