Policy analysis for complex and uncertain systems through computational experiments

Most model based methodologies for engineering or policy design depend upon possession of a model that is predictively accurate. However, many systems of practical importance are now recognized as being both highly complex and composed of multiple interacting adaptive agents. For these systems, the construction of a model that accurately predicts the details of system behavior is essentially impossible, and any model for these systems is certain to be "wrong" in at least some details. If modeling and simulation is to be used to reason from what is known about such systems, we must understand how to think usefully with "wrong" models. This paper describes an approach based on exploratory modeling and adaptive strategies that can be used to make decisions about systems that are so complex their detailed behavior cannot be predicted. This approach seeks to find strategies that perform reasonably well over broad ranges of plausible futures-instead of devising strategies that are optimal for some particular best estimate model.