Jeremy Attard

Complexity is often framed as an ontological property of systems, with profound epistemological and methodological implications–particularly in the social sciences, where it is invoked to justify fundamental limits on the applicability of natural science frameworks. Building on previous discussions by McIntyre, Ruphy, and more recently Francq, this paper challenges this ontological premise by arguing that complexity is best understood primarily as an epistemic property–that is, a feature of the models we use to represent and study systems, rather than an intrinsic characteristic of the systems themselves. Moreover, complexity depends on pragmatic elements, such as the goals at hand or the specific questions being asked. The foundational premise of our argument is that claims about a system’s complexity are invariably grounded in comparative judgments between models, whether formal or intuitive. Such judgments, in turn, depend critically on pragmatic and epistemic goals (e.g., the questions asked or the scale of analysis). We take this idea seriously and formalize it by defining the epistemic complexity of a system with respect to a given goal as the complexity of the least complex model capable of achieving that goal. Our account relies on a complexity score applicable to models, which we develop specifically for the restricted class of agent-based models (ABMs) in the social sciences. By reframing complexity as primarily an epistemic property, we dissolve persistent ambiguities in the literature–particularly the circular reasoning that arises when ontological claims about complexity are used to justify alternative epistemological frameworks. Our approach shifts the debate from metaphysical assertions to empirical questions about model performance, offering a constructive path forward for complexity research in both the natural and social sciences.

In this paper, I tackle one important obstacle to the minimal epistemological unity required in the interdisciplinary dialogue within complexity sciences, namely the ‘‘epistemological pluralist thesis’’. According to this view, there is no and can’t be any general epistemological framework within which all kinds of scientific production can be accounted for: each discipline, somehow, should be understood within its own epistemological perspective. I address this question in the specific case of the social sciences. I first synthesis typical pluralist arguments, defending the impossibility (or lack of relevance) to search for a common epistemological framework between all discipline. I then defend a monist position by opposing the pluralist theses both with theoretical and concrete (that is, manifest-counter-examples-based) arguments. I finally explore that question in the more specific context of complexity science.

Le présent travail s'inscrit à l'intersection de deux problèmes épistémologiques majeurs. D'une part, le problème de la démarcation scientifique, qui consiste à identifier ce qui distingue intrinsèquement un système (un énoncé, une théorie,...) scientifique d'un système non scientifique ou pseudo-scientifique. D'autre part, le problème de l'unité épistémologique des sciences, qui consiste à se demander si toutes les disciplines à vocation scientifique peuvent être vues comme des instanciations d'une notion unique de la scientificité. Ces deux problèmes ont soulevé de nombreux débats ayant mis en évidence un ensemble important de difficultés. Le terme « scientifique » désigne en effet des méthodes empiriques, des constructions théoriques et des pratiques de recherche si hétérogènes qu’il semble voué à l’échec d’en chercher une définition aisée à circonscrire. De plus, les objets des disciplines scientifiques sont eux-mêmes de natures très diverses, ce qui semble rendre pareillement caduque toute recherche d’un concept unique de science qui pourrait s’appliquer indépendamment de la discipline en question. Dans cette thèse, je me propose de prendre à contre-pied cet état de fait en soutenant la possibilité et la pertinence d'un modèle unitaire de la scientificité, tout en me restreignant à une approche épistémologique comparée entre les sciences physiques et les sciences sociales. Pour défendre mon propos, j’ai mobilisé deux types de réponses pouvant être opposées au constat présenté plus haut. D’une part, des réponses de principe, où j’examine et m’oppose à des arguments théoriques soutenant l’impossibilité ou en tout cas la difficulté de définir la scientificité en général, et la nécessité pour les sciences sociales de jouir d’une définition à part. D’autre part, je mobilise également des réponses par l’exemple. J’étudie alors plus en détail l’approche dite « analytique » en sociologie. Ce courant a cela d’intéressant pour mon propos qu’il ne semble pas nécessiter d’épistémologie alternative à celle ayant cours, par exemple, en physique ou en biologie, tout en prétendant bien produire des connaissances sur le monde social. Il s’agit donc d’un contre-exemple concret et manifeste de la thèse soutenant que la sociologie ne peut pas jouir du même type d’épistémologie que les autres disciplines. Plus concrètement, j’élabore un (méta-)modèle unitaire de la scientificité en me concentrant sur une unité d’analyse bien circonscrite : les modèles. Je distingue chez ses derniers, classiquement, deux composantes principales : une composante empirique, qui a pour vocation à identifier des régularités dans le réel auquel on a accès à travers des données, et une composante théorique, à visée explicative et classificatoire. Je propose alors de construire formellement un degré de scientificité global dans lequel se combinent la maximisation d’une certaine quantité d’information définie sur la composante empirique et un critère d’invariance structurelle défini sur la composante théorique. Ces diverses constructions, bien que formelles, permettent d’éclairer efficacement les questions épistémologiques que je me suis données au départ de mon travail, ce dernier ayant vocation à constituer une étape supplémentaire vers un modèle unitaire de la scientificité.

In the social sciences, within the explanatory paradigm of structural individualism, a theory of action – like rational choice theory – models how individuals behave and interact at the micro level in order to explain macro observations as the aggregation of these individuals actions. A central epistemological issue is that such theoretical models are stuck in a dilemma between falsity of their basic assumptions and triviality of their explanation. On the one hand, models which have a great empirical success often rest on unrealistic or even knowingly false assumptions; on the other hand, more complex models, with additional more realistic hypotheses, can (trivially) adapt to a wide range of situations and thus loose their explanatory power. Our purpose here is epistemological and consists in wondering to which extent demanding realistic assumptions in such cases is a relevant criterion with respect to the acceptance of a given explanatory model. Via an analogical reasoning with physics, we argue that this criterion seems too strong and actually irrelevant. General physical principles are not just idealized or unrealistic, they can also be formulated in many different yet equivalent ways which do not imply the same fundamental unobservable entities or phenomena. However, the classification of phenomena that such principles allow to highlight does not depend, at the end, on any particular formulation of these basic assumptions. This suggests that some hypotheses in theoretical models are actually not genuine empirical statements that could be independently tested but only substrates of modeling embodying a classification principle. Thus, we develop a structural invariance criterion that we then apply to rational choice models in the social sciences. We argue that this criterion allows to escape from the epistemological dilemma without condemning formal approaches like rational choice theory for their lack of realisticness nor being stuck to any antirealist viewpoint.

The problem of demarcating between what is scientific and what is pseudoscientific or merely unscientific - in other words, the problem of defining scientificity - remains open. The modern debate was firstly structured around Karl Popper's falsificationist epistemology from the 1930's, before diversifying a few decades later. His central idea is that what makes something scientific is not so much how adequate it is with data, but rather to what extent it might not have been so. Since the second half of the century, and in the wake of criticisms, such as the Duhem-Quine thesis, that were raised against falsificationism(s), most approaches to the problem of scientific demarcation are now multicriteria and holistic. However, the approach presented in this paper does not follow the same guideline. The present work can be seen as an attempt to adapt Popper's (sophisticated) falsificationism to a model-based view of scientific knowledge. Using formalization and focusing on a particular epistemic unit of analysis (namely, empirical models) allows to properly define the popperian corroboration degree and to view scientificity as the maximization of this degree of corroboration over all available models and data. We eventually recover, in a natural way, well-accepted scientificity criteria: empirical adequacy, Lakatos' progressive problemshifts, balance between strength and simplicity, parsimony, and coherence as special cases of this general scientificity principle. From this viewpoint, the language dependency of our empirical knowledge no longer appears as a limitation of falsificationism but as one more reason to take it as a good epistemological framework.

The deductive nomological (DN) model has been the basis for discussions about scientific explanations for decades. The overcoming of the logical empiricist program together with the raise of several counter-examples to the DN model have progressively led to a renewal of the reflections on this topic. The first step of this paper is to clarify the framework in which the epistemological question of scientific explanation is adressed. We make a proposal for a universal structure of scientific models, which constitute the basic epistemic unit of our analysis. It is then possible, within this framework, to clarify some discussions about scientific explanation, and to offer a new account of it. The latter tries to benefit from the advantages of the DN model, as resting on a law, together with neutralizing some of its typical criticisms. The work presented here is both abstract, exhibiting formal structures as general patterns of explanation, and concrete with real examples taken from actual scientific disciplines.