An emergence of coordinated communication in populations of agents with evolution simulated by genetic algorithm

摘要

The purpose of this communication is to demonstrate that in a population composed of agents that are capable of simple cognitive activities spontaneously emerges coordinated communication between agents. Each agent is characterized byu meaning vectors (internal states) represented by n-dimensonal binary vectors. Cognitive activities of agents are performed by simpe formal device represented by mappings that map binary vectors onto symbol strings (signals) and conversely. An elementary communication act consists in (1) a random selection of two agents, where one of them is declared as speaker and the other one as listener, (2) the speaker codes its randomly selected meaning vector into a sequence of symbols and sends in to the listener as a message, and finally, (3) the listener decodes this received meassage into a meaning vector. A Darwinian evolution of population is simulated by simple version of genetic algorithms, whwere agent mappings are considered as chromosomes and their fitness is evaluated on the basis of distances between speaker meaning vector and listener meaning vector constructed from the received messages. If these distances are small, then for both agents, speaker and listener, fitness is increased. It is demonstrated that in the course of evolution agents gradually improve decoding of the received messages (they are closer and closer to meaning vectors of speakers) and all agent gradually start to use tightly related cognitive devices, i.e. all agents start to use the same vocabulary for common communication. Moreover, if agent meaning vectors contain regularities, then these regularities are manifested also in meassages created by agents-speaker, i.e. similar parts of meaning vectors are coded by similar symbol substrings. This observation is considered as a manifestation of an emergence of a grammar system in the common coordinated communication.