The Effect of Structured Interactions on the Development of Fairness in the Ultimatum Game

Abstract

Software agents are increasingly employed to negotiate on behalf of human users as a solution for human disadvantages. Agents are however less adequate at addressing social dilemmas in which cooperation as best outcome is complicated due to conflicting interests. Human behaviour in such situations is mediated by considerations of fairness. An often used model for social dilemmas that emphasises bilateral negotiation, is the Ultimatum Game. Previous attempts to accurately model human negotiation behaviour in the Ultimatum Game have proposed extensions that are exploitable or not realistically applicable to real negotiation situations. An extension endogenous to such situations is structure from complex spatial networks, which was shown in other social dilemma models to provide fairer solutions. Implementations in the Ultimatum Game are however limited. In this project, complex network structure is implemented as an extension of the Ultimatum Game. We study how clustering and degree-heterogeneity as network characteristics influence the evolution of fair negotiation strategies and allocation outcomes. To accomplish this, a multi-agent model is designed to simulate the Ultimatum Game played in a population with differing interaction structure. We find that clustering and degree-heterogeneity are favourable to the evolution of fair negotiation strategies. Additionally, for both network characteristics we find an increase in homogeneity for utility per interaction. Our findings show that fair negotiation behaviour emerges despite self-interest in structured interactions. They further suggest that an automated negotiation system embedded in a spatial interaction network is more appropriate for social dilemmas.