I work in the area of formal knowledge representation. My areas of research include reasoning about knowledge and belief, non-monotonic reasoning, and reasoning about time and planning. I'm especially interested in combinations of the above, such as non-monotonic temporal reasoning (including the infamous Yale Shooting Problem) and multi-agent non-monotonic reasoning.
Most of my work is theoretical, but I am also interested in applying
my research to commercial applications.
I have developed expert systems for benefits inquiry in medical insurance,
and for product configuration in life and property insurance
(joint work with
Moninder Singh).
I am currently investigating how this work can be extended to
intelligent help desks.
I am also developing an intelligent agent for browsing the web (joint
work with
Moninder Singh).
The focus is on using knowledge about a specific domain (such as real estate or
jobs) to comprehend arbitrary web pages.
In addition, I am working with Juhnyoung Lee, Edith Schonberg, and
Mark Podlaseck on an electronic commerce tool for website analysis.
Inheritance Comes of Age: Applying Nonmonotonic
Techniques to Problems in Industry
(Click here for the journal paper)
Why isn't nonmonotonic reasoning more visible in industry today?
Mostly
because researchers have focussed on difficult toy problems instead of
useful solvable problems, and because researchers haven't concerned themselves
with technique and tractability issues.
I discuss an application of nonmonotonic reasoning which was successfully
used in industry: an extension of
inheritance --- inheritance of well-formed formulae --- that is considerably
more powerful than standard inheritance with exceptions.
This is a good overview of the two
papers listed below. Click on the title for the postscript
version; click here for the dvi version.
(Conference paper:Invited talk,
Proceedings, Fifteenth
International Joint Conference on Artificial Intelligence, Morgan
Kaufmann, 1997
Journal paper: Artificial Intelligence, 1998)
An Expert System Using Nonmonotonic Techniques
for Benefits Inquiry in the Insurance Industry (with Moninder
Singh)
Describes BenInq, an expert system used in the medical insurance industry by
both customer service representatives, who answer questions about
the extent of a patient's coverage for medical care, and policy modifiers,
who frequently change coverage rules. Reasoning is performed by
inheriting business rules, represented as formulae of first-order logic,
in a semantic network in which formulae are attached to the nodes.
(Proceedings, Fifteenth
International Joint Conference on Artificial Intelligence, Morgan
Kaufmann, 1997)
Inheriting Well-formed Formulae in a Formula-Augmented
Semantic Network
What happens when we expand the notion of
an inheritance network so that we can inherit well-formed formulas as
well as attributes? This paper explores this issue in detail, and
introduces the notion of a FAN, a formula-augmented semantic network.
(Proceedings, Fifth International
Conference on Principles of Knowledge Representation and Reasoning,
, Morgan Kaufmann, 1996)
New Problems for Inheritance Theories
This paper explores several issues discovered in the context of
developing an expert system for a medical insurance company,
including the inheriting formulae in a network, the interaction of
subtyping and subevents, and non-unary inheritance.
Click here for the figures.
Click here for the dvi version (without figures).
(Working papers,
Third Symposium on Logical Formalization of Commonsense Reasoning)
Motivated Action Theory (with Lynn Stein)
Describes Motivated Action Theory, a non-monotonic temporal theory
that handles the Yale Shooting Problem and related issues. MAT is based
on the principle that models are preferred if they contain events
that happen for a reason. (Artificial Intelligence,71,1-42,1994)
A Proper Ontology for Reasoning about Knowledge
and Planning
Linear time doesn't allow hypothetical reasoning; branching time
doesn't handle prediction properly. Both prediction and hypothetical
reasoning are needed for good planning theories. This paper presents
an ontology, known as relativized branching time that can
handle both.
Click here for the dvi version.
(Working Papers,
TIME-94)
Epistemic Logics for Multiple Agent Nonmonotonic
Reasoning I (with Ramiro Guerreiro)
Much commonsense reasoning requires that agents reason about the ways
in which other agents reason. Thus, there is a need for theories
that can handle multiple agent (or nested) non-monotonic reasoning.
This paper describes MANML, a theory that supports such reasoning.
Click here for the dvi version.
(Working Papers, Second Symposium on Logical Formalizations of
Commonsense Reasoning, 1993)
Knowledge and the Frame Problem
Extends Motivated Action Theory to the multiple-agent case;
first investigation of the issues involved in multiple-agent
non-monotonic reasoning. Click here for the
dvi version.
(International Journal of Expert Systems, 3(4), 1991.
Also in K. Ford and P. Hayes (eds): Human and Machine Cognition:
The Frame Problem in Artificial Intelligence, JAI Press