Sort out Your Neighbourhood: Public Good Games on Dynamic Networks

Axelrod (1984) and others explain how cooperation can emerge in repeated 2-person prisoner's dilemmas. But in public good games with anonymous contributions, we expect a breakdown of cooperation because direct reciprocity fails. However, if agents are situated in a social network determining which agents interact, and if they can in uence the network, then cooperation can be a viable strategy. Social networks are modelled as graphs. Agents play public good games with their neighbours. After each game, they can terminate connections to others, and new connections are created. Cooperative agents do well because they manage to cluster with cooperators and avoid defectors. Computer simulations demonstrate that group formation and exclusion are powerful mechanisms to promote cooperation in dilemma situations. This explains why social dilemmas can often be solved if agents can choose with whom they interact. More than 20 years after Axelrod's seminal computer tournaments and the discussion about direct reciprocity in repeated games [Axelrod1984], we still haven't understood all mechanisms leading to sustained cooperation. Most settings researchers have looked at are based on two unrealistic assumptions. Firstly, most models to explain cooperation comprise games with two players. 2-person prisoner's dilemmas, in particular, have received much attention. But more realistic settings involve more than two persons. Cooperation becomes much harder in multi-person dilemmas. Under realistic assumptions, it breaks down because targeted reciprocation against defectors is not possible. Secondly, most models do not take the e ects of social structure into account. They assume random matching of strategies or a tournament of pairwise matching. But the reality of human interaction looks quite di erent: People are situated in a social network. They meet some people much more often than others. Moreover, people can in uence with whom they interact, thereby changing the social structure that determines who interacts with whom. On the one hand, moving from 2-person to multi-person games makes the emergence and maintenance of cooperation harder. On the other hand, social structure often makes it easier. When we allow agents to change the social network, cooperation can emerge, even in settings quite hostile to cooperation. In this paper I present a model where agents are situated in a network they are able to change over time. They play public good games with their neighbours,