Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

SUO: RE: RE: A New Fundamentally Different Formal Motion




Eric,

At 12:53 2003-06-11, Eric Peterson wrote:
>Hi Randall;
>
>Please see below:
>
><snip>
>
>
> > It seems to me that what's needed is a way to experiment readily with
> > different axiomatizations, not just talk about them. If Cyc's logic
> > were better specified, it would constitute a platform for such an
> > experimental effort, but since it differs so widely and unspecifiedly
> > from FOL or any other well-defined logic, it's value in experimental
> > ontology design is limited as well.
>
>[ELP] Cyc is not on the table for discussion.  OpenCyc is.  I haven't
>bumped into any higher ordered axioms in the OpenCyc knowledge base.

They are part and parcel of Cyc--through and through. And OpenCyc is 
just Cyc with a pared-down knowledge base. The logic, inference engine 
and so-called "upper" ontology is the same.


>Is anybody aware of anything in OpenCyc that is anything but a 
>syntactic variant of FOL?

Temporal logics. Default logics. Axiom schemas. Lots of quantification 
over functions and predicates.


>-Eric
>
>P.S. I think you would be hard pressed to back up your claims against
>the Cyc logic.  I had to look hard to find higher order axioms and found
>few.  I just avoid using them and do fine.  They didn't at all appear to
>be entrenched in the code.  But I think you are mixing up their
>inference methods with their axioms.

Their axioms have no meaning independent of a Cyc server. Only with a 
Cyc inference processor can they be used to support inference as 
intended by their authors.


>Default logic is their default for certain types of axioms and they 
>use something more exotic than negation as failure.

No. Cyc's own default logic is the default unless you assert an axiom 
in monotonic mode.


>But we are in the axiom business here.

This is my point. Just pulling things out of your mind as 
axiomatizations without testing has no hope of succeeding beyond 
tightly circumscribed contexts. It's just as with conventional 
programming: The artifact must be subject to testing before one knows 
if it works correctly (i.e., as intended) or not.


>Have you ever used it?

Yes, I have. I have taken Cycorp's training and have used Cyc for what 
I suppose passes for real-world knowledge engineering work: HPKB.


>Your criticisms smack of the attenuated reliability of multiple 
>hearings and retellings and little to no verification.

I resent that.


>I assume that such attenuation is one of the important reasons that 
>hearsay isn't allowed in the courtroom.

Have you used Cyc to do knowledge engineering? Not just as a repository 
for you axiomatizations, but to produce functional inference-based programs?


Randy