Ricambiare lo sguardo delle macchine. Dietro gli impliciti della face recognition attraverso le training images

Facial recognition technologies, which are increasingly part of our experience of daily interaction with artificial intelligence, are often presented, both by developers and by the companies and institutions that adopt them, as pure utilities in a context of simple and effective automation (Floridi 2019, 2020). The vast field of automated facial recognition (AFR) includes all those technologies that apply algorithms to the human face and facial expressions, from face recognition apps to CCTV and police cameras, decoding features and characteristics through a purely artificial vision. What happens, however, if we focus on the peculiar visual AI’s way of seeing and, at the same time, classifying the human face? Which are the (social, political, cultural, and semiotic) implications of automated facial recognition, machinic vision with the human face as its object? A number of artistic projects have turned attention to facial recognition technologies by reflecting on the “invisible” images (Paglen 2019) on which computer vision based on convolutional neural network (CNN) feeds in order to function, the so–called training images. After some notes on AFR as a form of artificial intelligence, this paper aims to examine some of such artistic projects, specifically Trevor Paglen’s ImageNet Roulette and From “Apple” to “Anomaly” (2019), considering them a possible way to access a greater understanding of facial recognition as well as computer vision as a peculiar phenomenon of the contemporary visual landscape.