Accepted paper:

Facial recognition technology as software of categorization

Authors:

Abigail Nieves Delgado (Ruhr University Bochum)

Paper short abstract:

This paper focuses on the use of normative face templates in the development of facial recognition technologies. Based on an historical case study it analyzes the expectations underlying the introduction of such templates and discusses software-based identification and categorization practices.

Paper long abstract:

In recent years, the use of facial recognition technologies has gained popularity in public and private security sectors. These technologies work by capturing an individual's face in order to authenticate identity or identify a person in surveillance scenarios. The function of these technologies, involves several standards that introduce into the software specific normative values on how a face is supposed to look like. In other words, different facial templates and standards are used in facial recognition technologies to "teach" algorithms what a "normal" face is. This normativity turns recognition into categorization. This talk investigates the factors influencing the establishment of facial standards and discusses their consequences. Therefore, I present a case study from the 1990s known as FERET (Facial Recognition Technology). FERET was a competition organized by the Department of Defense in the U.S. to evaluate the state of the art in facial recognition technologies and to foster research in the field. By looking at the research teams and algorithms participating in FERET, this research explores how the different facial templates used in algorithms affect processes of human recognition and categorization. Who can be recognized or authenticated depends on the kind of information introduced during the algorithmic learning phase. As this case study shows, this normative dimension in the development of facial recognition technologies is of high political relevance, as their assumed objectivity and neutrality in public and private contexts fall into question. Lastly, this historical case seeks to provide insight into the functioning of ongoing software-based identification practices.

panel E03
Software sorted subjectivities