(WITH APOLOGIES FOR MULTIPLE POSTINGS)

We write to ask for non-binding expressions of interest in a possible AAAI
Fall Symposium on Chance Discovery,  proposed to be held in the northern
hemispheric Fall of 2002 on the east coast of the USA.

Chance events are rare or novel events with potentially significant
consequences for decision-making in some domain.   It is planned that this
symposium will be devoted to two primary questions:  How may we predict,
identify or explain chance events and their consequences? ("Chance
Discovery") and How may we assess, prepare for or manage them? ("Chance
Management").   Further information on the subject of the proposed symposium
is attached below this email.   In addition, a review of a recent workshop
in Japan on this topic can be found at:

 http://www.csc.liv.ac.uk/~peter/cd2001.html


The American Association for Artificial Intelligence (AAAI) conducts two
sets of symposia, in the Spring and Fall of each year, with the aim of
discussing emerging subjects which do not yet have their own regular
conferences.   These meetings generally consist of 25-70 people, and
encourage inter-disciplinary themes and participation, both within AI and
with other disciplines.   Further information about the AAAI Symposia can be
found on the Association's pages at:

 http://www.aaai.org/Symposia/symposia.html


We seek expressions of interest from anyone potentially interested in
attending or contributing to the proposed symposium.   These expressions are
for planning purposes only, and will not be taken as firm commitments to
attend or to participate.   In due course, if the proposed symposium is
approved by the AAAI, a Call for Participation will be broadcast.

Please send expressions of interest by email to Peter McBurney at:

 [log in to unmask]


With thanks,








Yukio Ohsawa, Simon Parsons and Peter McBurney

****************************************************************

Peter McBurney
Agent Applications, Research and Technology (Agent ART) Group
Department of Computer Science
University of Liverpool
Liverpool L69 7ZF
U.K.

Tel:  + 44 151 794 6768
Email: [log in to unmask]
Web page:  www.csc.liv.ac.uk/~peter/

*****************************************************************************



PROPOSED AAAI FALL 2002 SYMPOSIUM:

A. TITLE:  Chance Discovery:  The Discovery and Management of Chance Events

B. DESCRIPTION:

Chance events are rare or novel events with potentially significant
consequences for decision-making in some domain.   This symposium will be
devoted to two questions:  How may we predict, identify or explain chance
events and their consequences? ("Chance Discovery") and How may we assess,
prepare for or manage them? ("Chance Management").


A robot or agent engaged in planning needs to adopt a view of the future:
In order to decide its goals, and then to decide which is the best sequence
of actions to achieve these goals, the agent must make some assumptions --
explicitly or implicitly -- regarding the evolution (or constancy) of its
environment and the impact of its planned actions on that environment.   Any
rational action requires the prior adoption of a forecast; moreover, because
the environment may respond in unexpected ways to the actions of the
agent, the agent's forecast also needs continual revision in the light of
events.


But how can an agent (or a group of agents jointly) discover rare or
novel events and forecast their consequences?  Their very unlikeliness makes
them difficult to predict or explain by methods which use historical data or
pattern-matching -- the methods of time series analysis, data mining or
association rule detection.     One can think of this prediction or
explanation problem as a search for a global maximum (or minimum) of a
surface whose shape and features are unknown, in a space whose dimensions
may also be unknown.   What search algorithms are known to work well in
conditions of such ignorance?


And once a rare event has been identified, what are its consequences for the
agents concerned?   A planning agent cannot usually ignore these events, as
their consequences may significantly impede or facilitate the achievement of
its goals.   For example, although strong earthquakes occur in major urban
centres only rarely (relative to all the earth tremors that occur around the
world), such earthquakes tend to have human and economic consequences well
beyond that of the typical tremor.  A rational public safety body for a city
in an earthquke-prone area would plan for such contingencies even though the
chance of a strong quake is still very small.


In talking of chance events, the word "chance" is used, rather than (say)
"risk" because such events are not necessarily bad in their consequences.
Forecasting market demand for innovative products is an example.  We can ask
potential customers what they may think of a new product, but if they have
no experience of it, or of anything similar, their responses may not be
meaningful.  Nor may past demand for similar or substitute products provide
us with much guidance:  it makes no sense, for instance, to forecast the
demand for a new bridge by counting the numbers of people who currently swim
across the river it will span.    And what appears as a positive chance
event to some -- bridge builders, for example -- may be catastrophic to
others -- e.g. operators of river boats.  Thus, the perception and
management of chance events may be crucial to their definition as chance
events.


And, for many domains, the interactive nature of the relationship between an
agent and its environment may generate conceptual and modeling subtleties,
such as self-fulfilling and self-denying prophecies.  If an agent is very
powerful relative to the other entities in its environment, as is the case
for a sole provider of a highly-demanded software product for instance, its
predictions of the future may be realized by the very nature of its
environmental power rather than being evidence of its prognostic
capabilities.


Although of importance to rational agents or robots engaged in planning,
these examples demonstrate that the topic of Chance Discovery/ Management
covers a range of problems that have already arisen in other fields.
Examples include:

(a) demand forecasting for what marketers call "really new products";

(b) opportunity identification in business and marketing;

(c) hypothesis discovery in scientific theories;

(d) risk assessment and management in systems engineering, in environmental
management and in medicine;

(e) natural disaster prediction and management in public safety;

(f) identification of emergent behavior in systems of agents;

and

(g) the relationships between individual agents and the system in complex
systems theory.

However, the subject of Chance Discovery/Chance Management is not only
concerned with the techniques used in particular domains, but also with the
overall  relationship of an agent with its environment as the two, working
jointly, discover and manage chance events.   This holistic focus on the
agent and its environment as one, interacting, system is another point of
difference between the domain of Chance Discovery/Management and traditional
approaches to forecasting or pattern matching.


This proposed AAAI symposium will seek to bring together members of the AI
community with people from various application domains to share methods and
approaches to this set of problems.  Because many of the techniques in risk
analysis, demand forecasting, etc, were first developed by applied
practitioners in industry and government, it is hoped that members of these
professional communities will also participate, in addition to the
respective academic communities.


Moreover, as well as the sharing of approaches from different disciplines,
it is hoped that the Symposium will encourage and facilitate alternative
formalizations of the problems of Chance Discovery/Management and their
possible solution.  Formal methods of prediction and management of rare
events will be required if these techniques are to be adopted by planning
agents and robots acting in the world.    Because our interest concerns the
relationship between an agent and its environment, research in
human-computer interaction (HCI) may also be relevant, since the design of
an interface may facilitate (or inhibit) the discovery and management of
chance events.


Because this a new and multi-disciplinary domain, no regular conferences
or meetings devoted to it yet exist, although ad-hoc sessions on chance
discovery have been held in association with other international
conferences.  Recently, a one-day international workshop on Chance Discovery
was held as part of the 15th Annual Meeting of the Japanese Society for AI
(JSAI-2001), in Matsue, Japan, in May 2001.  This workshop was organized by
one of the co-chairs of this proposed AAAI Symposium, Yukio Ohsawa, and
attracted 25 participants;  a review of that meeting, written by Peter
McBurney, will appear shortly in the journal "Knowledge Engineering
Review".   This review is available on line at:

  http://www.csc.liv.ac.uk/~peter/cd2001.html



C. PARTICIPATION:

The organizers of this Symposium encourage participation from different
communities, both academic and industrial, including:

- Artificial Intelligence (particularly the multi-agent systems and planning
communities)
- Knowledge Discovery and Data Mining
- Information retrieval and analysis
- Human-Computer Interaction
- WWW Awareness
- Marketing theory and demand forecasting
- Risk analysis, prediction and management
- Social trends analysis
- Social psychology
- Management and decision sciences
- Operations research
- Statistics and data analysis
- Complex systems theory and application
- Philosophy of forecasting and risk.



D. FORMAT:

It is proposed that the format of the symposium will be a combination of:

- 1 or 2 Invited talks (each c. 45 minutes, including discussion)

- Presentations on contributed papers (c. 30 min)

- For each presentation, a short critique by a reviewer, nominated before
the conference (10 min),  The reviewer will have had time before the meeting
to read and reflect on the paper being critiqued, and so may be able to
provide a more considered response to it.

- Shorter presentations of posters (5-10 min)

- Panel discussions (c. 30 min)

- Parallel break-out sessions to discuss particular problems, e.g.
forecasting demand for hi-tech products; risk analysis for complex systems;
integration of technical and social risk analysis (c. 60 min).



E. ORGANIZING COMMITTEE:

Yukio Ohsawa
Graduate School of  Systems Management
University of Tsukuba
3-29-1 Otsuka, Bunkyo-kyu
Tokyo 112-0012
Japan
Tel:     + 81 3 3942 7141
Email:   [log in to unmask]


Simon Parsons
Agent Applications, Research and Technologies (Agent ART) Group
Department of Computer Science
University of Liverpool
Chadwick Building, Peach Street
Liverpool L69 7ZF
U. K.
Tel:     + 44 151 794 6760
Email:   [log in to unmask]


Peter McBurney
Agent Applications, Research and Technologies (Agent ART) Group
Department of Computer Science
University of Liverpool
Chadwick Building, Peach Street
Liverpool L69 7ZF
U. K.
Tel:     + 44 151 794 6768
Email:   [log in to unmask]


**************************************************************************