Risk management in intelligent agents
- Publication Type:
- Issue Date:
This thesis presents the development of a generalised risk analysis, modelling and management framework for intelligent agents based on the state-of-art techniques from knowledge representation and uncertainty management in the field of Artificial Intelligence (AI). Assessment and management of risk are well established common practices in human society. However, formal recognition and treatment of risk are not usually considered in the design and implementation of (most existing) intelligent agents and information systems. This thesis aims to fill this gap and improve the overall performance of an intelligent agent. By providing a formal framework that can be easily implemented in practice, my work enables an agent to assess and manage relevant domain risks in a consistent, systematic and intelligent manner. In this thesis, I canvas a wide range of theories and techniques in AI research that deal with uncertainty representation and management. I formulated a generalised concept of risk for intelligent agents and developed formal qualitative and quantitative representations of risk based on the Possible Worlds paradigm. By adapting a selection of mature knowledge modelling and reasoning techniques, I develop a qualitative and a quantitative approach of modelling domains for risk assessment and management. Both approaches are developed under the same theoretical assumptions and use the same domain analysis procedure; both share a similar iterative process to maintain and improve domain knowledge base continuously over time. Most importantly, the knowledge modelling and reasoning techniques used in both approaches share the same underlying paradigm of Possible Worlds. The close connection between the two risk modelling and reasoning approaches leads us to combine them into a hybrid, multi-level, iterative risk modelling and management framework for intelligent agents, or HiRMA, that is generalised for risk modelling and management in many disparate problem domains and environments. Finally, I provide a top-level guide on how HiRMA can be implemented in a practical domain and a software architecture for such an implementation. My work lays a solid foundation for building better decision support tools (with respect to risk management) that can be integrated into existing or future intelligent agents.
Please use this identifier to cite or link to this item: