Agents That Plan
7.1 Memory Versus Computation
The action functions of reactive agents (Figures 2.2, 5.1, and 5.3) do very little computation. Essentially, these agents have their actions selected for them—either by their designers, by learning, by evolutionary processes, or by some combination of these sources. The action functions themselves can be implemented by tables, production rules, or combinational logic circuits that prescribe actions given a feature vector. Such implementations lie toward the “space” end of the classic space-time tradeoff in computer science. They are space- or memory-based implementations—compilations of the designer’s knowledge.
A broadly competent reactive machine, able to perform complex tasks in complex environments ...