Reference: Hayes-Roth, B. A Multi-Processor Interrupt-Driven Architecture for Adaptive Intelligent Systems. June, 1987.
Abstract: An adaptive intelligent system reasons about and interacts with other dynamic entities in real time. Such a system requires capabilities for: (a) perception, to gain knowledge of other entities; (b) action, to affect other entities; and (c) cognition, to reason about and control its perceptions, actions, knowledge, and cognitions. Since, an adaptive intelligent system can perform many alternative operations - varying in complexity, urgency, and time requirements - at each point in time, it also requires an attentional mechanism to allocate its computational resources effectively in real time. This paper describes a class of MI architectures hypothesized to support a wide range of adaptive intelligent systems. MI architectures incorporate the basic reasoning capablilities of the BB1 blackboard control architecture with two innovations. First, to enable a system to perceive, act, and reason asynchronously and in parallel, MI architectures exploit multi-processor foundations. Second, to enable a system to guarentee a maximum latency to begin each successive operation, MI architectures utilize interrupt-driven execution cycles. The paper also discusses the relationship between Mi architectures and other AI research.
Notes: 19 pages.