Machines that skip the words
I recently read about a new research paper, and it stayed with me. Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the MIT-IBM Computing Research Lab presented SwaT, short for Swarm Telepathy. It is a method that lets AI models from different families exchange their internal states directly, instead of translating them into text and back. The idea behind it is hard to ignore: machines may not need our language to talk to each other.
Long ago, it crossed my mind that language, as we know it, might one day become obsolete. I no longer think it will simply disappear. But I do think it may change in ways we have not yet imagined, because that is what systems of signs do when they hit their limits.
When signs hit a wall
Pictograms hit a wall: they could readily show a bird or a house, but abstract ideas, such as justice or yesterday, were much harder to express. Letting signs represent sounds greatly expanded what writing could convey. Pictograms did not vanish, though. We still use them in road signs, icons and emoji, wherever a picture says it faster than words.
Latin is another case. It was once the language of an empire, of the church and of scholarship. It did not simply die. It dissolved into Italian, Spanish, French and the other Romance languages, and remained in use in law, medicine and scientific naming, where precise and stable terms mattered.
And whenever natural language proved too vague, people built more precise languages alongside it. Mathematics is one. Programming languages are another. Neither replaced everyday speech. They took over the tasks where vagueness was too costly.
The new interpreter
For decades, programmers were the interpreters between people and machines. They took what we wanted a computer to do and translated it into a language the computer could execute. That made them valuable, and in a sense, powerful.
With LLMs, many of us now talk to computers directly, in plain language. The benefit is obvious: far more people can turn their ideas into working solutions.
But the interpreter did not disappear; it changed. The LLM is the new interpreter. We express our intentions in everyday language, while the precision of code remains necessary underneath. In a way, we traded precision for accessibility.
Anyone who works seriously with these tools soon feels the limits of that trade. Part of the problem is on our side: our vocabulary is not always rich enough to say exactly what we mean. But part of it lies in language itself. Compared with a computer program, language is vague, and that vagueness leads to iteration after iteration before a solution is right. It costs more time and energy than an optimal process would.
The new scribes
If machines can increasingly communicate without words, what might that change for us?
History suggests the answer matters. Scribes in ancient Egypt and Mesopotamia held a privileged position because few people could read and write. In medieval Europe, Latin played a similar role for the clergy. Control over a system of signs has often been a source of power, and it took the printing press and mass education to spread it.
AI has lowered one barrier: the need to know a programming language to create certain kinds of software. At the same time, competitiveness will increasingly depend on the ability to use AI effectively, and the gap is growing between those who use these tools superficially and those who use them with skill. For example, two people can ask AI to draft the same document. The one who can define the task, provide relevant context and judge the result is more likely to get a useful document in less time.
However, the new scribes may not be those who speak the machine’s language, but those who own the machines: the models, the data and the computing power.
For anyone working in development, this is not an abstract question. Will AI let smaller economies, organizations and entrepreneurs skip a stage of development, or will it open a new gap faster than the old ones are closing?
Nuance for all
There is one more thing worth weighing. The vagueness of language is not only a cost. It is also one of the sources of new ideas: a word that means slightly different things to different people leaves room for misunderstanding, but also for discovery. And living with that vagueness trains us to recognize nuances of meaning, in words, in people and in situations.
Machine language promises efficiency, but those who own the machines may capture a disproportionate share of its benefits. Human language is vague, but it belongs to everyone, and its vagueness teaches us to notice nuance.
So perhaps the real question is not whether we will lose our language. It is whether the gains in efficiency will be shared, or whether they will concentrate power in a few hands and narrow the space for nuance and creativity.
I don’t know the answer. But it is a question worth asking while we still have the words for it.