To Infinity and Below!
From Humboldt to Generative AI
Returning from a short trip in the south, as the train moved through the fading light of late afternoon, the woman seated next to me was reading Wilhelm von Humboldt. One of his ideas came back to mind: the human capacity to produce an infinity of meaning from a finite number of signs. From a tiny alphabet, a brief memory, and a limited lifespan, we bring forth languages, works of art, systems of knowledge, entire symbolic worlds. Perhaps our strength lies precisely in this strange power to make meaning proliferate out of almost nothing.
If we now turn to generative AI, the idea of “infinity” appears again, though within an almost opposite dynamic. Here we encounter a machine built on immense scale, billions of parameters, petabytes of data, colossal computational power, producing from almost everything localized responses, synthetic forms, finite and probabilistic outputs.
Humans create infinity from almost nothing. Machines absorb almost everything in order to return the world to its deepest proximities. Seen this way, debates about the anthropomorphization of machines may lose much of their relevance. Humans and generative AI do not belong to the same dominant regime of meaning production.
Humans constantly exceed the signs they produce. A few words are enough to open onto infinite interpretations, myths, theories, works, sometimes even entire worlds. Human language always surpasses its own material conditions. It proliferates, branches out, drifts, and becomes charged with affects, contradictions, and ambiguities. We spontaneously inhabit a symbolic beyond of the sign. Of course, humans never entirely escape cultural correlations or automatisms. Yet even our repetitions seem capable of reopening interpretive possibility.
The machine proceeds differently. From a vertiginous accumulation of texts, images, music, and human traces, it calculates neighborhoods, stabilizes correlations, and draws together forms that become commensurable within the same computational space, what researcher Jean Rohmer calls a “computearth”
The machine does not seem to move through signs toward what humans experience as a transcendence of meaning. Rather, it navigates the dense web of relations linking signs to one another. A medieval image, a scientific diagram, and an advertisement can now be drawn together within the same latent geometry. Perhaps generative AI makes visible something that had never fully appeared before: the infra below of forms from which humans have always produced their excess of meaning. In the end, this may be far more unprecedented than a machine capable of thinking like us.
MD