A Theory of Reflective Agent Evolution
Murdock, J. William
MetadataShow full item record
Intelligent agents typically operate in complex, dynamic environments. Such environments require that an agent be able to adapt to meet new demands. One promising strategy for such adaptation is to have an agent reason about itself using an explicit model of the tasks it performs, the methods it uses to perform those tasks, and the knowledge which those tasks and methods manipulate. I propose to investigate the application of this reflective self-modification process to agents situated in interactive domains. This investigation will consist of three major components: (i) implementing an agent architecture which provides autonomous support for model-based evolution, (ii) instantiating a variety of specific agents into this architecture, and (iii) systematically varying both the descriptions of the agents and the design of the architecture itself. I expect this research to provide insight into the nature of agent evolution and the potential role of model-based reflection.