Epistemic Interfaces of Visualization and Interpretation: a possible resistance to the myth of autonomous agency in AI

Abstract

Anthropological studies on agency and cognition have used theoretical frameworks of co-dependency and composition (between subjects and objects) to explain the technical production processes of knowledge. In this context, ethnographic studies were carried out on digital technologies (Horst y Miller 2012; Geismar y Knox 2021), namely on Artificial Intelligence (AI) (Forsythe 2001; Suchman 2007), both in re- lation to itsproduction — with its designers and programmers — and in its imple- mentation — on its uses and appropriations. However, according to authors such as Pink (2021) or Drucker (2020), we find less research that bringsthese anthropologi- cal theories to the design of the technologies that determine our current cognitive ecology, as is the case of deep learning. Thus, this paper’s contribution lies in the development of a collaborativeethnography — with a deep learning engineer — to generate two experimental human-AI interaction interfaces that offer possible re- sistance to the myth of autonomous agency that characterizes the official discourseof AI; a mythical speech (Barthes 1957) that, as will be argued through the description of the ethnographic work, does not correspond to the action of programming AI. Ha- ving this as a starting point, we presentthe hybrid method that guides this research between deep learning programming for artificial vision, computational interpretability techniques and the use of artistic strategies —, a methodology that allowed the incorporation of the ethnographic knowledge ofthe field work into the interface design, where the anthropological concepts of social agency (Gell 1998; Latour 1999) and user-maker (Ingold 2013) stand out.