This work will study the evolution of communication in populations of agents, situated in simple, artificial environments. The primary research methodology is experimental and task-oriented in nature, involving the simulation of populations of male/female ANNs, using genetic operators (i.e. recombination, mutation and selection) that are applied to initially random populations of genotypes (which specify ANN connection weights and biases). Current simulations are being done both on computer workstations and a 16K Connection Machine (CM-2), whose massive SIMD parallelism is particularly suited to such simulations. These initial simulation studies will be headed in several direction -- ones that will support the evolution of more complex forms of communication. These directions include: (a) evolving more complex ANN architectures, (b) introducing multiple, distinct species in the environment, (c) adding simple forms of learning, metabolism and morphology, (d) augmenting our instrumentation for analysis and interpretation of the "meaning" of the dialects that evolve, and (e) increasing the complexity of the terrain and physics of the environment.
