The book The Politics of Modelling has its origins in the COVID crisis of 2020, when modeling, more than ever before, became a practice that guided and legitimized political decision-making. The COVID crisis was also the moment when the public became aware of the impact of modeling on political decision-making and the perception of “reality.” Given the controversial potential of the increased importance of modeling, the editors of the volume, Andrea Saltelli and Monica Di Fiore, together with other researchers, published a manifesto in Nature in 2020. The aim of the manifesto was to develop guidelines for responsible modeling, allowing for a critical use of models and their results.

Following the publication of the manifesto, the editors decided to collect different visions on the subject from lawyers, statisticians, economists, and philosophers from academic fields as diverse as governance, operational research, cognitive sciences, ethics, environmental sciences, civil engineering, and epidemiology. They invited these experts to give their views on what happens when a model is immersed in the system for which it was created, as well as on its performative properties and ethical implications.

The volume is divided into three sections. The first is devoted to the “nature of models” and includes the editors’ introduction and a contribution by Philipp Stark on the (weak) connection between models and the phenomena they describe. In their introduction, Saltelli and Di Fiore point to what they see as the core questions of modeling: Does a model assume neutrality or does it declare its normative stance? What is its level of abstraction? How attentive is it to the representation of uncertainty (p. 7)? Normativity, reduction of complexity, and uncertainty are the central dimensions of the book and the thread that connects the different contributions. The papers in section II discuss modeling in terms of normative frameworks and assumptions of practice, the implications and consequences of the mathematical structure of models, and the problem of unknowns. Section III focuses on the rules in practice and the methods to assess the performance of the model to measure uncertainty or its impact and consequences in decision-making.

Most of the contributions deal in one way or another with the limits of quantification and the problem of overreliance on numbers. In their joint contribution, political scientist Wolfgang Drechsler and philosopher Lukas Fuchs argue for a “metric fixation” that is intertwined with widely accepted norms of objectivity. Quoting Theodore M. Porter, they point to the core problem of quantification: reducing a decision to numbers is “a way of deciding without seeming to decide” (p. 92). Models, as number-producing devices, are no longer representations of the world that support the decision-making process but become the centerpiece of that process. While giving the impression of an “exhaustive representation of the phenomenon,” they often serve laissez-faire policies that seek to preserve the status quo (p. 93). According to Drechsler and Fuchs, the overreliance on numbers is accompanied by a voluntary ignorance of the caveats, ambiguities, and uncertainties that characterize the process of quantification and often dramatically limit the depth of our understanding and the range of policy alternatives (p. 86).

The caveats of quantification are closely linked to the problem of complexity and uncertainty. Models are only as good as the scientific understanding of the phenomenon and as reliable as the data provided to train them. If the understanding of the complexity of the phenomenon is poor, or if there aren’t enough data to test the model, the predictions won’t live up to expectations. Ting Xu, professor of law at Essex Law School, argues for a taxonomy for understanding and assessing the degree of uncertainty (p. 154), which she believes could help provide a framework for a better understanding of the types of uncertainty that modelers and regulators are dealing with (p. 162). Xu refers to Anthony Giddens’s distinction between external and manufactured risk and John Kay and Mervyn King’s distinction between resolvable and radical uncertainty. Both concepts distinguish between events that can be predicted in probabilistic terms because they occur regularly and often enough, and events that cannot be predicted because history provides no or insufficient experience to formulate them in probabilistic terms (pp. 157–159). In short, uncertainty accounts for all events that haven’t happened yet or are beyond our comprehension. Uncertainty is what makes modeling and prediction both difficult and necessary.

Although the contributors to the book agree that it is very important to assess the use and impact of modeling in policymaking, few of the contributions include case studies that would provide insights into how models have been used in a specific context. This is surprising, not least because the COVID pandemics that prompted the book project demonstrated numerous examples of the use of modeling and the interplay between modelers and policymakers. The question of what happens “when a model is immersed in the system for which it was created” remains largely unanswered. Highlighting the demand side of modeling would not only help to understand the growing importance of modeling as a purposeful scientific practice, it would also help to understand modeling as part of a “governmentality” that prioritizes the reduction of complexity through quantification. In Michel Foucault’s understanding, “governmentality” comprises a set of different techniques of governance, rules, routines, and institutions that together form a political climate that also influences individual ways of thinking and acting. In this concept, science and politics are not only intertwined but form a system in which power cannot exist without knowledge and vice versa. Seeing modeling as part of such a governmentality would help to better understand its normative and historical dimensions.

The governmentality of the “metric fixation” and the claim to reduce complexity that characterizes the demand for and use of modeling are historically specific. It would therefore be important to examine the historical context in which modeling arose. Who has benefited from number-based policies? Who didn’t? What has been lost along the way? Modeling, as a particular way of representing the world, cannot be dissociated from a period in which economic, social, and technological conditions are rapidly evolving, requiring a new form of orientation towards a future that may be quite different from the present. In this specific historical context, uncertainty becomes synonymous with historical change, the origins and dynamics of which we don’t fully understand. In this perspective, modeling could also be seen as a communication tool that helps to address our changing perceptions and assessments of “reality.”