unexpected behaviors: high mutation rates ( μ ) day
HYPOTHESIS
4. Experiments and Simulations
4.1. Some Preliminary Points About Experiments and Simulations
4.1.1. Simulations or Models?
Before moving on to discuss the arguments for the epistemic superiority of experiments, I should address a question that might come up: Why am I talking about simulation rather than modeling? The short answer is that I am effectively talking about both, because I take simulation to be the practice of studying models. But this sort of question might arise because people have talked about the relationship between simulation and modeling in different ways. The most common way to think of this relationship, and the way I think about it, is that a simulation is a study of a model, generally involving some dynamic temporal element (studying what happens to the model over time). Most people talking about simulations understand them in roughly this way. Peschard, for example,
exemplifies this view when she defines a simulation as “the manipulation of a putative model of the target system” (2012, p. 12). Weisberg (2013) says that simulations are ways to compute the behavior of a model using a particular set of initial conditions, contrasting
simulation with mathematical analysis as methods for investigating models. Winsberg defines simulation as the “comprehensive process of building, running, and inferring from computational models” (2003, p. 107).
There can be confusion here, though, because people understand ‘simulation’ in different ways, sometimes contrasting it with modeling. For example, a recent paper on simulation talks about it as follows: “Recent years have seen a developing discussion on the role and epistemology of simulation in modern scientific practice, as a subject worthy of its own attention, distinct from experimentation and modeling” (MacLeod &
Nersessian 2013, p. 533). Others talk about simulation as a particular kind of model, rather than an activity which one does with (or to) a model. One way this gets fleshed out is by saying that simulations are a particularly realistic kind of model, which involves an explicit dynamic element of modeling a system’s states over time. Peck and Lenhard, for example, both seem to endorse this kind of view when they talk about “simulation models.” Peck (2004) contrasts these with “simple models,” by which he means non- computational mathematical models, such as the basic models of population genetics. He separates simulations from other kinds of models, and says that simulations should actually be thought of as experiments, while (simple) models are a contrast class to experiments:
The kinds of experiment done with the simulation model give insight into future data-gathering efforts, test hypotheses that would be impossible to test otherwise and inform researchers about the implications of theoretical insights contained in the causal story that the model represents. Simulation is another experimental system with which to explore theories about how the real world works, using an artificial world that researchers can control. (Peck 2004, p. 533)
In a related vein, Lenhard (2007) talks about simulations as a special kind of model, rather than something one does with or to models, though unlike Peck he thinks that simulations are in a different category from experiments. He says that philosophers of
science have viewed the models that do the “real work” in science as the continuous mathematical functions characteristic of theoretical models, and simulations as just discretized versions of those functions. Simulations, he says, are formulated in a different way from other models, so deserve their own classification; it is incorrect to assume that there is a prior model and the simulation is just a “realization” of it. He argues that “simulation models,” in fact, should be thought of as a separate category of model:
Simulation models do not just apply the brute force of the computer in order to squeeze results out of [theoretical models consisting of continuous mathematical functions]. Instead, they require their own new kind of modeling… This type of discrete model defines the spectrum of potential models anew, motivated by the specific requirements of the computer. The reason for this is that the generative mechanism selected to imitate a certain dynamic has to “run” on the computer; that is, it must not require excessive computing capacity and must, above all, not become unstable, because, for example, discretization or truncation errors will build up. (Lenhard 2007, p. 187)
It is worth noting these different views on the relationship between simulation and modeling, and how these relate in turn to experiments. In any case, I do not have a
particular stake in how this definitional issue is ultimately settled, and will set it aside. I am interested here in the methodological and epistemic contrasts people have made between studying experimental systems and studying models, and I am focusing on simulation here, understood as the activity of studying models, because that is the main activity people have contrasted with experimentation. Perhaps it makes sense that this is a cleaner grounds for comparison than experiments versus models, because simulations (understood thus) are about studying models, and experiments are about studying physical systems in a laboratory, in the field, etc., both with the goal of learning about some target system. Construed thus, both methodologies are about manipulating an object to learn about a target.