Beyond transparency? YouTube and epistemic algorithms.

YouTube is often used as a reservoir of amateur footage for the constitution of research corpuses. Once constituted thanks to an extraction of videos from the platform, these corpuses are archived and submitted to different heuristic, hermeneutic and epistemic editing operations. Research is thus generally conducted outside the platform or “in spite of” the algorithms that determine its use. Starting from the method of “reasoned wandering” (Riboni 2016) that I applied in my recent research on the videos of the so called Yellow Vests protests (Hamers 2022), my paper aims at taking a step backwards to identify how the platform itself can be considered as an epistemic device insofar as it always performs editing operations that cannot be reduced to partly opaque and commercially (or ideologically) oriented processes. This project requires the researcher to stop considering the algorithm responsible for the platform's selections and suggestions as an obstacle to overcome (in line with the instructions the “support” or “faq” pages of the platforms themselves offer to the user, blurring the distinction between critical media education and functional media literacy ), or as a process to become aware of (in a media critical perspective) towards more transparency. Only then can the researcher identify new epistemic potentialities through possible dysfunctions or productive coincidences generated by the platform. Rather than reflecting on what kind of epistemic practices the platform hinders and how to bypass its algorithmic selection processes, I will try circumscribe to what extent we can consider it as an epistemic media, not from the point of view of the individual user feeding the machine with data that enables its function (Fisher 2021), but from the point of view of research on the circulation of visual motifs. In other words, I will try to show that against the possible (and sometimes fantasized) transparency that should allow the researcher to de-opacify the functioning of the platform, the outputs of the device itself can be handled as symptoms of a system driven by algorithms yet productive of critical knowledge. In order to outline such a possible epistemic agency of what I would like to consider as an interface of “visual study” (Brenez, 2009), some concrete outputs of the video selection processes operated by the platform will be confronted with the conception of a “heuristic montage” (Didi-Huberman 2009/2010/2011, following Warburg’s atlas method) as well as with Chris Marker’s and Harun Farocki’s related practices of the synthetic montage of (dis)continuity.