An 18-year-old asked us not to use AI. She was right.

A comment appeared on our TikTok a few months ago. An 18-year-old, a few years older than our target audience, wrote to tell us she thought we were being disrespectful. She had two concerns. First, that using AI to represent cultures we do not come from is a form of disrespect to those cultures. Second, that AI is an environmental cost the planet cannot afford.

I did not respond defensively. I sat with it. Because she was right to raise both.

This post is for every thoughtful parent who has the same worry. Not a defence. An honest account of the tension, what we did about it, and what we still cannot fully resolve.

What is the actual problem with using AI to show cultures?

AI models are trained on the internet. The internet is not a neutral record of human culture. It skews heavily towards Western, English-language sources. The further a culture sits from that centre, the thinner and more distorted its representation becomes in a general model.

So when you ask an AI to describe daily life in rural Benin, or explain the meaning of a Navajo weaving pattern, or narrate a Japanese seasonal festival, you are not pulling from a deep well of lived knowledge. You are pulling from whatever fragments made it into the training data, filtered through a Western lens, smoothed by a model trained to sound confident.

That is a genuine problem. Not a hypothetical one. And we are building a platform designed specifically to help children understand cultures that are not their own. The tension is real and it sits at the heart of what we do.

Why we built it anyway

Not building was also a choice. And I had to look at that choice clearly.

Children are already forming their understanding of other cultures through screens. Mostly through algorithms designed to hold attention, not broaden perspective. Those algorithms have documented bias of their own, and no checks on what a child absorbs. The alternative to us was not a neutral, culturally respectful void. It was more of what already exists.

A small team cannot cover the whole world without AI assistance. We are not a network of a hundred journalists and documentary-makers with cameras in every country. To build something genuinely global, we had to use the tools available. The question was never whether to use AI. The question was whether to use it carelessly.

We chose to build anyway. And then we built protections in.

Three layers we built before we let Una speak about a culture

The first layer is what we call cultural guardrails in the pipeline. These are review agents trained to flag specific patterns: cultural appropriation, imperialist framing, the flattening of a complex culture into a single story. Content does not move forward if those flags are raised. This does not catch everything. It is a first line, not a complete solution.

The second layer is a validated knowledge base. Before Una draws on a general model, she draws on a knowledge base built with cultural anthropologists, psychologists, and intercultural communication experts. That base is our attempt to give her a richer, more grounded starting point than the raw training data can offer. It is still imperfect. It reflects the choices of the experts we were able to reach. But it is far better than none.

The third layer is human. We are building a Fellows programme: a network of cultural representatives, people who belong to the cultures Una speaks about, who review what she says about their communities. They are the check that the first two layers cannot be. They notice when something is technically accurate and still feels wrong. That instinct matters.

None of this makes the trade-off disappear. It reduces the harm we might cause. It makes us more honest about what we are doing. That is the most we can claim, and we claim it carefully.

What about the environmental cost?

The 18-year-old's second concern deserves its own paragraph. AI inference has a real energy cost. Large model training is particularly intensive, and the data centres that run AI at scale draw enormous amounts of power. These are not abstract concerns.

We use AI inference, not continuous training. We have chosen smaller, more efficient models where possible. We do not run processes at scale that a targeted call could handle. That is not a complete answer. It is the honest picture of where we are.

As the field moves towards more efficient architectures, and as more data centres move to renewable energy, the cost calculation will shift. We will keep looking at it. We are not dismissing it by saying that.

The most useful thing about that comment

Unomundi is a cultural-discovery app where children aged 6 to 12 explore countries and cultures through stories, games, and conversations with Una, a warm AI guide. That is what we are building. We are building it because we believe the cost of not building it is children left with a narrower, more fearful view of the world.

But the 18-year-old who pushed us is not our opponent. She is our conscience. The people who care enough to challenge us publicly are exactly the people we are accountable to. They are part of what we are building, even if they do not know it yet.

Being honest about the tension is not a weakness in the product. It is the product. If we pretended the problems did not exist, we would lose the one thing that makes us worth trusting: the willingness to keep looking at it.

If you are a parent who thinks about this kind of thing, you are the person who should be using Unomundi. Not because we have solved it. Because you will help us get closer.

Sign up to test our beta now

FAQs

Is it disrespectful to use AI to represent other cultures to children?

It is a real risk. General AI models are trained on internet data that skews Western and English-language, which means they can distort or flatten cultures they represent less well. Unomundi addresses this through a specialist knowledge base built with cultural experts, content guardrails, and a Fellows programme of human cultural reviewers.

Does Unomundi use AI?

Yes. Unomundi uses AI because a small team cannot cover every culture in the world without it. The alternative was leaving children's cultural understanding to attention-driven algorithms with no cultural checks at all. We chose to build with AI and to build protections in.

What is Unomundi?

Unomundi is a cultural-discovery app where children aged 6 to 12 explore countries and cultures through stories, games, and conversations with Una, a warm AI guide. There are no ads and no addictive design patterns.