Reference: Hayes-Roth, B. An Architecture for Adaptive Intelligent Systems. 1995.
Abstract: Our goal is to understand and build powerful agents that function effectively in challenging niches. In particular, we identify a class of niches to be occupied by "adaptive Intelligent systems (AISs)," present a small class hierarchy of AIS niches, and our analysis with a particular niche: ICU (intensive care unit) monitoring. AIS niches are considerably more challenging than the niches occupied by typical AI agents. They require performance of diverse tasks. They provide variable resources for performing tasks. They present stressful contextual conditions. They impose conflicting performance criteria. Therefore, to function effectively in AIS niches, an agent must be highly adaptive. It must adapt its perceptual strategy to its dynamic cognitive requirements. It must adapt its control mode to uncertainty in its environment and constraints on its actions. It must adapt its reasoning tasks to demands and opportunities presented by its environment. It must adapt its reasoning methods to the available resources. It must adapt its meta-control strategy to its dynamic configuration of potential activities. We have designed and implemented an agent architecture that supports all of these different kinds of adaptation with a single underlying theoretical concept: An agent dynamically constructs explicit control plans to guide its choice among situation-triggered possible actions. The architecture has been used to build experimental agents for several AIS niches. We illustrate the architecture and its support for adaptation with examples from Guardian, an experimental agent for ICU monitoring.
Notes: Updated December 1993. Also as STAN-CS-93-1496. Updated April 1995.
Full paper available as ps.