Re: Directions for future research
Pat,
PC> OK. Where would you start?
Good question. For starters, the best way to address
any problem is to recognize that there is a problem.
Next, follow Peirce's general advice for any research:
"Do not block the way of inquiry." That implies that
we should examine current approaches and ask whether
they might facilitate or hinder progress toward a more
general solution.
One example is chess-playing programs. That was
suggested in the early 50s as a challenging problem
for AI, which, it was hoped, might shed some light on
the general problem of intelligence. But when that
problem was solved, it didn't lead to any great leap
in our ability to solve other difficult problems.
As for ontologies, one suggestion was to develop a
giant ontology of everything. Cyc has come the closest
to doing that, and so far, there is no evidence that it
has simplified or facilitated the more general problem
of designing a flexible learning system that facilitates
interoperability, knowledge discovery, knowledge acquisition,
or natural language understanding. Maybe Cyc can eventually
help, but it hasn't yet done so, and it's not clear whether
it is addressing the right questions.
One reason for stating the Grand Challenge proposal as
the task of integrating language with perception is that
it does not favor or discourage any particular approach.
It doesn't suggest that one should or should not use the
Cyc approach or any of its results. It doesn't claim that
Chomsky or Schank was right or wrong. It merely says that
here is a challenging problem that every child solves at
a very early stage in life, and it seems to be fundamental
to essentially every aspect of intelligence.
PC> If you think that the VAE could be expanded to include
> other capabilities, would it be possible to make some
> version of that public (source code and all), with an
> indication of what you think should be added first?
I actually didn't say that it could be extended to solve
the full range of issues involved in the knowledge soup.
But I did show it as an example of a flexible method of
dealing with knowledge and learning that is different
from what people have been doing. And I wouldn't even
claim that VAE is what should be done first.
The paper I'm writing for ICCS'05 will address some
more general issues that include VAE as a special case.
Even then, I would not claim that as the best possible
direction for future research. It's just one more
approach. Much more important than any proposed
solution is to recognize the general problem:
How can you design systems with the flexibility
of natural language that can have the perceptual
and learning ability of a two-year-old child?
That's the real problem. I don't claim to have the
answer, but I am concerned that the focus of much of
the current research is not addressing that problem.
Instead, some of it may make it harder to solve.
John