Emergence in and Engineering of Complex MAS
We discuss the implications of emergence and complexity for the engineering of MAS. In particular, we argue that while formalisms may play a role in specification and implementation of MAS, they do not provide predictive power in complex systems. Thus we must look to other tools for understanding these systems. Statistical methods may go some way to helping us, but they too have their limitations. Rather, we argue that we must revert to classic scientific method: observing the systems, positing theories and hypotheses, then testing and improving them. That the study of complex MAS has to be more of a natural (as opposed to formal) science. This has the consequence that there will be severe limitations on the extent to which one can “design” complex MAS. We illustrate this case with two examples of MAS, both of which display system level dynamics that cannot directly be predicted from the behaviour of individuals. We call upon those in the ESOA community to explicitly reject those tenets that are only useful with simple MAS.