Learning Models of Macrobehavior in Complex Adaptive Systems

Macrobehaviors in complex adaptive systems are highlevel behaviors that arise from a combination of low-level interactions and characteristics of individuals. Over time, individuals participating in these complex systems make choices that influence the behavior of other individuals. These types of interactions are present in both human and computational systems. For example, complex interactive behaviors are demonstrated as audience members enter a talk (Schelling 1978) and software agents negotiate over shared resources (Klein, Metzler, & Bar-Yam 2005). Macrobehaviors arise when groups or aggregations of individuals develop grouplevel structure through interactions of individual characteristics and actions. One complex system that is familiar to academics is the pattern of citations in a given field or subfield of science (e.g., (McGovern et al. 2003)). Citation networks exhibit many interesting behaviors: researchers rise and fall in prominence, papers get cited at different rates at different times, research topics become popular then fade out over time, etc. One example of a macrobehavior is the rise to prominence and ongoing success of a research group. A typical university research group contains a few faculty (usually one or two), graduates students studying with those faculty, and perhaps additional research or technical staff. Presumably, these researchers have joined together because of shared research interests and the opportunity to build ideas collectively. Citation frequency of both the individual researchers and the whole research group depend on the richness of the interactions within the group. Intuitively, we recognize that groups with the right mix of people and talents will tend to be cited more than the sum of those same individuals working separately. Macrobehaviors, also called emergent behaviors, can lead to mutually enhanced behavior (as in the case of research groups), mutually detrimental behavior, or simply neutral behavior that is otherwise remarkable to system observers. Mutually detrimental behaviors, often called social pathologies or social dilemmas, are of particular interest to researchers. Jensen and Lesser define a social pathology as a “system behavior in which two or more agents interact such that improvements in local performance do not improve system performance” (Jensen & Lesser 2002). Examples