Rule(s) of Prediction
Contemporary societies increasingly rely on predictions based on machine learning in manifold settings, including media distribution, policing, economy, etc. Many scholars are invested in researching the consequences of these transformations. However, the question is mainly tackled from an output-oriented question, emphasizing bias or the lack of transparency of these systems. A question hardly asked, though, is the one of the conditions of possibility of these systems, or: how machine learning algorithms are becoming integrated into the newly emerging social figurations. In my PhD project, I therefore raise the question, what epistemic conditions and practices of knowledge production are needed to enable prediction in socio-technical systems. The increasing reliance on Predictive Analytics and Machine Learning Systems results in new social figurations, accommodating possibilities of tackling false positives of the prediction but also, even before the algorithms computed one single number, reduces contingency in interpretation for data production. In both cases the epistemic necessities implicitly formulated by algorithmic systems therefore produce new forms of socio-technical order and a system of control as new forms of socio-material interdependencies, resulting in what could be called “doing predictions”.