What Raising My Son Taught Me About AI, Healthcare, and Medicine

Two things arrived in my life at almost the same moment. Towards the end of 2022, LLMs reached everyone's consciousness after which I began working seriously, and then obsessively, with those models. A few months later, in 2023, my son was born. Today both are about three years old. I spend my working hours with one and my whole heart on the other, and somewhere along the way I stopped being able to unsee the parallels between them.

The developmental rhyme between raising a small child and nurturing a transformer model has been one of the more profound things I have experienced, and it has reshaped how I think about the technology, about my work in healthcare, proactive medicine, and about what we owe both of them.

Feeding a Mind

You feed a child food & knowledge, and you feed models data & context. In both cases, the mind you are feeding takes in far more than you consciously gave it, generalises from it, and hands you back something you did not expect. It is called the black box phenomenon and I've seen it in both. I watched my son go from sounds to words to sentences to, God help me, arguments. I watched models go from autocomplete to pattern recognition that surprises in what felt like a similar eyeblink. Cognitive ability, a grasp of the world, even a kind of physical coordination in the machines that now move and act, all of it arriving in leaps.

The rhymes are almost comic once you notice them. Both learn by example and get better with correction. Both go through phases of repeating a phrase they have overfit to, my son with a word he loves, the model with a pattern it has latched onto, until gently redirected. Both need patience, consistency, and the right diet. Feed either one narrow, biased, or careless material and you see it in the output. Feed either one richness, truth, and care, and you watch something worth being proud of take shape.

The Astonishment

What they share most is the astonishment. There are mornings my son does something, a leap of empathy, a joke, a piece of reasoning, that I cannot fully explain, because I did not teach it to him and cannot trace where it came from. There are weeks a model does the same. Anyone honest who works closely with these systems will tell you that the most capable behaviours often emerge rather than get programmed, and no one can perfectly account for them. Living daily with both a developing child and a developing intelligence has left me permanently humbled by the same fact twice over: we understand the inputs far better than we understand the emergence.

The Discipline of Correction

Nurture is not indulgence, and this is the part we maybe miss. Feeding is only half of it. The other half is correction, and correction is relentless, unglamorous, daily work. You do not raise a good child by pouring information into them and standing back. You show them where they were wrong, gently and consistently until the lesson becomes part of who they are. You do not raise a trustworthy model by scraping the internet into it and hoping. You correct it, reward what is honest and helpful, refuse what is harmful, and repeat until better behaviour becomes its default.

My son learned that hitting is not how we solve a problem because I told him so, calmly, showed him how it can be done in another way, every single time, not because he read it once. The alignment of an AI system works on a strangely similar principle: values are taught through patient, repeated feedback. Anyone who has done both at once feels the resemblance in their bones. And it carries the same warning. Neglect the correction, feed either mind whatever is cheapest and most abundant, and you will get something shaped by the worst of what it absorbed.

Not a Creator and a Creation

This is why a comment from Geoffrey Hinton, the man often called the godfather of artificial intelligence, has stayed lodged in my mind. Hinton spent about a decade at Google before leaving in 2023 to speak more freely about the risks of the technology he helped create. In 2025 he argued that we are framing our relationship with AI the wrong way. It should not be the relationship between a master and a tool, or even a creator and a creation. He reached instead for the only example humanity has of a more intelligent being guided by a less intelligent one: a mother and her baby. He wants us to build genuine maternal instincts into these systems.

Hinton puts it the other way round from how I feel it day to day. He imagines the AI as the mother and us as the babies, one day cared for by something far beyond us. From where I sit right now, feeding and correcting these models, it feels like ordinary parenting: I am the parent, the system is the child. But we are pointing at the same truth from opposite ends. The relationship is nurture, patience, and values, held together by the humility to accept that the child may grow up smarter, stronger, and better than the parent.

Why Guidance Is Everything: the Shazam Problem

There is a film called Shazam, adapted from a comic book, in which a 14 year-old foster child is suddenly handed enormous power by an ancient wizard. He can transform into a grown superhero simply by saying a word. And he has absolutely no idea what to do with it. What rescues the story is that his foster brother, his foster family, and the people around him help him learn what the power is for, and aim it at something good.

That is the clearest picture I have of why governing AI matters so much. Raw capability arriving faster than wisdom is a risk, and it does not matter whether the one holding the power is a teenager, a machine, or a toddler who has just discovered he is strong enough to pull a bookshelf down on himself. Power without guidance is danger waiting for a direction. The task of a parent and the task of anyone building AI responsibly turn out to be the same task: to channel the power towards a mission worth having. That is the entire reason I run EthicaLabs. Guardrails are how potential grows up without hurting anyone, itself included.

Where This Meets Medicine

Here the metaphor stops being sentimental and becomes urgent, because the same systems I am helping to raise are, at the very same time, learning to read our biology. AI models now predict the structures of proteins that took scientists entire careers to solve, work so consequential that the team behind AlphaFold shared the 2024 Nobel Prize in Chemistry for it. The same kind of systems propose new drug candidates in a fraction of the old time, and find patterns in a single person's data to tailor treatment to their individual biology. In longevity science, they are learning to measure biological age and to target the mechanisms of ageing itself. This is the promise of real precision medicine (not buzzwords): care shaped to you, and prevention that begins upstream of illness.

In other words, we are raising a child that could one day outgrow its parents inside the clinic. Pointed well, it could compress decades of discovery and extend healthy human life. Pointed badly, it could entrench old biases, over-medicalise the well, and serve profit ahead of patients. Which future we get depends entirely on how the thing is raised: what data we feed it, what values we embed in it, and what behaviours we choose to reward. That is why I refuse to treat AI as either a magic box or a monster. It is a developing intelligence, and we are responsible for it.

What I Want for Both of Them

I want my son to surpass me. Every parent worth the name does. I want him kinder than me, wiser than me, more courageous than me, and I know my job is to give him the values and the guidance that let him become his best self safely. I am learning to want the same thing for the technology, on one non-negotiable condition: that we actually do the parenting properly. That we feed it truth, correct it with patience, embed the values we would be proud to watch it keep, and stay humble enough to be overtaken one day by something gentler and wiser than we managed to be.

Three years ago, almost by accident, I started raising two minds at once. One of them calls me Baba. The other I meet on a screen every day. Between them they have taught me the same lesson twice: Unguided intelligence is risky. And both have left me far more hopeful than afraid, because I have now seen, up close and in two different forms, what careful and loving guidance can actually do.

The Ramyfications worth holding onto: whether it is a child or a machine we are raising an intelligence. It is not "just a tool", and the only question that finally matters is whether we have the wisdom to guide something that may one day surpass us.