Paper short abstract:

This study draws on three year ethnographic work and 154 interviews to explores data annotation reintermediation in China, highlighting mediator organizations’ role in re-embedding data annotation in local societies and shaping human-algorithm complementarities in data annotating process.

Paper long abstract:

Data annotation, often unseen yet vital for data-centric technologies, is the focus of this investigation (Gray and Suri, 2019). Drawing on three years of ethnographic research and 154 interviews within China's annotation industry, this study adds to the debate on data work reintermediation. It responds to the emerging trend of reintermediation, as identified by Graham's team, which emphasizes the role of intermediary institutions in balancing algorithmic control and labor, challenging previous literature favoring disintermediation and its technological advantages (Graham & Lehdonvirta, 2017).

The concept of Complementary Organizations to Algorithms (COTAs) is introduced, drawing on human-computer interaction research. Supported by economists and sociologists, such as Autor (2015) and Shestakofsky (2017), it highlights a move towards complementarities but not substitutions between human and computer. The study examines how COTAs mitigate computerization’s and algorithms’ shortcomings by providing organizational resources for China's data supply chain and data annotation industry. Additionally, it illustrates that local governments, NGOs, and vocational training institutions can act as COTAs. For instance, the Guizhou local government functioned as a COTA, creating mechanisms to stabilize demand fluctuations for annotations from tech hubs and adopting a "people-optimization" approach to aid "product-optimization" in data annotation ecosystem.

Data collection involved thorough three-year fieldwork across seven data annotation centers in China and semi-structured interviews with stakeholders from major tech firms like Alibaba, TikTok, Tencent, and Baidu, data bureau and exchange center. This comprehensive method uncovers the complex interplay between human labor, technology, and the evolving data annotation landscape, highlighting COTAs' role in connecting these areas.