What is it？
Invisibility of Computer Vision is an ongoing research project on how scientists train AI models with their own eyes.
Entering the laboratory
I conducted nine-month laboratory ethnography as my master dissertation on computer vision (CV), a dominant subfield of AI on how to enable machine to see like human.
By tracing the journey of papers in this field from initial inspiration to acceptance at top AI conferences, I find image data mediated by GPU and computer screen became “invisible” and unexplainable; to address this technical issue, scientists usually modify, debug, and validate the model through human eyes instead of quantitative metrics. Through the AI lab study, I practice situating this sensory reasonableness of knowledge into the tradition of scientific image making, while it challenges the philosophical concept such as objectivity, theory, experiment, and validation as well as anthropological concern on inscription.
