I just read The Explanatory Potential of Artificial Societies that criticises agent-based modelling for not being able to provide explanations. It focusses upon the example of the simulation of the Anasazi. The author( Till Grüne-Yanoff) is a philosopher. The main argument he makes is too many different models could produce the same aggregate results and hence too many possible explanations means we don’t have any explanation at all. Edmund Chattoe-Brown and Corinna Elsenbroich have written a good response commentary.
My feeling is that when there is little known to ground the behaviours of the individual agents the criticism is strong. But as Chattoe-Brown and Elsenbroich say even in such cases ‘AS is still “the best of a bad lot”’.