Abstract

Proceedings Abstracts of the Twenty-Third International Joint Conference on Artificial Intelligence

Macau: A Basis for Evaluating Reputation Systems / 191
Christopher J. Hazard, Munindar P. Singh

Reputation is a crucial concept in dynamic multiagent environments. Despite the large body of work on reputation systems, no metrics exist to directly and quantitatively evaluate and compare them. We present a common conceptual interface for reputation systems and a set of four measurable desiderata that are broadly applicable across multiple domains. These desiderata employ concepts from dynamical systems theory to measure how a reputation system reacts to a strategic agent attempting to maximize its own utility. We study a diverse set of well-known reputation models from the literature in a moral hazard setting and identify a rich variety of characteristics that they support.