Autonomous Group Formation of Heterogeneous Agents in Complex Task Environments

Individuals cannot solve complex tasks by themselves due to their limited capabilities. By self-organizing into groups, individuals with different capabilities can overcome their limitations. Individuals and groups often change over time: The individuals that form the group learn new ways to solve the task, while groups adapt their composition in response to the current needs of the task. The latter is driven by the differing characteristics of the individuals, as some of them might be better adapted at a particular point in time but do not participate in the group. By self-organizing, groups absorb these individuals within their ranks, so they have the best-adapted members. However, there is a lack of consensus on whether changing a group's composition over time is beneficial or detrimental to task performance. Moreover, previous research has often assumed that agents are homogeneous. We implement an adaptation of the NK-framework using agents with heterogeneous capabilities, which includes an individual learning mechanism and a second-price auction mechanism for group self-organization. Heterogeneity in the agents' capabilities ensures that groups have an incentive to change their composition over time. Our results suggest that group self-organization can improve task performance depending on task complexity and how prominent is individual learning.