Architectures for Modeling Emotion:
Cross-Disciplinary Foundations
http://homepages.feis.herts.ac.uk/~comqlc/ame04
American Association for Artificial Intelligence (
AAAI
)
Stanford University, Palo Alto, California, March 22-24,
2004
The question is not whether intelligent machines can have
emotions, but whether machines can be intelligent without any emotions.
Marvin Minsky, The Society of Mind
This page last modified: March 18, 2004
Recent years have witnessed increased interest
in modeling emotion within cognitive and behavior-based (software and
robotic) agent architectures and cognitive models of human
behavior. This interest results in part from advances in agent
technology, cognitive neuroscience and emotion research that make such
models possible, and in part from maturing applications that require,
or benefit from, the inclusion of different emotion-related aspects
(e.g., adaptive human-computer interfaces, social and expressive
robots, autonomous agents, decision support systems, etc).
This surge of interest has led to a number of
emotion-based architectures and applications. However, this work is
often carried in an 'ad hoc' manner since, due to the short history of
the field and the lack of appropriate frameworks for common
reflection, there is a still very limited understanding of the
mechanisms underlying such architectures, and of standards for a sound
validation practice. Researchers in this area increasingly perceive
the need to move in this direction to make work in the field progress
beyond mere engineering applications and towards a more scientific
discipline. The aim of the proposed symposium is to take a step in
this direction.
Back to Index
The objective of this symposium will be to
provide a forum focusing on mechanisms underlying agent architectures
that include or emphasize emotion. In particular, we will solicit
review and state-of-the-art contributions and will focus on two
aspects not contemplated in previous symposia and workshops:
validation of emotion models and architectures , and
relevance of recent findings from affective neuroscience
research, in addition to existing research in
psychology. Specifically, we wish to explore the ways in which
neuroscience and psychology results can motivate and inform the design
of emotion models and architectures, constrain specific mechanisms and
processes within these models, serve as a source of data for model and
architecture validation, and benefit from the feedback provided by
computational models and tools. To this end, we propose to bring
together researchers from a variety of backgrounds, in particular:
computer science, artificial intelligence, robotics, neuroscience,
cognitive science, and psychology. More precisely, symposium
objectives include:
- To bring together researchers within the focused area of emotion
models and architectures, in contrast to previous related symposia
which had a much broader scope.
- To enhance the general level of knowledge of psychological
and neuroscience research relevant to emotion architectures through a
series of focused keynote addresses by established experts in these
areas.
- To discuss the relevance of recent findings in
neuroscience and psychology emotion research to computational
models of emotion.
- To focus on issues of model validation and coupling of computational
modeling with empirical studies.
- To provide opportunities for discussing specific core issues regarding
mechanisms underlying emotion architectures that encompass different relevant
disciplines, as represented by the questions listed below.
- To provide a forum for presentation of on-going and emerging research
focusing on these core issues, and opportunities for discussion and feedback
from established experts for specific research efforts.
Back to Index
Some of the specific research questions and topics we propose to address
include the following:
- Levels of model granularity:
- What are possible alternative levels of granularity for modeling
emotion, and emotion-cognition interactions, in integrated architectures
(e.g., neuroanatomical circuits and components, aggregated functional
units)?
- When are lower-level / higher-level (functional) models more
appropriate?
- What research questions are best answered by models at distinct
levels of granularity?
- What are the time scales of emotion and of emotional processing
(e.g., "hot" emotions vs. moods)? Do they require different architectures?
Do they have the same functions?
- Availability of empirical data
- What is the availability of empirical data required to generate
models at different levels of granularity?
- Does neuroscience provide adequate data to allow modeling
these phenomena at the circuit and neuroanatomical components?
- Does psychology (and neuroscience) provide adequate data
supporting empirically-motivated choices for representation and inferencing
involving internal mental constructs and processes (e.g., goals, expectations,
plans)?
- Model and architecture design and development
- Which psychological theories, computational methods and representational
formalisms are appropriate for different types of architectures and architecture
components, at the different levels of granularity?
- What are the advantages/disadvantages and uses of different
'design' approaches (e.g., engineered, developmental, evolutionary)?
- Model and architecture validation
- How are the validation criteria and metrics different for different
levels of granularity? For different types of architectures? For different
phenomena modeled?
- What are the best ways of coupling computational modeling and
empirical research approaches, for the purposes of data generation
and hypothesis validation?
Back to Index
To support the objectives listed above, we propose
a focused and highly interactive symposium. Rather than a standard symposium
format, consisting of large numbers of short presentations and one or two
keynote addresses, the symposium will be organized around a series of:
- Working groups: 1 or 2 sessions in which the participants will split
into small groups to discuss particular issues and specific questions, decided
in advance (summaries from these groups will be presented to all symposium
participants during the 'general discussion sessions').
- Moderated panels.
- Invited keynote addresses from different disciplines.
- Poster sessions.
- General discussion sessions: one or two general sessions to present
and discuss the results from the working groups and to promote the exchange
of ideas among all participants.
- System demonstrations to provide hands-on experience with any working,
portable systems.
Working groups, panels and keynote talks will
each address a subset of the core questions of interest to the
symposium. The questions will be prepared by the chairs and organizing
committee (taking into account relevant feedback from the
participants) and distributed beforehand. Speakers, working groups and
panelists will thus be asked to focus on particular issues, while
general discussion sessions will have a more informal nature. In
addition, poster or demo sessions (about two) will be held to allow
participants to present their work, paying particular attention to how
this work addresses issues discussed at the symposium.
Back to Index
- Jean-Marc
Fellous, The Salk Institute for Biological Studies, USA. From
Human Emotions to Robot Emotions
-
Gerry Matthews, University of Cincinnati, USA. Designing
Personality: Cognitive Architectures and Beyond
- Aaron Sloman,
University of Birmingham, UK. What are emotion theories about
-
Craig Smith, Vanderbilt University, USA. A functional
perspective on emotion elicitation: Some considerations for the
development of emotional architectures
List available as PDF file
Updated schedule (version of March 18)
Registration
brochures are available from the AAAI web site.
Back to Index
- October 3, 2003: Submissions due
- November 7, 2003: Notification sent to authors
- January 5, 2004: Application deadline for student
financial aid
- January 12, 2004: Notification of financial aid awards
- January 15, 2004: Camera-ready papers e-mailed to chairs
- January 30, 2004: Electronic submission of
camera-ready papers to AAAI.
- February 13, 2004: Registration deadline for invited
participants
- February 20, 2004: Hotel reservation cut-off date
- February 27, 2004: Final (open) registration deadline.
- March 22-24, 2004: Symposium dates
Back to Index
Eva Hudlicka
(Chair), Psychometrix Associates, USA.
Lola Cañamero
(Chair), University of Hertfordshire, UK.
Cynthia Breazeal, MIT, USA
Jean-Marc Fellous, The Salk Institute, USA
Joseph LeDoux, NYU, USA
Jonathan Gratch, USC, ICT, USA
Christine Lisetti, University of Central Florida, USA
Gerry Matthews, University of Cincinnati, USA
Paolo Petta, ÖFAI, Austria
Fiorella de Rosis, University of Bari, Italy
Craig Smith, Vanderbilt University, USA
Back to Index
Back to Index
Go to Lola
Cañamero's
Home Page