Agents in Interaction

Call For Papers

Full-day Workshop
May 10, 1998
Minneapolis/St. Paul

AGENTS IN INTERACTION - ACQUIRING COMPETENCE THROUGH IMITATION

Associated with
Second International Conference on AUTONOMOUS AGENTS (Agents '98)
Minneapolis/St. Paul, May 10-13, 1998

Co-organisers: Kerstin Dautenhahn * and Gillian Hayes**
*Department of Cybernetics, University of Reading, UK
**Department of Artificial Intelligence, University of Edinburgh, UK

Submission Deadline: 15 January, 1998

New: Program

The scope of the workshop encompasses social learning and imitation as a means of one agent, software or embodied, learning an individual behaviour pattern or utterance from a member of the same or a different species and including it in its own behavioural repertoire. The workshop is intended to attract people from different communities where social learning and imitation is involved, i.e. where agents learn from each other or their users through interaction.

AREAS OF INTEREST

IMITATION AND SOCIAL LEARNING IN AGENTS

Imitation is supposed to be among the least common and most complex forms of animal learning. It is found in highly socially living species which show, from a human observer point of view, 'intelligent' behaviour and signs for the evolution of traditions and culture. There is strong evidence for imitation in certain primates (humans and chimpanzees), cetaceans (whales and dolphins) and specific birds like parrots. Recently, imitation has begun to be studied in domains dealing with such non-natural agents as robots, as a tool for easing the programming of complex tasks or endowing groups of robots with the ability to share skills without the intervention of a programmer. Imitation plays an important role in the more general context of interaction and collaboration between agents and humans, e.g. between software agents and human users. Intelligent software agents need to get to know their users in order to assist them and do their work on behalf of humans. Imitation is therefore a means of establishing a `social relationship' and learning about the actions of the user, in order include them into the agent's own behavioural repertoire.

The role of imitation as an effective learning mechanism is important in engineering domains. The main aims are to develop imitation as a machine learning method for an agent, allowing it to learn from one or very few examples which are performed by a model, and to facilitate indirect knowledge transfer from one agent to another. The latter becomes more and more interesting for scenarios where interactions between heterogeneous agents are studied, because in these situations the simple transfer of a successful control program from one agent to another is often impossible because of great differences in construction and behavior characteristics. Examples include the use of imitation of movements by a robot to learn a navigation task, and the acquisition of a synthetic robotic language by observation. The obvious extension is to situations where the agent has to learn from a human `model' in, for example, the context of service robots which must adapt to humans and cooperate/work hand-in-hand together with humans. In Artificial Life research on individualized robot societies, imitation is used as a social mechanism for identifying and building up social relationships towards robot group members. In Software Agent research, imitation is used as a means of enabling agents to adapt to one another and develop a coherent group behaviour.

As a research topic imitation tackles such fundamental problems as sensory intelligence, motor control, real-time learning architectures, intermodal representation, social interactions, motivational and emotional control of behavior, and scaling-up from sensorimotor intelligence to cognitive systems. Generally, different mechanisms are studied in these different domains, so the problem arises of integrating them in a common framework. The topic of imitation is broad enough to cover all these interesting issues.

The aim of the workshop is to draw together researchers working in software, hardware and wetware fields with the common goal of understanding the role of social learning in making agents useful, believable, acceptable or simply natural.

WORKSHOP FORMAT

The Workshop will comprise a few keynote talks, a panel discussion with participants from different research areas, and sessions with presentation of state-of-the-art social learning and imitation research.

SUBMISSION DETAILS

People who are interested in participating in the workshop are asked to submit an extended abstract (not more than 4 pages). Please submit 4 hardcopies to:

Kerstin Dautenhahn
Department of Cybernetics
The University of Reading
Whiteknights, PO Box 225
Reading, RG6 6AY, United Kingdom
tel: +44 (0) 118 931-8219 or -6372
fax: +44 (0) 118 931-8220
K.Dautenhahn@cyber.reading.ac.uk

An email submission in plain ascii format is also possible, postscript submissions cannot be accepted.

IMPORTANT DATES

January 15, 1998: Workshop papers due
February 28, 1998: Notification
March 30, 1998: Final Copies for Workshop Notes Due

The workshop participants are required to register for the AA'98 main conference. Workshop registration will be handled by the AA'98 Committee along with the main conference registration.

PROGRAM COMMITTEE

Kerstin Dautenhahn (University of Reading, UK)
Gillian Hayes (University of Edinburgh, UK)
Roy Middleton (Edinburgh Virtual Environment Centre, UK)
Peter McOwan (University of Reading, UK)
Simon Penny (CMU, USA)
Paolo Petta (Austrian Research Institute for Artificial Intelligence)
Angi Voss (GMD, Germany)

10th of May 1998, Hyatt Regency Hotel

Morning Session 9:00-12:30

9:00-9:25

Luc-Laurent Salvador (Neuchatel, Switzerland):
The mechanical and universal nature of imitation

9:25-9:50

Chrystopher Nehaniv and Kerstin Dautenhahn (Aizu, Japan and Reading, UK):
Mapping between dissimilar bodies: affordances and the algebraic foundations of imitation

9:50-10:15

Gal A. Kaminka and Milind Tambe (University of Southern California, USA):
What is wrong with us? Improving robustness through social diagnosis

10:15-10:40

Mathias Bauer and Dietmar Dengler (DFKI, Germany):
TrIAs - an architecture for trainalbe information assistants

10:40-11:10 COFFEE BREAK

11:10-11:35

Winton Davies and Peter Edwards (Aberdeen, UK)
Communicating inductive inferences in agent societies

11:35-11:50

Gary Boone (Georgia Institute of Technology, USA)
Imitation in information management agents

11:50-12:30 Discussion

Afternoon Session: 14:00-17:30

14:00-14:25

Brian Scassellati (MIT, USA)
Imitation and mechanisms of shared attention: a developmental structure for building social skills

14:25-14:50

Cynthia Breazeal (Ferrell), MIT, USA:
Regulating human-robot interaction using "emotions", "drives" and facial expressions

14:50-15:10

Jim Wyatt, Kerstin Dautenhahn, David Keating (Reading, UK):
Proposed scenario for the study of social learning and imitation amongst agents in a co-habited mixed reality scenario

15:10-15:35

John Demiris and Maja J. Mataric (Edinburgh, UK and Los Angeles, USA):
Perceptuo-motor primitives in imitation

15:35-16:05 COFFEE BREAK

16:05-16:15 Discussion paper by Angi Voss (GMD, Germany)
Behavior imitation- how could it function, what could be gained?

16:15-17:30 Panel and General Discussion

17:30 END OF WORKSHOP



Kerstin Dautenhahn, 23/4/1998.