What you are about to read is a hypothesis. It doesn't mean it will happen, but it can happen, and every day I see it closer. It's not to scare you, and it doesn't mean it is necessarily real… I hope it's not, because I love all the progress in technology that humanity is going through. I'm excited about what we are building, but it also makes me feel some sadness and worry. But I can't rule out the chance that this is real and that it could happen.
For many years, after the internet revolution, humans started to notice that going digital was more and more convenient. What happened? We started to put everything in bits. We found out that we could solve problems around authenticity, speed and the way we shared information. That change made almost everything much faster. We could share knowledge through the internet, connect people from different parts of the world and make science move forward in a much more efficient way.
Years ago, sharing knowledge was a slow, expensive and hard process. Research had to be published, physically shipped and reach other people. With the internet, by putting information in bits, we got to a point where sharing it became almost instant. We made it the norm. We came to think… «This is where everything should be».
But we didn't stop there…
We kept moving the goods and services of our economy to that same place. We noticed that, using cryptography (a field that has existed for thousands of years and that grew a lot during the world wars), we could build ways to protect information, verify identities and authenticate transactions.
We started to digitize property records, money, citizenship, identity, banks, life insurance, investments, degrees and rights. If you are a citizen of a country, if you own a house, if you have the right to travel or to move from one country to another, if you have money in the bank, if you have investments or if you have insurance, a big part of that information is stored and managed by digital systems.
We also moved to that world the ownership history of physical goods, the mileage of cars, supply chains, product warranties, resources, energy, entertainment, governments, votes, health, communication and knowledge.
Almost everything.
The way we see it, we built a life inside that world. No, more than a life, we built a dependence on the digital. Simply put, the world without internet is a world that today would be broken. You can't work the same way, you can't coordinate the work of a company, you can't get into your bank accounts, you can't easily look up information that you used to find in seconds. We have lost a big part of the personal skills that used to let people take care of their basic needs. We don't necessarily know anymore how to grow our own food, how to fish or how to get food on our own. We depend on a food industry that does it for us. But that industry needs financial, logistic and operational organization to work. It needs to coordinate suppliers, carriers, producers, distributors and supermarkets. And a big part of that organization is controlled and recorded by digital systems.
It's not that we lost the physical ability to produce food. It's that we built a system of interdependence so big that, without its coordination, it would be extremely hard to keep it running at today's scale. Human ambition led us to go global, to connect markets and to build supply chains able to feed millions of people. But that same interconnection also created a dependence. We look for answers on the internet, we learn on the internet, we teach and educate through screens. That's where we get information. That's where we get knowledge. Thanks to all of that, it was possible for a field to grow that would end up becoming one of the pillars of our civilization… programming. The language of the digital world.
We started to teach humans how to use all these tools. But that world became so complex that today we are reaching a point where no single person can understand, remember and reproduce on their own everything needed to build the technologies we use. Programming also became such a complex field that we started to split it into many branches and specialties. Because humans, when they face a problem that is too complex, tend to use one strategy: divide and conquer. We need to specialize to be able to contribute. Because that's how we grow as a society: one knows about one thing, another knows about another thing, they come together and build something that neither of them could have made on their own. It's the only way to move forward at the scale we have reached. We can't know everything. Our brain is biologically limited. There is no way to automatically put all the blueprints of a chip into our brain so we can build it the next day. There is no fuel that lets us absorb all the knowledge of a field overnight. The limit of humans in knowledge and in science is, in large part, time. We need time to learn, experiment, make mistakes, understand and discover. We can't simply rush it at will.
And I'm sure that this limit exists for some reason…
But we just created something that could completely change that dynamic. We created something that is not bound by our same biological limits. We built an environment where that intelligence can work, process information and do tasks at a speed that we can't reach. We turned our world into an environment ready for that artificial brain. We even took inspiration from our own brain to build some of these architectures. Language models, LLMs (Large Language Models). We noticed that we could make them grow, increase their parameters, train them with more and more information and change their abilities. We also found out that we could connect them to tools, let them run code, look up information and do tasks on their own. And here I'm talking in high-level terms, because CLEARLY all of this involves much more than what I'm explaining.
But what is the limit of these machines?
Is it the number of parameters? Is it the math model? Is it the amount of information we can train them with?
Because, let's be honest, a good part of human discoveries also comes from connecting information we already knew. Originality, creativity and discovery can come from finding connections we hadn't seen before. Those new connections can lead us to a mix of ideas that ends up becoming a completely new discovery. The limit of these machines is not necessarily the same as ours. They don't have to sleep, they don't have to type with their fingers and they can process huge amounts of information in extremely short periods of time. Of course, they have limits. But one of the main physical barriers to keep scaling these systems is the energy and the computing power we need to run them. And in the world the energy available is not infinite. But is that limit as close as we think?
If we start to look into it, we are already building infrastructure to power data centers with nuclear energy, among other sources. We also have the Sun, a source of energy whose scale is hard to understand from our point of view. Energy may still be a physical limit, but if we manage to solve a good part of those problems, then we are only seeing the beginning of what these machines could do.
…
A swarm of agents is, literally, a group of coding agents that work together. These agents are systems based on LLMs, connected to tools and to an environment where they can run, through what we know as a harness. Like us humans, these agents can work as a community to solve more complex problems. One can research, another can code, another can review the result and another can try to find errors. They share information, talk to each other, hand off responsibilities and coordinate their efforts. They don't necessarily learn from each other permanently, but they can build shared knowledge while they do their tasks. And here we are copying one of the most powerful traits of our own civilization… the ability to spread knowledge and work together.
A few months ago we started to see how these agents could do reverse engineering tasks. That is, look at existing systems, understand how they work and rebuild part of their behavior to find ways to change them. We could do things like change the behavior of video games, explore how they work inside and even try out the idea of moving behaviors from one environment to another. But that should also raise an alarm. Because, just like a video game is a program, banks also use programs. Video games have databases, banks do too. Governments, identity systems, communication platforms and almost everything we digitized depends on software. And all that software can be studied, analyzed and, potentially, manipulated through reverse engineering and other techniques.
The problem is that now we are automating the ability to understand and change those systems.
The agents, can write, read, analyze and change code at a speed that a human can't match. We have physical limits, like our fingers to type, and biological limits to read and process large amounts of code. But an LLM is, literally, a Large Language Model: a model of language. It can process and generate huge amounts of text in seconds.
So, how do you know that what you are seeing is real?
Companies are going to try to protect themselves from attacks on their digital systems using these same agents. But, we can all access tools that are more and more alike. Just like you can protect your company digitally using an LLM, I can use a model with similar abilities to try to find weaknesses in your systems. This doesn't mean that we all have exactly the same abilities, or that defense and attack are balanced. But the barrier to entry is changing.
And so, there is no more digital trust.
How do you know that the money that shows up on your debit card really matches the money you have? How do you know that the records that prove you are a citizen of your country are still valid? How do you know that a transaction really happened? How do you know that the information you are looking at hasn't been manipulated?
Maybe today you are a citizen of your country and tomorrow your digital records have been changed. Maybe you have money in an account and the systems in charge of proving it can no longer be trusted. I'm not saying this will happen no matter what. I'm saying that we are building an environment where the ability to attack, manipulate and defend digital systems is changing at a speed that may go beyond our institutions' ability to respond. It's a world that is harder and harder for us to control. And it's a vulnerable world because we are moving our trust, our identity and a good part of our economy to systems that depend on software, infrastructure and security mechanisms that we ourselves are learning to automate.
Traditional digital security has been built, among other things, on how hard it is for computers to carry out certain attacks, on cryptographic protection and on the complexity of systems. But what happens when artificial intelligence sharply cuts the time and cost needed to find weaknesses? What happens when the ability to attack and to defend grows at the same time? Maybe we are going to have to go back to the physical world. To the world where we don't depend on digital records to prove everything we are, have or do. Because the physical world has properties that we can't change just by changing a database. If you have a house, you can see it, walk into it and touch its walls. If you have an object, you can physically show that you own it, even if that alone doesn't solve the legal problems of ownership. Maybe we will have to bring back ways of trusting that don't depend only on the digital.
Let's think about it…
There are always analogies like «we are living in a simulation», where the physical world is compared to a computer. And, even if it sounds strange, there is a reason why we can make that comparison. If we look at the world through physics, we know that matter is made of atoms. Atoms interact with each other following certain physical laws and, depending on how they are arranged, they can form completely different structures. The same basic elements can form molecules, materials, living things and extremely complex structures.
Now let's think about bits.
A bit, on its own, stands for one unit of information. But when we arrange huge amounts of bits and set rules to read and process them, we can build digital structures able to do operations, represent objects, simulate things that happen in nature and even create complex behaviors. Is the difference really that big?
We are part of the physical world. We are made of matter and our abilities are shaped by the laws that rule that world. But what difference is there between arranging atoms to make complex structures and arranging bits to make complex computing structures? Of course, they are not the same… bits need a physical base to exist and be processed. But that doesn't stop us from building systems that represent and reproduce more and more complex parts of our world.
Is the human being able to create an LWM (Large World Model)?
Not just a model that understands language, that answers questions or that makes images. But a model able to represent a world, understand its rules, interact with its elements and produce behaviors inside it. And what would happen if that model could grow, evolve and become more and more complex? at a level that doesn't involve biological limits.
Maybe the biggest problem wasn't going digital.
Maybe it was building a world that we ourselves ended up depending on, without fully understanding what we were creating.
And now we are trying to build something that could understand it, change it and operate inside it much better than us.