NEMAS: Norm Entrepreneurship in Multi-Agent Systems

Abstract

We propose a framework that integrates norm entrepreneurship from human society into the dynamics of an agent society. Most work in agent coordination in a distributed environment studies norms that are provided to agents as part of their specification or distributed from centralised agents. Exploring an alternate perspective, we focus on peer-to-peer interaction by providing the agents with the freedom to initiate norm creation in demanding situations like a potential interference. This paper explores the concept of norm entrepreneurship through proactive establishment and emergence of norms by agents. A common approach in prior work focuses on coordination problems that are reduced to simple game theory models involving the simultaneous performance of a single action by each agent. Instead, we define the concept of a local coordination plan (LCP), which is a sequence of actions from each agent to cope with an interference in their normal course of action. We identify LCPs across various scenarios and abstract these plans using coordination state machines (CSMs). A coordination state machine contains a separate state machine for each agent where the states encapsulate the potentially constrained and suboptimal movement options agents have at a given time. We also explore how multiple LCPs lead to a coordination state machine of the same format and how a coordination state machine can abstract across multiple scenarios.