When can a Computer Simulation act as Substitute for an Experiment? A Case-Study from Chemisty |
| Table of Contents |
It would be rather surprising if despite these similarities there were no differences between simulations and experiments. For, if there were no substantial differences why would people still invest large sums into highly sophisticated experimental set-ups like that of a particle collider when they could just as well buy a super computer? Some philosophers believe that it is rather difficult to draw a sharp line between simulations and experiments (Morrison 2009, Winsberg 2010, Parker 2009), but we believe there are at least three fundamental differences between simulations and experiments which are highly relevant for the epistemic status of either category.
First of all, experiments can provide us with new empirical data, computer simulations cannot. While it is true that computer simulations deliver results that may not have been known or expected by us beforehand, computer simulations can by their very nature only produce results that are implied by the premises on which the simulation is built. It is important here to understand the difference between a) things that are not logically implied in our prior knowledge, b) things that are logically implied in our prior knowledge but unknown to us and c) things that are logically implied in our prior knowledge and also known to us. For category a, simulations cannot help us; only experiments can help. For category b, simulations and experiments can help us. And for category c, neither is needed, because we know it already. Another way of putting it would be to say that simulations can only deliver us results that fall within the deductive closure of our prior knowledge.[3]
Therefore, if the term “empirical data” is understood as data of empirical origin then computer simulations do not generate new empirical data. Sometimes the term “empirical” is used in a wider sense. Barberousse et al. (2009, p. 560), for example, speak of the data that is generated by simulations as data about empirical systems. But they, too, do not consider it to be new data of empirical origin.
Another important difference is that some experiments operate directly on the target system, while computer simulations never do. More precisely, the kind of relation that subsists between the experimental object and the target system is typically one of the following: identity, being an instance, being a part. For example, if a car tester wants to know whether a car can drive faster than 100 mph and, in order to find out, accelerates the car to that speed then the object is identical with the target system. If physicists want to know whether white light is composed of different colors, they can let a beam of light fall through a prisma to find out. In this case the object (a particular beam of light) is an instance of the target system (light). If an archaeologist intends to determine the age of an old building and takes a stone of that building to submit it to certain tests than the object is a part of the target system.
There are also experiments that do not operate directly on the target system. If one experiments with an electrical harmonic oscillator in order to learn something about a mechanical oscillator (Hughes 1999, p. 138), then this experiment does not operate on the target system itself. But the fact that some experiments operate on the target system or on an instance or a part of the target system, suffices to set the two categories of experiment and simulation apart. Because experiments can operate on the target system, the experimental method has an epistemic reach beyond that of simulations.
As our case study below shows, there are also application cases where simulations have an epistmic reach beyond that of experiments. However, in this case the limitations of the experimental method are more a matter of practical constraints, while in the opposite case there are principle reasons why the epistemic reach of simulations cannot have the same extent as that of experiments.
Finally, experiments can be used for the testing of fundamental hypotheses (experimentum crucis), which, again, computer simulations cannot. It is obvious that a simulation cannot be used to test fundamental hypotheses. For, the outcome of the simulation would simply depend on the very hypothesis upon which the simulation is built. It would be impossible to replace an experimentum crucis like Young's double-slit experiment that demonstrated the wave nature of light
by a simulation, because the results of the simulation would merely reflect which of the competing theories was programmed into the simulation.
Summing it up: Despite many striking similarities there are several features of experiments that clearly set the experimental method apart from the simulation method. This is true, even if in some cases simulations can act as an surrogate for experiments. We will discuss one such case further below.
In the following we look at one concrete example case and examine under what conditions simulations can serve as a surrogate or for experiments. The question has both an epistemological interest and practical relevance. It has an epistemological interest, because it touches on the relation between theoretical reasoning, experimental testing and empirical observation in science. The question has practical relevance, because it is important to understand when one can trust the results of a simulation that is offered as a surrogate for an experiment or measurement.[4]
|
[3] Winsberg apparently holds a different view when, referring to a particular example, he says: “To think it is true is to assume that anything you learn from a computer simulation based on a theory of fluids is somehow already `contained' in that theory. But to hold this is to exaggerate the representational power of unarticulated theory. It is a mistake to think of simulations as tools for unlocking hidden empirical content.” (Winsberg 2010, p. 54) While it is true that the presuppositions of a simulation are not only formed by theories but by theories plus further modelling assumptions, there is no way around the restriction that a simulation cannot deliver results that go beyond what is implied in the presuppositions. In this sense simulations are indeed just tools for unlocking hidden content.
[4] This question aroused public attention in the aftermath of the eruption of the Iclandic volcano Eyjafjallajökull in 2010, when airlines were ordered to stay on the ground on the basis of simulation of the spreading of the volcanic ashes. This was criticized by representatives of the airlines who claimed that computer simulations were an insufficient basis for this decision (Tagesschau 2010/04/18). Similar questions are raised by environmental protection policies justified by climate simulations.
|