From Genetic and Evolutionary Optimization to the Design of Conceptual Machines

This paper considers some of the methodological lessons the author has learned in moving from applications of genetic and evolutionary optimization (GEO) to the design of GEO algorithms that work. Speciically, the cultural divide in modeling methodology between those who design \conceptual" machines|entities such as programs and algorithms|versus those who design \material" machines| entities such as mechanical and electrical devices|is crossed, and is replaced with an economic model of the modeling process. For designers of GEO and related procedures this suggests that modeling and experimentation costs should not be ignored, and that a level of modeling and experimental rigor should be adopted that appropriately beneets the advance of the technology. Two kinds of \little" models borrowed from the engineering of material machines are used to show their utility in a GEO design problem: the design of a selectorecombinative genetic algorithm. This style of modeling yields remarkably powerful analytical assistance at very low cost in derivation and use. The paper concludes by abstracting the lessons of this approach and recommending their use across the spectrum of conceptual machine design.