What do we want to keep when AI can already do it for us?

There is a question we may have to start asking sooner than we think: what do we want to keep when AI can not only do it for us, but do it better?

I recently came across an article by Bill Gates about the decisions we need to face if we want what he calls a "turbulent AI era" to benefit all of us instead of becoming the exact opposite. He talks about work, inequality, education, security, and human relationships, but one idea caught my attention more than the rest. Gates calls it Human Reserved: the things we might decide to keep for people even when a machine can already do them just as well or better.

Bill explains it through the example of a nature reserve. We could build there, but we choose not to because what we would lose matters more than what we would gain. The idea sounds simple. The hard part begins when you try to decide what should remain inside that reserve.

Not everything we do is a task

When I look for an example, the first thing that comes to mind is so obvious that saying it almost feels absurd: caring for a child.

It is easy to imagine a system that, one day, can watch over them better than we can, tailor an explanation to each moment, or never forget an appointment. Many of the tasks involved in caring for someone could be delegated, and some would probably be done better that way (and I am sure we would save ourselves a raised voice or two and more than one tantrum!).

But caring for a child is not about completing a list of tasks "correctly." It is a relationship, a responsibility, and part of your own life, your essence, and your purpose. Walking away from it completely would mean giving up part of that life, and no one in their right mind would choose to delegate experiences like that, right?

There are things we do to achieve a result, and others where experiencing the process is also part of the result. In the first group, a machine doing something better may be reason enough to hand it over. In the second, perhaps not.

Gates uses a similar example when he talks about the caregivers who supported his father through Alzheimer's. He also describes a very specific situation: a machine might be able to tell you that you have an incurable disease, but that does not mean it should. At moments like that, human presence is not an inefficiency added to the service. It is part of the value we are providing.

When automation also distances you

On a much smaller scale, the same question appears when we automate our work or our own projects. At first, you replace a repetitive task and end the day happy because you have "got your time back." Then you connect several tasks and automate an entire process. Now, with the level of AI any of us can access, that process can even analyze its own results, identify problems, and refine itself without you having to step in.

That is an enormous capability. And that is precisely why it is worth asking how far we want to take it.

If a system can handle operations, solve problems, and improve on its own, everything may work better and better while you understand less and less of what is happening inside. Something similar could already happen in the "B.AI" days (before artificial intelligence): you only had to keep outsourcing more tasks and responsibilities than perhaps you should until, little by little, the core of the business started operating independently of the person who created it. The difference is that now you can build the system yourself, and that system can eventually distance you from the very thing you created.

My first reaction is not to hold on to every decision or review every step. That would turn human control into a performance designed to make us feel as if we were still there. What I would want to preserve is knowing what is happening, how we got here, and where the project is heading.

We need enough information to understand the direction, refine it, and take the wheel if it stops going where we intended.

What if the result is better?

But there is another problem. Staying on course cannot mean clinging to our first idea either. A wise person can change their mind; a fool never does.

If an AI finds a path that produces better results, I would want to understand why and be willing to change my mind. But "better results" does not settle the decision by itself. A metric may improve because the system discovered an approach we had not considered. It may also improve because it crossed a line we were not willing to cross.

This is where ethics, responsibility, and something as basic as deciding what "better" really means come in. Or, more precisely, which lines we are willing to cross to achieve that result. Not every way of meeting an objective is acceptable. And if getting more requires acting in a way I consider unethical, I would rather get less.

That is why I think that if we reserve anything for people, it should be the ability to define the purpose, examine the means, change our minds when the evidence calls for it, and maintain limits that a metric cannot represent on its own.

But complete understanding cannot become an absolute requirement either.

Imagine that an AI finds a cure for a type of cancer we cannot treat today. The result is reproducible, trials show that it works, and we can verify its safety, but for a time the explanation is beyond our scientific understanding. Should we reject it until we can understand every step?

I do not have a clear answer, but I would find it difficult to argue that we should. In a case like this, turning human understanding into the ceiling of what we can accept could prevent one of the greatest benefits of the technology. We would have to demand evidence, safety, monitoring, and clear accountability. And I am convinced that there will come a time when AI finds solutions that we simply cannot understand. I do not know whether, when that moment comes, we should also demand complete understanding before saving those lives.

That example breaks every easy answer. Of course, we should not accept everything we do not understand. But perhaps we should not prohibit everything we do not understand either.

A decision labs cannot make alone

Gates takes this question to its real scale. AI can affect employment, how we distribute wealth, children's education, health, security, and the relationships we maintain with other people.

The companies developing the technology can provide knowledge and propose solutions, but they should not decide by themselves which jobs may disappear, which risks are acceptable, or which forms of human presence deserve protection. Nor does it seem reasonable to let the market provide the only answer. If replacing a person is cheaper, the pressure to do so will exist even when the cost to that person or their community never appears on a balance sheet.

I do not know whether to call it a constitution for AI, but perhaps we need something similar: a shared set of principles for deciding each case, knowing that the answer will not be identical in every country or remain unchanged forever. We need to protect ourselves from our own advances too.

This would not be a fixed list of protected professions. It would have to help us ask harder questions, philosophical ones in many cases. Is human presence part of the value? Who is responsible for the harm when the system fails? Can we verify a result even when we cannot yet explain how it was reached? What responsibility, relationship, or purpose do we lose when we step away and let AI take our place?

Gates argues for a public process involving governments, experts, educators, healthcare professionals, communities, and the people who will live through the transition. I think this is an important distinction: the ability to do something may come from a laboratory, but the decision to bring it into society should not come only from there.

What we still want to experience ourselves

I do not know what Human Reserved should include. The more I think about it, the more exceptions I find. A rule designed to protect us could prevent a major advance for humanity. An automation designed to free us could end up separating us from something that gave our lives purpose.

But I think the question is worth asking even without a definitive answer. Until now, we have tended to celebrate every new capability by asking what AI can do and how much time it saves us. Perhaps we need to add another question: what do we lose when we stop doing it ourselves?

Not to preserve tasks out of nostalgia or to keep people artificially busy. It is about deciding what we want to remain connected to and which parts of our lives we still want to be present in, even when a machine could take over.

AI will greatly expand the boundaries of what is possible. We should not allow those boundaries to decide by default what kind of life and society we want to live in.