Workshop at WWW2006, Edinburgh, UK, May 2006
Reasoning on the Web (RoW06)

Invited talk by Ed Barkmeyer, National Institute of Standards and Technology 

Title:
Rules on the Web - Why?

Abstract:
40 years of research into simulating the human reasoning processes with
computer algorithms has given rise to many different approaches.  The
existing Semantic Web languages - RDF and OWL, and ISO "common logic" -
are designed primarily to support two of those approaches: description logic
and "pure" first-order logic.  By comparison, most of the successful
artificial intelligence applications of the last 30 years have used two
entirely different approaches: logic programming and production rules.

The original purpose of the Semantic Web technologies was to enable human
researchers and their software agents to quickly find relevant material
among the billions of documents available on the World Wide Web. For Web
searches, and many other applications, these technologies would appear to be
perfect: they capture invariant truth, they separate knowledge from its
application, and they separate the skills required for knowledge modeling
from those for automated reasoning. By comparison, "rules" technologies
encode knowledge in a way that directs the reasoning machine in making the
useful inferences, which requires both skills in the knowledge engineer.
Why would we want to standardize, or even use, rules technologies?

There has always been a broad spectrum of applications for knowledge
engineering.  When humans can state, in computer-processable form, exactly
what they know, searching the Web for information is just the tip of the
iceberg.  The standardization of languages on which automated reasoning can
be based supports solving complex problems by both aggregating the relevant
knowledge from several sources and decomposing the problem into subproblems
for diverse agents.

This talk identifies a number of shared application problem spaces in which
the "rules" approach is likely to remain superior.  It discusses the
advantages of "rules" systems over "monotonic logic" systems in these areas.
And, while the traditional rationale for choosing rules technologies has
been performance, this talk argues that the benefits of sharing
application-specific and time-specific ("non-monotonic") knowledge go well
beyond that.