SBS AI Series | Q&A with Hsain Ilahiane: How Culture Shapes AI

Artificial intelligence is changing rapidly, but understanding its impact requires looking beyond the technology itself. Across the College of Social and Behavioral Sciences, our faculty are examining AI through the lenses of ethics, language, culture, human behavior, policy and society — and asking fundamental questions about what it means to be human in an age of intelligent machines. As part of our exploration of AI and its many facets, this series features in-depth conversations with SBS scholars whose expertise brings distinct perspectives to this rapidly evolving field. These conversations go beyond the headlines to explore the questions, possibilities and consequences of AI — and what they can tell us about ourselves and the world we are building.

In this first of three Q&As, Hsain Ilahiane professor of Middle Eastern and North African Studies and associate director of the School of Global Studies, brings his expertise in applied and business anthropology to the conversation about AI. Ilahiane's teaching and research explore how technology shapes culture, communities and everyday life. In this Q&A, he explores how culture, politics and local contexts shape the technology — who gets to shape it, whose knowledge is represented, and how different societies may use and understand AI in very different ways. 

When we talk about AI, the focus is almost always on the technology itself. As an anthropologist who studies how societies adopt new technologies, what do you think those conversations are missing?

I think the first thing we need to recognize is that “AI” is not one thing. Broadly speaking, artificial intelligence refers to a constellation of computational systems, algorithms and models trained on data, that can perform tasks we associate with human intelligence, such as recognizing patterns, generating language and images, making predictions, or assisting with decisions. Those models operate through material infrastructures, computers, phones, sensors, robots, data centers, and through institutions and organizations that decide how they will be designed and used. So, when we talk about AI, we are really talking about a sociotechnical system, not simply an intelligent machine.

And that is what I think is often missing from the public conversation: the social life of the technology. We tend to talk as though AI arrives from somewhere outside society and then produces effects upon us. But technologies never enter an empty social space. They enter families, workplaces, schools, markets, governments, and communities that already have their own histories, inequalities, values, and ways of knowing.

I saw this very clearly in my research on mobile phones. The phone itself might be technologically similar wherever it travels, but people make very different things out of it. In some communities, mobile phones became ways of maintaining kinship across migration networks; elsewhere they became tools for conducting business, moving money, sharing information, or coordinating agricultural work. The technology did not dictate its social meaning. People incorporated it into relationships and practices that already existed.

I expect something similar with AI. Take agriculture: An AI system might analyze satellite images, weather data and crop records and recommend when a farmer should plant or irrigate. But the farmer may possess generations of local knowledge about soils, rainfall, plants, animals, and subtle environmental changes that may never have entered the datasets used to train the model. The anthropological question is therefore not simply, “Is the algorithm accurate?” It is also: What happens when algorithmic knowledge encounters local knowledge? Which knowledge is trusted? Which becomes authoritative? And what knowledge may be lost or marginalized because it cannot easily be converted into data?

Or, consider education: The same generative AI system might be used by one student as a shortcut to complete an assignment and by another as a tutor, a translation tool, or a way of gaining access to knowledge in a language or educational environment where other resources are scarce. The underlying model may be the same, but its social meaning and consequences are not.

As an anthropologist, I am interested not simply in what an AI model can do, but in the larger sociotechnical world around it: Who built it? What data and assumptions went into it? Who controls it? Who has access to it? How do people actually use it? Whose knowledge does it recognize, and whose knowledge does it leave out?

Can AI ever be culturally or politically neutral, or does it inevitably reflect the values, assumptions and priorities of the people who build it?

I don’t think AI can ever be completely culturally or politically neutral. But that does not mean that every algorithm is deliberately designed to promote a particular ideology. Consider what must happen before an AI system ever reaches us. Someone decides what problem the system should solve, what data it should learn from, what categories it should use, what languages should be represented, what counts as a successful outcome, what constitutes an error, and what kinds of risks are acceptable. None of those decisions is entirely neutral.

Language models offer a good example. If a model is trained primarily on material available in English and other widely digitized languages, it will inevitably know those cultural and linguistic worlds better than languages with much smaller digital footprints. An Amazigh-speaking community in Morocco, for example, may possess extraordinarily rich oral traditions, ecological knowledge and historical memory, but much of that knowledge may simply not exist in the digital datasets available to an AI system. The model’s limitations are therefore not simply technical. They reflect inequalities in whose knowledge has historically been recorded, digitized and made computationally accessible.

There is also a deeper historical dimension. AI did not emerge in a social vacuum. As the history of the field shows, some of its foundational developments were closely connected to military research, Cold War institutions, universities, corporations and government funding. Those institutional environments helped shape which problems received attention and resources. That does not make AI inherently militaristic or capitalist, but it reminds us that technologies have histories, and those histories leave traces in what gets built and what societies imagine technology should do.

So rather than asking whether we can create a perfectly neutral AI, which I don’t think we can, I would ask: Can we make the assumptions and values embedded in AI more visible and contestable? Can we diversify the people and forms of knowledge involved in building these systems? And can the communities affected by AI participate in deciding how it should be used? For anthropology, these are the more interesting questions. 

Your research has examined how technologies like mobile phones have been adopted in different cultural settings. Do you expect AI to follow a similar pattern, with different societies embracing and using it in very different ways?

Absolutely. Although I would use the word “appropriation” rather than simply “adoption.” 

For example, in communities with extensive migration networks, the mobile phone became much more than a communication device. It became part of maintaining kinship across enormous distances. A migrant living abroad could remain involved in family decisions, send money, receive news from home, and participate in the everyday life of a household thousands of miles away. The technology became embedded in existing obligations of kinship, reciprocity and belonging.

Consider agriculture again. A farmer in Arizona, Morocco or India might have access to similar AI tools for predicting weather, identifying plant diseases or managing irrigation. But those systems will enter very different agricultural worlds. Farmers will combine, or sometimes reject, the recommendations of AI according to local knowledge, experience, water availability, land tenure, economic conditions and ideas about risk. The algorithm may be similar, but the social practice surrounding it will not be.

Education offers another example. A university student in the United States might use generative AI to brainstorm an essay or explain a difficult concept. Somewhere with fewer educational resources, the same technology might function as a tutor, translator, writing assistant or even provide access to forms of expertise that were previously difficult to obtain. At the same time, communities may have very different ideas about whether using AI constitutes assistance, collaboration, authorship or cheating. Those categories themselves are cultural.

But AI introduces something that makes this relationship more complicated than the mobile phone. A mobile phone largely connected people to one another. Generative AI increasingly mediates people’s relationship with knowledge itself. It can summarize, translate, recommend, interpret and generate information. AI may simultaneously reshape how people write, learn, communicate, work, and decide what counts as credible knowledge or expertise.
 

"I don’t expect one global AI culture to emerge — I expect many AI cultures. The technologies may circulate globally, but their meanings and uses will be produced locally."

Who gets to shape the future of AI? What communities or perspectives risk being overlooked as the technology evolves?

At the moment, the power to shape AI is highly concentrated. A relatively small number of technology companies, engineers, investors, universities and governments have enormous influence over what gets built, what problems AI is asked to solve, what data it learns from, and ultimately what kinds of technological futures become possible.

From an anthropological perspective, the issue is not that these people necessarily have bad intentions. It is that no small group of people, however talented, can represent the extraordinary diversity of human experience. When AI is developed primarily within particular economic, linguistic and cultural environments, the assumptions of those environments can easily begin to look universal. What AI knows is not the same thing as what humanity knows.

The communities most at risk of being overlooked, therefore, are often those already underrepresented in the institutions and datasets shaping AI: people in the Global South, Indigenous communities, speakers of less digitally represented languages, rural populations, poorer communities, and people whose knowledge is transmitted orally, through practice or through forms that do not easily become data.

This raises what I think is one of the most important anthropological questions about AI: What happens to knowledge that cannot easily be digitized?

Anthropologists know that an enormous amount of human knowledge is embodied, contextual and relational. A farmer may recognize subtle changes in soil, plants or weather that are difficult to put into words. An elder may carry histories transmitted through generations that have never been written down. These are not inferior forms of knowledge simply because they are difficult for a computational system to process.

My concern is that as AI becomes increasingly involved in education, government, medicine, business and everyday decision-making, what is computationally legible may gradually acquire greater authority than what is not. We could begin confusing what is available to AI with what is worth knowing.

But I don’t think the answer is simply to criticize AI from the outside. We need to broaden who gets to participate in imagining, designing and governing these systems. Engineers and computer scientists are essential, but so are anthropologists, historians, linguists, artists and other ways of understanding human experience. Most importantly, the communities affected by these technologies need a meaningful voice in deciding how they are developed and used.

That also means involving people earlier. It is not enough to build an AI system and then ask anthropologists or communities to identify its biases afterward. By then many fundamental decisions have already been made. Different forms of expertise need to be present upstream, when the problem itself is being defined.

Ultimately, the future of AI should not simply be something that happens to societies. People should have a role in deciding what parts of human life we do not want to delegate to algorithms at all. The future of AI is too important to be left to any one discipline. We all have a stake in it, and many more voices need to be in the room.

##