Formal models can facilitate the development of organizational theory. The complex, adaptive, non-linear nature of human endeavor makes computational models a particularly useful type of formal model for exploring organizational behavior. Researchers, however, rarely contrast such models with empirical data. Further, researchers tend not to contrast such models with other such models. This paper presents results from the artificial organization project and demonstrates how contrasting models with each other and with empirical data can facilitate the development of more veridical organizational models. Specifically, this paper examines whether certain organizational models, differing only with respect to the cognitive limitations of, and adaptability or general intelligence of the agents are better or worse predictors of the behavior of similar organizations composed of humans. It is shown that, not only do different agent models predict different levels of organizational performance, they also predict different relative standings for different organizational structures. Finally, the adequacy of organizations composed of simple computational agents for predicting the behavior of organizations composed of humans appears to increase as the complexity of the organizational structure increases.
