A Functional Modeling Architecture for Re ective Agents

Consider the topic of reeection: how can a system understand its own reasoning? This has always been a central concern in the eld of Artiicial Intelligence. With the advent of software agent technologies, this topic becomes even more prominent. We claim that systems performing autonomous or semi-autonomous tasks in complex, dynamic, highly interactive environments can beneet enormously from a rich understanding of their own reasoning processes. This claim suggests two major issues: (i) how to represent a description of an intelligent agent, and (ii) what to do with such a description. To address (i), we have turned to past research on functional models in the domain of physical devices; we treat intelligent agents as abstract devices and our models of these devices are inspired by existing models of physical devices. This approach, in turn, suggests some approaches to (ii); existing model-based reasoning techniques may be adapted for use on models of agents. We introduce a system named Sirrine2 which provides not only an architecture for constructing and modeling intelligent agents but also a set of mechanisms which use these models. We describe an example of the system's use in the domain of meeting scheduling.