The pragmatics of correlation or how models reshape the government of technical objects

In this paper we focus on how models carry out correlations and how correlations are used as techniques of government. Correlations are fragile exercises which put together two phenomenon, e.g. one nanoparticle and one toxic effect, without linking them with a straight causality. We examine correlations in the situated case of QSAR (quantitative structure-activity relationship) models applied to the toxicity of nanoparticles. Using scientific and regulatory documentation related to QSAR, and the preliminary results of an empirical study of a QSAR project conducted in France, we analyze the initial expectations, i.e. providing knowledge on nanoparticles useful for regulatory purposes, and how this ambition was confronted to the refusal of nanoparticles to feed the QSAR models in a satisfying way. We argue that this led to "thinking on the edge" the wonders and worries of correlation. This calls for a pragmatics of correlation: what does it do? Which are the effects and consequences of a "co-relating" approach? Exploring these questions leads us to examine the type of knowledge produced through models applied to large sets of data, and how it contributes to certain types of regulatory objectivity, particularly in the European legal context. Thus, we discuss the ways in which a pragmatics of correlation offers analytical entry points for the study of the contemporary transformation of the public administration of technical objects.