The Labor of Training Artificial Intelligence: Data Infrastructure, Mobility, and Marginality

Machine intelligence relies on Al (artificial intelligence) trainers, workers who perform labor such as data annotation and algorithm optimization. However, the promise of Al does not often benefit workers equally; instead, it puts them in precarious situations, such as low wages and subordination to machines. My dissertation takes an interdisciplinary approach to draw attention to these pressing issues by exploring the sociotechnical, cultural, and economic dimensions of this emergent technology-mediated labor in the context of large data infrastructures. My arguments and proposed concepts (e.g., sociotechnical/algorithmic mobility) respond directly to the under-theorization of mobility research and ecologically unequal exchange theory in HCI. In my dissertation, I argue that the Al trainers, who often work in developing regions of western China, are shouldering the burdens of (1) alleviating China’s poverty through Al for development programs, (2) sustaining Eastern China’s platform economy as key participants in large-scale data infrastructure projects, and (3) promoting global Al advancement by providing disembodied labor on products such as high-quality training datasets through repetitive and low-paying work. Using multi-sited ethnography and participatory design methods, my dissertation describes the experiences of under-resourced and under-studied Al trainer communities and the effects of AI on them. It will also offer context-sensitive design recommendations for supporting emergent technology-mediated labor and policy interventions for ethical and sustainable Al training practices.