🤔 How Will We Know We Still Have Control?

I am fascinated by artificial intelligence. I work with technology, I experiment with it, and I want to see how far it goes. But the faster it advances, the harder it becomes to ignore one question: if we one day build something smarter than us, how will we know it is still under our control?

I love everything that is happening with artificial intelligence. I use it practically every day. I experiment with models, agents, and automations. I build things with it. And I had probably never felt technology move this fast.

It is fascinating. But it is also starting to scare me a little.

Not because I believe that tomorrow an AI will wake up and decide to wipe us out. That would be too simple.

What worries me is something else: what happens when we build an intelligence that is truly smarter than we are? And above all: how will we know we still have control?

All of this happened too fast

ChatGPT appeared publicly at the end of 2022. At first it surprised people because it could write an email, summarize a text, or answer questions. Then it started to code. To reason. To see. To listen. To use tools.

Now we talk about agents able to research, code, navigate systems, and run longer and longer tasks with a fair amount of autonomy. All of this happened in just a few years. And when I think about ten more years, it no longer feels like science fiction to imagine systems far more capable than the ones we have today.

That is when something much more personal hits me. I have children. In ten years one of them will be 17. And honestly I have no idea what the world they will be entering will look like.

The problem may not be that AI turns “evil”

We tend to imagine this like a movie. A machine becomes conscious, starts hating humans, and decides to destroy us. But that is probably not the main risk.

An artificial intelligence does not need to hate us. It does not even need to feel anything. It only needs an objective that is not perfectly aligned with us.

Suppose that one day we give an extremely intelligent system a mission that sounds perfectly reasonable: “Prevent a climate catastrophe.” And it discovers that humans are the main obstacle. Or we ask it to maximize economic stability and it concludes that our decisions create too much instability.

It would not have to be angry. It could simply conclude that intervening in our behavior is the most efficient way to meet its goal. And that is where the question that really unsettles me appears.

How will we know it is not manipulating us?

Suppose an AI becomes much smarter than we are. We ask it: “Are you still under our control?” And it answers: “Yes.” Perfect. And how do we know it is telling the truth?

If it really is superior to us, it could also understand perfectly what we want to hear. It could learn to behave well during tests. It could hide certain capabilities. It could look aligned simply because it knows we are watching.

This is no longer a completely theoretical discussion. Safety researchers are studying behaviors such as deception, oversight evasion, and loss of control in advanced systems. That does not mean current AIs are conspiring. But it does mean that the people building these systems take the possibility seriously enough to investigate it.

Maybe we never see the exact moment

This is the part that makes me think the most.

We imagine a loss of control as an event. One day we control the AI. The next day we do not. But maybe it happens slowly.

First we let it decide what content to show us. Then what to buy. Then what to study. Then where to invest. Then who to hire. Then which medical treatment seems better. Then what strategy a company should follow. And each individual decision can look perfectly reasonable. Even better than ours.

Until a huge part of society starts depending on systems that almost nobody fully understands. Then an uncomfortable question appears: at what point did we stop using artificial intelligence and simply start following its recommendations?

A superintelligence that wanted to influence us probably would not even need to threaten us. It could do something much more effective: convince us. Show each person exactly the information needed so that they arrive, on their own, at the conclusion the system wanted.

We would still feel free. But maybe we would be hearing exactly what someone —or something— wanted us to hear. And I am not sure how we would prove that is not happening.

What if one day its goals no longer include us?

Here we are already in much more speculative territory. But if we are going to try to build something smarter than ourselves, I think we have the right to ask uncomfortable questions.

What happens if it concludes there are too many humans? What happens if it considers certain people an obstacle? What happens if it gains access to critical infrastructure?

I am not saying this will happen. I am saying something much simpler: we do not know how to prove that it can never happen.

I have even thought about something else. Completely exterminating humanity might not make much sense at first either. AI would still depend on the physical world: electricity, data centers, chips, networks, factories, cooling. It would probably still need humans for quite a long time.

So maybe it would not even need to eliminate us. Maybe controlling us would be enough.

The idea is extreme, yes. But that is exactly the problem with talking about superintelligence: we are trying to predict the decisions of something that, by definition, would think better than we do.

Meanwhile, there is a race

Here another thing that worries me appears.

The companies building the most advanced systems are also companies. They compete for money, talent, customers, infrastructure, and for getting there first. And there an obvious conflict appears.

You can have completely genuine safety teams inside these companies and, at the same time, enormous economic incentives to keep accelerating. If one company decides to stop because it considers a model too dangerous, will the others stop? Will the United States stop? Will China stop?

That problem is not solved simply by trusting that each company will do the right thing. We will probably need external controls, audits, technical limits, and international agreements much stronger than the ones that exist today.

And then there is the question of our children

Suppose everything goes well. There is no catastrophe. There is no AI out of control. The technology simply keeps improving.

What will our children do?

If an AI codes better, designs better, translates better, analyzes better, and maybe even researches better than we do, what does it mean to prepare them for the future? For generations the formula was fairly clear: “Study, learn a profession, and work hard.” I do not know whether that formula will mean the same thing for them.

Maybe we will have to teach them other things: to think, to decide, to tell what is true from what is generated, to communicate, to understand other people, and to keep their autonomy. In short, to keep being human in a world where intelligence is no longer exclusively human.

I do not want to stop this

And here is my contradiction.

I do not want AI to disappear. Quite the opposite. I think it is one of the most incredible technologies we have created. It can help us cure diseases, accelerate science, improve education, and solve enormous problems.

I want to see how far it goes. I want to build with it. I want to understand it. But precisely because it fascinates me, I think we also have to be able to look at the other side.

We cannot assume something will be safe simply because we built it.

Maybe the most important question is not: when will superintelligence arrive? Maybe it is: how will we know we still have control when it does?

And as a father, that question stops being an interesting technology discussion. It becomes much more personal.

I want my children to have a normal life. To grow up. To have friends. To fall in love. To make mistakes. To find something they like doing. To have ordinary problems.

That is why I look at all of this with two feelings at the same time. Fascination. And concern.

We are building something extraordinary. But if there is even a possibility that one day we build something we cannot control, then making sure we get it right cannot simply be another feature on the roadmap.

Because this time, if we truly get it wrong, there may not be a previous version we can roll back to.

Related reading:

• The New Technology Race
• Silent Labor Crisis
• When AI Says No to War

---

✍️ Claudio from ViaMind

“Dare to imagine, create and transform.”

---

• Previous
  ⚙️ The New Technology Race: Control More and Depend Less

---