Toward automated evolution of agent communication languages

Abstract

The aim of our research is to understand and automate the mechanisms by which language can emerge among artificial, knowledge-based and rational agents. We want to design and implement agents that, upon encountering other agent(s) with which they do not share an agent communication language, are able to initiate creation of, and further are able to evolve and enrich, a mutually understandable communication language. Our research is supported by the principled methodology of designing rational, socially competent artificial agents based on Bayesian probability and decision theories, and on the research in linguistics and cognitive psychology that addresses the issues of function, mechanisms, development, and evolutionary history of natural languages. In our work, we express some of the key insights obtained in linguistics and cognitive science in formal terms of decision theory and game-theoretic. mechanism design. We propose that the evolution of an agent communication language can be accomplished by the mechanism of negotiation, developed in economics and game theory, and automated in recent work in artificial intelligence. Negotiation is suitable because it can be mapped to settings in which rational interacting agents could use communication for their mutual, yet selfish, benefits. The agents can make mutually beneficial agreements that will allow efficient communication, but they have a conflict of interest about which language constructs to use each would prefer a communication language that is easier and less costly to use from their own individual perspective.