Community-Centred Agents and Emergent Social Conventions

In this paper, we examine techniques by which a community of autonomous agents can reach global agreement upon the use of social conventions in coordination and cooperation games. The merits of examining such convention acquisition is twofold: it can help us learn more about how humans behave in social systems, and it can explain how agents should behave in artificial social systems. We notice that one of the most established convention acquisition strategies seems to rely on some tenuous assumptions and does not represent human behaviour very well. Therefore, we invent new, community-centred convention acquisition strategies. A new environment that takes location into account is defined and these new strategies are examined against other well-established strategies in terms of convergence, stability and social welfare generation. The new strategies are concluded to be optimal in our environment and give rise to interesting human-like behaviours. Suggestions for making the simulation environment more realistic are given and future work is suggested.