Chinese artificial intelligence is often described through models, chips, surveillance, patents or state capacity. That image is powerful, but too narrow. When seen through social and cultural anthropology, labour sociology and science and technology studies, another scene appears: delivery workers negotiating with algorithms, factories replacing bodies with machines, inland territories becoming data-labelling bases, disabled workers entering digital value chains, and local governments translating AI into development policy.

The question changes. It is no longer only what kind of AI China produces, but what kind of social life is needed to produce it.

Three researchers help shift the frame. Sun Ping (孙萍), based at the Chinese Academy of Social Sciences and the University of the Chinese Academy of Social Sciences, studies platform labour through food delivery work. Huang Yu (黄瑜), based at Minzu University of China, brings the discussion into industrial automation and the policy language of “machine replacement” (机器换人). Tongyu Wu (吴桐雨), based at Zhejiang University, opens the human infrastructure of data annotation (数据标注), the work without which intelligent systems do not learn to recognise the world.

These researchers are not merely windows through which an outside observer can understand China. They take part in academic, labour and public debates within China itself. The most interesting gesture here is not to use China as a “case” for theories produced elsewhere, but to see how Chinese fieldwork corrects and displaces inherited frameworks.

The algorithm in the street

Sun Ping (孙萍) has spent years studying platform labour in China. Her recent book, 《过渡劳动：平台经济下的外卖骑手》, carries the official parallel English title Transitional labour: food-delivery workers in the platform economy of China. The title already contains a central intuition. Delivery work is not simply another precarious job, nor a brief step towards something else. It is a suspended, mobile and unstable form of labour, organised by platforms that promise flexibility while adjusting time, routes, incentives and penalties.

In her work on food delivery riders, Sun does not treat the algorithm as an abstract black box. She observes it in the street, in the body that accelerates, in the vibrating phone, in the map that reorganises the city, in the calculation of minutes, penalties and rewards. The platform does not simply replace a human boss. It redistributes command through instructions, scores and expectations that appear neutral because they arrive through an interface.

This is the first anthropological turn of the radar. AI does not begin when a machine “thinks”. It begins earlier, when everyday life becomes legible to a system of calculation. In delivery work, the city becomes an optimisable surface; time becomes compressed; the worker becomes a moving piece within an algorithmic choreography.

But Sun Ping also shows that delivery workers are not passive victims. They learn shortcuts, negotiate with the app, share knowledge and develop forms of counter-algorithmic practice (逆算法). The system calculates, but workers calculate too. In that friction, a more interesting anthropology of AI appears: an AI that commands, but still needs to be interpreted, bypassed and inhabited.

The robot in the factory

Huang Yu (黄瑜) moves the scene from the street to the factory. Her work on automation in South China allows us to read the policy of “machine replacement” (机器换人) not as a simple story of technological modernisation, but as a social reorganisation of labour.

Robotisation is often presented as inevitable progress. In industrial development discourse, replacing human labour with machines can be described as efficiency, upgrading or the leap towards a new economic model. But Huang Yu asks a more uncomfortable question: who gains power through that change, who loses it, and what happens to the migrant workers who sustained Chinese manufacturing growth for decades?

The promise of a “robot dividend” replaces the older “labour dividend”. For years, industrial competitiveness relied on large pools of low-cost migrant labour. When the discourse shifts and the robot appears as the new source of productivity, the politics of labour does not disappear. It changes form. The machine does not enter an empty space. It enters a factory already shaped by hierarchies, migration histories, inequalities of skill and state expectations of development.

This matters for AIthropology. AI and robotics are not just technologies added to the economy. They are ways of imagining the social future. In the Chinese case, industrial automation condenses an ambition for modernisation, but also an anthropological question about the place of bodies in production. Which bodies are considered replaceable. Which skills are recognised. Which working lives remain caught between village, factory and service economy.

Seen this way, robotisation is not a story of machines against people. It is a story about the redistribution of value, prestige, risk and expectation. The automated factory does not eliminate the social. It rearranges it.

Data as global infrastructure

Tongyu Wu (吴桐雨) provides the third movement: looking at the human infrastructure of data. If Sun Ping lets us observe the algorithm in the street and Huang Yu the robot in the factory, Wu brings the question to data-labelling bases, institutional circuits and the workers who make it possible for AI systems to recognise images, texts, gestures or patterns.

It is important to avoid an easy reading here. Data annotation is not “China’s backstage”. It is AI’s backstage. It also exists in the United States and Europe, connected to digital platforms, outsourcing centres and global geographies of precarious labour. The importance of the comparative work by Tongyu Wu, James Muldoon and Bingqing Xia lies precisely in showing that China and the United States organise differently an infrastructure that belongs to the global AI industry.

In that contrast, China does not appear as an exotic exception, but as a producer of analytical categories. While many American and European companies outsource data annotation through digital platforms or business process outsourcing centres, the Chinese ecosystem shows another mode of organisation: data-labelling bases in inland cities, articulated with local governments, technology companies and territorial development policies. The difference matters because it prevents one experience, that of Western platforms outsourcing work to the Global South, from becoming universal.

The work of Xia and Wu on disabled workers sharpens this reading even further. The data industry can open forms of employment and recognition, but it can also produce new dependencies and new peripheries. Disabled People’s Organisations, local subsidies, firms and platforms are assembled into an ecosystem that promises inclusion while reproducing vulnerability.

The metaphor of the tree helps, but only if used precisely. Xia and Wu take José van Dijck’s image of the “platformisation tree” and correct it from the Chinese field. Where the European framework tended to foreground the role of the central state and platforms, their research shows something decisive: the mediating role of local governments. In their adaptation, the central government is the root; negotiation between local authorities and technological capital forms the trunk; Disabled People’s Organisations, or DPOs, are the branches; and workers’ negotiation and resistance appear as the leaves.

This difference is not decorative. The trunk is not simply “the companies”. It is the friction between local governments and technological capital. That is where decisions are made about whether a data-labelling base is installed, what support it receives, which workers are incorporated, what promise of inclusion is activated and what forms of dependency remain underneath. Chinese fieldwork does not illustrate a European theory. It corrects it.

Three scenes of the same infrastructure

Sun Ping’s street, Huang Yu’s factory and Tongyu Wu’s data infrastructure are not three separate topics. They are three scenes of the same social transformation.

In the first, the algorithm organises urban movement. In the second, the machine reorganises industrial production. In the third, labelled data prepares the world so that it can be processed by AI systems. Together, these studies show that artificial intelligence is not only a laboratory technology, nor only a competition between firms and states. It is a way of composing society.

That is why it matters to look at China through these researchers. Not because they offer a more benign or more critical version as a block, but because they shift the frame. China stops being merely “the other technological superpower” and appears as a theoretical and ethnographic field where labour, state, market, body, territory and future intersect.

There is also a methodological warning. The “anthropology of AI” does not always call itself that. In China, it may circulate through terms such as 人类学, 民族学, 传播学, 社会学, 科技社会学, 平台经济, 数字劳动 or 数据标注. If we only search for the Anglophone label, we see very little. If we follow practices, ethnographic methods and questions about social life, a richer field appears.

Closing

Chinese AI does not live only in model labs, data centres or state strategies. It also lives on an electric bike crossing the city under algorithmic pressure, in a factory imagining its future through robots, in a room where someone labels data so that a machine can learn to see.

Seen through these three studies, artificial intelligence is not an artificial mind floating in the cloud. It is a complete social chain. Someone delivers, someone competes with a machine, someone classifies the world so that the system can recognise it.

Perhaps that is the most fertile question for an anthropology of AI: not how intelligent a machine is, but how much social life it needs in order to appear so.

Sources