Article Details

Computer vision research feeds surveillance tech as patent links spike 5×. A bottomless appetite for tracking people as "objects". A new study shows academic computer vision papers feeding surveillance-enabling patents jumped more than fivefold from the 1990s to the 2010s. The researchers, including Stanford University's Pratyusha Ria Kalluri and Trinity College Dublin's Abeba Birhane, collected more than 19,000 research papers and more than 23,000 patents. In a manual content analysis of 100 computer-vision papers and 100 citing patents, they report that 90 percent of the papers, and 86 percent of the patents, extracted data relating to humans. The survey of academic research, published this week in the journal Nature, found that ambiguous language sometimes masked the surveillance implications of the research by referring to humans as objects. “Through an analysis of computer-vision research papers and citing patents, we found that most of these documents enable the targeting of human bodies and body parts,” it said. Far from a few rogue researchers, the study found that “the normalization of targeting humans permeates the field. This normalization is especially striking given patterns of obfuscation. "We reveal obfuscating language that allows documents to avoid direct mention of targeting humans, for example, by normalizing the referring to of humans as ‘objects’ to be studied without special consideration. Our results indicate the extensive ties between computer-vision research and surveillance,” the researchers wrote. In an accompanying article, Jathan Sadowski of Monash University in Melbourne argued it was unlikely to be coincidental that computer vision research ended up serving the needs of military, police, or corporate surveillance. "These technologies are created in a political and economic landscape in which the interests of massive corporations and military and policing institutions have a huge influence over the design and use of AI systems," he wrote. Furthermore, "I would argue that computer vision's focus on human data extraction does not merely coincide with these powerful interests, but rather is driven by them. The language surrounding this research is a powerful form of object-oriented obfuscation that abstracts any potential issues related to critical inquiry, ethical responsibility, or political controversy," he added. Sadowski argued the research should underpin more critical forms of inquiry and policy “focused on blocking the pipeline that supplies the military–industrial surveillance complex with such powerful systems.”