Abstract

What is the method of Critical AI Studies? So far, this emerging discipline has almost exclusively resorted to porting methodological gambits from established disciplinary configurations and apply ing them onto the new object/assemblage/practice that is generative AI. In this talk I examine the repercussions of this methodological conservatism. More specifically, I uncover three casuistries in how the concepts of benchmarks, black boxes and stacks are considered by humanistic studies of artificial intelligence, leading to a critical landscape that hopelessly lags behind the technical status quo. I finally call for, and point towards, a more technically sound set of methodologies that might take into account humanistic strengths in the “close reading” of technical objects without foreclosing the potential offered by technical specificity.