Re: Some references about ontology and analogy: SUO redux
As one who has invested a lot of effort in AI and related
subjects, I really do hope that projects like SUO, Cyc,
Project Halo, and the Semantic Web will succeed -- at least
to the extent that the produce a positive ROI, even if
they don't satisfy all the hopes and claims of their
PC> ... we are still missing a standard upper ontology.
But at the same time, the evidence suggests that many
of the currently proposed directions are overoptimistic.
As examples look at slides #17 and #18 of the knowledge
What Kant said in 1800 was repeated in different words and
from different perspectives by Wittgenstein, Waismann, and
Whitehead in the early 20th century. There's no evidence
from anything being done today that would suggest any other
conclusions. From Kant:
"Since the synthesis of empirical concepts is not arbitrary
but based on experience, and as such can never be complete
(for in experience ever new characteristics of the concept
can be discovered), empirical concepts cannot be defined."
"The topic of every science is an abstraction from the full
concrete happenings of nature.... But there can be no
logical test for the possibility that deductive procedure...
may introduce into relevance considerations from which the
primitive notions of the topic have been abstracted."
As for Project Halo, I think that something useful can be done
to make knowledge acquisition somewhat easier. It's worth
exploring, but as I said in the challenge talks, the methods
of deduction are already fairly good. The areas with the
greatest need for improvement are induction and abduction.
That's where the greatest opportunities lie.
Noah used the term "document centered". I like the emphasis
on natural language texts as the starting point for knowledge
acquisition, and that fits with my long-term belief that all
knowledge representation languages -- and, in fact, all notations
for mathematics or any other subject -- are supplements to natural
languages. Every one of them is defined directly or indirectly
in terms of some NL, and every one them is taught and explained
in a NL. In Wittgenstein's terms, they are just language games
that supplement some NL.
One point about the Project Halo discussion is the argument for
chemistry instead of physics: chemistry problems can be stated
entirely in terms of formulas, but physics problems nearly always
involve diagrams. But that leads into another issue: I don't
believe that natural language understanding is possible without
relating the terms to the geometry of the visual, tactile, and
motor mechanisms. In fact, that is my recommendation for a
"Grand Challenge" proposal:
Integrating geometrical pattern recognition
with natural language question answering
I believe that this is one of the most critical problems that
must be solved before true AI is possible.