Simulations need knowledge about human behavior in order to feed the simulation models with realistic rules. In that sense, simulation projects are customers of the results of most of the research presented in this book. Simulations are however not simple translations of human behavior into computer models - rather, modeling, being ``as much an art as a science'', often involves reducing the evidence provided by psychological experiments into a few simple rules. This is particularly true for large scale simulations, where the computation of individual strategic decisions cannot take more than a few seconds per agent. The restriction of computational resources also explains why large scale simulations put as much emphasis on fast relaxation than on realistic modeling on human behavior. This emphasis is reinforced by the fact that fast relaxation of agent-based simulations is similar to fast relaxation of traditional assignment.
This paper then looks at some issues of true agent-based implementations, such as agent-oriented plans storage, or an agent database to remember more than one strategy per agent. These modifications move iterated transportation simulations away from the framework of non-linear optimization and into the realm of complex adaptive systems. As pointed out in the text, these changes sometimes make the simulations more robust and more stable, as does the agent database. Often however, they make the simulations less stable in the sense of larger fluctuations of Monte Carlo runs. This seems to be particularly true when agents become ``smarter'', meaning that they in some sense use more knowledge about the system.