Tools for Evaluating the Consequences of Prior Knowledge, but no Experiments. On the Role of Computer Simulations in Science
|Table of Contents|
|2 Common features of simulations and experiments|
|3 Distinguishing features of experiments and simulations|
|4 Borderline cases|
|5 Summary and conclusion: Computer simulations as a tools for drawing conclusions from prior knowledge|
Regarding the epistemic status of computer simulations, the above reasoning leads to the conclusion that computer simulations are in no way an empirical method of science but that they clearly belong on the theoretical side of science. This result is not surprising, because, after all, computers are logical machines that only allow us to draw the consequences of given premises. Therefore, the result of a computer simulation can have no more empirical content than can logically be derived from the input data and the modelling assumptions of the simulation.
Computer simulations can be likened to experiments, because for certain types of experiments - like the wind tubes that von Neumann refers to (see section 4.1) - their empirical character is of no particular importance. But other than experiments, computer simulations can under no circumstances gather new empirical data, operate directly on an empirical target system or be employed for the testing and possible falsification of fundamental natural laws.
As a consequence, it is at best a metaphorical way of speaking if computer simulations are described as “computer experiments” (Gramelsberger 2010). And it is misleading if computer simulations are described as a “third way of doing science” (Axelrod 2003, p. 90) (Rohrlich 1990, p. 507) (Kueppers/Lenhard 2005), if by this it is meant that computer simulations stand somewhere between the theoretical side and the empirical side of science, because computer simulations are a purely theoretical device. At most it can be conceded that the description as a “third way” and the metaphor of a “computer experiment” capture certain phenomenological aspects of computer simulations which concern the way simulation studies are set up, conducted and evaluated. In this respect computer simulations may indeed appear experiment-like. For they allow us to try things out and they may lead, just like experiments, to surprising and unanticipated results. But computer simulations do so only in an artificial environment in the computer and not in an empirical environment.
This has important consequences for the epistemic status and, in particular, the justification and validation requirements of computer simulations. When Wendy Parker concedes that “especially when scientists as yet know very little about a target system, their best strategy may well be to experiment on a system made of the `same stuff' [rather than with a computer simulation] ” (Parker 2009, p. 494), then this is true, but the emphasis is misplaced. For, it is in fact only when we already know very much about a target system in terms of comprehensive and empirically well-confirmed background theories that we can safely rely on computer simulations as a substitute for experiments. It is therefore no surprise that computer simulations are most successful in those areas of science where we have powerful background theories, like quantum chemistry for example.
By the same token, the fact that computer simulations are a merely theoretical tools renders understandable their lack of success in those areas of science that have to do without empirically well-confirmed background theories such as the social sciences, where the lack of empirical content and proper validation of simulations is frequently lamented (Heath et al. 2009). There is a danger of underestimating the paramount importance of empirical validation of computer simulations, if they are conceived of as some kind of experimental tool. It is therefore best to think of computer simulations not as “virtual experiments” but as tools that allow us to evaluate the consequences of theories and prior knowledge.