What Can We Learn from Genetic Algorithms: A New Perspective for Search Methods

摘要

This paper is concerned with search methods, in particular with geneticalgorithms (GAs) and their status in artificial intelligence (AI). We describe and study GAs in the context of classic AI search techniques, such as depth-first or A*. Systematically studying similarities and differences between general search methods and genetic algorithms we claim a place for GAs in the AI search family. GAs are distinguished by relying on a fixed data structure of the search space and a standard child construction mechanism. In the second part of this paper we surpass GA traditions concerning these aspects and generalize along two axes. We discuss non-standard data types and child construction mechanisms with an arbitrary number of parents. Hereby we also achieve a general framework, a shell, to define basic search operators. Finally, we summarize what we can learn from genetic algorithms and sketch the outlines of a scientific program for developing and investigating new kinds of search methods.