Introduction
REM is an operating environment for intelligent agents. The focus
of the REM project is on reflective evolution. Agents in REM use
reasoning about their own processing (i.e., reflection) to make
adjustments to this processing (i.e., evolution). REM is a product of
the
Design & Intelligence Laboratory
research group in the
College of Computing at
Georgia Tech.
Participants in the REM project include
Ashok K. Goel
and
J. William Murdock.
Research on REM involves topics in
Artificial Intelligence and
Cognitive Science.
Theory
Human beings are tremendously effective at adapting to novel,
unexpected situations. Compare a human accountant to an accounting
program on a computer. The accounting program may be able to perform
millions of computations per second, vastly more than the human can.
However, even a very minor change to the format or organization of the
information that an accounting program receives will typically render
it unable to do anything at all. For example, if a typical accounting
system is designed to accept a set of numbers (such as employee pay
rates and hours worked) separated by commas and it receives this
information separated by spaces, it will simply fail. In contrast, a
human accountant would have no problem dealing with such a change. We
hypothesize that this human ability is derived, at least in part, from
the human's comprehension of his or her own reasoning.
A human working on some process understands the goal of the
process, the techniques used to accomplish the goal, and the nature of
the information that is used to support those techniques. In this
example, the accountant knows that the numbers are inserted into the
equations for computing wages and that the commas were present merely
to separate the numbers. This information is also present
implicitly in the accounting program (since it is able to
perform these computations). However, the information is not
represented in a way that the program can manipulate and alter.
Furthermore, the program has no procedures for performing such
alterations. Thus, unlike the human, the program is not able to infer
that the spaces in the new records serve the same purpose as the
commas in the old records and that the numbers in the new records
should simply be used in the same way that the numbers in the old
records were. The REM operating environment supports agents which are
able to make modifications of this sort to themselves. Agents encoded
in REM contain a model of how they work and can use this model to
adapt to novel situations.
Papers
The following documents provides a comprehensive overview of the
theoretical agenda behind the REM project:
-
J. William Murdock's Ph.D. Thesis
-
Meta-Case-Based Reasoning: Self-Improvement through Self-Understanding.
J. William Murdock and Ashok K. Goel.
Journal of Experimental & Theoretical Artificial Intelligence, 20(1):1-36, March 2008.
(PDF, WWW)
The following papers provides detailed information about individual aspects of
the REM project:
-
Localizing Planning with Functional Process Models. J. William Murdock
and Ashok K. Goel, Proceedings of the Thirteenth International
Conference on Automated Planning & Scheduling (ICAPS'03). Trento,
Italy, June 9-13, 2003.
(PDF)
-
Meta-Case-Based Reasoning: Using Functional Models to Adapt Case-Based
Agents. J. William Murdock and Ashok K. Goel, Case-Based Reasoning
Research and Development. Aha, D.W., Watson, I., and Yang,
Q. (Eds.). Proceedings of the 4th. International Conference on
Case-Based Reasoning (ICCBR'01). Vancouver, Canada, July 30 - August
2, 2001.
(PS,
PDF)
Published in a subseries of the Springer-Verlag
Lecture Notes in Computer Science series.
Finally, these aditional papers do not focus on REM, per se,
but do provide some relevant additional background:
-
A Theory of Reflective Agent Evolution. J. William Murdock,
Ph.D. Thesis Proposal, Georgia Institute of Technology College of Computing
Technical Report GIT-CC-98-27, 1998.
-
Modeling Computation: A Comparative Synthesis of TMK and ZD
J. William Murdock, Georgia Institute of Technology College of Computing
Technical Report GIT-CC-98-13, 1998.
-
Prolegomena to a Task-Method-Knowledge Theory of Cognition.
J. William Murdock, pp. 746-751, Proceedings of the Twentieth Annual
Conference of the Cognitive Science Society (Cogsci'98), Madison, WI,
August 1-4, 1998.
-
Semi-Formal Functional Software Modeling with TMK. J. William
Murdock, Georgia Institute of Technology College of Computing
Technical Report GIT-CC-00-05, 2000.
-
An Adaptive Meeting Scheduling Agent. J. William Murdock and
Ashok K. Goel, Proceedings of the First Asia-Pacific Conference on
Intelligent Agent Technology (IAT'99), pp. 374-378, Hong Kong,
December 15-17, 1999.
-
Towards Adaptive Web Agents. J. William Murdock and Ashok K. Goel,
Proceedings of the Fourteenth IEEE International Conference on
Automated Software Engineering (ASE'99), pp. 335-338, Cocoa Beach, FL,
October 12-15, 1999.
Links
-
REM uses Loom
for its underlying knowledge representation system.
-
Some of the reasoning in REM is done by generative planning, using Graphplan.
-
SIRRINE
is a predecessor of REM.
-
AHEAD is a project
involving the analysis of detected terrorist threats. It
uses a modeling language similar to the one employed in REM.