Sociotechnical challenges to the technological accuracy of computer vision: The new materialism perspective

Computer vision is a subdomain of artificial intelligence widely utilized in autonomous vehicles, digital health, and image processing at large. Despite significant breakthroughs, concerns persist regarding the technical challenges computer vision may encounter in the real, open world. Drawing on qualitative interviews and content analysis, this article examines the diverse sociotechnical challenges that affect the technical accuracy of computer vision systems in real-world settings. From a new materialism perspective, this article defines the technological accuracy of a computer vision system as its ability to successfully enact an objective reality of things through the maintenance of a stable sociotechnical assemblage comprising various components, including things, sensors, data, algorithms, experts, institutions, and more. Computer vision-based artificial intelligence systems evolve as dynamic, emergent entities responsive to contextual changes across phases, domains, and environments. However, when these changes are inadequately considered, computer vision may produce unintended performances or encounter technical dilemmas. Ultimately, the success of computer vision hinges on its capacity to adapt to an open world characterized by diverse contexts or, alternatively, to flatten those contexts into a uniform, closed world. The conclusion proposes future research and policy recommendations for achieving technological accuracy in computer vision.