Variable coupling of agents to their environment (abstract)
George Kiss · ACM SIGOIS Bulletin · 1992
The paper identifies generality and power (processing work per unit time) as two major but conflicting requirements for the design of autonomous intelligent agents. A separation of these two concerns leads to the notion of a variable degree of causal coupling between parts of a mechanism and its environment in terms of space and time. An implementation strategy for variable coupling can be a layered architecture, where higher layers support generality and lower layers support power. The consequences of these distinctions for the choice of representations and processing strategies is discussed and illustrated. Some speculations are offered about the need for an agent self subsystem in the architecture in order to support mechanisms for full intentionality.