How Digital Computer Simulations Explain Real‐World Processes

Abstract

Scientists of many disciplines use theoretical models to explain and predict the dynamics of the world. They often have to rely on digital computer simulations to draw predictions fromthe model. But to deliver phenomenologically adequate results, simulations deviate from the assumptions of the theoretical model. Therefore the role of simulations in scientific explanation demands itself an explanation. This paper analyzes the relation between real‐world system, theoretical model, and simulation. It is argued that simulations do not explain processes in the real world directly. The way in which simulations help explaining real‐world processes is conceived as indirect, mediated by the theoretical model. Simulacra are characterized further, and turn out to be a priori measurable. This gives a clue to a better understanding of the epistemic role of computer simulations in scientific research.

Acknowledgements

The work presented here was carried out at the Konrad Lorenz Institute for Evolution and Cognition Research, Altenberg, Austria. Support by the institute is gratefully acknowledged. An earlier version of this paper was presented at the workshop Models and Simulations, Paris, June 2006. I would like to thank the discussants for their helpful comments. I also owe thanks to the two anonymous referees of this journal for valuable hints that helped further to improve the paper.

Notes

[1] See Glennan (1996, 2002) for an account of mechanisms that does not include activities. I stick with the account of Machamer et al. (2000), which fits better to the scientific models in question.

[2] It should be noticed that a scientific model is not just a mathematical structure (in contrast to the use of the term ‘model’ in model theory: Balzer et al. 1987), but includes what in the structuralist jargon is called an ‘intended application’ (Sneed 1971) by providing an interpretation of the variables (Krohs 2004).

[3] While digital simulations are usually performed on a computer, a human calculator, for example, is also a possible ‘device’ for running such simulations, though with limited calculating capacity. Consequently, computer techniques applied in simulation are regarded by Humphreys (2004) as extensions of our computational abilities, comparable to the technological enhancement of our sensory apparatus by scientific instruments.

[4] Construed like this, it turns out that models may mediate between two phenomena (processes), allowing for understanding one as an imitation of the other. This is an additional kind of mediation by models, supplementing that between phenomena and theory (Morgan and Morrison 1999).

[5] This is the sense in which also Nancy Cartwright uses the term ‘simulacrum’, though with respect to theoretical models, not to simulations (Cartwright 1983, 152–153). Applied to simulations only, my use of the term does not commit one to a Cartwrightian anti‐realistic account of scientific explanation but remains neutral in this respect. See Stöckler (2000) for a defence of the view that the use of simplified models and simulations could be reconciled even with a full‐fledged realism.

[6] The numerical values are definite even if they are not read out, so one might be tempted to speak of measured worlds instead. However, to count as measured, the values had to be not only definite but also recorded, by an observer or by the computer. Since this need not happen, no more than measurability can be claimed.

[7] I am not concerned with the issue of the indeterminacy of certain parameters, which may not be overcome by any advance in measuring technology. Indeterminacy of this kind has to be regarded as a feature of the real world and therefore needs to be reproduced in some way in a simulation of a process, where it is relevant.

[8] In many cases it may be possible for an experienced modeller to ‘see’ this ‘immediately’, but this means nothing more than that one is able to conceive the model or central features of it without writing down the equations.

[9] This general claim holds a fortiori with respect to any particular parameter of the simulation. We know, in advance of any experience, that the parameter is measurable without restrictions.

[10] Several kinds of mechanical calculating machines, in particular pinwheel machines (Martin 1992), seem to be an exception, but these have to be properly classified as mechanically working digital machines, not as analog ones.