for UCSB workshop “ Agent - based modeling of complex spatial systems ”

Early successful examples of spatial agent-based models were often highly abstract, and were used to demonstrate how particular macro-scale outcomes could emerge as the result of decentralized interactions of autonomous decision-making agents. Examples include the demonstration of segregation provided by Schelling's early models (Schelling 1971) and the emergence of power-law distributions of wealth from Epstein and Axtell's Sugarscape model (Epstein and Axtell 1996). Such models had a very simple role: to extend theory by demonstrating that a particular set of behaviors and interactions could generate a particular macro-scale or emergent outcome (Epstein 1999). As the discipline of spatial-agent based modeling matures, an increasing number of highly empirical models (in the style of " Cell 4 " models as described by Parker, Berger, and Manson (2002) are being constructed. These models are often developed to support policy analysis, and are motivated by a belief that models that include complex dynamics and interactions are likely to produce " better " results that models that omit those dynamics. While my discussion focuses on models of human-environment interactions (Bousquet empirical spatial agent-based models are also being developed in other areas of social science, including epidemiology, sociology, and political science (Castle and Crooks 2006). Many scholars have voiced concern that such highly empirical models of complex systems are in danger of being black boxes in which the relationships between inputs and outputs, and even the behavior of the model itself, are not well understood. Some argue that for these reasons agent-based modeling is appropriate only for highly abstract demonstration of theoretical outcomes, and that the push to increase complexity in these empirical models may run counter to scientific principles of parsimony in modeling. Fundamental questions are raised about where and how agent-based modeling fits into the scientific method. In the contrary view, others argue that the policy environment is complex and therefore demands development of complex models, and that policy questions we now face, such as potential contributions of land-use and land cover change to global climate change, and the potential transmission paths of pandemics, are so pressing that any and all practical approaches to obtaining answers must be tried. A separate thread argues that spatial agent-based models may contribute to the development of an interdisciplinary theory of spatial social science. Yet, whatever one's view of the appropriate role for ABM in spatial social science, tensions exist between the need to simply build better empirical …