RE: The inevitability of death, or the death of inevitability?

From: Ben Goertzel (ben@goertzel.org)
Date: Sat Dec 29 2001 - 21:12:31 MST


> Ben Goertzel wrote:
>
> > the semantic
> > markup is going to mainly be inserted by (initially primitive)
> AI's, not by
> > humans.
> >
> > AI's that are costly to operate can read a Web page once and
> insert semantic
> > markups.
>
> If one has AIs that can usefully extract the appropriate semantics from
> unstructured data, what's the benefit to having them then do
> structured markup?
> If you've got the former, the latter seems like a performance
> hack at best, a
> waste of bits and cycles at worst.
>
> I.e., the semantic Web already exists. :-)

Well, I tried to explain this, but apparently I was not clear enough.

The point is, I predict that there are going to be many years during which
the following
two statements hold:

1) There are AI's that can usefully extract the semantics from unstructured
data

2) Such AI's are very expensive to operate, requiring highly expensive
computing resources

Hence, there will be a benefit in having such an AI read a piece of text
*once*, outputting
the semantic information it has recognized IN A FORM THAT CAN THEN BE
UNDERSTOOD BY
CHEAPER, STUPIDER AI PROGRAMS.

In fact, the years during which my two statements hold *are now upon us*, in
the sense that there
exist relatively simple but computationally expensive NLP systems that can
usefully extract a lot
of information from unstructured text. LexiQuest, for example, is a firm
that markets such systems.

The current non-existence of the "semantic web" is therefore, in my view,
basically due to sociological
reasons rather than due to the contradictoriness or unproductivity of the
idea.

I.e., no one has created a business model founded on using
computationally-costly AI tools to mark up the
Web, and then allowing a variety of cheaper tools written by various people
to read these mark-ups. Which is too
bad because this would significantly increase the intelligence of the
Internet.

-- Ben



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT