After Babel

There is something strange about the moment we are living through. For most of my life, the world moved toward ever deeper integration. Companies shifted production to wherever it was most efficient, countries specialized, and components travelled back and forth across continents before becoming a finished product. Meanwhile the internet connected people and markets in ways that would have been hard to imagine only a few decades earlier.

Now the movement has partly reversed. We are talking again about tariffs, trade wars, strategic raw materials, domestic production and self-sufficiency. Europe wants to manufacture its own semiconductors, the US wants to reduce its dependence on China, and China is building its own technological ecosystems. Energy, minerals, semiconductors and data centers have become security policy. The physical world is, at least in places, coming apart.

At the same time, something in the digital world is moving in almost exactly the opposite direction. For the first time we are building machines that can not only translate our languages but also help us reach knowledge that used to be out of range. While the material world fragments, the world of information appears to be integrating. It is a strange simultaneity — and the story of the Tower of Babel may say more about our situation than you would first expect.

A very old problem

The story of Babel is thousands of years old. Whether you read it religiously, historically or philosophically, it contains a fascinating idea. People speak the same language and decide to build a city and a tower reaching toward heaven. It is usually described as a story about human hubris, but there is another detail that is at least as interesting: what makes the construction possible is not really the tower. It is the language.

Because they understand one another, they can organize, divide the labor and combine the knowledge and effort of many people into a single project. Their shared language becomes a kind of infrastructure for their civilization. Then the language is confused. Nobody takes away their intelligence, their hands or their tools. The bricks are still there. What disappears is their ability to coordinate — and so the building stops.

Read that way, Babel is not just a story about a tower, or even about pride. It is a story about the relationship between language, cooperation and human capability. An individual can achieve a great deal, but civilization only emerges when the knowledge of many people can be combined.

No single person can build an iPhone

Look at a modern phone. No individual knows how to build one from scratch. There probably isn't even a single company that could independently recreate the whole chain from raw material to working device. Someone designs the processor, someone else builds the machine that manufactures the processor, minerals are mined elsewhere, and the operating system is developed by thousands of people. The communication standards were created collectively over decades, and the mobile networks are built by yet more companies and people.

The strange thing, then, is that the phone exists at all. It is the product of a civilization that has become extraordinarily good at dividing knowledge. A person no longer needs to understand the whole. She needs to understand her part and be able to trust that others understand theirs. That is specialization, and specialization has been one of civilization's most powerful engines.

But specialization has a price: we become dependent on one another. The more advanced our systems get, the more other people, companies and countries have to function for the whole to function. That vulnerability has become increasingly visible as geopolitics has moved back into the economy.

When the world starts reversing

In a line of reasoning about the risk of a future civilizational crisis, Professor Jiang puts the relationship roughly like this: technological progress = specialization × globalization. It is not a law of nature, of course, but as a mental model it is interesting. The more the world trades and cooperates, the more we can specialize. Sweden doesn't need to be able to manufacture everything Sweden needs. Japan doesn't need all its raw materials inside its own borders. The Netherlands can become exceptionally good at one small but decisive part of the semiconductor industry.

But what happens if the world becomes less global? Then countries have to do more themselves. Efficiency is partly traded for resilience, specialization for redundancy. Factories and supply chains start being judged on security rather than cost alone. Jiang's conclusion is pessimistic: if globalization declines, technological development will slow with it.

It is far from obvious that he is right about how far this goes. But the question he raises matters. What happens to a civilization whose technical capability rests on extreme specialization when the political world starts making that specialization harder? And here a counterforce appears that did not exist before: artificial intelligence.

A new kind of translator

When we say a language model knows languages, we tend to think of translation. I write Swedish, someone else writes Japanese, and the AI translates between us. Useful, but not really revolutionary — machine translation has been around for a long time. What is new is that language models are also beginning to translate between different ways of thinking.

A lawyer and a programmer may both speak Swedish and still use entirely different languages. The same goes for the physician and the statistician, the engineer and the economist, the researcher and the politician. Our specialization has created thousands of small Babels. Scientific language has become so specialized that researchers can struggle to understand research outside their own field. Software has its language, law has its own, economics another.

In between those worlds we are now placing the language model. I can hand it a legal text and ask what it means from an IT architecture perspective. I can read research in a language I don't speak and discuss the content in Swedish. I can describe an idea in ordinary words and get help expressing the same idea as code. That does not make me a lawyer, a researcher or a programmer. But the distance between those worlds of knowledge has shrunk.

That is why a language model is something more than a translation machine. It is starting to act as a semantic middle layer between people and between fields of knowledge. It does not have to make us experts in everything to change society. It is enough that it makes it substantially easier for different kinds of expertise to meet.

Two worlds moving in opposite directions

This is where the paradox I find so fascinating appears. In several domains the material world is moving toward greater separation, while the digital world moves toward greater connection. Countries are trying to reduce their dependence on one another, while people find it ever easier to use each other's knowledge. We are building more national borders around production while the borders between our fields of knowledge grow thinner.

Geography matters more for where semiconductors, energy and raw materials come from — and is nearly irrelevant for an idea. Someone in Sweden can sit at the kitchen table discussing a Japanese research paper in Swedish, compare it with European legislation, and minutes later test the reasoning in code. What only a few years ago could have required several people with different languages and competencies can, in some situations, be done by one person with access to a language model.

It is hard not to see something Babel-like in this, only in reverse. Just as the global economy risks fragmenting, technology is reducing the significance of another kind of fragmentation: the division of human knowledge.

What if globalization was only ever a tool?

That leads to an interesting thought. Perhaps it was not globalization itself that drove our technological development. Perhaps globalization was above all a way of achieving something more fundamental: coordination. In which case Jiang's model could be rewritten as: technological progress = specialization × the ability to coordinate knowledge.

For a few hundred years, trade, companies, universities, standardization, telephony, aviation and the internet have made that coordination steadily cheaper. Now language models arrive and lower the cost further. If a person can use a language model to move more easily between fields of knowledge, the friction around specialization drops.

We still need experts. AI does not make thirty years of experience irrelevant. But expert knowledge can become easier for others to find, understand and combine with other knowledge. That can make small groups strangely capable. Small organizations can gain access to capacity that used to require much larger ones, and people can combine fields that were previously separated simply because no individual had the time or opportunity to master the language of all of them.

Digital integration could, in this way, offset part of the material fragmentation. Not by replacing raw materials, factories or energy. A language model cannot conjure lithium, build a semiconductor fab or move grain across a blockaded sea. But it can change our ability to organize the knowledge needed to deal with such problems.

Babel in reverse

This is where the old story takes on new meaning. Babel is about a humanity that can first understand one another and therefore carry out an enormous shared project. When that shared understanding disappears, the people scatter and the project ends. We may be at the beginning of the opposite movement.

We are not getting a common language back. Swedish will remain Swedish and Japanese will remain Japanese. Lawyers will keep speaking law and programmers code. But between all these languages a new layer is emerging: a machine that can stand in the middle and try to understand what one side means and express it in a way the other side can grasp.

The consequences are hard to survey. We often discuss AI through the question of when machines become more intelligent than humans. That may not even be the most interesting question. The truly significant change could come much earlier — not when machines become as intelligent as people, but when people, with the help of machines, become far better at using each other's intelligence.

But who owns the shared language?

There is a considerably darker question here too. If language models really do become a shared layer between people and fields of knowledge, we have to ask who controls that layer. Babel's common language belonged to the people themselves. Our new semantic middle layer may instead end up controlled by a small number of companies and states.

If a growing share of our search, programming, research, translation and decision support passes through language models, whoever controls the models occupies a peculiar position. They don't need to own the information to hold power over how we reach it. They may come to control the interface itself between people and accumulated knowledge.

That gives the term digital sovereignty a deeper meaning. It is no longer only about where our data is stored or where our servers stand. It is also about who builds the layer through which we interpret the world and understand each other.

Perhaps that is why our era is so hard to read. We are seeing two apparently opposed forces at once. Geopolitics, raw materials, security and national interest are pulling the material world apart. At the same time, language models are making it easier for ideas to cross languages, professions, disciplines and cultures.

The material world may be heading toward less globalization while human knowledge heads toward something resembling hyperglobalization. One movement says we must become less dependent on each other. The other gives us a greater opportunity than ever to understand and use each other's knowledge.

Maybe these forces cancel out. Maybe they will instead amplify each other in ways we don't yet understand. But there is a thought in the old story of Babel that still feels oddly current: human strength lies not only in what an individual knows, but in our ability to understand one another well enough to build something together.

For thousands of years, language has both made that possible and set its limits. That limit is now starting to move. And perhaps the most interesting question about AI is not how intelligent the machine can become, but what people become capable of as Babel slowly starts running in reverse.