This project focuses on the philosophical implications of models and model-based predictive practices operating in social contexts. My approach is motivated by the need to respond to the problem of Two Cultures (a term coined by C. P. Snow in 1959, referring to the polarisation of Western intellectual life) by making stronger connections between the sciences and the humanities, as well as escaping the limits of critique. Consequently, this thesis engages in an in-depth technical analysis of predictive systems, and questions their social impact by using philosophcal tools. My method stems from the ‘non-standard’ philosophy and epistemology practiced by François Laruelle and Anne-Françoise Schmid respectively, and frames machine learning models as ‘integrative objects’. Schmid advocates that when faced with contemporary complex phenomena, any singular discipline is unable to fully grasp them and assess their implications. An integrative object is therefore an object exceeding disciplinary boundaries and requiring a complex approach in order to be productively understood. I argue that machine learning models represent a similar phenomenon – a central component in machine learning, a model represents patterns found in the training data, and subsequently serves to process new inputs. When operationalised in contexts such as policing, this process can have severe social consequences. Schmid emphasizes that interdisciplinarity is not enough and instead suggests ‘collective intimacy’ as a practice of disparate disciplines in true collaboration. As defined by Schmid: “This new mode of visibility supposes an ‘outside-discipline’ and a new conception of the scientific object, previously occulted by the epistemology of theories, too fascinated as they are by the criteria of the true.” (Schmid and Hatchuel 2014). The ‘outside discipline’ posited by this project emerges at the intersection of computer science, social science, as well as continental and analytic philosophy. Drawing on insights on the functions, characteristics, and limitations of models as theorised by the philosophy of science, I propose a framework for determining the usefulness of machine learning models implemented in social contexts. Rather than rejecting those systems wholesale as inadequate, and widening the gap between the critique of predictive systems and their engineering side, predictive models could be used as tools for further research and to help advance our understanding of modeled phenomena.