Human labour behind AI: Rethinking artificial intelligence and work in China
Agencies, Funds & Programmes
Date: 10 September 2026
Language: English
Transcript: https://transcripts.un.org/en/asset/k13/k13dpjbthh
Transcripts available through this tool are created by using automatic speech recognition and are not official records nor official documents of the United Nations. Official records and official documents are available on the Official Document System of the United Nations.

---

Moderator [0:10]:

Turn this on. Okay, perfect. So, ladies and gentlemen, good morning. Welcome back to the second keynote session of this conference. My name is Maria Maksi. I'm a senior advisor at the Geneva Graduate Institute's task platform. and I have the pleasure of moderating this keynote session today. Well, before we begin, I would like us to take a moment to recognize the scale of what has been achieved with this conference. and especially with the International Network on Digital Labor, bringing an international community of researchers and activists together, sustaining it across nine years, nine editions of this conference, and creating a space for dialogue, collaboration, and cross-fertilization of research is an extraordinary achievement. And behind it are years of strong commitment, dedication, enormous amount of work, and intellectual leadership. So I would like to ask you to join me in giving the whole INL team A very warm round of applause. And special thanks. Special thanks to Uma, Paola, Antonio, and all of you contributing your research and ideas. So now it's my great pleasure to introduce Professor Julie Yujie Chen. Associate Professor of the University of Toronto's Institute of Communication, Culture, Information and Technology in the Faculty of Information. Julie's work has reshaped how we think about the relationship between labor, digital technologies, capitalism and globalization. She is the founding editor of Platforms and Society, co-editor of the SAGE Handbook of Digital Labor, and a contributor to Notes Toward a Digital Workers Inquiry. She is currently writing a book on AI data workers in China. And her keynote today starts from a Chinese expression, which is, Rencong, meaning both artificial and human labor. And as the Chinese saying goes, no human labor, no intelligence. So we very much look forward to hearing from you, Julie, how this perspective and especially the China as a case study can help us deepen our understanding of AI supply chains. The floor is yours.


Associate Professor [3:51]:

Thank you. Thank you. Can you all hear me OK in the back?

OK. Good morning. Thanks so much to the organizing committee and especially Uma for having me today. So can we get started with my slides? So while we're waiting, if you are studying platform labor, I highly encourage you to submit to our journal, which is called Platforms and Society. Okay, thank you. That's why AI cannot replace human. So there is a widely spread saying goes like this in Chinese, well known in the AI industry in China. no human labor, no intelligence. The saying basically involves a very witty wordplay of the Chinese character, artificial, which means both human labor and human made. In China, there are two different variations, both means no human labor, no intelligence. The origin of this phrase is untraceable, but it has been invoked across different contexts. In my interviews across China, in the introductions of the companies where I did my ethnography, in the news, textbooks, trade publications, tutorials for AI trainer, and at high profile industry events, DeepSeek's CEO, Liang Wenfeng, was reported to have done data labeling himself for the sake of data quality. More than a more than a aphorism, this Chinese expression encapsulates indigenous insights into the essential role of human labor in AI production. It becomes a sense making device to serve diverse rationales, define the value of data work, and cultivate ideal working subjects in China. Using this phrase as my point of departure, I ask, instead of invisible labor, I ask what it means to be a data worker when the job becomes an ordinary, salient, or even state-promoted occupation. Given the heterogeneity of AI workforce, I'm particularly interested in the driving forces behind the diverse work arrangements. For me, why and how individual workers end up in specific types of annotation jobs, like office-based versus home-based, is just as important as the labor process that deprive them of their autonomy. So my central questions are, how is data work produced in China? What makes a flexible supply chain of data labor? My central government my my central argument for today is that digital configurations of data work has set in motion a hierarchical labor market with standard employment, formalized work at the top and multilayered system of flexible data worker below like informal office workers and home-based workers. The seemingly contradictory trends of labor formalization and flexibilization suggests a conjunctural convergence of, uh, conjunctural convergence in China between state led institutional efforts to develop the AI industry and an ongoing labor precaritization process jointly driven by the state, market and the workers themselves. This labor precaritization process has expanded beyond the historically marginalized groups of workers, such as migrant workers and the stay-at-home moms, to incorporate university graduates and white collar workers who are under growing pressure to find an acceptable job or a secondary source of income. So to make my argument, I will first introduce my conceptual framework very briefly, drawing on my analysis of three work configurations, I'll make three points. First, about the multifaceted roles of the state, second, about the impact of existing in- institutions and technologies, uh, technological conditions, and third, about the varying meanings of data work, uh, ascribed to data work. I will conclude with some reflections on rethinking AI and labor from and in China. I developed the concept of digital work configuration to capture how different social organizations and technologies are at play in mobilizing data workers. The concept recognizes that digital technologies function as infrastructures and structural power in the mobility and regulation of data labor. The notion of digital work configurations extends sociologist Sarah Schreider's insightful point about the links between how workers find their jobs and how their employment relations are regulated to the domain of AI data work. It allows me to examine technologies as embedded forces rather than a uniform deterministic force indifferent to social differences. As I will show, this concept provides analytical position to flesh out varying roles of institutions, social actors, and technologies in shaping the hierarchical data labor supply. My first point, with a pro-tech development strategy in China, the state plays a strong role in institutionalizing and formalizing data work. However, that's a big however. Other enduring social and economic institutions contribute to producing informal and precarious workers anyway. So there are estimated over 9 million data workers in China. Echoing global trends, China's AI data industry is shaped by just-in-time production and the rising demands for data accuracy and quality. The domestic, uh, specialist crowd work as well as the data suppliers have been mushrooming. Here is just a list, you don't have to recognize the companies. Um, so basically this suggests there, uh, this just an absence of monopolistic or dominant, uh, annotation platforms or data firms within the industry. What further distinguishes China's AI data industry is the central government's shift towards a data, a tech-driven development agenda since mid-2000s. Under what sociologist, the Leijawen, described as techno-state capitalism, alliances between the rising tech capital and the government at different levels have actively constructed and differentiated high-tech or high-end sectors, firms, workers from the so-called low-end. the latter are deemed as obsolete or even unwanted. AI data annotation is now positioned as part of, or at least adjacent to, the high-end industry. The formalized data work stems from this institutional orientation. So I define formalized data work as employment relationship characterized by a written labor contract and a comprehensive social insurance coverage. In China, labor contracts is the most important legal document determining, uh, formal employment relations, it enti- it entitles workers to the employee scheme social insurance coverage, which has broader coverage and, and much better benefits than the resident scheme. Formalized data work mainly draws from the local urban youth labor market. Workers have fixed schedules and receive monthly pay based on completed projects. They are generally young, 20 to 25 years old, and hold associate or bachelor's degrees. The gender ratio is roughly equal. The formalization of data work represents the state's institutional efforts to develop the AI industry with a strategic emphasis on data and data production. So these institutional initiatives includes the recognition of AI trainer as occupation in 2020, the publication of the national occupational skill standards in 2021, establishment of training programs across China, publications of textbooks, and most recently, designation of state-backed data annotation centers. And I can go on, but I must emphasize two things. First, the data annotation sector in China has already flourished before the central government made the institutional move. Second, as scholars like Zhou Xueguang and Yuan Yuan An have argued, geographical disparities in China and the tensions between centralized policy mobilization and decentralized local and often flexible implementations often produce significant variations in practices and results. Therefore, the data labor supply chain is more complicated than a state-led, uh, account would suggest. That being said, I argue these initiatives have successfully constructed data work as a viable occupation with technical or even skilled, uh, connotation by its association with AI. Vincent Mosko, uh, would describe this process as a construction of digital sublime, a s- a symbolic value creation technique. More importantly, I want to shift attention from AI industrial policies to the labor market reality facing university graduates and role of China's employment policies. The predominance of young, well-educated workers in the formalized data work is the result of a deliberate recruitment strategy. As an HR at one of the company I did my field work said that they hire only students and fresh graduates. This approach has established a university to enterprise pipeline, supplying qualified job candidates, and it is also developed against the backdrop of a tough, tough used labor market. And so these are the two statistics. Since late 1990s, China has undergone a massive massification of higher education. The university enrollment rose from 2.1 million in 1990 to nearly 11 million in 2022. This year, the number reached a record 12.7 million. So accompanying this rapid expansion of tutorial Tertiary education is rising youth unemployment and underemployment rate, you can see in the, uh, chart A. Uh, according to this recent study by, uh, Li Yunhao and Lu Yao, the unemployment rate of young people with associate degree and above has been higher than the high school graduates, uh, persistently since 2000. So in response to this persistent youth employment challenges, the central government has introduced a wide range of policies and incentives since 2008. Most relevant to today's topic is the expansion of the graduate traineeship program in 2009 following the global financial crisis. Different from student internship, traineeships are only open for unemployed university graduates aged 16 to 24 within two years of graduation. The program encourages qualified enterprises to create positions and expand their workforces through subsidies and other policy incentives. There is no national standard, so eligibility and stability levels vary across provinces and cities, for example. In one of my field site, qualified companies receive a subsidy of 1,200 Chinese yuan per month for each intern or trainee for up to six months, as long as they met requirements such as job creation targets and the minimum conversion rate from interns and trainees to full-time employees. Just to give you a sense, most workers at that company earn around 4,000 Chinese yuan per month. To provide points of comparison, in Chongqing, a major city in southwest China, qualified companies receive 1,000 yuan per trainee per month. And in other provinces, Shaanxi offered 3,000 yuan per month per intern, while Hunan province offered half of that. In both provinces, however, eligibility was limited to small and micro enterprises. In short, my point is, In addition to the student intern system, the graduate trainees, uh, programs has become an integral component of China's policy responses to ongoing youth employment challenges. In their study of China's manufacturing sector, Jenny Chen and colleagues discovered that the vocational schools have been mobilized by factories and local government to supply cheap student labor. They argue the student internship system has been turned into a precarious labor regime serving industrial production. Under the intern labor regime, students often perform unpay, unpaid, and underpaid work, and are subject to triple management from managers, uh, teachers, as well as the local officials. There are notable differences between studen, students interns in the factory and data workers in the university enterprise pipeline. In my study, neither interns nor trainees reported being managed other than their, uh, by other than their employ- employers, and they were generally paid the same piece rate as regular workers. Nevertheless, trainees share a vital lack of legal employment status with students interns. During internships, students remain enrolled in school and they are not eligible to sign a labor contract, so they are not legally recognized as employee. Similarly, trainees do not enter formal employment relations. And instead, they sign a trainee agreement, a traineeship agreement under which employers are exempt from contributing to their social insurance. This puts them in a legally vulnerable position while reducing the cost for the employers. Lei Yaven pointed out that under the techno-state capitalism, high tech, high technology has become a core indicator in social and economic evaluation systems. As a result, local government's performance metrics, which includes the job creation, have been geared toward, uh, increasing the proportion of the high tech. In the case of data work, state institutional measures have produced and assign a high tech property to both the occupation and the sector. They actually helped align, formalize the data work with the local techno-developmental agenda. However, I argue, oh, sorry, John. I argue the graduate traineeship system represent a variation and an expansion of the intern labor regime that renders young job seekers legally unrecognized as employee for a period of time. In doing so, it serves out, it carves out a new form of state capital coalition on the techno-developmentalism that manifests precarious labor conditions even as data work is simultaneously being institutionalized. The enduring influence of social and economic institutions is also evident in other data work configurations such as remediated informal work. I use the term to describe mainly informal office-based work without labor contract and the limited or no social insurance coverage. This work is remediated because workers are often hired by subcontractors or unregistered family run workshops that secure project from data contractors or from crowd work platforms. To stay flexible, some subcontractors maintain a hybrid workforce, combining in office and remote workers. This informal data labor supply stems from the entrenched system of subcontracting in China. However, the informal labor market was reported to count for 47% of the data production in 2021. The labor market is local, urban, and semi-urban, and under-regulated. Workers vary significantly in age and educational background. The youngest informal office worker I met just turned 18, and the oldest were in their 40s. Educational backgrounds range from middle school graduates to college degree holders. The workforce was dominated by women, especially those with care responsibilities. The locations of these workers are particularly noteworthy. Echoing David Harvey's idea of capital's spatial fix, sociologists Tong Wu Yu and colleagues note that in China, data work is outsourced to inland regions with lower labor costs. These destinations are mostly third or fourth tier cities and counties, and in China, by the way, county is the lowest urban administrative unit just above the rural level. I would like to push this argument even further by connecting the relocation of data work to China's longstanding migrant labor regime. In 2020, there are more than 1800 counties and county level cities in China. About 90% of them experienced a net population outflow, which means these places are sources of domestic migrant labor. Two of my field sites are among them. The cornerstone of China's migrant labor region is the household registration system known as Hukou. Initially designed to restrict the rural to urban mobility, the Hukou system has become a powerful institution that has created socioeconomic and political hierarchies between the urban population and the migrant workers. The latter faces structural discriminations in access to urban employment opportunities, equal pays, labor protections, and public services such as education. Scholars like Dorothy, Solinger, and Panai argue that the migrant labor regime sustains a steady supply of young workers in the prosperous industrialized cities in China through spatial divisions and social constructions of differences. I wanted to argue the flip side of the forever young workforce in the destination cities is the hollowed out hometowns. The young workforce exodus and the stagnant local economies with limited job opportunities reinforce each other, creating highly mobile labor markets. One manager in China complained to me that in their worst months, 40 out of 50 data workers left. He said, "It's hard to keep people." Those who are more likely to stay are women with caregiving responsibilities, as one data worker named Ray said, After all, in our hometown, most young people have left for big cities on the run to make a living. Those who remain are mostly like me, mothers staying behind for their children. Rui actually had a degree in medicine. She has spent many years in Guangzhou as a migrant worker. She had to return to her hometown because only one of two of her children was eligible to attend public schools there. The family's decision to have her instead of her husband return obviously reflected the patriarchal expectation of women to be the primary caregiver. In this regard, remediated informal data work is intertwined with the migrant labor regime by capitalizing on the relative immobility of the returnee migrants, or most of them are women. Over the past decade, Digital platforms have transformed urban services sectors like for delivery, domestic work and logistics, absorbing large numbers of migrant labor leaving manufacturing. While migrant workers job in destination cities are increasingly platformized, I wanted to point out that data work landed in their places of origin. These trends, in my opinion, are two sides of the same coin, which I call a digital fix to the migrant labor regime. It makes migrant labor continuously indispensable in the country's digital development project. The digital fix is operationalized through not merely the uneven regional development, but also social differences and gendered divisions of labor in the social reproduction. Another noteworthy characteristic of the informal data labor market is the role of social media in recruiting and regulating workers, and this leads to my second point. Social media has become informal infrastructures for data labor mobility, contributing a fragmented structure of the data labor supply chain. Chinese workers access data jobs and the projects through a wide range of channels, including professional recruitment platforms, crowd work platforms, classified ads website, social media platforms like WeChat, Douyin, which is the domestic version of TikTok, and Rednote, which you might have heard of, and QQ. Formal data workers are often recruited through professional platforms like Boss Zhipin. In contrast, social media is central in configuring data work in the informal market. So using the third work configuration, which I call networked individual hustling, as an example. So this work configuration mainly includes home-based workers with flexible schedules who are mostly not protected by labor contract laws or covered by employment-based social insurances. Many work on multiple platforms and for several clients through subcontractors. The labor market is geographically dispersed, translocal, and unregulated, and sometimes it's even illegal. The workforces includes both part-time and full-time workers from diverse demographic and educational backgrounds. Some workers I interviewed are white collar professionals and civil servants, whose Most primary and stable jobs provided social insurance benefits for them so that they do data work as a secondary income source to maintain their middle class lifestyles. This demographic diversity and precarious work conditions conquer with the existing literature. However, I used the term networked individual hustling to highlight two additional features of Chinese digital home workers. First, the use of social media platforms has exceeded the function of mutual support groups as documented in the literature. Take Tieren as an example. She began as a quality inspector for a team of 29 remote data workers for a speech AI company. By 2022, she has established her own online group, over 20 people, including her relatives like Gaozi, who is her cousin, Gaozi's cousin, and other newly made online contact. Her way of expanding connections beyond the kinship and local ties is emblematic of what sociologists would call networked individualism. Networked individualism was also seen in Xiumi from South China. She joined the shoes network of more than 20 social media groups, but later developed her own team of eight. So for Chinese home-based data workers, their access to data work is far from being monopolized by BPO data suppliers or crowd work platforms. Instead, these extended online networks are their typical entry point, which formed one of the most diverse, individualistic, and fragmented technological infrastructures for home-based data workers. For the all, social media has been used systematically for recruitment and work management in the informal labor markets. Jobs are published on short video platforms. If we had time, I can play this, the short video, during Q&A. And jobs posted on regular social media platforms, here is an example on QQ. Job posted on those social media tend to use informal language, emphasize time commitment, number of workers, and pay cycles. So if you have ever visited a physical day labor market in any of the China's manufacturing hubs, this social media, uh, mediated virtual market would appear strikingly familiar to you. Much like its offline counterpart, workers recruited through social media would join project-specific online groups where they receive login credentials to access data, project requirements, and more often than not, they face human managers and get paid through those social media groups. And their group would dissolve once their project is finished. In this regard, I argue these technological project, um, practices have transformed the social media from spaces of mutual support into a self-organized secondary virtual labor market, which integrates, uh, which basically spread as spending from, um, recruitment to payment. Social media enable what, uh, Sherry Arnold called, uh, re-outsourcing, whereby worker entrepreneurs, uh, subcontract parts of the, uh, project to their relatives and friends. However, because open calls on Chinese social media reach a broader, unspecified pool of online workers than the kinship networks seen in re-outsourcing, I call this practice re-crowdsourcing. The second distinction I want you to capture in networked individual hustling concerns is technological and the cultural roots. All the social media mentioned so far have over 90% penetration in China. So basically, they serve as an infrastructure for social life in China. More importantly, their algorithms are designed to foster local connections and community development. These pro-local algorithmic recommendations were found particularly effective in the informal labor market for recruitment. Social media sinking into the infrastructure of informal labor market for data work in China demonstrates their deep enchantment in the daily lives of the working of Chinese working class. As Jack Chew argued, working class technological culture is characterized by informal appropriation of digital technologies for affordability, grassroots innovations, and informal entrepreneurship. QQ in particular, carries strong cultural resonations and social associations with China's working class and youth. Its widespread adoption and anonymity have, uh, have facilitated the informal labor market for earlier forms of digital work, such as ghost writing, click farming, remote e-commerce, uh, customers, customer services. Many home-based workers are veteran online gig workers for them. Data annotation is just another avenue of earning money online, one that is unstable, not categorically different from other or earlier forms of digital work that they are engaged in. Gaozi, for instance, I mentioned earlier, juggled between multiple offline and online temporal jobs. In 2024, she spent more time working as a part-time chef in the local restaurant than as a data worker online. For Gao and other veteran home workers, precarity is the norm and a state of living instead of a condition introduced by data work. Their hustling as a way of life reveals a profound level of both insecurity and agency, which we have been witnessed among historically marginalized workers in China, but also not just in China. To paraphrase anthropologist Sean Bell's argument, frenetic entrepreneurial energies coexist with political resignation in the hustling lifestyle. These contradictions makes precarious data workers simultaneously hopeful, resourceful, and indifferent to collective actions. As discussed earlier, the state seeks to assign a sense of high techness or like skilled to data work through institutionalization, certification, and formalization. Home workers, however, view data annotation as merely another income generation opportunity, a perception that couldn't be further from the state's vision. Citing market volatility, Tianran put it bluntly, working on it full time is not worth it. It's a waste of time. So her judgment brings us to my final point. the meanings of data work are shaped by socioeconomic circumstances, organizational positions, and cultures of ren gong. The cultural constructions of data work is rife with ambiguity, ambivalence, and conflicts. Take remediated informal work as an example. As I have argued, a female-dominated workforce in hinterland is produced by a digital fix to the migrant labor regime. Yet, disproportionate number of women has led to a gendered discourse that rendered the job as not a man's job. As a manager in South China put it, rationalized the gender imbalance in this way, there are more women than men because women may not have high salary expectations. Sounds familiar? They generally face less financial pressure than men who are expected to buy a house, a car, and support a family. Beyond imposing differential wage expectations on women and men, gender discourses are reinforced in frequent references to stereotypically feminine qualities. Many believe that women are more suitable for the job because they could sit long. As Lisa Nakamura criticized, the claim that women's nimble fingers made them physically natural for electronic assembly work was used to recruit women of color at lower wages. In the same vein, gender discourses portraying women as sedentary, meticulous, quiet, easy to manage, and subject to less financial pressure were utilized to justify wage suppression and insufficient social insurance coverage. Such narratives were, uh, especially popular in the county level where informal economy operates, uh, and where women are made to be less mobile than men to begin with. On the other **** **** in economically stagnant counties present opportunity for women caregivers to build social connections and against self-confidence. Several informal workers described how being a primary caregiver at home had left them feeling socially disconnected from the world and how data work in their hometown meant more than earning income while caring for their family. As Rui noted, doing the job actually exposed her to all kinds of data and tasks with specific content, which made her believe that there is nothing she couldn't learn. Similar ambivalence emerged among the university graduates in the formalized work. There is a combination of appreciation, aspiration, and a sense of inferiority caused by the employment challenges. They generally enjoyed the egalitarian workplaces, which can be attributed to its relatively heterogeneous workforce. But some of them are acutely aware of the declining value of their college degrees on the job market, such as Li Jie, the first one on the screen. Her coworker, Wang Yi, believed that their college training, even in sought after majors such as computer science, was superficial and outdated. Feeling left behind by several waves of industry change, Wang Yi eventually gave up hope of pursuing professional positions in tech industry. Nevertheless, former workers generally shared a strong faith in the future of the AI industry and recognized their contributions to make that happen. They refer themselves as unsung heroes and AI teachers. Yet few considered data work requiring technical skills. And regarded the skill certificate, which I mentioned earlier, they regarded skill certificate as remotely relevant to their everyday work. In this sense, former workers' perceptions of data annotation were also shaped by their own socioeconomic circumstances and organization positions, which also diverges from the state's discursive construction. So taken together, these contested meanings of data work reveal dynamic cultural constructions made by workers who are not passive recipients of the state-promoted discourses or propaganda. As apologist Anna Tsing made a convincing point about how diversity and the social differences are productive in supply chain capitalism. She wrote, I quote, "Diversity forms a necessary part of supply chain capitalism's scale making practices. The linking of diverse firms makes the supply chains big. Diversity conditions the responses of both capital and the labor to the problems of cutting labor costs and disciplining the workforce," end quote. My analysis of different types of worker configurations shows how diversity is produced and made to be continuously productive in China's AI data industry. So Chinese normative recognition of no human labor, no intelligence leads to almost past dependent, if you will, playbook now seen in the robotic industry, which mobilizes both university students and housewives to produce training data for robots. Again, labor diversification goes hand in hand with social differentiation. To conclude, the rise of formalized data work in China's tiered labor supply chain suggests that the state's efforts and ambition to standardize the sector, which serves the tech-driven development agenda. But the paradox is hard to ignore. To a certain degree, the expansion of the intern labor regime has also been institutionalized as a conditioned pathway to formalize the data work. This conditional, albeit temporary stage is created by state policies aimed at promoting job creation and skill development, and it is further exploited by the market actors and the local governments. The hierarchical data labor supply chain in my study reaffirms what Qin-Huang Li's argument that the state has continued to produce precarious workers and defining precarity through labor policies, social institutions, laws, and the pro-market developmental strategies. The proliferation of informal and precarious labor, as we see in the digital fix to the migrant labor regime, exposes the contradictions inherent in the state's embrace of a data-first approach to AI. These contradictions are urge us to place the project of digital labor pre-characterization as a broader historical process instead of brushing it aside in the name of development. As a conceptual tool, digital work configuration lends an analytical position for examining different roles of the state, the market, digital technologies, the existing structure of social inequality across diverse work arrangement. Such a position also helps fresh out the dynamic and the specific expressions of workers in intersectional challenges, precarities, agency, and desires as they encounter their own technological condition as well as the top-down technological ambitions. Mapping data workers' technological and social conditions calls into question any assumptions about the socio-technological infrastructure of AI data labor supply, both in China and globally. Its defining characteristics is that it is a patchwork of infrastructures, which extends far beyond the dominant annotation platforms, tech companies, and intermediary firms to include diverse technological systems and other social organizations. This patchwork takes shape through labor diversification, layered forms of mediation that combines formal institutions and informal infrastructures, as workers and intermediaries repurposes existing technologies to shape their own work conditions and access otherwise denied data jobs. This patchwork character is particularly relevant to countries where informal economies are undergoing digital transformation. Finally, from annotation platforms to social media, digital systems started to embody the critical sides of inequalities and the divergence between formal and informal workers, between tech-enabled discrimination and the new possibilities. However, new economic possibilities, such as formalized data work, shouldn't hide the reality of expansion of precarity, which now includes university graduates and white collar workers. My study of AI data labor in China does invite all of us to question the leapfrogging province of digital developmental agenda, which are increasingly adopted across developing countries. Which is most striking is the extent to which digital developmentalism depends on existing social structures and economic institutions to produce labor diversity, precarity, and new forms of inequality. Consequently, the state promotion of digital development and the practices by workers to shape their own infrastructure and cultural meanings of data work only points to the ongoing complexity, contradiction, and the contingency in the global AI supply chain. Thank you very much. And here is an article which I share. I go into more details for the empirical findings just in case you're interested.


Moderator [42:32]:

Fascinating. Thank you so much for this rich and multidimensional analysis. We have 10 minutes left. Questions? Yes, please. Just for practical reasons. Can you please also introduce yourself?


Speaker 4 [42:46]:

Yeah, no, but that's not really, well, Nicolas Mestromontat, but there are many sessions starting at noon, so shall we say that they start at 15 past 12, and then we can stay with you, or?


Republic of Korea [43:08]:

Hello. I really enjoyed your presentation. It was fascinating. Thank you very much. My name is Sophia Lee from Chungang University in South Korea. My question is, I really enjoyed the configurations, learned a lot, and I wonder whether there is, I mean, among data workers, there are different types of data work. So it must be the skill level required for data work must be different, if I'm right. So I wonder if there's a relationship between this formalized, labor informal, so these types of different groups, and then the skill level of data work, and then whether for the university students, especially, whether this experience as an intern actually contributes to skill formation. Thank you.


Moderator [44:01]:

I think the first person. You can take some questions. Other questions at the back, very back.


Speaker 7 [44:12]:

Hello, I'm Marco Marone from the University of Salento and thank you very much for this very inspiring keynote. I found particularly interesting the parallel with another, let's call it this way, another nodes of the high value chains, which is that of Foxconn and the digital manufacturing. But I was wondering if you encounter also some similarities of another key concept of that literature, which is that of the dormitory labor regime, because some cases in Europe or something similar somehow came out. And I was wondering also if you met, and if you don't, why? Because it seems very effective also to solve one of the main problem, which is that to keep data workers. Yeah.


Moderator [45:04]:

Excellent. Yes, please.


Speaker 9 [45:06]:

Yeah, thank you so much for your presentation. My question is, you mentioned it was non-elite college graduates that entered these. So what do elite college graduates do? And do you have a sense of if there's a devaluing of university education with that growth of these sort of jobs? Like, is it making it, going to university, less attractive for high school students, for example? Thanks.


Romania [45:33]:

Thank you. I am from Arba Juhia University in Romania. I have one question and a curiosity. Let's start with curiosity. I'm curious to know what happened before 2020. You said that they started recognizing data workers in 2020, if I understand correctly. My question, it is, since 2020, China is-- unemployment is growing. which means the value of university or college degree is decreasing. What do you think? What do you see as benefit or hindrance due to this recognition of underemployment?


Speaker 11 [46:15]:

OK, yeah.


Moderator [46:16]:

More questions.


Speaker 13 [46:20]:

Thank you for your presentation. I want to ask one question because-- yeah. because you get those data worker from different province. I'm just wondering is there any difference from this worker from different government province and also except for the gender age dimensions and also education that you mentioned in your study. Some other factors like ethnic group might be considered in your research. Thank you. Yeah.


Moderator [46:53]:

More and more. Yes, please.


Speaker 15 [46:57]:

Thank you for your presentation. My question was regarding the institutionalizing of the Zeta work. Like you said that before the state institutionalized it, it was already flourishing in China. What changed after the state institutionalized it? Yeah, thank you.


Associate Professor [47:14]:

Okay, okay. Okay, okay. Yeah, I will go very quick. So first, I think the last question is kind of like similar to your question about what happened around 2020. So it's basically the sector has already developed its own way. The formalized work, which is the employment relations, have already exist in some of the companies. But I would say the state certification of the job or recognize AI trainer as occupation at a central government level signals a strong political signal and also allocate certain money so that they can support the creation of this job in certain provinces. However, how that got implemented to the fifth one about the differences in provinces, how the central government's incentives and policy support got implemented depends very much on how the local dynamics. Maybe the local, they have like formal enterprises who are willing to provide formalized work, which is one of my side, but there are also other firms that their firm are formal firms, but they hire workers as informal workers. They don't protect their workers. So there are a lot of local disparities like across the board. And for the dormitory, labor dormitory regime, I didn't observe that in the relocated data work because the job already relocated to their hometown. So they live in their homes, they are not living in the dormitories. The existence of the labor, dominated labor, labor regime adopted by Foxconn is because those workers are migrant workers leaving their hometown. They do the work in the cities. And for Sophia's question, which is about skill levels and whether university students contribute to their skill formation, so I think The concept of skills is getting very messy when it comes to data work because the nature of data work is getting fragmented and highly decontextualized. And it is designed in such a way. So even for very complicated jobs, such as for autonomous driving, or I have interviewed those who engaged in annotation for medical AI. Within the company, they actually went through trainings for months. However, they still consider that's actually not really related to their skill development because it's very specific about the project. If their employer, the company, lost the clients, there is no transferable skills whatsoever. And the concept of skill also gets messy. And my personal take on this, I'm not sure skill is the right word or frame to examine the phenomenon. Also because, for instance, I have encountered People speaking dialect like Cantonese, their language skills, and the university students who are hired to do annotation jobs because they could speak like a... like Laos and Vietnamese, those foreign languages. It's not necessarily a skill, it's their cultural capital in general. So it's complicated. And then for the skill formation, university graduates, we can unpack that into two different types. First one is actually those who are hired, well, they themselves are university students attending elite universities. So they tend to be hired, like for instance, by the AILAC. from the same university, that actually would add on their resume. So they basically use this internship, although all the internships are highly legally vulnerable, they could just add an item on their resume so that they can go on to pursue a very good career. But then for those from non-elite universities, is not really helping them because it is very, very competitive on the labor market in China. If you wanted to pursue like the most sought after like tech companies, it's mainly basically the some, some, some basically implement open rule, but some are implemented the rule unspoken, which is they only hire those graduated from the very top elite universities like project 211 or project 985, about like 100 universities in China. So, yeah, again, it really depends on the socioeconomic conditions of the workers. And then I think there is a question about non-elite universities. Yes, that's why, because non-elite university, so the youth labor employment pressures are more accurately filled by non-elite universities graduates. Although the intensification of competition in the labor market also affects the graduate from elite universities, that's because they are all competing for the top tech companies. But those tech companies are not even look at the resumes from the non-elite universities.


Moderator [52:28]:

Okay, excellent. And with that, we can conclude now. Thank you so much.


Associate Professor [52:32]:

Thank you so much, everyone.


Moderator [52:33]:

Thank you all for your participation. Thank you. Thank you. It's incredible.