Distributed Distributive Justice

Abstract

Some open systems must address a standard resource allocation problem: how to collectivise and distribute a set of common-pool resources, with respect to multiple criteriasuch as fairness, inclusivity and sustainability. Previous work in self-organising multi-agent systems formalised Nicholas Rescher's theory of distributive justice so that agents could self-organise the allocation according to contextualised legitimate claims. However, although this presented a viable solution in principle, in practicethe process contained two 'expensive' voting rounds and an(albeit elected) centralised 'head' role, compromising parallelism, scalability and security. This in turn could be a hindrance to its application in large, real-life systems. In this paper, we propose an alternative solution to the problem, based on direct interaction between agents in a decentralised manner. Instead of voting, we develop an autonomous mechanism of trust and reputation calculated independently by each agent from its local observationsand interactions. Simulations demonstrate the efficiencyand effectiveness of the new solution to find fair allocations indiverse scenarios, including situations with non-compliant agents. Moreover, the effect of topology on our algorithms capacity isexamined, showing that it is possible to achieve fair solutions even when agents do not have direct access to the whole network.