From: Stephen Reed (reed@cyc.com)
Date: Fri Nov 05 2004 - 12:49:14 MST
On Fri, 5 Nov 2004, Ben Goertzel wrote:
>
> Hi Stephen,
>
> Indeed, Cycorp has done an excellent job of bringing in government funding
> ;-)
>
> Unfortunately though, as I've expressed before, I'm pretty sure the Cyc AI
> design is incapable of achieving a high level of general intelligence...
>
> How are things with Cognitive Cyc?
>
> ben
Ben,
Correspondence with James Albus at NIST was encouraging but they are short
of sponsorship funds due to the war in Iraq. At Cycorp there is modest
low-level interest in my work so far with an Albus-inspired
hierarchical, learning control structure, but I hope to get some
effort-hours allocated from a sponsored project in the next few months. I
have already developed an agent control language that can implement Albus
nodes in java. The language statements are represented in CycL (predicate
calculus) as scripts, then compiled out into a java classes for
execution. The behavior methods are organized into scalable nodes having
sensations and performing actions through message-passing. The upper
levels of the node hierarchy implement a Friendship Architecture, i.e.
BeFriendly is the uppermost node.
Currently, I am excited about importing all of WordNet 2.0 into Cyc as
concepts. That's part of a DARPA sponsored project we call Accessible
Research Cyc, that will deliver a very large subset of Cyc to Universities
and companies under a research license. Importing WordNet is attractive
because our text parsers will then have much better recall (percent
coverage), with the drawback of much more word-sense ambiguity to deal
with. I expect that we will add 80,000 concepts (lexicon plus some
definitional assertions) to Cyc in this effort.
Agreed that a commonsense knowledge base is incapable of achieving a high
level of general intelligence, provided that a deductive inference engine
is its only active facility. But Cycorp believes that the hard part of AI
is constructing a fleshed-out commonsense conceptual vocabulary, whose
later additions (hopefully soon) include scripts for intelligent behaviors.
As we have discussed, other AI designs, given a certain maturity, could
simply read Cyc to acquire all that we have hand coded, as a short cut to
understanding dictionaries and encyclopedias. On the other hand, we think
that once Cyc is capable of learning by reading, learning by being taught
by us using dialog, and capable of learning from experience, then we can
subsequently plug in probabilistic adapters to ground out symbolic
concepts from real-word sensations (robotics).
There are AI projects based on probabilistic learning working towards
knowledge richness, and alternatively there is Cyc, based upon first
achieving commonsense knowledge richness (e.g. recently we added a
sophisticated temporal reasoner) including how-to knowledge, and then
working towards self-improvement and probabilistic learning.
The beauty of our approach is that knowledge base improvement is a
virtuous circle attracting sponsors.
Cheers.
-Steve
--
===========================================================
Stephen L. Reed phone: 512.342.4036
Cycorp, Suite 100 fax: 512.342.4040
3721 Executive Center Drive email: reed@cyc.com
Austin, TX 78731 web: http://www.cyc.com
download OpenCyc at http://www.opencyc.org
===========================================================
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:50 MDT