Ethnic AI

There is not a single, universal form of artificial intelligence. What we call “AI” is in fact a constellation of systems, models and cultural infrastructures, each shaped by different datasets, technical choices, visual traditions and implicit worldviews.

This becomes particularly evident in image-generating A.I.. American or Western-based systems such as DALL·E, GPT Image, Midjourney, Runway Gen-4.5, Google Veo, Nano Banana, FLUX, and other platforms often produce results that reflect specific aesthetic conventions: cinematic realism, advertising language, digital illustration, Western art history, commercial photography and contemporary design culture. 

Chinese systems such as Kling, Seedream, Seedance, Vidu and others may respond to the same prompt in a different way, privileging other visual codes, narrative structures, symbolic references or ideas of beauty.

The question, then, is not simply whether these differences come from the culture or religion of the people who programmed them, or from different levels of sophistication in deep-learning algorithms. 

It is more complex. Generative A.I. is shaped by the interaction between training data, model architecture, censorship rules, market expectations, cultural memory, visual archives and the values embedded in the societies that produce and regulate these technologies.

Given the same prompt, two A.I. systems may generate different images not because one is “more intelligent” than the other, but because they have learned from different worlds. Their imagination is statistical, but the material from which that imagination is built is profoundly human.

This raises an important question: what kind of images would be generated by an A.I. developed within a Muslim, Hindu or African cultural context?

In many non-Western societies, religion is not separated from everyday life in the same way it often is in secular Western modernity. It is not simply a private belief system, but a way of interpreting existence, nature, the body, time, community, destiny and the invisible. An A.I. trained within such a context might therefore produce images in which the sacred is not an added decorative element, but an underlying structure of perception.

A Muslim generative A.I. might develop visual languages shaped by geometry, calligraphy, aniconic traditions, sacred architecture, light, rhythm, repetition and the metaphysics of divine order. A Hindu generative AI might draw from mythology, cosmology, ritual colour, cyclical time, divine multiplicity, temple sculpture, mandalas and the symbolic continuity between human, animal, cosmic and spiritual forms.

But the question becomes even more radical when we imagine a generative A.I. rooted in African tribal or indigenous cultures. What would an A.I. create if its visual memory came primarily from oral traditions, masks, ritual objects, ancestral symbols, body painting, textiles, animist cosmologies, collective ceremonies and non-linear concepts of time? Would it produce images not as individual artworks, but as visual events connected to memory, community, healing, trance, territory and ancestral presence?

In this sense, generative A.I. is not only a technological tool. It is also a mirror of the cultures that build it. Every A.I. carries within it a partial archive of the world. The real question is therefore not only what AI can generate, but whose imagination it is allowed to learn from.