Machine learning methods have recently created high expectations in the climate modelling context in view of addressing climate change, but they are often considered as non-physics-based ‘black boxes’ that may not provide any understanding. However, in many ways, understanding seems indispensable to appropriately evaluate climate models and to build confidence in climate projections. Relying on two case studies, we compare how machine learning and standard statistical techniques affect our ability to understand the climate system. For that purpose, we put five (...) evaluative criteria of understanding to work: intelligibility, representational accuracy, empirical accuracy, coherence with background knowledge, and assessment of the domain of validity. We argue that the two families of methods are part of the same continuum where these various criteria of understanding come in degrees, and that therefore machine learning methods do not necessarily constitute a radical departure from standard statistical tools, as far as understanding is concerned. (shrink)

Because they contain idealizations, scientific models are often considered to be misrepresentations of their target systems. An important question is therefore how models can explain the behaviours of these systems. Most of the answers to this question are representationalist in nature. Proponents of this view are generally committed to the claim that models are explanatory if they represent their target systems to some degree of accuracy; in other words, they try to determine the conditions under which idealizations can be made (...) without jeopardizing the representational function of models. In this article, we first outline several forms of this representationalist view. We then argue that this view, in each of these forms, omits an important role of idealizations: that of facilitating the identification of the explanatory components within a model. Via examination of a case study from contemporary astrophysics, we show that one way in which idealizations can do this is by creating a comparison case that s... (shrink)

Projections of future climate change cannot rely on a single model. It has become common to rely on multiple simulations generated by Multi-Model Ensembles (MMEs), especially to quantify the uncertainty about what would constitute an adequate model structure. But, as Parker points out (2018), one of the remaining philosophically interesting questions is: “How can ensemble studies be designed so that they probe uncertainty in desired ways?” This paper offers two interpretations of what General Circulation Models (GCMs) are and how MMEs (...) made of GCMs should be designed. In the first interpretation, models are combinations of modules and parameterisations; an MME is obtained by “plugging and playing” with interchangeable modules and parameterisations. In the second interpretation, models are aggregations of expert judgements that result from a history of epistemic decisions made by scientists about the choice of representations; an MME is a sampling of expert judgements from modelling teams. We argue that, while the two interpretations involve distinct domains from philosophy of science and social epistemology, they both could be used in a complementary manner in order to explore ways of designing better MMEs. (shrink)

Mathematical models are often expected to provide not only predictions about the phenomenon that they represent, but also explanations. These explanations are answers to why-questions and particularly answers to why the predicted phenomenon should occur. For instance, models can be used to calculate when the next total solar eclipse will happen, and then to explain why it will take place on July 2, 2019. In this regard we can obtain explanations from a model if we can solve the model equations (...) which govern the phenomenon under study. But some equations have no explicit solution or are too complicated to solve. In these cases it is difficult for a... (shrink)

Non-epistemic values pervade climate modelling, as is now well documented and widely discussed in the philosophy of climate science. Recently, Parker and Winsberg have drawn attention to what can be termed “epistemic inequality”: this is the risk that climate models might more accurately represent the future climates of the geographical regions prioritised by the values of the modellers. In this paper, we promote value management as a way of overcoming epistemic inequality. We argue that value management can be seriously considered (...) as soon as the value-free ideal and inductive risk arguments commonly used to frame the discussions of value influence in climate science are replaced by alternative social accounts of objectivity. We consider objectivity in Longino's sense as well as strong objectivity in Harding's sense to be relevant options here, because they offer concrete proposals that can guide scientific practice in evaluating and designing so-called multi-model ensembles and, in fine, improve their capacity to quantify and express uncertainty in climate projections. (shrink)

An important task in mathematical sciences is to make quantitative predictions, which is often done via the solution of differential equations. In this paper, we investigate why, to perform this task, scientists sometimes choose to use numerical methods instead of analytical solutions. Via several examples, we argue that the choice for numerical methods can be explained by the fact that, while making quantitative predictions seems at first glance to be facilitated by analytical solutions, this is actually often much easier with (...) numerical methods. Thus we challenge the widely presumed superiority of analytical solutions over numerical methods. (shrink)

Recent years have seen a notable increase in the production of scientific expertise by large multidisciplinary groups. The issue we address is how reports may be written by such groups in spite of their size and of formidable obstacles: complexity of subject matter, uncertainty, and scientific disagreement. Our focus is on the International Panel on Climate Change, unquestionably the best-known case of such collective scientific expertise. What we show is that the organization of work within the IPCC aims to make (...) it possible to produce documents that are indeed expert reports. To do so, we first put forward the epistemic norms that apply to expert reports in general, that is, the properties that reports should have in order to be useful and to help decision-making. Section 2 claims that these properties are: intelligibility, relevance and accuracy. Based on this analysis, section 3 points to the difficulties of having IPCC reports indeed satisfying these norms. We then show how the organization of work within the IPCC aims at and to a large extent secures intelligibility, relevance and accuracy, with the result that IPCC reports can be relied on for decision-making. Section 4 focuses on the fundamentals of IPCC’s work organization--that is, division of labour within the IPCC--while section 5 investigates three frameworks that were introduced over the course of the functioning of the IPCC: the reviewing procedure of IPCC reports, the language that IPCC authors use to express uncertainty and the Coupled Model Intercomparison Project. Concluding remarks are offered in section 6. (shrink)

Empirical agreement is often used as an important criterion when assessing the validity of scientific models. However, it is by no means a sufficient criterion as a model can be so adjusted as to fit available data even though it is based on hypotheses whose plausibility is known to be questionable. Our aim in this paper is to investigate into the uses of empirical agreement within the process of model validation.

Convergence of model projections is often considered by climate scientists to be an important objective in so far as it may indicate the robustness of the models’ core hypotheses. Consequently, the range of climate projections from a multi-model ensemble, called “model spread”, is often expected to reduce as climate research moves forward. However, the successive Assessment Reports of the Intergovernmental Panel on Climate Change indicate no reduction in model spread, whereas it is indisputable that climate science has made improvements in (...) its modelling. In this paper, after providing a detailed explanation of the situation, we describe an epistemological setting in which a steady model spread is not doomed to be seen as negative, and is indeed compatible with a desirable evolution of climate models taken individually. We further argue that, from the perspective of collective progress, as far as the improvement of the products of a multi-model ensemble is concerned, reduction of model spread is of lower priority than model independence. (shrink)

The assessments issued by the Intergovernmental Panel on Climate Change (IPCC) aim to provide policy-makers with an objective source of information about the various causes of climate change, the projected consequences for the environment and human affairs, and the options for adaptation and mitigation. But what, in this context, is meant by ‘objective’? In practice, in an effort to address internal and external criticisms, the IPCC has regularly revised its methodological procedures; some of these procedures seem to meet the requirements (...) of objectivity, at least as understood in a specific sense, but the relationship between objectivity and value-neutrality requires further investigation. The aim of this paper is to offer an appropriate philosophical account of objectivity, reconcilable with the fact that the IPCC is not value-free. I argue that Sandra Harding’s notion of strong objectivity is particularly well suited to this goal, and I examine the extent to which the current IPCC procedures match her account. (shrink)

It is often said that computer simulations generate new knowledge about the empirical world in the same way experiments do. My aim is to make sense of such a claim. I first show that the similarities between computer simulations and experiments do not allow them to generate new knowledge but invite the simulationist to interact with simulations in an experimental manner. I contend that, nevertheless, computer simulations and experiments yield new knowledge under the same epistemic circumstances, independently of any features (...) they may share. (shrink)

Whereas experiments and computer simulations seem very different at first view because the former, but not the latter, involve interactions with material properties, we argue that this difference is not so important with respect to validation, as far as epistemologyEpistemology is concerned. Major differences remain nevertheless from the methodological point of view. We present and defend this distinction between epistemology and methodology. We illustrate this distinction and related claims by comparing how experiments and simulations are validated in evolutionary studies, a (...) domain in which both experiments in the lab and computer simulations are relatively new but mutually reinforcing. (shrink)

It has been argued that the Duhem problem is renewed with computational models since model assumptions having a representational aim and computational assumptions cannot be tested in isolation. In particular, while the Verification and Validation methodology is supposed to prevent such holism, Winsberg argues that verification and validation cannot be separated in practice. Morrison replies that Winsberg overstates the entanglement between the steps. The paper aims at arbitrating these two positions, by stressing their respective validity in relation to domains of (...) application. It importantly argues for an increasing use of formal methods in verification, that makes disentanglement possible. (shrink)

Wagenknecht a récemment introduit une distinction conceptuelle entre dépendance épistémique translucide et dépendance épistémique opaque, dans le but de mieux rendre compte de la diversité des relations de dépendance épistémique au sein des pratiques collaboratives de recherche. Dans la continuité de son travail, mon but est d’expliciter les différents types d’expertise requis lorsque sont employés instruments et ordinateurs dans la production de connaissance, et d’identifier des sources potentielles d’opacité. Mon analyse s’appuie sur un cas contemporain de création de connaissance scientifique, (...) à savoir le traitement de données astrophysiques. (shrink)

Like any science marked by high uncertainty, climate science is characterized by a widespread use of expert judgment. In this paper, we first show that, in climate science, expert judgment is used to overcome uncertainty, thus playing a crucial role in the domain and even at times supplanting models. One is left to wonder to what extent it is legitimate to assign expert judgment such a status as an epistemic superiority in the climate context, especially as the production of expert (...) judgment is particularly opaque. To begin answering this question, we highlight the key components of expert judgment. We then argue that the justification for the status and use of expert judgment depends on the competence and the individual subjective features of the expert producing the judgment since expert judgment involves not only the expert's theoretical knowledge and tacit knowledge, but also their intuition and values. This goes against the objective ideal in science and the criteria from social epistemology which largely attempt to remove subjectivity from expertise. (shrink)

Parameterization and parameter tuning are central aspects of climate modeling, and there is widespread consensus that these procedures involve certain subjective elements. Even if the use of these subjective elements is not necessarily epistemically problematic, there is an intuitive appeal for replacing them with more objective (automated) methods, such as machine learning. Relying on several case studies, we argue that, while machine learning techniques may help to improve climate model parameterization in several ways, they still require expert judgment that involves (...) subjective elements not so different from the ones arising in standard parameterization and tuning. The use of machine learning in parameterizations is an art as well as a science and requires careful supervision. (shrink)

In empirical modeling, mathematics has an important utility in transforming descriptive representations of target system into calculation devices, thus creating useful scientific models. The transformation may be considered as the action of tools. In this paper, I assume that model idealizations could be such tools. I then examine whether these idealizations have characteristic properties of tools, i.e., whether they are being adapted to the objects to which they are applied, and whether they are to some extent generic.

Climate science is expected to provide usable information to policy-makers, to support the resolution of climate change. The complex, multiply connected nature of climate change as a social problem is reviewed and contrasted with current modular and discipline-bounded approaches in climate science. We argue that climate science retains much of its initial “physics-first” orientation, and that it adheres to a problematic notion of objectivity as freedom from value judgments. Together, these undermine its ability to provide usable information. We develop the (...) notion of usability using work from the literature on adaptation, but our argument applies to all of climate science. We illustrate the tension between usability and the objective, physics-first orientation of climate science with an example about model development practices in climate science. For solutions, we draw on two frameworks for science which responds to societal challenges: post-normal science and mandated science. We generate five recommendations for adapting the practice of climate science, to produce more usable information and thereby respond more directly to the social challenge of climate change. These are: (1) integrated cross-disciplinarity, (2) wider involvement of stakeholders throughout the lifecycle of a climate study, (3) a new framing of the role of values in climate science, (4) new approaches to uncertainty management, and (5) new approaches to uncertainty communication. (shrink)

Some philosophers see an analogy between simulation and experiment. But, once we acknowledge some similarities between computer simulations and experiments, can we conclude from them that simulations generate empirical knowledge, as experiments do? In this paper, I argue that the similarities between simulation and experiment give the scientist at most the illusion that she is conducting an experiment, but cannot seriously ground the analogy. However, it does not follow that experiments are always epistemologically superior to simulations. I analyze the cases (...) when simulations and experiments equally yield new empirical knowledge. (shrink)

In philosophical studies regarding mathematical models of dynamical systems, instability due to sensitive dependence on initial conditions, on the one side, and instability due to sensitive dependence on model structure, on the other, have by now been extensively discussed. Yet there is a third kind of instability, which by contrast has thus far been rather overlooked, that is also a challenge for model predictions about dynamical systems. This is the numerical instability due to the employment of numerical methods involving a (...) discretization process, where discretization is required to solve the differential equations of dynamical systems on a computer. We argue that the criteria for numerical stability, as usually provided by numerical analysis textbooks, are insufficient, and, after mentioning the promising development of backward analysis, we discuss to what extent, in practice, numerical instability can be controlled or avoided. (shrink)

In 1959, mathematician Mark Kac introduced a model, called the Kac ring, in order to elucidate the classical solution of Boltzmann to the problem of macroscopic irreversibility. However, the model is far from being a realistic representation of something. How can it be of any help here? In philosophy of science, it is often argued that models can provide explanations of the phenomenon they are said to approximate, in virtue of the truth they contain, and in spite of the idealisations (...) they are made of. On this view, idealisations are not supposed to contribute to any explaining, and should not affect the global representational function of the model. But the Kac ring is a toy model that is only made of idealisations, and is still used trustworthily to understand the treatment of irreversible phenomena in statistical mechanics. In the paper, my aim is to argue that each idealisation ingeniously designed by the mathematician maintains the representational function of the Kac ring with the general properties of macroscopic irreversibility under scrutiny. Such an active role of idealisations in the representing has so far been overlooked and reflects the art of modelling. (shrink)

On July 5, 2012 the Investigation Committee on the Accident at the Fukushima Nuclear Power Stations of the Tokyo Electric Power Company (TEPCO) issued a final, damning report. Its conclusions show that the human group – constituted by the employees of TEPCO and the control organism – had partial and imperfect epistemic control on the nuclear power plant and its environment. They also testify to a group inertia in decision-making and action. Could it have been otherwise? Is not a collective (...) of human beings, even prepared in the best way against nuclear risk, de facto prone to epistemic imperfection and a kind of inertia? In this article, I focus on the group of engineers who, in research and design offices, design nuclear power plants and model possible nuclear accidents in order to calculate the probability of their occurrence, predict their consequences, and determine the appropriate countermeasures against them. I argue that this group is prone to epistemic imperfection, even when it is highly prepared for adverse nuclear events. (shrink)