Prototype: “Photogénie: img_004_1.jpg”
The first public demonstration of facial recognition software (FRS) was designed by Takeo Kanade and displayed at the 1970 World Fair in Osaka, Japan: “Nippon Electric Company ran an attraction named ‘Computer Physiognomy’: a person sits before a Picamera, the picture of his face is digitized and fed into the computer…[and] his face is classified into one of seven categories, each of which is represented by a very famous person”. That the tech demo was based on celebrity faces fits within the historical trajectory of cinematic history, wherein thinkers like Walter Benjamin, Béla Balázs, and Jean Epstein wrote about the magnetic qualities of cinematic faces in close-up. Later, it was the celebrity face that fascinated Warhol in his Screen Tests (1964-66), films that capture some of Kanade’s rationale for using famous faces: by the early 1970s, the celebrity face became completely symbiotic with the technology that produced and distributed it.
The faces contained within FRS machine learning (ML) training databases are essential to the application of the technology on the various populations under the action of its gaze. Importantly then, many contemporary databases, including YouTubeFaces in the Wild, Celebrities Images, and ibug, use celebrity headshots for ML and facial recognition. The ibug facial landmarks dataset, in particular, was essential in establishing the 68 specific points that modern FRS use to identify and verify unique faces.
The use of celebrity faces deeply troubles the “recognition” element of a facial recognition apparatus. In one sense, the type of recognition a facial recognition-enabled (FRE) camera possesses defines the type of spectator the FRE-camera is, and, from this, we should ask how training the artificial intelligence (AI) portions of the FRE-camera on celebrity faces optimizes for certain types of “recognizable” faces (i.e. the types of populations that are most likely to be granted fame and celebrity).
In another sense, a famous face is “recognizable,” and with that recognition comes slippery blurrings of public and private: the celebrity face represents a specific individual’s identity; however, it is also public, in that it is produced for mass engagement and circulated across mass media. By treating every person that the FRE-camera looks at as “recognizable,” the apparatus is creating similarly dangerous notions of private individuals as public commodities through which each person’s private and individual biometrics become nodes in public models of data processing that can range from mall kiosks, to job hiring, to law enforcement and border security.
Reflecting the above arguments, using the ibug dataset, my in-progress research creation project “Photogénie” uses the faces within the dataset, processes them through custom-built FRS, to produce pieces that display the multiple ways that FRS “sees” a human face and the central place the celebrity face has played in the problematic advances of the technology.