IEEE Annals of the History of Computing

The detection and removal of child pornography online is a vexing process, weighted with moral urgency. In recent years, this process has become increasingly digitally automated. In 2016, the National Center for Missing and Exploited Children (NCMEC) reported that they had reviewed over 500,000 image and video files suspected to contain child pornography; and since 2002 they had reviewed an aggregate of 192 million images in order to identify 12,500 child victims.¹ NCMEC is a nonprofit that serves as the United States’ main clearinghouse for child exploitation content, and works directly with federal and local law enforcement to produce case files in criminal investigations of missing children, child abuse, and child exploitation recorded for the intent of commercial distribution. Under a 2008 bill,² electronic communication service providers such as Facebook and Microsoft are also mandated to report suspected child abuse images to NCMEC. My research tracks how this legislation has marked a turning point in the reporting and classification of child pornography and its bracketing under the recent political movement against child trafficking. As electronic media companies’ stake in exploitation cases grows, corporate research teams have begun developing tools to automatically detect images that might potentially contain nudity or child pornography. NCMEC and law enforcement, similarly, have begun using software to match flagged images against databases of known missing children or criminal records.

Historians of computing and software might notice that the “new” technologies I am following have a much older precedent, one that relies upon a different sort of representational craft—forensic image sketching. Policing institutions have long maintained artists on forensic teams to assist in producing composite images through oral description, to do age progressions on missing persons, or to produce facial reconstructions based on postmortem skeletal remains. The composite sketches produced of a victim or a suspected offender are the result of a collaborative and discursive process between investigators, lab experts, witnesses, and family members before becoming manifest through the artist’s pencil. The production of a forensic sketch is thus a highly collectivized task, one oriented toward the “apprehension” of a potential victim or offender. Apprehension implies the prospect of arrest or rescue, and simultaneously the process of making apparent through the crafting or procurement of identifying data to build a corpus of evidence. I add to this that in order to know what to detect one must become knowledgeable about the object (e.g., data, face, person) sought. I suggest we may view the work of digital image forensics in child exploitation cases as similarly heteromated,³ a practice of distributed cognition⁴ between computer vision algorithms and the humans who interpret algorithmic outputs. The notion of heteromation acknowledges the necessity of human mediation in a technical process and suggests the distributed nature of apprehension.

Over the past three decades, forensic investigations have become increasingly specialized to fall under the domain of technical personnel. Ericson and Shearing⁵ describe this process of professionalization as a “scientification” of the police, with the division of policing into public safety and criminal investigations. While “street cops” devoted to public safety routinely patrol areas, investigative police are trained to work more acutely on long-term cases and to use forensic technologies, from fingerprinting to DNA testing, to obtain evidence and bolster claims in legal proceedings. Scholars in social studies of science and technology have documented the historical and cultural contexts of reliance on technological methods in forensic investigation—especially the symbolic value afforded to technological evidence and those who can wield it—in a society that increasingly upholds the power of scientific truth and production.⁶ Historians of technology have also described how representational devices—technologies, practices, objects—have been made central to the structuring of scientific practice.⁷ The discourse around digital image forensics mirrors that of DNA forensics, with a “belief in science and technology as a way to attain the truth and effectiveness in criminal investigation.”⁸ Forensic technologies have an “aura of infallibility.”⁹ The marketing of image forensics and classification software often touts its high accuracy and low rate of false positive results; that...