Send message to Author

Paper short abstract

Drawing on a historical overview of computer vision’s role in contemporary AI, the research traces the material conditions of its production, from platform-based gig work to business process outsourcing companies in the global South, operating under the narrative of impact sourcing.

Paper long abstract

Computer vision is the field that has primarily driven the contemporary expansion of deep learning. While its success in the late 2000s was certainly grounded in the innovative architecture of convolutional neural networks, it was also the outcome of a new form of labor organization: the large-scale image annotation carried out by Amazon Mechanical Turk microworkers, one of the first online platforms to place a strong emphasis on AI training.

Since then, data annotation for AI training has expanded significantly, not only through the proliferation of different platforms, but also by redirecting a substantial portion of the business process outsourcing industry—previously engaged mainly in operations such as customer service—toward the training and fine-tuning of neural networks. This production regime is often framed as impact sourcing, i.e., the practice of recruiting highly marginalized social groups with low levels of education and limited access to formal employment. Behind the narrative that portrays this production model as a vehicle for emancipation and empowerment often lies a business strategy that exploits the limited bargaining power of these social groups and their abundant labor supply.

Building on a historical and socio-technical analysis of AI, the proposal examines the case of companies in India that recruit marginalized groups along gendered, cultural, and religious lines to train computer vision systems. Drawing primarily on semi-structured interviews with data workers, the analysis addresses the concrete material conditions under which these AI models are produced.