From: Ben Goertzel (ben@goertzel.org)
Date: Fri Nov 05 2004 - 17:18:56 MST
Hi Stephe n,
> Agreed that Cyc is not designed using the human mind as a model.
My statement was stronger than that, though. I think that there are
probably some universal properties of strongly-resource-limited minds, and
that Cyc
a) doesn't have many of these properties
b) is very awkward to incorporate into any system that does have these
properties
I have tried to enumerate these properties in my prior writings on cognitive
science, under the name "the psynet model of mind"
> Rather
> Cyc uses a philosophical approach (e.g. James Allen's time relations,
> Davidson's approach to representing events).
Agreed, but the philosophy underlying Cyc is sorely limited, missing out
for instance on the ideas raised by the complex-systems philosophy of mind,
by the philosophy of uncertainty-management (localized Bayes net models are
not philosophically satisfactory, really, are they?), by the philosophy of
embodiment and mind-body relations, etc. Cyc is based on some sound
philosophy, but only on philosophy that deals with a small and narrow subset
of mental phenomena.
> I hope that the impedance
> match between Cyc's vocabulary and English will be improved after the
> import of WordNet 2.0, opening a wide door for practial applications in
> the field of text knowledge management.
The mismatch between Cyc and English doesn't have to do with vocabulary.
It has to do with things like the meanings of prepositions (not covered in
WordNet) and subject-argument relationships (not covered in WordNet). And
these aspects of linguistics, in which Cyc most badly mismatches English,
are the precise aspects of linguistics that most closely tie in with the
mind's relationship with the body and with other minds (see e.g. Calvin and
Bickerton, "Lingua ex Machina", for a treatment of the grounding of these
linguistic phenomena in social psychology.)
I'm aware that this conversation is being had at too high a level of
abstraction -- it's really a technical discussion, but getting into the
nitty-gritty details of linguistic, inferential and perceptual/active
knowledge representation in emails to SL4 probably isn't going to work...
-- Ben Goertzel
This archive was generated by hypermail 2.1.5 : Tue Feb 21 2006 - 04:22:48 MST