Metareasoning: Thinking about thinking
July 13-14, 2008, Colocated with AAAI 2008, Chicago, Illinois.
Accepted Papers (new)
Workshop Program (new)
The 21st century is experiencing a renewed interest in an old idea within artificial intelligence that goes to the heart of what it means to be both human and intelligent. This idea is that much can be gained by thinking of one's own thinking. Metareasoning is the process of reasoning about reasoning itself. As shown below, it is composed of both meta-level control of computational activities and the introspective monitoring of reasoning to evaluate and to explain computation. Meta-level control is the ability of an agent to efficiently trade off its resources between object level actions (computations) and ground level actions to maximize the quality of its decisions. Introspective monitoring is necessary to gather sufficient information with which to make effective meta-level control decisions or to explain failed object-level reasoning. This workshop will explore the implications of this model by examining the various aspects of metareasoning and models of self and their role in single-agent and multiagent applications.
The above figure and an accompanying manifesto are available on the web (www.mcox.org/Metareasoning/Manifesto). To increase coherence of the workshop sessions and to help attendees to relate heterogeneous positions, all authors are encouraged to include and reference at least one of a set of provided figures, either positively, negatively, or as a contrast to their own alternative models. The goal is to use this as a unifying theme. This is especially important because authors of selected papers will be invited to prepare chapters based on their workshop contributions for a forthcoming book on metareasoning. The workshop will not simply collect a loosely affiliated set of technical papers; rather, the objective is to present a cohesive story of what metareasoning is, what its limitations are, what benefits it promises, and how these promises can be implemented computationally.
This two day workshop will include a number of short paper presentations, thematically organized discussion sessions, a break-out problem-solving session with discussion, and two speakers. Professor Don Perlis from the University of Maryland, College Park, will present on day one and Professor Aaron Sloman from the University of Birmingham, U.K. will present on day two. We also will include panel discussions after each group of paper presentations so that the audience can ask follow up questions that compare and contrast the related positions. Finally a special track will be targeted for the topic of evaluation of metareasoning systems.
Topics of Interest include:
Michael T. Cox (co-chair) Anita Raja (co-chair)
Senior Research Scientist Assistant Professor
Intelligent Computing Department of Software & Information Systems
BBN Technologies University of North Carolina at Charlotte
Cambridge, MA 02138 Charlotte, NC 28223
(617) 873-3632 (704) 687-8651
Vincent Conitzer, Assistant Professor, Duke University
Stefania Constanini, Professor, Univ. of L'Aquila, Italy
Ed Durfee, Professor, University of Michigan
Stan Franklin, Research Professor, University of Memphis
Andrew Gordon, Research Assistant Professor, University of Southern California
Eric Horvitz, Principal Researcher, Microsoft Research
Victor Lesser, Professor, University of Massachusetts
Paul Robertson, Senior Research Scientist, BBN
Lenhart Schubert, Professor, University of Rochester
Steve Smith, Research Professor, Carnegie Mellon University
We encourage the submission of
high quality, original papers that are not submitted for publication elsewhere.
The submission should not exceed eight pages in the AAAI style (www.aaai.org/Publications/Author/author.php),
either in PostScript or PDF format. Surface mail address, e-mail addresses
should be included for all contributing authors. Submissions must be emailed to
either chair (firstname.lastname@example.org
by the deadline period and must include and reference atleast one of the figures from
statements are also accepted.