Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!Germany.EU.net!EU.net!sun4nl!cs.vu.nl!embronne
From: embronne@cs.vu.nl (Bronneberg EM)
Subject: Re: Question about Cyc's assertions
Nntp-Posting-Host: galjas.cs.vu.nl
References: <ADPYyblG_T@loz.intes.odessa.ua> <3nfup5$aus@lyra.csx.cam.ac.uk>
Sender: news@cs.vu.nl
Organization: Fac. Wiskunde & Informatica, VU, Amsterdam
Date: Wed, 26 Apr 1995 11:16:22 GMT
X-Newsreader: TIN [version 1.2 PL2]
Message-ID: <D7n3BB.7Ft.0.-s@cs.vu.nl>
Lines: 42

Ralph Becket (rwab1@cl.cam.ac.uk) wrote:
: In article <ADPYyblG_T@loz.intes.odessa.ua>,
: Vitaly S.Lozovsky (7:0482) 24-6184 <loz@loz.intes.odessa.ua> wrote:
: >[...]
: >
: >Someone formulated the Pig's principle: "If something is good, the
: >more of it -is better". Probably the same situation with the number of
: >productions: I don't believe that this is the right way to Heavens.
: >The thinking machine must be much more complicated in architecture,
: >and it must reason about many complicated things: scenarios, plans,
: >theories, models, and what else! Would it be natural if somebody
: >taught his/her child by demonstrating series of situations paired with
: >the 'proper'reactions? It looks rather idiotic, excuse me.

: The Cyc knowledge base is not a giant list of (stimulus/response) pairs.
: It is an attempt to formalise and codify basic common-sense knowledge
: that is required in order to reason about the real world.  People 
: acquire this knowledge first through experience and secondly by being 
: taught.  Both of these approaches are still beyond the state of the 
: art (although Cyc is an improving approximation of the latter).  As 
: Cyc learns more, it becomes easier to teach.  It is easier to describe 
: things at a higher level, which can only be done with a sufficiently 
: expressive lower level framework.  Moreover, as Cyc grows it is increasingly
: able to help its knowledge enterers describe and organise new information. 
: I believe that some of Cyc's learning is already being done through 
: semi-supervised digestion of natural-language texts.

Is this really happening? I know that these are two of the goals Lenat and
Guha are trying to reach with Cyc, and especially the first one (that it
becomes easier to teach Cyc something) should happen if the project is
succesfull. But I never read anything, where this is actually described. So I
wonder whether it is. If so, can anybody please tell me where I can find more
on this. For example a description of the process by Lenat and/or Guha?

: [....]

: >Vitaly Lozovsky

: Ralph

Miel Bronneberg,
embronne@cs.vu.nl
