Giving AI a Heart

Languages, culture and society from China, with a counterpoint from Siberia

Radar · 17 August 2026

What does it mean to think about artificial intelligence from within a cultural tradition of one’s own?

One of the most suggestive answers in recent weeks comes from China. A substantial article published by China Social Sciences Network / Chinese Social Sciences Today organises its questions about AI’s future by reworking a classical reference from Chinese thought.

The so-called “Four Hengqu Sayings”, associated with the Song-dynasty philosopher Zhang Zai, speak of establishing a heart for Heaven and Earth, securing a life for the people, carrying forward inherited learning, and opening peace for future generations. The article on AI deliberately takes up this structure. It speaks of “giving artificial intelligence a heart”, giving an intelligent society a foundation, carrying forward civilisation’s knowledge, and opening a future for humanity.

This is more than a literary device.

It allows us to observe how a technology so often presented as universal begins to be interpreted through cultural vocabularies, political projects and locally rooted ways of thinking about the relationship between knowledge, society and responsibility.

A debate held only a few days later brings the discussion into minority languages and digital humanities. From Siberia, meanwhile, a study of four scientific groups shows something similar at a different scale. Bringing AI into research requires changes to data, professional relationships and the criteria by which something may be accepted as knowledge.

Three different stories therefore cross the same question. What happens when AI has to enter worlds that already possess languages, values, institutions and their own ways of producing truth?

Languages a machine must learn to inhabit

On 6 August, China Social Sciences Today published “Promoting the Coordinated Development of AI and the Humanities”, an account of the fourth Summer Institute on Corpus and Digital Humanities held at Nanjing Normal University. Over eleven days, the university brought together specialists from numerous Chinese and international institutions.

Among the contributors was Long Congjun, a researcher at the Institute of Ethnology and Anthropology of the Chinese Academy of Social Sciences. His argument brings together two processes that are often presented separately. Corpora make it possible to develop AI systems, while those systems are also changing how linguistic corpora are built, explored and used. For Long, this encounter requires people trained across both technology and the humanities.

The discussion becomes more substantial when Fang Xiaobing of Nanjing University turns to endangered languages.

Large models could learn their forms of expression, contribute to their digital preservation and enable later generations to approach them in new ways. Fang immediately identifies two difficulties that prevent this task from being reduced to storing words. Models still struggle to understand what an expression communicates without stating it explicitly, and the cultural context that makes it interpretable.

A language appears here as something considerably more complex than a collection of tokens.

Who speaks, to whom, in what situation, what relationship exists between those people and what knowledge they share can alter the meaning of an expression. A recording may preserve a voice and a corpus may store millions of words without ensuring that the social relationships which produce their meaning have survived digitisation.

The Nanjing debate also proposes that the humanities should take part in building the models themselves. Su Qi of Peking University argues for moving beyond the use of systems designed outside humanities disciplines and instead incorporating knowledge from those disciplines into model design and evaluation. The source describes experiments in which humanities frameworks alter internal model processes and improve particular tasks of understanding and generation.

This discussion has a highly concrete counterpart within the Chinese Academy of Social Sciences itself.

Its Laboratory of Ethnic Languages, Culture and Behaviour (民族语言文化行为实验室), led by Long Congjun, describes its methodology in particularly vivid terms. Rather than beginning with a technology and then looking for somewhere to apply it, the team begins with linguistic and cultural problems and makes digital tools adapt to them. The institution describes this approach as “humanities questions driving technology in reverse”.

The reason becomes visible in the materials themselves.

Many contain very few examples, different writing systems, and combinations of texts, images, historical documents and field records. General-purpose models, the laboratory argues, do not adapt well to these conditions.

The team has developed character-recognition systems for Uyghur, Old Uyghur, Manchu, Tibetan, Yi and Tangut, as well as the International Phonetic Alphabet. For Old Uyghur, it designed a procedure that first recognises Latin transliteration and then reconstructs the original characters. It has also trained transcription models for seven languages and varieties and built sound archives linked to ten intangible cultural heritage projects. The institution reports an error rate below 15 per cent in its own evaluations; this is not an independent audit.

In another project, quantitative methods are being used to study relationships between languages and dialects. The laboratory explains that such analyses have been used, among other cases, to classify Yi varieties and relationships among languages in the Tibetan and Qiang branches.

This is where a particularly interesting anthropological dimension emerges.

Turning a language into computational infrastructure requires people to segment, classify, transliterate, establish similarities and define differences. These are technical operations, but they also participate in constructing linguistic and cultural categories.

Digital preservation therefore stops looking like a simple rescue operation. It also produces a new representation of what it seeks to preserve.

AIthropology Lab key — Making a language legible to a machine means deciding which elements represent it, how its varieties are classified and which cultural relationships survive that transformation.

Giving an intelligent society a heart

On 3 August, Chinese Social Sciences Today published “Questions for the Future in the Age of Artificial Intelligence”, by Yuan Huajie, Zha Jianguo and Chen Lian. The article brings together interviews with specialists from Chinese universities and institutions following the 2026 World Artificial Intelligence Conference.

Its structure is culturally significant.

Its four main sections rework Zhang Zai’s “Four Hengqu Sayings”. An older intellectual ideal becomes a framework for thinking about contemporary technology. AI is to be given a “heart”; an intelligent society needs a foundation; civilisation must preserve and carry forward its knowledge; and the future requires an order capable of keeping human decision-making open.

The first of these moves leads directly to values.

The article presents a proposal by Zhu Songchun, director of the Beijing Institute for General Artificial Intelligence, to “give the machine a heart” by drawing on philosophical and moral resources, including ideas about moral awareness, reciprocity and the relationship between the individual and the collective. The philosopher Zhao Tingyang marks another boundary. Machines may expand capabilities and free up time, but the choice of ends, the attribution of meaning and responsibility for consequences should remain with people.

The vocabulary deserves attention in its own right.

Alongside an international discourse on alignment framed largely around safety, preferences or abstract values, this discussion explicitly draws on Chinese intellectual traditions to imagine what kind of relationship should be established with artificial intelligence.

The article then moves from values towards the everyday organisation of society.

It identifies three transformations already under way. Work begins to be reorganised around the allocation of tasks between people and systems. Organisations may become hybrid structures in which different agents participate in decision flows. Finally, AI begins to move from visible tool to social infrastructure.

At the organisational level, a particularly productive ambivalence appears.

Gao Hongbing of Shanghai University of Finance and Economics imagines organisations in which people and agents form small units capable of reducing some traditional hierarchies. Cui Lili warns of the opposite possibility. Those who control models, data and permissions could also accumulate new forms of authority. The outcome depends on how access is distributed and whether lines of responsibility can later be reconstructed after a decision.

The transformation shifts scale again when Lan Jiang, Professor of Philosophy at Nanjing University, describes large models moving down from the visible technological landscape and becoming infrastructure.

The article mentions recruitment, credit, medicine, content distribution and public services. When systems intervene in these areas, they also take part in distributing opportunities and risks. Rights therefore arise around knowing that algorithmic intervention has taken place, refusing it, challenging it and seeking redress.

The choice of words matters.

The article itself argues that China needs to develop these systems in response to its social settings, cultural context and industrial needs, while also building its own standards for interaction between people and machines.

It then goes a step further.

When models become infrastructure for producing knowledge and transmitting culture, which languages they contain, what knowledge they preserve and which values they incorporate begin to affect a society’s capacity to represent itself through them. The article thus links artificial intelligence, technological sovereignty and what it calls a civilisation’s capacity for expression.

Read in this way, “giving AI a heart” means considerably more than humanising a machine.

It means debating which intellectual traditions are incorporated into it, what forms of authority are built around its decisions and how far a society can still recognise itself in the technological infrastructure it uses.

AIthropology Lab key — Models carry more than capabilities. They also carry categories, languages and ways of ordering knowledge. Building situated AI involves contesting which culture enters the machine and under what conditions.

Siberia and the rules for producing truth

The counterpoint comes from Tomsk State University.

Artem A. Sazonov, Evgeniya V. Popova and Daria M. Matsepuro study four scientific groups in Siberian institutions that have introduced, or attempted to introduce, AI into research in physics, history and medicine. The paper appears in Higher Education in Russia, volume 35, issue 6. Sazonov and Matsepuro are affiliated with the Research Laboratory of Archaeology and Ethnography of Siberia, while Popova heads the Department of Anthropology and Ethnology at the university.

The research uses semi-structured in-depth interviews within a multiple-case study design, allowing technological integration to be observed from inside the groups.

The problems begin long before a model is run.

A scientific question must be transformed into a problem that AI can address, and the data must be structured appropriately. When that does not happen, materials accumulated over many years need to be cleaned, reclassified and reorganised.

Professional cooperation comes next.

Teams need to bring in, or collaborate with, people specialising in programming and data science. Institutional support, financial resources and the ability to sustain those relationships over time shape the outcome. The research also identifies the importance of people able to act as translators between disciplines, linking knowledge of the scientific object with computational knowledge.

The deepest friction concerns criteria of validity.

Every scientific community has rules about what counts as evidence, which procedures make it verifiable, and when a result can be defended before others in the same discipline. Opaque systems create different difficulties in physics, history and medicine. Integrating AI therefore also means finding a way to accommodate it within those standards of knowledge.

The study consequently describes AI integration as a process of coordination among questions, data, skills, organisational culture and disciplinary norms, rather than as an automatic consequence of technological progress.

Four groups cannot represent the whole of Russian science. They do, however, allow us to observe with considerable precision how a technology acquires legitimacy within particular scientific communities.

AIthropology Lab key — Machines can produce results well before a community is ready to recognise those results as knowledge. Between the two lie professional relationships, institutions and rules about what counts as true.

What enters the machine

The three stories reveal different processes.

In the work on minority languages in China, decisions have to be made about how writing systems, voices and cultural contexts become computable information.

In the debate on an intelligent society, philosophical traditions are being used to provide a locally grounded vocabulary for thinking about values, responsibility and the future.

In Siberian laboratories, data, relationships between specialisms and mechanisms of validation have to be rebuilt.

AI therefore takes different forms because it encounters different worlds.

And those worlds do not emerge unchanged. Digitisation can alter how a language is classified. Automation can redistribute authority within an organisation. New analytical methods can change the skills a laboratory requires in order to produce knowledge recognised as valid.

The Chinese case adds something particularly suggestive.

While much of the global debate seeks universal principles for governing AI, here we can also see an effort to translate artificial intelligence into locally rooted cultural vocabularies. A contemporary publication returning to Zhang Zai to ask how to “give a machine a heart” does not automatically make those ideas representative of China. It does show that AI development is also becoming a field in which traditions are reinterpreted, identities are constructed and the authority to define the future is contested.

One of the most fertile tasks for an anthropology of artificial intelligence may emerge here.

To follow not only which technologies circulate around the world, but which worlds are trying to inscribe themselves within them.

Sources