Distributed search and conflict management among reusable heterogeneous agents

The current state of knowledge-based technology is such that most application systems are built from scratch. In order to move beyond the prohibitive cost of constantly reinventing, rerepresenting, and reimplementing the wheel, researchers are beginning to examine the feasibility of building application systems with heterogeneous and reusable agents. In this dissertation, we explore a comprehensive model of distributed-search systems comprising multiple agents where each agent is a complete and independent system that represents a specific area of expertise. Our approach acknowledges the inevitability of conflict among the agents, and exploits that conflict to drive agent interaction and guide local search. A primary objective of the research is to understand how to build distributed-search systems that can integrate multiple heterogeneous computational agents. These systems must be able to: (1) solve large-scale problems; (2) take advantage of knowledge embodied in existing reusable agents; (3) effectively coordinate the actions of agents to achieve efficient and coherent problem solving; and (4) effectively manage conflicts among agents, allowing them to find high-quality solutions despite inherent inconsistency. A second objective is to understand how to build expert reusable computational agents that (1) use the most appropriate representations, algorithms, and inference engines for their area of expertise; (2) coordinate with other agents to effectively work within a team; and (3) are developed outside the scope of particular application system. In this dissertation, we introduce the conceptual model we have developed for distributed search and conflict management among reusable heterogeneous agents. We present an implemented framework that provides flexible architectural support for agent integration, coordination, conflict management, and for multiagent solution evaluation. We also describe two implemented application systems built within the framework, and report results from our experimental investigation of the efficiency and effectiveness of agent configurations within those systems.