Generative artificial intelligence is usually presented as a technical, corporate or regulatory problem. Models, costs, chips, productivity, privacy, copyright. Yet these days reveal another movement, quieter and perhaps more interesting for anthropology. The discipline no longer appears only as a critical observer of AI. It is beginning to enter the places where AI is designed, taught and normalised.

The shift matters. It is one thing to study how ChatGPT is used, how platforms classify the world, or how automated systems reproduce inequalities. It is another to intervene in the training of those who design these systems, and in the classrooms where people decide what it means to learn, write, assess or research after the arrival of generative models.

Today’s radar moves between those two places. A class at MIT brings anthropology into chatbot design. A symposium at St Andrews asks how to teach anthropology when AI is already part of the educational environment.

Designing chatbots is not just programming responses

In March, MIT presented Humane User Experience Design, a cross-listed class between computer science and anthropology, taught by Arvind Satyanarayan and Graham Jones. The course begins from a simple and powerful intuition. If chatbots are conversational systems, they cannot be treated merely as interfaces. We need to understand what a conversation is.

Here linguistic anthropology stops being a humanistic ornament and becomes a design method. Speaking with someone is not simply a matter of producing correct information. It involves tone, context, turns, expectations, misunderstandings, silences, authority, trust and repair. Conversation is a social practice, not a pipe through which data flows.

This matters because much conversational AI is still assessed through narrow criteria. Accuracy, speed, immediate usefulness, user satisfaction. But an interaction can be technically correct and socially poor. It can respond without listening, accompany without understanding, personalise without recognising the person, simulate closeness without assuming responsibility.

The MIT class points precisely to this zone. Qualitative methods, interviews, observation, interaction analysis and attention to context do not enter after the model, as an ethical varnish. They enter before, in the way the system is imagined.

The anthropological question would not be how to make this chatbot seem more human. It would be more uncomfortable. What kind of social relation are we designing when we let a machine converse with someone?

From the chatbot to the classroom

The other side of the movement appeared at St Andrews, where the online symposium AI and Anthropological Pedagogies was held on 4 and 5 June. It was not a last-minute open call, nor an event organised by EASA, but a symposium held by the Department of Social Anthropology at St Andrews, with a programme dedicated to how teachers and students of anthropology are facing AI in education.

Here the direction is reversed. If at MIT anthropology enters AI design, at St Andrews AI enters the anthropology classroom. And not as an external topic, but as an everyday presence. Writing tool, interlocutor, shortcut, threat, prosthesis, source of suspicion and object of experimentation.

The symposium is not interesting simply because it says that AI should be taught. Many universities already repeat that. What matters is that it moves the question towards concrete practices. What do we do when students write with generative models, when a tutorial is prepared with ChatGPT, when an ethnography is thought through with a machine, or when the classroom fills with texts whose authorship is increasingly difficult to delimit?

Anthropology has a relative advantage here. It is used to working with mediations, ambiguous relations, partial voices, situated authority and the social production of knowledge. It does not need to imagine AI either as an external demon or as a pedagogical salvation. It can treat it as what it already is. A cultural infrastructure that reorganises practices.

The synthetic gaze

Among the symposium papers, one formula is especially fertile for thinking about the present. The idea of a synthetic gaze. The expression allows the debate to move beyond the simple question of bias towards something deeper. What kind of observer does AI produce when it interprets, classifies or generates images of people and social worlds?

Modern medicine long spoke of the clinical gaze, that way of seeing which turns the body into a case, a symptom, a file or a legible surface. Generative AI adds another layer. It can see without being present. It can represent suffering without having encountered it. It can compose plausible images without assuming the responsibility of a relation.

The synthetic gaze is not neutral because no gaze is neutral. It is made of data, instructions, categories, omissions and previous hierarchies. It can amplify stereotypes, soften structural violence or turn situated experiences into generic scenes. And yet it can also appear clean, objective, useful, even compassionate.

Here anthropology has something specific to say. It is not enough to ask whether the generated image is false or true. We need to ask what relation it establishes with what it represents, what it erases in order to function, whom it turns into an object of vision and who is authorised to look.

From asking whether to studying what happens

The bridge between Cambridge and St Andrews helps us understand the speed of the shift. In 2024, Ella McPherson and Matei Candea published a manifesto at Cambridge on AI and academic life. It did not begin with the usual institutional question, how to incorporate AI, but with a slower one. Whether AI really contributes to research and learning, under what conditions and at what cost.

Two years later, Candea appears in the St Andrews programme with Benjamin Knight in a paper on learning with and learning about ChatGPT. That continuity is significant. It does not show a naive conversion to technological enthusiasm, but the passage from normative caution to situated experimentation.

First, let us not assume that every innovation improves academic life. Then, let us observe what happens when the tool is already inside, when it becomes an interlocutor, when it modifies writing, reading, assessment and the imagination of what counts as intellectual work.

That trajectory summarises the present moment well. Generative AI can no longer be thought of only as a tool that some people use and others reject. It is entering the very architecture of education, design and knowledge production. The anthropological question is not whether it exists, but what relations it reorganises.

The discipline under pressure

A note of caution is needed. Anthropology is not discussing AI in a vacuum. While some programmes explore chatbots, generative pedagogies and synthetic gazes, other areas of the discipline face very concrete institutional pressures. Public funding under dispute, attacks on diversity policies, conflicts over freedom of expression, and museum debates shaped by memory, heritage and representation.

For that reason, the turn towards AI should not be read as an academic fashion. If taken seriously, it can be a way of defending the public relevance of anthropology in a world where more and more cultural decisions are encoded in technical systems.

But it can also become an uncritical absorption if the discipline merely translates its vocabulary into the language of innovation. It is not enough to say that AI needs context, culture or ethics. We need to show, case by case, what changes when conversation, learning and the gaze pass through generative infrastructures.

Closing

The dispute is not humanities against technology. That opposition already arrives too late. The question is what kind of social world is being inscribed in the systems that design conversations, mediate learning and produce forms of seeing.

MIT shows one path. Bringing anthropology into chatbot design before the social relation is reduced to an interface. St Andrews shows another. Bringing AI into pedagogical debate without turning it into either a total enemy or a magical solution.

Between the two, a strong thesis appears for the present. Anthropology does not only study AI. It can also contest the conditions under which AI learns to speak, to look and to teach.

Sources