The Dark Side of the Force: When computer simulations lead us astray and "model think" narrows our imagination. (revised version, October 2006)

Eckhart Arnold

1 Introduction
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”
5 Conclusion
Bibliography

5 Conclusion

Quite a few lessons can be learned from the previous examples of failures of Axelrod style computer models. Some of them are truisms, but as they are often neglected they are important nontheless.

First of all, if our models are to be explanatory then the establishment of a close fit between model and reality is at least as important as the construction of the model itself. The biological examples such as Milinski's and Parker's studies on predator inspection suggest that establishing this fit may even be much harder and more time consuming than constructing the model itself.

Secondly, when there is no close fit between model and reality then the model has hardly more epistemological strength than a mere metaphor. Therefore, one must be very careful when drawing conclusions from them. Computer generated metaphors are no better than ordinary metaphors. At best one can regard these conclusions drawn from them as mere hypotheses that still require an independent empirical confirmation. Without this empirical confirmation explanations based on computer simulations amount to nothing more than model based story telling. Such computer simulations are in a way comparable to non falsifyiable theories, because there is no way to test whether they simulate correctly the empirical process they are meant to simulate.

Finally, we should be aware of the fact that although the ease and power of formal modeling has been greatly increased with the advent of the computer, there still remain scientific areas where the advantages of formal modeling are doubtful or where it is not possible at all. Computer simulations are just one scientific tool among others. It is helpful in some situations but useless in others. Where computer simulations cannot not go beyond a merely metaphorical resemblance of empirical reality their use is probably not worthwhile.

t g+ f @