Emergent social norms and their interaction with other means of behavioral control in complex adaptive agent societies

In agent societies, social norms emerge from interactions and sharing of information between members of a society [1]. Social norms play a crucial role in many contexts in which (human) agents interact and make decisions, such as politics or organizations. Topics of emergence and enforcement of such norms have been studied in various fields, e.g. multi-agent systems [2]. The economic literature has studied the interplay between exogenously defined social norms and incentive mechanisms, but has widely failed to properly address the emergence of social norms [3]. We follow this line of research and aim at (better) understanding how social norms that emerge in agent societies interact with other means to control behavior in agent societies.
Based on the NK-framework [4], we set up a complex adaptive system that represents an agent society with multiple interacting entities working on a complex set of interdependent binary decision tasks. Agents operate on pairwise-correlated performance landscapes which together form the task environment of the agent society. Agents interact with their peers in social networks and share information about their past actions, which creates desirable behavioral patterns. We refer to these patterns as emergent social norms and model agents to include them in their decision rules. Along with complying to emergent social norms, agents aim at maximizing their performance-based incentives, and apply the approach of goal programming to balance the two objectives. In the proposed complex adaptive system, social norms are recursive in that they emerge from the agents’ previous actions, but, once formed, can feed back to influence the further decision-making process of agents.
Our results suggest that the emergent social norms tend to have an adverse effect on the system’s performance, unless the agents are operating on highly correlated performance landscapes. If parameterized properly, incentive mechanisms can help to offset potential performance loss: for agent societies facing highly (moderately) complex tasks, incentives based on team (individual) performance tend to lead to higher system-level performance.