A Von Neumann Approach to a Genotype Expression in a Neural Animat

This paper describes an approach to evolution upon an artificial insect existing in a simulated two-dimensional world, comprised of a “spreading-activation” neural network. Study ofthe difficulty of hand-building neural networks has resulted in a genetic expression, based on the notion of the Von Neumann computer architecture and the biological principles of DNA, which incorporates both operations and data in a simulated DNA strand. Work in progress indicates that by embodying behavior parameters and actual neural connections in a genotypic “language,” and by expressing that phenotypically as an animat, the computer simulation appears able to evolve a better species of animat through mutations upon the genotype.