Lists Home |
Date Index |
Too far. It did work last time. AI systems, notably expert
systems written in languages suited to the task such as Prolog
or LISP work. SGML works too. (Corrupt history leads to
corrupt conclusions, and that is the SemWebs biggest and
most intractable obstacle.) What was learned is that AI is tedious
to write and difficult to scale. Say: expensive.
Following these threads, the point of the SemWeb seems to be:
1. One language moreorless, so as cheap as possible.
2. One language moreorless, so scaling comes of linking and that
is what networks do.
3. Applications of the linked information: TBD.
I agree that the frustration of STimBL down to Elliotte is in
item three. Systems doing this work do it without items
one and two so items of type three never get on the radar.
This may come down to 'not enough customers really want
machines to do this work' for reasons which are not
coupled to the technology. See para 1 above.
The 'telephone to financial conversations' app is
called a 'link analysis' application. Those are
used in the industry I work in. Abstractly,
mining hidden links among loosely coupled processes can
be applied to other information domains. How to treat
these as so-called 'proofs' is interesting because the
measure of 'proof' always takes in more 'proof systems'
and rules. The measures of proof for a technician repairing
an engine and that of a judge deciding if a warrant is
merited are not the same.
From: Miles Sabin [mailto:email@example.com]
For seconds, I think it's quite reasonable to question whether it's
worth even bothering to read TFM if it's more or less the same old
stuff that didn't work last time with no reasonable expectation that
the new twist (angle brackets and URIs) is going to help find a useful
route out of the blind alley.