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

Re: [CL] Re: Decidability



Chris,

I completely agree with your conclusion.  If all the
logicians working on the Semantic Web would say that,
I would be happy:

 > And I say again, I use decidable and undecidable
 > reasoners all the time, and find them all useful.
 > As with all these silly kinds of arguments, it boils
 > down to having a good set of tools available, and
 > knowing their characteristics.  Decidability is
 > one characteristic.  It's useful to know.  That's
 > all.  No big whoop.

What started this series of notes was that original
quotation I circulated, which was triggered by some
people who have been trying to make decidability
a nonnegotiable requirement for any version of
logic used on the Semantic Web.

I agree with you that it's important to know,
and to know when it's relevant or irrelevant
to a particular kind of problem.

However, I do want to clarify the relationship:

 > What your diagram seems to imply is that decidability
 > is somehow related to N, which it is not.  An undecidable
 > problem can manifest in a small data set.  THAT is the
 > problem and why it is useful to know.

Please note the two endpoints in the diagram, both of
which are independent of N:  a constant-time algorithm
is finite and unchangeable for any N, and an undecidable
problem takes infinite time for any N.

I think that's a useful way to look at the relationship:
undecidability is the upper limit of the whole range
of complexity classes.

John

GIF image