unexpected behaviors: high mutation rates ( μ ) day
HYPOTHESIS
4. Experiments and Simulations
4.5. Experimenters Almost Never Study Th eir Targets Directly
There is a view which is in the background in arguments for the materiality thesis and the surprise claim, but which one could hold without endorsing either. This is the relatively common view that a key difference between experiments and simulations is that
577) quotes Gilbert and Troitzsch as holding this view: “The major difference is that while in an experiment, one is controlling the actual object of interest (for example, in a
chemistry experiment, the chemicals under investigation), in a simulation one is experimenting with a model rather than the phenomenon itself.”
This kind of view underlies the claim that reproduction versus representation characterize object–target relationships in experiments versus simulations, respectively, discussed in Section 4.2 above. It also comes up for proponents of the surprise claim, and could be a line of response to my response to the surprise claim in Section 4.4.1. The idea would be that when you discover a hidden feature in an experimental object of study, you are learning something about your target. But when you discover a hidden feature in a simulation you are learning something only about your model. For that reason, the objection goes, experiments still put us in an epistemically privileged position with respect to their capacity to surprise us. This objection is not necessarily about how often
experiments versus simulations lead to surprises, or how often they lead to the discovery of genuine hidden features. The objection regards the epistemic payoff of those hidden features. That is, it says that the productivity of surprises from experiments is worth more because it tells us about targets in the world. The claim is that when a researcher discovers a hidden feature in a simulation, the burden is on him to show why it tells us anything about the world outside the model.
This view is wrong. Well, it is not entirely wrong: It is true that simulationists do not study their targets directly. They study models, which stand in for their targets. But experimenters almost never study their targets directly, either. In physics and chemistry, objects of study in the laboratory are often instances of the target entities or phenomena in the natural world, like particular subatomic particles or elements or kinds of reactions.
But even then experimenters need to do some work to show why the conditions that apply to those entities or phenomena in the laboratory apply to them in general out in nature as well, or (sometimes) whether they are even identifying the correct entities or phenomena in the laboratory (on this latter point see related discussion in Galison 1987). In biology and biomedical research, there is often even more work to be done to show why inferences from the object of study to a target of inquiry in the natural world are licensed, as when using Drosophila to study genetics in general, or a small group of clinical trial volunteers to study potential future drug-takers in general. When computer models are the objects of study, scientists always have to do the work to show why inferences to the world outside the object of study are valid. When experimental systems are the objects of study, they almost always do too.
Only in very rare cases can experimenters be said to study their targets directly. The only sorts of cases where this applies are cases where (1) the target is delineated in particular rather than general terms and consists of a small, clearly delimited set of entities, and (2) the experimenter is studying exactly that set of entities as her object. Cases where this might hold include studies of very small populations, for example, certain field studies in anthropology, where the goal is to say something about a small, clearly delimited population of humans and researchers engage with every member of that population as their object of study; or biomedical studies of the only 50 people in the world with an extremely rare genetic disorder for the purposes of making inferences about people with that genetic disorder.29 Other rare cases that might meet these conditions include chemistry experiments in which researchers create a new synthetic element in the laboratory, for the purpose of making an inference about how that element behaves, and
29 Even here, though, there could be problematic issues involved in making inferences about past or future
the only existing instance of that element is the one which they are studying. For example, in 2014 researchers claimed to have created element 117 (ununseptium) (Khuyagbaatar et al. 2014); it is very difficult to create this synthetic element in the laboratory and the element exists for only a fraction of a second before falling apart. But people created it and wrote a paper about it. In these sorts of cases, where the only proper instance of the target in the universe (as far as we know) is exhausted by the experimenter’s object of study, it is correct to say that the experimenter is studying her target directly. In the vast majority of experiments, this is incorrect.
These sorts of cases are rare exceptions, and not at all the norm. In all other cases of scientific research, the object of study is standing in for the target of inquiry. There are many ways for one thing to stand in for or represent something else, and it is beyond the scope of this discussion to enumerate all these various ways and exactly how to
understand them (but see Frigg 2006, Van Fraassen 2008 and others). The key point here is just that a scientist always has to do some extrapolation work to show why her object is an appropriate stand-in for her target. This applies whether her object is a physical system in a laboratory, a population in the field, or a model visualized on a computer screen. This applies whether the difference between her target and object is that one is a biological population in nature and the other is computer code, or that one is a biological population in nature and the other is a biological population in a test tube. If we were to define experiments as cases where researchers study their targets directly, then very few cases of scientific research would count as genuine experiments.
It is worth mentioning another exception to the view that experimenters study their targets directly, simulationists do not: There are also rare cases where simulationists
ultimate target of inquiry is not some system in the natural world, but the model under study itself. Weisberg calls these cases “targetless models,” where “[t]he only object of study is the model itself, without regard to what it tells us about any specific real-world system” (2013, p. 129). The Game of Life (see discussion in Section 4.4.1) is a perfect example. The model is not meant to represent some particular target system in the natural world. Rather, it is studied as an interesting case in its own right of properties of interest to researchers in artificial life and computer science, like emergent dynamics and universal computation.