Organizations of self-interested agents

Abstract

Existing agent-based systems apply the productive power of Multi-Agent Systems (MAS) and Distributed Artificial Intelligence (DAI) techniques. However, little work has demonstrated agents autonomous enough to form concurrently operating non-trivial, non-predetermined organizations where agents form such organizations specifically to improve their individual welfare through improving their collective productivity. In my work, organizations are defined as sets of autonomous agents that reach, and generally abide by, agreements to exchange, over periods of time, computational services for value, and where such agreements are voluntary by each agent. The dissertation1 identifies and explores techniques by which autonomous agents may form organizations that persist over time, in dynamic environments, and interact with agents outside the organization. Curiosity about the computational basis of persistent organizations is a driving force behind this work. Extensive writings in Economics, Sociology, and Political Science purport to explain why certain human organizations operate the way they do. From a computational view, and in particular from an Artificial Intelligence perspective, there is no welldeveloped modeling approach that supports building models with the large number of agents necessary to model human social behavior using the terms of these other areas of study when they appeal to methodological individualism on a large scale. If we project recent rates of advancement in computational power, we find we can project the power to support detailed and explicit models of the behavior of each of the individual agents within an organization or society. My dissertation addresses the methods needed to harness that growing power to the creation of large-scale, detailed, models of interrelationships among large numbers of individual decision units, which decision units are “agents” in Distributed Artificial Intelligence work. My thesis is that the formation and operation of organizations is rational and computable, and can be modeled using Artificial Intelligence tools. I extend Multi-Agent Systems technologies by introducing procedures by which sovereign computational agents may form organizations. A sovereign