From: ben goertzel (ben@goertzel.org)
Date: Wed May 22 2002 - 10:51:01 MDT
hi,
***
5. Although Albus design comes out of the Robotic/Numerical Control field,
I agree that his architecture is generally applicable. The RCS library
seems tailored to numerical control so I would not be immediately
interested in an interface whose purpose was to make Cyc more
intelligent. Rather, I would enhance Cyc along the lines of the Albus
Reference Model Architecture, and at the lowest hierarchical levels,
explore and use the world of external knowledge sources.
***
I can see you could make a Cyc-ish implementation of the Reference Model
Architecture...
Ultimately though, you've got to deal with the problem of mapping big
numerical data tables (coming in from sensors) and numerical control
parameters (going out to actuators) into Cyc propositions, in a way that is
useful in real-time....
I suspect that, in dealing with this problem within Cyc, Novamente or
anything else, plenty of aspects of the RCS library will be very useful
guidance, *in addition to* the overall RMA design
***
6. The most encouraging aspect of what I read, is that Albus' architecture
has been proven on manufacturing cells and self-guided vehicles. If I
dream that someday Cyc will be the intellect of robots, then it must at
the minimum be able to perform as well as RCS, or be able to program an
RCS configuration.
***
Yes, but this is obviously a weak ambition!
RCS works a lot worse than humans so far, and (as you know) a lot of what
is needed to make it better than it is, is incorporation of inference on
relatively abstract declarative knowledge -- knowledge learned from
experience, and knowledge transferred from humans.
I think that integrating RCS-type knowledge with Cyc-type knowledge is a
very deep and hard problem... and of course I think that Novamente's
knowledge representation constitutes *one among many possible paths* to
solving it....
-- ben
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:39 MDT