Can agent-based models assist decisions on large-scale practical problems? A philosophical analysis

The use of predictive agent-based models as decision assisting tools in practical problems has been proposed. This article aims at a theoretical clarification of the conditions for such use under what has been called post-normal problems, characterized by high stakes, high and possibly irreducible uncertainties, and high systemic complexity. Our argument suggests that model validation is often impossible under post-normal conditions; however, predictive models can still be useful as learning devices (heristic purposes, formal Gedanken experiments). In this case, micro-structurally complex models are to be preferred to micro-structurally simple ones; this is illustrated by means of two examples.