Tuesday, November 29, 2005
Center of Excellence Proposal
à
The Taos Institute
(on the possibilities)
(on how to move forward an
understanding of future “machine intelligence”
[bead thread on curriculum reform]
Communication from Gary
Berg-Cross à [251]
Note from Jeff Pollock
Paul-
One of the things missing in any such dialog today is an unambiguous and exhaustive description of
(a) what is possible in various logic languages today (eg: the facts of the matter) as a grounding for a discussion about
(b) what data is optimally described by some representation (eg: opinions with measures).
I find it really hard to respond to a broad generalizations about what the future of machine intelligence *should* be, then delineate various observations about the limitations of x [rdf,owl,cg,scl,flora,prolog,etc.], to produce a useful "gap analysis" about the steps that need to be undertaken.
I would go so far as to agree with you in principle that that current semweb [rdf/owl] approaches are insufficient, and I would permit that you may be right that they are not directly on the path to "machine intelligence nirvana," whatever that is since we've not agreed, but I cannot agree with an assertion that they are a useless waste of time.
Given the state current IT practices with RDB, XML, and Java, this sort of discussion smacks of mechanical engineers circa 1900 driving around in carriages arguing about the design of spaceships. Personally, I have opinions about what the spaceship might look like, but I have facts that RDF and OWL are indeed much better subsystems for certain things (federated systems metadata for one) than RDB, XML, or Java. In other words, we can argue about the spaceship vision, but don't dismiss a better suspension system just because you don't see why a spaceship would need it - current carriages might benefit greatly from it.
I'm interested in grounding a discussion in analysis that takes current practice [xml, java, rdb] and contrasts with emerging practice [rdf,owl] to distinguish clear value, then goes further to contrast emerging practice [rdf,owl] against future goals [x,y,z] to evaluate gaps.
One such attempt, for the W3C rules workshop, is attached. I think it neatly describes capabilities and characteristics for OWL and FOL style systems. Allowing that it doesn't address your concerns, how can you add to the discussion in a similar framework that helps lay-people conceptualize the differences in KR approaches? (eg: examples of abductive reasoning questions and non-answerability in DL/FOL systems would be a start)
-Jeff-
BCNGroup response à [254]