Artificial Intelligence through Simulated Evolution
Abstract¶
The advent of high speed computers has made radical changes in information technology. Yet, almost all computer utilization is unnecessarily restricted to the explicit calculations indicated by the equations that describe the solution to the problem of interest. Artificial Intelligence Through Simulated Evolution suggests and develops additional ways to make use of the computer. Specifically, the book describes a general technique which may be used to address a number of the fundamental problems of information technology including prediction, detection, discrimination, pattern classification, identification and the control of an unknown transducer. In the book, the authors show how the random mutation of an arbitrary logical “organism” yields an “offspring.” Both of these “machines” react to the available history and are evaluated in terms of their individual worth with respect to the given goal. The better of these machines is selected to serve as the new “parent.” Such fast-time mutation and selection is continued with real-time decisions being based upon the logic of the machine which survives at the time those decisions are required. The efficiency of evolution is improved by introducing a cost for the complexity of each machine. In addition to offering a new approach to some of the practical problems of information technology — evolutionary programming, as described here — opens the door to self-programming of computers and even to an automation of the scientific method.
Summary¶
Introduces evolutionary programming as an artificial intelligence method that treats intelligence as adaptive behavior produced by simulated evolution. The book develops populations of finite-state predictor machines, mutates their transition structures, and selects them by predictive performance, establishing one of the founding branches of evolutionary computation alongside later genetic algorithms and evolution strategies.