The Dark Side of the Force: When computer simulations lead us astray and "model think" narrows our imagination. (revised version, October 2006)
|Table of Contents|
|2 Different aims of computer simulations in science|
|3 Criteria for explanatory simulations|
|4 Examples of Failure: Axelrod style simulations of the “evolution of cooperation”|
Computer simulations can be employed in science not only for generating explanations but for various different purposes. They can, for example, be used to merely express certain theoretical assumptions or concepts. In this sense they provide a sometimes weaker and sometimes stronger but usually simpler and more flexible alternative to mathematical modeling. Or they can be used to prove the “logical possibility” of certain general assumptions such as the assumption that cooperation is possible among egoists. Or they can be used to explore the possible consequences or implications of certain assumptions. All of these previously mentioned uses of computer simulations can be subsumed under the general title of exploratory simulations or, as these are sometimes also called, speculative simulations. It is the distinctive mark of this type of simulations that the simulations do not need to resemble empirical reality. If there exists any resemblance at all then it is typically vague and consists in the plausibility of the assumptions.
Another - potentially more important - class of computer simulations are predictive simulations. The purpose of predictive simulations is to generate accurate predictions for some empirical process. An example might be simulations in meteorology that predict how the wheather is going to be in the future. The assumptions that enter into predictive simulations do not need to be in any way realistic. As long as the predictions prove to be reliable, it is permissible to use strongly simplified assumptions about the modeled process or even assumptions which are known to be false. This shows that just because a simulation produces successful predictions it does not necessarily also provide an explanation for the predicted phenomena, even though successful predictions may be one among several indicators for a simulation to be explanatorily valid.
The most desired case, however, would be that of an explanatory simulation that is a type of computer simulation that actually allows us to explain the empirical phenomena that are modeled in the simulation. It is this class of simulations that I am concerned with in this paper.