Socially-Inspired Mechanisms for Restricting Exploitation in Artificial Agent Societies

Multi-agent systems;Social simulation

Savarimuthu, Amala Arul Sharmila ; Purvis, Maryam ; Purvis, Martin ; Savarimuthu, Bastin Tony Roy

University of Otago

关键词: Multi-agent systems;    Social simulation;   

Others  :  https://ourarchive.otago.ac.nz/bitstream/10523/1961/3/SavarimuthuAmalaAS2011PhD.pdf

美国|英语

来源: Otago University Research Archive

【 摘 要 】

Human societies have long cultivated the ability to organise themselves into groups and have also established formal or informal rules of behaviour that are expected within these groups. In the field of multi-agent systems, researchers are inspired by this ability of human societies to form groups and establish social control, and they have applied them to solve some of the problems in artificial agent societies. One of the problems in artificial agent societies is the problem of non-cooperation, where individuals have motivations for not cooperating with other agents. An example of non-cooperation is the issue of freeriding, where some agents do not contribute to the welfare of the society but do consume valuable resources. This can be likened to the ;;commons” problem. The way to address this problem is by imposing strict rules by centralised institutions. However, centralised solutions suffer from performance bottlenecks, and their scalability is poor. Towards this end, our first objective of this thesis is to investigate decentralised mechanisms for facilitating social control in agent societies. Our second objective is associated with an important attribute of modern artificial societies, which is the openness of such societies. Agents may join and/or leave these societies at any time. Towards this end, our second objective of this thesis is to investigate mechanisms which can handle the dynamism of open agent societies. Another key aspect in facilitating social control lies in employing appropriate mechanisms that can facilitate such control. In this thesis we are inspired by decentralised social practices found in human societies. This thesis investigates mechanisms that contribute towards the formation (via self-organisation) of different groups in an agent society based on their cooperativeness. It demonstrates that these mechanisms help in achieving the separation of good agents (cooperators) from bad agents (noncooperators) without expelling them from the society. It demonstrates how the concepts of tags can be used for group formation and how the information about the cooperativeness of agents in the society can be spread based on using socially-inspired mechanisms. It also investigates how monitoring and control mechanisms such as referrals, voting, gossip, resource restriction, and ostracism can be used in artificial agent societies. Thus our focus of this thesis is to develop socially-inspired mechanisms to facilitate self-organisation of groups in agent societies to restrict exploitation. We demonstrate that the formation of groups shield ;;bad” agents from taking advantage of ;;good” agents. We also demonstrate that the society is better off if the groups are organised based on their cooperativeness. Overall, the goal of this thesis is to investigate and demonstrate the new socially-inspired mechanisms for the self-organisation of groups in open, decentralized agent societies. This thesis initially systematically explores closed, centralized societies and gradually moves on to open, decentralised societies, since many real-life societies lie somewhere between these two ends of the open spectrum, with more and more societies lying closer to the end of full openness. We believe the mechanisms explored in this thesis can be applied to open, decentralized agent societies, such as electronic file-sharing societies to help avoid the problem of freeriding. The mechanisms proposed in this thesis could also be applied to organise agents into groups based on their behaviour, in virtual worlds and other online communities.