Selected Papers of #AoIR2024:  

The 25th Annual Conference of the  
Association of Internet Researchers 

Sheffield, UK / 30 Oct - 2 Nov 2024 
 
 

 

Suggested Citation (APA): Tracey, P., Zhang, B.Z., Garcia, P., Haimson, O.L., & Thomas, M. (2024, 
October). The Technopolitics of Waiting: Case Studies of AI Training in China and Homeless Services 
Systems in the U.S. Paper presented at AoIR2024: The 25th Annual Conference of the Association of 
Internet Researchers. Sheffield, UK: AoIR. Retrieved from http://spir.aoir.org. 

THE TECHNOPOLITICS OF WAITING: CASE STUDIES OF AI 
TRAINING IN CHINA AND HOMELESS SERVICES SYSTEMS IN THE 
U.S. 
 
Pelle Tracey* 
University of Michigan, School of Information 
 
Ben Zefeng Zhang* 
University of Michigan, School of Information 
 
Patricia Garcia 
University of Michigan, School of Information 
 
Oliver L. Haimson 
University of Michigan, School of Information 
 
Michaelanne Thomas 
University of Michigan, School of Information 
 
*The first two authors contributed equally to this work. 
 
Introduction 
 
Many theorists of the information economy have argued that digitization has resulted in 
a “speeding up” of our experience of time (i.e. Gleick, 1999). In contrast, we contend 
that for many, especially those with less power, the techno-utopian vision characterized 
by datafication and Artificial Intelligence (AI) instead produces a state of prolonged 
waiting. In doing so, we contribute to a growing body of work (Slota et al., 2022; Miceli 
et al., 2020) investigating how “logistical media” (Peters, 2015) order the lives of people 
working on, and worked on by, emerging technologies. We draw from two 
“chronographies of power” (Sharma, 2014) to illustrate how, even across differences in 
context, data and AI produce experiences of time characterized by the persistence of 
waiting. In doing so, we demonstrate how the “long-standing but mistaken belief, 
hegemonic in Silicon Valley, that automation will deliver us more time” (Wajcman, 2019) 
belies the highly stratified temporal impacts of automation, datafication, and AI. 



 

 

 
This work draws from two ethnographic studies examining the production and 
implementation phases of data-driven technologies. Specifically, we focus on the 
contexts of AI training in China, and the deployment of homeless services technology in 
the U.S. In the rapidly evolving world of AI, AI trainers—the people who meticulously 
perform tasks like data annotation—are vital, yet overlooked and unacknowledged (Wu, 
2023; Zhang, 2023). The promises of AI bypass these contributors, relegating them to 
precarious conditions, a subordinate role to machines, and, importantly, endless 
waiting. On the other side of the Pacific Ocean, the homeless services system is 
increasingly sociotechnical, deploying algorithms and dashboards to help confront a 
massive and growing crisis. These tools promise to increase speed and efficiency; 
however, in practice, they leave many waiting for resources that may never arrive. 
Weaving together both narratives, this paper articulates how attending to waiting 
provides a new lens to understand how power operates across sociotechnical contexts, 
while contributing novel empirical work in key sites for understanding datafication and 
AI. 
 
Methods 
 
Ethnography reveals the intricacies, significance, and politics surrounding both the 
development and implementation of data-driven systems. Over 9 months of in-person 
ethnographic fieldwork, the second author studied AI production in China, a pivotal site 
in global AI production. To explore AI trainers' everyday professional experiences, they 
relied on multi-sited ethnographic fieldwork conducted in several underdeveloped 
regions in China, including working as an AI trainer intern at an AI data annotation 
center for 9 weeks. Their work role involved hands-on participation, observing different 
projects and tasks, attending team meetings, and dining with colleagues. The data for 
this paper is grounded in 16 formal interviews with AI trainers and over 150 pages of 
field notes and online archives. 
 
The second study, conducted by the first author, examined data-driven governance in 
the homeless services system in an East Coast U.S. city. Through 13 months of 
in-person ethnography and 60 formal interviews, the first author investigated the 
everyday functions and impacts of data-driven governance at an outreach center, while 
volunteering at the site preparing food, talking to guests, staff, and other volunteers, and 
observing data work. Following an ethnography of infrastructure approach (Star, 1999, 
382), the first author examined field notes and transcripts alongside policy and technical 
documents, and histories of the homeless services system. Both authors anonymized 
all data in this paper to protect interlocutors’ privacy and ensure their safety. 
 
Waiting derived from “hybrid management” and the promise of accuracy 
 
Study one contributes the concept of “hybrid management” as a hybrid form of labor 
control. Managing technical accuracy in AI training represents a case of hybrid forms of 
control, as workers are being controlled and managed by both traditional factory-like 
criteria and gamified platforms but enjoy the benefits of neither. The lack of security and 
benefits in traditional settings and the flexibility of the platform settings, leads to 
additional and excessive waiting among interlocutors. 



 

 

 
One of the field sites also implemented a facial recognition clock-in system and 
mandated workers to use the system to record their attendance. The record both 
captures attendance and qualifies workers for the 200 yuan ($27.80) monthly “perfect 
attendance award.” Such incentives are important, as Business Process Outsourcing 
companies often have a strong desire to retain workers to ensure consistency in task 
performance and data quality. Such clock-in systems ensure that workers are physically 
present in the data center for a prolonged period, even though their pay is often based 
on piecework. Workers tolerate spending idle time in order to get the monthly bonus and 
be assigned to new tasks. This idle, unpaid time in the center can amount to days or 
even weeks in the center not actively working. In addition to waiting for new tasks, 
workers also wait for feedback from more powerful actors such as quality inspectors, 
stand by for rules to be established, and wait to be assigned to new projects. To 
achieve a desirable technical accuracy, some clients demand workers to prove they 
have spent a certain amount of hours on given projects. Even if they finish early, 
workers still need to ensure that they log enough screen hours resulting in additional 
unpaid waiting. 
 
Waiting for housing 
 
People experiencing homelessness are constantly being made into data. The 
interlocutors the first author spoke with described being counted as they access 
services and during annual Point-in-Time counts; measured and sorted as they seek 
subsidized housing; and timed as they navigate the shelter system. At every turn, 
someone was handing them a form, or asking them questions, and many were unaware 
of the extent to which their lives were tracked and digitized. While such 
recordkeeping—the assemblage of “file selves” (Harre, 1984)—is characteristic of 
bureaucracy, data have proliferated and taken on new urgency in an era of automation 
and data dashboards (Tracey, Garcia, and Punzalan, 2023). This datafication promises 
to make homeless services more efficient. But for interlocutors, waiting remained a 
persistent feature of daily life. 
 
People visiting the outreach center described waiting for a shelter bed; for a shuttle to 
take them from the shelter to an intake center; for their caseworkers to process forms; 
for the soup kitchen to serve the coffee; and perhaps above all else, for the black box of 
the housing system to work. Coffee might be ready in a few minutes, while the housing 
system could take months or even years. While someone curious about the wait time for 
coffee could get an answer easily, the length and structure of waiting processes for 
housing and shelter were typically a mystery. This information asymmetry charged the 
experience of waiting, as often interlocutors had no idea how long they would be waiting 
for, making it difficult for them to make plans or long-term decisions. 
 
Conclusion 
 
Waiting, in both contexts, carries multiple meanings. In one sense, interlocutors waited 
for some immediate outcome, such as a shelter bed or the assignment of a new task. 
But in another, they wait in the hope of something more abstract and substantial—a 
slightly better future. Our findings suggest that the technologies which mediated the 



 

 

experience of waiting in the first, more immediate sense, also impacted how 
interlocutors conceptualized the future. Our work also illuminates that despite the 
promise of data-driven technologies, pervasive waiting serves as evidence of an 
enduring residue—an unequal power structure. While data and AI create immediacy for 
some, for others, they produce a sense of being outside of time, waiting indefinitely for a 
better world that may never materialize. 
 
References 
 
Charmaz, K. (2006). Constructing Grounded Theory: A Practical Guide through 

Qualitative Analysis. SAGE. 
 
Gleick, J. (1999) Faster: the Acceleration of Just About Everything. Pantheon Books. 
 
Harre, R. (1984). Personal being: a Theory of Individual Psychology. Harvard University 

Press. 
 
Miceli, M., Schuessler, M., & Yang, T. (2020). Between Subjectivity and Imposition: 

Power Dynamics in Data Annotation for Computer Vision. Proceedings of the 
ACM on Human-Computer Interaction, 4(CSCW2), 115:1-115:25. 
https://doi.org/10.1145/3415186 

 
Peters, J. D. (2015). The Marvelous Clouds: Toward a Philosophy of Elemental Media. 

University of Chicago Press. 
 
Sharma, S. (2014). In the meantime: Temporality and cultural politics. Duke Univ.  

Press. 
 
Slota, S. C., Fleischmann, K. R., & Greenberg, S. R. (2022). A Brief Typology of Time: 

Temporal Structuring and Dissonance in Service Provision for People 
Experiencing Homelessness. In M. Smits (Ed.), Information for a Better World: 
Shaping the Global Future (pp. 211–224). Springer International Publishing. 
https://doi.org/10.1007/978-3-030-96957-8_19 

 
Star, S. L. (1999). The Ethnography of Infrastructure. American Behavioral Scientist, 

43(3), 377–391. https://doi.org/10.1177/00027649921955326 
 
Tracey, P., Garcia, P., & Punzalan, R. (2023). Recordkeeping, logistics, and translation: 

A study of homeless services systems as infrastructure. Archival Science. 
https://doi.org/10.1007/s10502-023-09410-0 

 
Vora, K. (2015). Life support: Biocapital and the new history of outsourced labor. 

University of Minnesota Press. 
 
Wajcman, J. (2019). How Silicon Valley sets time. New Media & Society, 21(6), 

1272–1289. https://doi.org/10.1177/1461444818820073 
 
Wu, D. (2023). Good for tech: Disability expertise and labor in China’s artificial 



 

 

intelligence sector. First Monday. https://doi.org/10.5210/fm.v28i1.12887 
 
Zhang, B. Z. (2023). The Labor of Training Artificial Intelligence: Data Infrastructure, 

Mobility, and Marginality. Companion Publication of the 2023 Conference on 
Computer Supported Cooperative Work and Social Computing, 459–461. 
https://doi.org/10.1145/3584931.3608926