Emergence of Protocols and Languages in Multiagent Systems

The objective of this research is to develop a fundamental understanding of the role, usefulness and emergence of commonality among autonomous interacting agents. The main thesis of this work is that the role and usefulness of emergent, and build-in, commonality can be properly investigated using the paradigm of utility maximization as defined by decision theory. Expected utility can be applied by individual agents as guide to how to achieve their goals; it is then referred to as the agents' rationality.

In this context we propose, and seek to verify during this research, that to understand the role of commonalities, such as rules of behavior, social laws, and communication languages, one should view them as being of value in decision-theoretic sense, either from the perspective of the individual agents, or from the perspective of the agents' designer.

The above point of view enables us to answer a number of important questions relating to protocols and social laws: How can the autonomous agents exploit regularities of their domain and develop a shared understanding of desirable rules of interaction by themselves? What learning methods can facilitate this development? Why should the autonomous agents adopt a shared rule of interaction? How can the agents estimate the utility of an emergent commonality with respect to their own goals? Also, given a rule of behavior that can be imposed on a multi-agent system by the designer, what is its global utility given the, likely incomplete, knowledge of the future parameters of the domain?

Due to similarities between physical interaction and communicative behaviors, the insight into commonalities in physical interaction sheds light onto emergent and imposed commonality in communicative behavior: Can agent communication languages arise in multi-agent societies without them being imposed in any form by the designer? What is the impact of knowledge representation and inference mechanisms used by the agents on their most suitable communication? What is the utility of imposing on the agents a particular communication protocol or a given communication language? Do the rational agents always have to follow an externally imposed communication convention? What mechanisms should be used by the agents to modify the imposed communication conventions?