We choose who we build AI with. Here is why that matters for your kids.
Updated: Jul 24
When Geoffrey Hinton won the 2024 Nobel Prize in Physics, he used the moment to warn that we may be building AI faster than we can handle. Hinton is one of the researchers who built the foundations of modern AI. He spent decades advancing it. Then, in 2023, he left his role at Google so he could speak freely about where it was heading.
That takes a rare kind of integrity. I found it striking. And I kept thinking about it in the context of my own work.
Because every day we build Unomundi, my co-founder Matt and I make choices about who we trust. Which AI providers we use. Which partners we build on top of. Which voices we listen to. Those choices are not visible to you when your child opens the app. But they shape everything your child encounters inside it.
Why does it matter who makes the AI in a children's app?
Most parents think about AI safety at the level of content. Will the AI say something harmful? That is the right question. But it is not the only one.
There is a quieter question underneath it. Who built this model, and what did they optimise for? A model built to maximise engagement behaves differently to one built with long-term human wellbeing as the goal. The outputs may look similar. The trajectory is not.
When you choose a children's app, you are not just choosing features. You are inheriting the values of every company in that app's supply chain. Most parents never see that chain. We think they should, at least in outline.
Does it matter who builds AI while governments are still working it out?
Regulation is coming. I believe that. But the honest answer is that governments are still trying to understand what is possible. The people writing the rules are doing their best, but they are working with incomplete information, and the technology does not wait. There is a genuine lag.
In that lag, the choices are made by builders, investors, and buyers. Not by regulators. Every company that uses an AI product has already voted. Every investor who funds an AI lab has already voted. Every family that downloads an app has already voted.
We can defer that responsibility to bodies that may not yet fully grasp what is possible. Or we can make the choice now. We chose now.
What choosing safety-first actually looks like in practice
It is not a badge. It is not a line in the mission statement. It is a set of small, often inconvenient decisions.
It means choosing partners who do not prioritise short-term engagement over long-term safety. Sometimes that partner is slower to build with. Sometimes it is more expensive. Sometimes the safer option has fewer features than the flashier alternative.
We have walked away from tools that were impressive but built on foundations we did not trust. We have taken longer paths. We have asked harder questions of the companies whose technology runs inside ours. That is not heroic. It is just what we think responsible building looks like.
When we pick the models that Una runs on, we choose for the provider's approach to safety, not for raw capability. We look for published, long-term thinking about how to build AI responsibly. We did not choose the flashiest option available. We chose the one we felt most comfortable putting near a seven-year-old, with a human reviewing what reaches a child.
The steering starts with us
Hinton's point was not that AI is doomed. It was that direction is a choice. It can be steered. But only if the people building it, and the people funding it, and the people buying it are willing to steer it.
For us, that steering is not abstract. It shows up in a child's conversation with Una. It shows up in what Una does when a question gets complicated. It shows up in how we handle the data my daughter generates when she explores Japan or Bolivia inside the app.
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. Una does not chase engagement. She follows curiosity. That distinction comes from choices we made before the first child ever opened the app.
I think about this as a parent, not just as a builder. I am a solo mother. I do not have a partner checking the things I miss. I have to trust my own judgement about what reaches my children, and I have to be able to live with it. That accountability is clarifying. It makes the uncomfortable choices easier to make.
What I want you to take away from this
I am not asking you to take our word for it. I am asking you to ask the question. Not just of us. Of every app on your child's screen. Who built the AI inside this? What did they optimise for? What do they walk away from when the safer path is harder?
Those questions matter for adults. They matter more for children. A child's first experience of AI shapes their sense of what AI is for. That is a significant thing to get wrong.
We are trying to get it right. We will not always succeed. But we will keep making the choice.
FAQs
Does it matter which AI company powers a children's app?
Yes. A model built to maximise engagement behaves differently to one built with long-term wellbeing as the goal. When you choose a children's app, you inherit the values of every company in that app's supply chain.
What AI does Unomundi use?
Unomundi chooses the models behind its guide Una for the provider's long-term approach to AI safety rather than raw capability, and a human always reviews what Una shares with children.
Why can't we just wait for governments to regulate AI in children's apps?
Regulation is coming, but there is a genuine lag between what technology can do and what rules can cover. In that gap, choices are made by builders and buyers, not regulators. Waiting means deferring responsibility to bodies that may not yet fully grasp what is possible.