RE: Hardware Progress: $212/GFlop/s

From: Stephen Reed (
Date: Wed May 07 2003 - 12:31:05 MDT

On Wed, 7 May 2003, Ben Goertzel wrote:
> > Likewise this year we have a team of programmers working on generalized
> > semantic integration of relational databases. The objective of our work
> > is to make structured information from databases and web services appear
> > to be Cyc assertions for the purpose of deductive inference.
> I am pretty skeptical that this can be done in a meaningful and generalized
> way...

The generalized part is the mapping vocabulary. Each structured knowledge
source requires semantic mapping by hand, using the generalized
vocabulary. So I agree with your point and regret not being more
specific. Cyc is far from being able to automatically understand an
arbitrary structured knowledge source.

> I've just gone through a long process of mapping a bunch of biology
> databases into Novamente's semantic structures... and it was not a
> straightforward task, it required a lot of human intelligence and some
> biological knowledge too, to get the mappings right.
> Of course, one can simply map the relational algebra of an RDB into
> predicate logic [or probabilistic combinatory term logic, in Novamente's
> case]. This may lead to something interesting, sometimes. But it will very
> very rarely lead to anywhere near the most natural logical representation of
> the RDB structures.
> Of course, a smart enough AI can read in the RDB in a simplistic
> mathematical way, and then on its own learn a better reprsentation,
> reformulating the RDB-derived knowledge intenrally. Novametne is not that
> smart yet, and I don't believe Cyc is either...

We are working on a tool which can use DB schema as the starting point for
the subsequent hand-mapping.

> But maybe you don't really mean a "generalized" way? I mean, if you're
> looking at a restricted class of DB's (say, transactional DB's that back-end
> web servers; or DB's that come with Siebel systems) you may be able to do
> something "generic" across that class. Even there, though some special-case
> translator tweaking will probably be needed for each DB you process, unless
> the DB's are *really* regimented (as in the case of Siebel DB's).

We are not being even this general in our semantic integration efforts.

The primary benefit of our work is to avoid having to hand-enter
assertions about a domain when there exists a structured source of
facts about that domain. We only have to map the schema for the
structured source and reuse the underlying general database/web service
connection infrastructure.


Stephen L. Reed                  phone:  512.342.4036
Cycorp, Suite 100                  fax:  512.342.4040
3721 Executive Center Drive      email:
Austin, TX 78731                   web:
         download OpenCyc at

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT