SSP Group Meeting

11am, 4 October, 2005
Room 4.03, Appleton Tower
CISA, School of Informatics
University of Edinburgh

Social Reasoning in Computational Multi-Agent Systems

Michal Pechoucek

The concept of multi-agent system contributes to the fields of computer science and artificial intelligence by a societal and distributed approach to computing. Non-trivial problems can be solved, complex systems can simulated and controlled by a community of autonomous computational entities. These entities can be geographically distributed across wider network infrastructures.
The computational entities in multi-agent systems need to perform specific types of social interaction in order to achieve collective behaviour and collective decision making. Socially oriented reasoning, the reasoning process that underlies rational interaction, is currently a subject of a deeper theoretical investigation and practical deployment in industrial applications.

We will introduce the concept social reasoning and social knowledge, comment various types of social knowledge and present two specific techniques for maintenance and exploitation of social knowledge in computational multi-agent systems: (i) the acquaintance models, the knowledge structures representing agent awareness of its interaction neighbourhood -- mainly in competitive and adversarial environment (ii) the stand-in agents, that are autonomous computational entities freely migrating within the network and providing vital social information in e.g. disruptive and partially inaccessible environment.