Tools for Evaluating the Consequences of Prior Knowledge, but no Experiments. On the Role of Computer Simulations in Science

Eckhart Arnold

1 Introduction
2 Common features of simulations and experiments
3 Distinguishing features of experiments and simulations
4 Borderline cases
    4.1 Experimentum crucis and analog simulations
    4.2 Simulation-like experiments
    4.3 Experiment-like computer simulations
    4.4 Hybrid simulation-experiments
5 Summary and conclusion: Computer simulations as a tools for drawing conclusions from prior knowledge

4.4 Hybrid simulation-experiments

A more complicated case is that of hybrid methods which fuse simulations and empirical methods such as experiments or measurements. It is the existence of these methods that Morrison (2009) mobilizes in order to dispute a clear cut distinction between simulations and empirical measurements. An example would be magnetic resonance imaging (Lee/Carroll 2010), because before the raw data obtained from the electromagnetic signals emitted by the previously stimulated protons of the body is turned into an image it runs through various highly sophisticated computations some of which are not quite unlike a computer simulation. This raises the question whether the procedure as a whole is more like a measurement and thus an empirical procedure or more like a computer simulation or, maybe, something different that resembles neither of these. The last answer represents the stance that Mary S. Morgan (2003) takes. If this answer is true then this would mean that the distinction between simulations and experiments (or measurements for that matter) is inevitably blurred.

In order to answer this question a few words need to be said about the nature of measurement. Any measurement procedure that makes use of a more or less sophisticated apparatus, and in particular those measurements that involve computations, produces data as its output that for the sake of the apparatus or the computations involved is best described as refined data. Now, it is still reasonable to classify refined data (in contrast to raw data which has not gone through any intermediate processing or computation steps) as empirical data or empirically measured data as long as the data describes the state of system from which the measurement was taken at the time of the measurement. Thus, if we measure the humidity and temperature today in order to compute the humidity and temperature of tomorrow this is not a measurement but a prognosis based on a measurement. But if we measure the volume and weight of a piece of metal and then compute its density, we can reasonably speak of an (indirect) measurement of the density, the density being the refined data, while the weight and volume is the raw data.[6]

But this means that even a highly sophisticated hybrid method such as magnetic resonance imaging can be classified as empirical measurement, because the refined data it produces is data about the same system from which the raw data was gathered and is at the same time dependent on the content of the raw data. In this particular case, the systematic link results from the fact that the refined data represents a structure (that of the part of the human body which are displayed on the screen) that is causally responsible for the raw data. Because of this causal link the refined data can be reconstructed from the raw data. The assessment of refined data as empirical data and of hybrid methods as empirical methods despite their containing a theoretical component is motivated by the following reasons:

  1. The content of the refined data depends on the content of the raw data and changes if the content of the raw data changes depending on the particular refinement process.
  2. Hybrid methods can have all of the features that distinguish experiments from simulations as described above (see section 3): They can provide us with new empirical data, they can operate directly on the target system and they can therefore be used to test fundamental hypotheses.
  3. In a sense there are no pure empirical methods, anyway. Even an observation with our barest senses is partly determined by the “apparatus” of our sense organs. But if this is true, then we have little reason to consider the measurements we make with the help of a sophisticated artificial apparatus as less empirical than the observations with our sense organs. At the same time, the distinction between purely theoretical methods (human reasoning, calculations or simulations etc.) and empirical methods (including hybrid methods) is fairly clear cut.

Therefore, the existence of hybrid methods that combine measurement with simulation leaves the distinction between simulations and experiments intact.

[6] In the literature on simulations also other definitions of hybrid methods can be found. Mary S. Morgan, for example, speaks of “the hybrid `simulation' form, which mixes mathematical and experimental modes” (Morgan 2003, p. 225). For her a simulation is already a hybrid if it uses empirical input data, in which case she uses the term “virtually experiments” (see her table 11.3 on page 231 of her paper). In the beginning of her paper she gives the example of two simulations of a bone where the model of the bone is based different kinds of empirical measurements. For her, both are hybrid methods which she locates between experiments and simulations. According to the stance adopted in this article both are simulations because the output data describing the strength of the bone is not causally responsible for the input data, which captures the geometrical structure of the bone.

t g+ f @