By: Anthony Aubel
An essential aspect of scientific inquiry involves the use of models. The study of complex systems often demands a simplified means by which data can be represented and helps scientists make sense of patterns in the data. Philosophers of science more or less agree that in order for models to have any epistemic value, they must involve some sort of representation although there are widely varying views on what exactly those representations involve. Tarja Knuuttila, in her paper “Modelling and representing: An artifactual approach to model-based representation”, however, questions the limitations of the representational view in scientific models while advocating a different approach that considers the concrete ‘materiality and media-specificity’ as more productive means of assessing the epistemic value of models. She argues that models are to be seen as epistemic artifacts that stand for some real-world phenomena (Knuuttila, 2011). Although she does not exclusively deny the role of representation in scientific modeling, she does support deflationary accounts of scientific representation (Knuuttila and Others, 2005).
In this essay, I will reflect on the readings from Knuuttila to defend the position that representation is a necessary and essential component in scientific modeling that we cannot do away with if we want to have a plausible explanatory account of phenomena under investigation. Specifically, I argue that representation serves to preserve the relationship between phenomena in the world and the underlying mathematical rules that explain their behavior while emphasizing how the process of simulation in particular plays a key role in assessing the epistemic value of models.
In scientific practice, direct observation is commonly used as a means to gather and interpret data at initial stages of exploration. But relying on this method alone runs the risk of failing to identify patterns and relations that could be crucial in our understanding of the phenomena under investigation. In order to study the behavior of phenomena in a more systematic way, models are constructed and used to map real-world phenomena to the underlying theoretical-mathematical structure. This may, for example, involve geometric representation that describe the behavior of observed phenomena in the form of linear equations (or non-linear ones in the case of more complex systems). The key here is really the representational capacity of models in some form or another that enables some degree of epistemic access.
The knowledge gained from such practice contributes significantly to the epistemic value of models. But evaluating how well a model mediates the relation between real-world and theory requires a reliable system of measurement. This is where simulation as a process can come into play.
Simulation can be used as a metric to assess the epistemic value of models as representation. Simulation can be distinguished from a model by virtue of its dynamic property. Unlike the rather static nature of a model as purely representational, simulation is a process that makes use of variables within a specified set of constrained parameters as represented by the model to study the behavior of phenomena in concordance to predictions made by that model. This is consistent with Knuuttila’s demand for “concretely constrained and manipulable nature of models” (Knuuttila, 2011).
One way that simulation helps in evaluation of models (and the scientific claims it entails) is to provide a means by which its accuracy can be assessed. The accuracy of a model’s prediction can be tested by simulation based on how well it maps onto the real world. Models often attempt to represent a distillation of only those salient features and variables that are constrained by mathematical rules. Since the underlying mathematical structures are abstract representations, it’s often difficult to assess those parameters strictly in terms of a model’s concrete artefactual features, such as ‘materiality’ or ‘media-specificity’, as emphasized by Knuuttila (Knuuttila, 2022). In abstract representations, the parts that are often filtered out are understood as not playing a significant role in the model's predictions.
The extent to which a model fails to produce a simulation that accurately predicts behavior approximating real world phenomena can be a measure of its representational capacity. This often goes at the cost of attempting to generalize the behavior of a system for the sake of gaining a better understanding. Here, simulation can be used to test the generality of a model’s predictions. For instance, researchers can evaluate the generality by running a series of simulations of a model under a broad range of conditions. This serves not only as a valuable metric for assessing the epistemic value of a model, but also helps make complex systems or phenomena more accessible and understandable. One classic instance of this is noted in studying the behavior of gasses by applying the ideal gas law. By running a series of simulations of the model under varying temperatures, pressures, and volumes, one can determine how well the model generalizes to different conditions. This has been instrumental in efforts to provide a plausible explanation of not only the general behavior of gasses, but also in specific cases where accurate predictions were made under non-ideal conditions.
Another critical factor in assessing the epistemic value of a scientific model is its robustness. A model is considered robust if its predictions remain consistent across a range of different conditions. Therefore, a good way to measure the robustness of a model is to expose it to varying conditions via simulation and observe whether it can produce similar results compared to standard conditions. In material sciences, for instance, computer simulations are often run under varying conditions (i.e. temp. Pressure, deformation rate, etc.) to evaluate the robustness of a model’s predictions of the specific properties of materials (i.e. strength).
Computer simulations in particular can offer a flexible solution to Knuuttila’s concerns about difficulties in understanding how models can give us knowledge “if they were merely abstract structures” (Knuuttila, 2011). By utilizing mathematical rules, computer simulations provide a space where parameters that go into its predictions can be constrained and allowed to be manipulable. Contrary to Knuuttila’s insistence, this does not necessarily require us to always have a tangible object present (in many instances this would be impossible). In fact, it would be a mistake to pose such a limitation since it may hinder deeper understanding of underlying mechanisms. Having a robust understanding of salient properties of the object is often sufficient to gain the relevant knowledge we desire.
Representation is an essential component of scientific modeling that plays a crucial role in preserving the relationships between real-world phenomena and their underlying mathematical rules. The process of simulation is particularly valuable in assessing the epistemic value of models by providing a flexible and manipulable space where the predictions of the model can be tested and evaluated under a range of conditions. Overall, representation and simulation are essential tools that allow scientists to gain a better understanding of complex systems and phenomena and generate valuable knowledge about the world around us.
References:
Frigg, R. and Hartmann, S. (2020) ‘Models in Science’, The Stanford Encyclopedia of Philosophy (Spring 2020 Edition). Spring 2020. Edited by E. N. Zalta. Available at: https://plato.stanford.edu/archives/spr2020/entries/models-science/.
Knuuttila, T. (2011) ‘Modelling and representing: An artefactual approach to model-based representation’, Studies in History and Philosophy of Science Part A, 42(2), pp. 262–271. doi:10.1016/j.shpsa.2010.11.034.
Knuuttila, T. (2022) ‘Models, Representation, and Mediation’, Philosophy of Science, 72(5), pp. 1260–1271. doi:10.1086/508124.
Knuuttila, T. and Others (2005) ‘Models as epistemic artefacts: Toward a non-representationalist account of scientific representation’.
Morrison, M. and Morgan, M.S. (1999) ‘Models as mediating instruments’, Ideas in context, 52, pp. 10–37.