Abstract
Starting from the idea that an epistemological shift from the identification of causes to spaces of massive data saturated with strong correlations characterizes both most of the algorithms encircling our lives - such as recommendation algorithms -, and generative artificial intelligence such as LLMs illustrate, this talk will propose a few concepts likely to draw a framework for making intelligible the new epistemic and political configurations opened up by AI.
I'll start with the concept of profile (defended in Les sociétés du profilage, Payot 2013) as a point in a hyperspace of data, then outline the related notion of ranking (probabilisticranking ), and that of 'score' which flows from it and floods our existences (credit score, social credit, polygenic risk score, benefit score, etc.).
Considering the ontological consequences of profiling, I will defend the relevance of the notions of loop and miscibility. I will show their relevance to the issues of generated images and algorithmic biases.
