Introduction
Following advancements in information technology, the scientific practice in the current century has increased the debate on whether experimentation is the same as simulation. In the past, scientists have relied on experimentation when carrying out different studies. This trend has begun to change since scientists are now combining experimental and simulation methods. This change has developed a new way to view and do science.
Besides, the new trend has “created opportunities to review the roles that experiment and simulation play in the scientific process” (Parke, 2014, p. 521). In this paper, I seek to support the widespread perception among philosophers of science that experiments have epistemic superiority over computer simulations. However, I also build on this debate by showing that the distinction between experimentation and simulation reveals more information much concerning epistemic value.
To defend the view that experiments have epistemic superiority, there are two main challenges that have to be addressed. The first challenge is to explore and determine the source of the epistemic privilege of an experiment. The second challenge is to show why this comparison is essential.
Background Information
An understanding of experiments as interventions in real-world systems differentiates them from simulations. Field experiments are differentiated from laboratory experiments simply according to their location (Barberousse, Franceschelli, & Imbert, 2008).
Unlike within the confines of a laboratory, where scientists have control over numerous variables, field experiments might entail manipulating a single variable, such as preventing a certain species from a given geographic location. In this light, experiments are viewed as generating more reliable and trustworthy scientific knowledge. Simulations are viewed as only necessary when experiments are too costly or impossible.
Even though they might be less comprehensive or fertile than experiments done for the generation of knowledge, simulations should not be viewed with suspicion and contempt because they have the potential to generate scientific knowledge. Parke (2014) claimed that such opinions on the inferiority of simulations to experiments are mere generalizations. Experiments allow for more direct and comprehensive analysis.
Experiment vs. Simulation
There is a widespread assumption that experiments are more precise than simulations, in the sense that they use tangible objects of study. This assumption plays a role in the concern among various experimental scientists that simulations are chosen merely because they are cheap. Consequently, this assumption might hinder the discovery of new knowledge about the world (Barberousse et al., 2008). According to Fallis (2007), simulations only focus on revealing the consequences of the knowledge that already exists. This assumption is backed by the idea that a computer-based simulation is simply revealing the consequence of various theoretical assumptions. Therefore, all information in the results of the simulation reflects what was present in the theoretical assumptions.
Various considerations play major roles in one’s ability to draw conclusions regarding epistemic superiority. First, how much knowledge is available about the object and the target? According to Parke (2014), when the background knowledge is limited, “a physical sample of the target or close approximation is an essential starting point, and in other contexts, such information is available from the world making simulations reliable methodology; for instance, the study of molecular bond angles in chemistry” (p. 325). Enough information is available about chemical bonding to allow questions to be answered through computer simulations concerning how atomic substitutions will affect the bond angle in a certain molecule. When very little is known, experiments tend to have epistemic value.
Parke (2014) also observed that simulations are seen to have less impact than experiments on the generation of scientific knowledge. Essentially, the claim is that experiments create better inferences about phenomena or natural systems compared to simulations. Experiments have proven to be reliable guides to a generation of scientific knowledge. Parke concurred that experiments have a certain level of epistemic value because she acknowledged that all scientific inquiry entails engaging with some objects of study—a model, a physical structure in the laboratory, or a combination of both—to gain insights about some target of research.
This aspect implies that the methodological difference between experiment and simulation is fundamental for compiling conclusions regarding epistemic value. However, Parke refuted this claim by showing that this situation is applicable only in a context-sensitive manner, but not as an abstraction across science. Based on this notion, Parke insisted that differences between experiments and simulations should not be employed as a foundation for generalizations.
It would be rather irrational for one to conclude that experiments had no epistemic advantages over simulations. If there were no significant epistemic privilege, why would scientists choose to invest large sums of money in experimental setups when they could buy a supercomputer for simulation? Even though some philosophers such as Parke (2014) have concluded that it is difficult to determine the epistemic difference between experiments and simulations, it is essential for one to understand the distinction between things that are not rationally implied in people’s prior knowledge.
In this case, simulations are not in a position to solve problems, but experiments are reliable. In other words, simulations can only assist in solving issues that exist within the deductive closure of one’s prior knowledge. Thus, Parke declared that computer “simulations do not produce new empirical data” (p. 326).
Moreover, experiments create higher inferential power. Gendler (2010) suggested that experiments are in a good position to generate credible views concerning their objects of study. This scholar also views experiments as powerful sources of meaningful surprises or valid novel knowledge. However, I do not agree with this claim for just because a case of scientific inquiry counts as an experiment or simulation, it is not proof of epistemic value.
Undoubtedly, there is a great deal of critical work that ought to be conducted if the inferences from objects of research to subjects of inquiry in the natural world are to be realized. By arguing that there is an epistemic value between experimentation and simulation, I do not imply that such a difference is impossible. There are ways to conduct meaningful computer simulations, but only because humanity has learned about the world via observation and experiment.
Focusing on the entire issue, empirical data are essential for solving scientific issues regarding the natural world. Nonetheless, there are two limitations to this claim. First, even though the information is present in theoretical assumptions, it is not obvious that it is known. Thus, even if a computer simulation is simply a calculation, it might generate new knowledge.
Inferential Power
Parke (2014) rejected the supposition that experiments possess epistemic value over simulation, claiming that this view is pervasive among philosophers of science. Parke further claimed that the epistemic superiority of experiments over simulations is founded on the claims of their inferential ability. Parke claimed,
Belief in the epistemic privilege of experiments over simulations is often grounded in ideas about their relative inferential power. In particular, the idea is that experiments lead to better inferences about natural systems or phenomena than simulations do (this is sometimes referred to as the issue of external validity). This difference has to do with the relationship between their respective objects of study and targets of inquiry (p. 75).
Greca, Seoane, and Arriassecq (2014) argued that the tangible objects of study are the distinguishing features of experiments. Parke referred to these widespread views as the “materiality thesis” (p. 75). On the contrary, “experiments’ elements or objects reproduce parts of the natural world while simulations’ objects represent sections of the world” (Gendler, 2010, p. 118). Because the form of a subject or thing in the laboratory is similar when outside the laboratory, experiments manifest more inferential power. This assertion holds because the objects under study are similar to what is in the real world.
This fact makes inferences about the world possible and simple. In this case, the knowledge has epistemological implications. Scientists tend to feel justified when they reveal new insights into the world through experiments because the world and experiments possess similar characteristics of objectivity.
Therefore, there is undeniably some veracity in the materiality thesis. Experimenting on a real specimen or sample in the natural world is the ideal way to get a grip on acknowledging the process when very little is known about the target under study. For instance, it would be more convincing and idealistic for scientists carrying out laboratory experiments to use living organisms like rabbit fro experiments as compared to using computer simulations of living systems. Computer simulations may not provide the objectivity found in physical objects of study.
Gendler (2010) posited that the “the fact that some experiments operate directly on the target system provides epistemic privilege over simulations” (p. 126). For instance, if physicists aim to know whether the white light is made of various colors, they can shine a beam of light over a prism to find out. In this experiment, “the object, which is the beam of light, is an instance of the target (light) and the fact that experiments can operate on the target system is enough evidence that they have epistemic reach beyond that of simulations” (Parke, 2014, p. 330).
The fact that simulations do not incur large sums of money is their only advantage, but it also means that the results might not reflect the real world. Parke (2014) does not concur with this perception and insists that similarity, not materiality, is what matters when justifying certain inferences concerning target systems. However, Parke failed to offer a thorough account of how to determine relevant similarities.
Experiments in “virtue of the nature of their objects can surprise in a way simulations cannot” (Parke, 2014, p. 331). The exponents of the surprise claim share the notion that experiments and simulations vary significantly, either by value or content. Although surprises arise in simulations, what transpires in a simulation is that familiar and new insights are not anticipated. Arguing from a qualitative viewpoint, Greenwald (2004) showed that although simulations may be in a position to surprise, experiments can both surprise and confound.
Parke (2014) insisted that various computer simulations undoubtedly share some epistemic functions with experiments by asserting that simulations are run to offer new data systems that are complex or impossible to investigate with the normal instruments like lab equipment. For example, neutron-matter interaction has been widely tested using molecular dynamics simulations.
Grune-Yanoff and Weirich (2010) suggested that to determine the distinction between simulation and experimentation, it is essential to compare the two methods with other constant factors. For instance, a researcher conducting a biology test might dissect a mouse to study the internal organs as well as the skeleton. The same researcher cannot obtain the same level of experience from a simulated dissection online.
Simulations manifest how the programmer thinks the actual world should be rather than how the real world responds when explored by a student. Besides, learning new knowledge in science entails being a skilled observer, and a precise recorder of every detail observed rather than being a participant in a computer game claiming to teach about the natural world. However, simulations and experiments can be viewed as complementary rather than competitive activities.
When viewed in this manner, simulations can be helpful when cheap and faster procedures are needed. Simulations offer a high degree of flexibility. Thus, numerical experiments can be accomplished by turning certain terms on or off and by varying input parameters. Both experiments and simulations are dynamic procedures.
Practical Consequences
Parke’s (2014) claims are based on an intuition that an experiment has epistemic privilege, not that it is free of pragmatic constraints. However, in this paper, my defense of epistemic value implies that the superiority claim is a principled one because it signifies that in a certain situation, there never exist actual choices between an experiment and an equivalent simulation. This suggests that in any given situation, the experiment overrides simulation in the realm of epistemic utility.
The difference between experimentation and simulation manifests epistemic value. Parke also claimed that the only difference is existing between the two models matters for pragmatic reasons. Based on this claim, it is a fact that performing an experiment is more costly than performing a simulation. The materials, reagents, and the workforce required to do an experiment tend to cost more than conducting a model on a computer. Under simulation, one can see the progress and results faster as compared to the normal process of carrying out an experiment.
Under certain circumstances, several logical simulations provide consistent results, but they will need more background information than experiments do. One of the ways to “determine the superiority’s practical salience is that relative to a given set of background information, there are cases when simulations will give arbitrary answers where else experiments provide defined answers” (Gendler, 2010, p. 122).
This observation implies that pragmatism is not the only essential factor considered when selecting between an experiment and a simulation. For instance, in a given test or question, one cannot justify selecting a simulation over an experiment on the grounds that it is less costly.
Even though such choices might not be made consciously, it is necessary to conduct careful research before deciding which model to adopt. In some cases, those who deny epistemic superiority argue that experiments calling for the deliberate interference of human beings are detrimental and unethical. However, even though the epistemic value might not outweigh the moral cost, it does not mean that there is no epistemic privilege.
According to Gendler (2010), arguing on a pragmatic basis could mean more barriers in obtaining reliable results to questions scientists choose to investigate. For example, in a case where a scientist conducts an experiment, resources could be saved by probing into questions that might require more than one simulation to attain reliable results. For clarity, take the case of smallpox, which was eliminated in the 1980s.
The scientific community has raised debates seeking approval to destroy the remaining stockpiles of the variola virus that causes smallpox in an attempt to minimize the chances of future infections. Recently, research by the World Health Organization has highlighted specific, medically essential research that could not be possible if the existing stockpiles of the virus were destroyed (Greca et al., 2014).
The reason is that there is not enough knowledge of the virus to answer questions solely through simulation. This case provides evidence that there exist some questions that can only be answered with experiments on the virus itself. However, I think it is justifiable to consider disposing of the stockpiles. This assertion implies that studying a “material system as opposed to a computer model manifests better inferences and opportunities to uncover surprises” (Greca et al., 2014, p. 899).
To this end, it is necessary to consider some of the concerns that Parke (2014) has raised about the materiality thesis and in particular, her view that material correspondence does not necessarily involve higher inferential power and that it is hard to make sense of the difference between material and formal object–target correspondence.
Winsberg (2008) is sympathetic to the conclusion that such differences exist, although they might not mean much in the present when simulations should be used to complement experiments. Parke (2014) showed that there are exceptions to the generalization that experiments have epistemic privilege, but thinking in terms of such generalizations is misleading when judging the epistemic privilege in cases of scientific inquiry.
Parke emphasized that researchers should not look to the experiment/simulation distinction to generalize anything in principle about epistemic value. Contrarily, I have shown that multiple cases in experiments are assumed to have the higher epistemological prerogative as compared to simulations. Even though inferential power and capacity to create surprise may not be broadly generalizable across science, they offer important cues to support the epistemic superiority of experiments over simulations. Studying a material system as opposed to a simulation might not necessarily lead to perfect inferences, but at least there is an increase in reliability.
Conclusion
In this paper, I acknowledge that simulation is a critical new tool for scientists. Even though simulation shares many aspects with experiments, experiments possess a dynamic epistemic capacity. Although the foundations of simulations are still shaky, experiments seem to have gained ground. Simulations can allow one to undertake a study within a very short time and with limited resources.
However, this practical advantage comes with an epistemological price. This assertion holds that studying a model as opposed to a material system entails sacrificing realism, which undermines epistemic value. Nevertheless, more information is essential before concluding that one model has epistemic value over the other. Therefore, more research is needed on this subject of the superiority of experiments over a simulation.
References
Barberousse, A., Franceschelli, S., & Imbert, C. (2008). Computer simulations as experiments. Synthese Journal, 169(3), 557-574.
Fallis, D. (2007). Attitudes toward epistemic risk and the value of experiments. Studia Logica, 86(2), 215-246.
Gendler, T. (2010). Intuition, imagination, and philosophical methodology. Oxford, UK: Oxford University Press.
Greca, I., Seoane, E., & Arriassecq, I. (2014). Epistemological issues concerning computer simulations in science and their implications for science education. Science & Education Journal, 23(4), 897-921.
Greenwald, M. (2004). Beyond benchmarking—how experiments and simulations can work together in plasma physics. Computer Physics Communications, 164(3), 1-8.
Grune-Yanoff, T., & Weirich, P. (2010). The philosophy and epistemology of simulation: a review. Simulation & Gaming, 41(1), 20-50.
Parke, E. (2014). Experiments, simulations, and epistemic privilege. Philosophy of Science, 81(4), 516-536.
Winsberg, E. (2008). A tale of two methods. Synthese Journal, 169(3), 575-592.