Socially Intelligent Agents

 

Latest Meeting dedicated to "Socially Intelligent Agents": 2000 AAAI Fall Symposium Socially Intelligent Agents - The Human in The Loop, 3-5 November, Sea Crest Resort, North Falmouth, MA, USA. Summaries of working groups are available here. If you would like to see the contents of the Proceedings (Technical Report FS-00-04) and/or order a copy from AAAI then please go to AAAI Press. This year's related event: AAAI Symposium "Emotional and Intelligent II: The Tangled Knot of Social Cognition",November 2-4, 2001.

 

New Interdisciplinary Journal: Interaction Studies: Social Behaviour and Communication in Biological and Artificial Systems, John Benjamins Publishing Company. Editors-in-Chief: Kerstin Dautenhahn and Harold Gouzoules.

This Webpage emerged from the Socially Intelligent Agents, AAAI Fall Symposium which was held in the Tang Center at MIT in Cambridge, Massachusetts, November 8-10, 1997. Information about the proceedings content and how to order it can be found here. Extended contributions are published in special issues "Socially Intelligent Agents", Applied Artificial Intelligence, Vol 12 (7-8), 1998, and Vol 13 (3) 1999. Related to the symposium is the book Human Cognition and Social Agent Technology published by John Benjamins Publishing Company end of 1999. In October 1999 (Vol 5. No. 3) Social Intelligence was published, a special issue of Computational and Mathematical Organization Theory. A double special issue of the journal Adaptive Behavior with the title Simulation Models of Social Agents appeared in 2000, Vol 7, 3/4, table of contents. A Special Issue of Journal of Artificial Societies and Social Simulation (JASSS) on "Starting from Society - the application of social analogies to computational systems" appeared on 31st January 2001, volume 4, issue 1. A double issue on "Imitation in Natural and Artificial Systems" appeared in 2001, Cybernetics and Systems, volume 32, number 1-2. MIT Press  published an edited book on "Imitation in Animals in Artifacts" in spring 2002. A Special Issue of IEEE SMC "Socially Intelligent Agents - The Human in the Loop" 31(5) has appeared, see list of contents. Imitation in Animals and Artifacts, Cambridge, Mass., USA: MIT Press, 2002, guest editors Kerstin Dautenhahn and Chrystopher L. Nehaniv. See the table of contents. Socially Intelligent Agents - Creating relationships with computers and robots, edited by Kerstin Dautenhahn, Alan Bond, Lola Caņamero, Bruce Edmonds, Kluwer Academic Publishers. See the description of content (with ordering information) and table of contents (pdf document). 

Please send suggestions and comments on this page to Kerstin Dautenhahn (K.Dautenhahn@herts.ac.uk) who maintains this page. Last update: 17 December 2003. 

A short summary of the AAAI 1997 Fall Symposium (contribution in AI Magazine)

The symposium discussed sociality in software, robotic and animal agents. Bringing together researchers from different fields resulted in cross-disciplinary discussions on how a single agent is embedded in a social and cultural environment, how this agent interacts and communicates with each other agents, and how societies of agents are formed. The majority of approaches referred to human-style forms of social interaction which are for instance required in agents which assist, cooperate with, or represent a human being. Despite of technical and methodological differences in dealing with robotic and software agents, the symposium identified themes which cross the natural boundaries of agent species, for example believability, narrativity, imitation, emotions, personality, cultural adaptation, and the coupling of internal and external dynamics. 

A particular focus was on the role of the "human in the loop" as observer, designer, or user of social agents, for instance as a programmer of agent products, experimenter in robotics, and social interaction partner in software games and service robotics. Generally, agency and sociality are conceived of as characteristics of a system which can be objectively described and engineered. Current developments in areas like believable agents, interactive art, personal software assistants, and virtual pets question this assumption and point towards works which have been done in philosophy, arts, cultural theory, and social sciences. 

The symposium discussed both rational and "irrational" (emotional, subjective, inconsistent) aspects of socially intelligent agents, in this way stressing the need of a symbiosis between engineering and humanities in order to build expressive, interactive and social agents. This first Socially Intelligent Agents (SIA) meeting did not attempt to achieve definitions of agent, sociality and social agents. However, we started to discuss the complexity of social agents and appropriate design criteria in different applications. The symposium also addressed risks and opportunities provided by social agent technology. 

AI Magazine 19(2) 1998

A short summary of the AAAI 2000 Fall Symposium (contribution in AI Magazine)

The Symposium "Socially Intelligent Agents - The Human in the Loop" gave an overview of the state-of-the-art of theory and applications in the very active and highly interdisciplinary area of Socially Intelligent Agents (SIA). Much of this work is strongly inspired by forms of natural social intelligence characteristic of social animals, in particular, human beings (e.g. communicating verbally and non-verbally, expressing and recognizing emotions, reading another agent's 'mind', etc.). Different from meetings in Multi-Agent-Systems (MAS) or Distributed Artificial Intelligence (DAI), this symposium discussed the design and evaluation of socially intelligent agents with the "human in the loop", i.e. systems that can establish, maintain and develop social relationships with human beings. Here, the human can find herself e.g. in the role of user, observer, assistant, collaborator, competitor, customer, patient, or friend of such agents. The importance of such work is demonstrated in application areas such as e-commerce, agents for training, learning and therapy environments, agents for entertainment, and others. In all these application areas the human user's attitudes towards the agent, in terms of believability, credibility, trust, etc., are important factors that determine the acceptance and success of such a system and its utility in real-world applications. Therefore, an in-depth study is required of theories and models originally developed in areas such as psychology, brain research, ethology, and other fields not traditionally linked to the domains of artificial intelligence and software engineering. Also, human-agent interactions need to be studied and evaluated carefully: making agents "just like we are" is not necessarily desirable and/or feasible in particular application domains, cf. discussions onautonomy versus control in human-machine interface design. 

Environments and devices that interface humans (preferably non-intrusively) with agents and computers were another main theme of the symposium. It was generally felt that the more information an agent can acquire about intentions, emotions, beliefs and other internal and mental states of a human, the better the agent can adapt to the human and predict her behaviour and possibly changing attitudes. This is particularly relevant in scenarios where agents are supposed to have 'long-term' relationships with human beings and requires adaptation and learning abilities for agents, i.e. being able to individually identify the human, as well as learning and acting upon interaction histories with humans. 

Particular research challenges presented at the symposium were projects that study heterogeneous agent societies, i.e. how communities of agents (hardware or software) can be integrated and perform useful tasks in human societies. In other projects the application area itself poses a particular challenge, e.g. agent systems that are used in social problem solving for adults or therapy for children with autism. In those application areas it is not only desirable but an explicit goal that the agents are 'persuasive', i.e. change behaviour and attitudes of the humans interacting with them, and therefore requires careful consideration of ethical issues.

Presentations, working groups and general discussions at the symposium demonstrated 1) significant advancements in the field, compared to the 1997 AAAI FS "Socially Intelligent Agents", 2) the importance of interdisciplinary work that could advance the development of a 'theory of social minds', both natural and artificial, and 3) the identification of particular research challenges (e.g. unconstrained scenarios, heterogeneous scenarios) and challenge scenarios that can help to further develop the field and systematically explore design spaces and spaces derived from application areas with particular requirements. Future developments in social, emotional and narrative intelligence research might lead to 'truly' socially intelligent agents, e.g. agents that have 'natural' (non-trivial) conversations with us, can recognise people as individuals and 'mental agents' with distinct emotions and personality, and ultimately agents that truly 'care about us'.

Links relevant to SIA research

Journals, General

·         User Modeling and User-Adapted Interaction, Kluwer Academic Publishers

·         WEB INTELLIGENCE AND AGENT SYSTEMS: An International Journal,  Publisher: IOS Press, ISSN 1570-1263

Journals, Special Issues

 

Books

Mailing Lists, Archives

 Conferences, Workshops in 1998, 1999 and 2000.

Conferences, Workshops 2001

Conferences, Workshops 2002

Conferences, Workshops 2003

Conferences, Workshops 2004

·        CHI2004 Workshop on Shaping Human-Robot Interaction Understanding the Social Aspects of Intelligent Robotic Products

Societies, Organisations, Networks

Links, Papers, Research Groups, Discussions (a bit out of date since I don’t check the URL’s regularly)


Kerstin Dautenhahn, 6/1/2003