W3C home > Mailing lists > Public > www-ws@w3.org > May 2003

Re: Meaning

From: Drew McDermott <drew.mcdermott@yale.edu>
Date: Fri, 23 May 2003 16:58:25 -0400 (EDT)
Message-Id: <200305232058.h4NKwP014867@pantheon-po02.its.yale.edu>
To: www-ws@w3.org

   [Dr. "Wex":]

   To sum up: you simply cannot reduce reality to a set of formalized
   universal and existential predicates.

I agree.  I don't really see how the Cyc contexts you mentioned will
help.  But there are certainly other useful tools that don't fit that
well into an axiomatic approach.  For instance, a company might have a
Bayesian model of trustworthiness.  You might have an axiom that says,
"If the model says A is trustworthy with probability > 0.95, then it
is okay to do business with A," but the model's "reasoning" is a black
box as far as the axioms are concerned.

   Which brings me to point 2.  Can we, or can we not, proceed?  The point of
   the Semantic Web, as I understand it, is to enable real people to use
   machine augmentation to assist their real world tasks using the Web's
   resources.  Can an agent with this level of understanding do that?

   I would like the answer to be "yes" but I fear it is "no," though not for
   any technical reasons.  Let us briefly take the case of a business that
   wishes to buy widgets.  We must convince the financial officers of the
   company (not the techies, mind you) that we can produce a system of
   sufficiently sound and complex reasoning that it can go out on the Web,
   find, and purchase widgets for their company.  That is, spend real money
   on real things.

   And we must build such a system knowing that it has an imbecile's notion
   of "own" and "buy" and simply no concept whatsoever of "fraud" or

   I imagine this to be a *very* hard sell.

Well, we just omit the part about the imbecilic degree of
understanding :)

I think you're a tad too pessimistic.  The issue is not whether we can
build a perfect system, only whether we can build a cost-effective
system.  If a company can save $XXX per year by replacing human widget
buyers with automated agents, it will do it, even if the $XXX includes
some losses due to fraud.  After all, human widget buyers can be
fooled, and they are susceptible to fraudulent schemes that computers
will be immune to.  For instance, Like-New Widgets might bribe the
buyer into buying from them.

Also, I don't see why the system must have no knowledge whatsoever
about fraud and deceit, even granted it can't know everything.

                                             -- Drew McDermott
Received on Friday, 23 May 2003 16:58:30 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 20:37:08 UTC