Fate and Language

Man has delegated his activity to the machines. He has relinquished in their favour the faculty of thought. And machines certainly think. Indeed, in the evolution of this thought they go beyond the limited function originally envisaged. For example, they have invented the inconceivable effects of speed which so modify anyone experiencing them that it would be difficult, indeed arbitrary, to say that that person is the same as the one who lived in a world of slowness. The sensation that grips man, then, confronted by this thinking derived from his own thinking process but out of his control and growing steadily, unstoppable even by his putatively creative will, is nothing less than panic terror, a terror whose traps he fondly imagined he had circumvented, presumptuous child who fondly imagined he could walk in darkness without its company. Once again, you will discover at the origin of this terror the antagonism between man contemplating himself, contemplating himself in the process of being, and his thought process which is constantly evolving. Tragic nature of all mythology. There is an essentially modern tragic symbol: it is a sort of large wheel which is spinning and which is no longer being steered by a hand.

— Louis Aragon, Paris Peasant (1926)

Archaeologists have suggested that the topmost level of the ziggurat of Etemenanki served as an observatory for the Babylonian astronomers the Greeks called “Chaldeans,” whose planetary observations were “interpreted as omens,” inseparable from divination (Curta 2016). The Tower reaches toward heaven not only to approach the gods but to read their intentions. The desire to manage uncertainty—to gather signs, to extract from their arrangement a knowledge of what comes next—is a project with no natural terminus. There are always more signs to read, more data to gather, more variables to incorporate and excuses for inaccuracy. The New Tower ascends because the only alternative is stoppage or collapse, and to stop is an ideological and material concession that cannot be accepted.

For Benjamin, the reading of mantic symbols such as signs and portents belongs to the ossified order of myth. Fate is the realm of guilt, debt, and predetermination—the religious domain in which human beings are powerless to interfere in the workings of a cosmic order. Susan Buck-Morss writes in The Dialectics of Seeing:

Within myth, the passage of time takes the form of predetermination. The course of events is said to be predestined by the gods, written in the stars, spoken by oracles, or inscribed in sacred texts. Strictly speaking, myth and history are incompatible. The former dictates that because human beings are powerless to interfere in the workings of fate, nothing truly new can happen, while the concept of history implies the possibility of human influence upon events (1989).

A large language model, or LLM—the technology at the center of the current “artificial intelligence” industry—is a neural network trained on enormous quantities of text that generates strings of text by predicting the most probable next word (or “token,” the smallest unit the model processes) from probability distributions derived from the statistical weight of everything it has ingested. Whatever else these systems do, their foundational operation is prediction. The infrastructure required to support this prediction is Babelian in scale—thousands of GPUs (the specialized processors, originally designed for rendering videogame graphics, that power machine learning), energy measured in the output of small power plants, supply chains reaching from rare earth mines to undersea cables to data centers arrayed across continents—and its logic demands perpetual expansion. Each generation of machine learning model is larger and more onerous than the last in terms of its material requirements. This extraction touches beyond the immediate and material. In my home state of Texas, data centers for AI are given more than $1 billion (USD) in tax breaks each year, preventing the communities affected by these projects from benefiting from their creation through those lost taxes (Cobler, Mahajan, and Tribune 2026). Negative local impacts of hyperscale data centers range from increased load on an already strained electrical grid and rising consumer electrical costs, increased water usage, air pollution, the creation of large “heat islands,” and intense and inescapable noise.¹

In The Task of the Translator (1921) Walter Benjamin argued that the act of translation reveals a kinship between languages, not by making the foreign familiar, but by allowing foreignness to transform the target language. He called the substrate of this kinship reine Sprache (pure language)—not a language anyone speaks, but the totality of what all languages, taken together, intend. What is meant always exceeds what can be communicated. Stephen Dougherty (Dougherty 2025) has argued that massively multilingual neural translation models operationalize something like this kinship as an engineering principle: rather than building pairwise systems between two languages, the model draws on the shared resource of all languages simultaneously, so that a lower-resource language like Yiddish benefits from co-training with related Germanic languages and dozens of others besides. Structural kinship is real but pattern recognition, however massive in scale, is not backed by any actual referent. Benjamin critiqued “information” as a sort of linguistic junk-food—explained, ready to use, pre-selected for relevance to the reader, and always plausible—whether true or not. The psychological connection arrives pre-made in an informational burst, such as a social media news post, as opposed to the story, whose staying power is connected to its restraint from self-interpretation.

“Meaning is mutable,” artist Everest Pipkin said in a 2016 talk, “it builds itself in the mind of the reader as words compound… we as humans are built-in pattern recognizers and storytellers and any text need only do half the work; we will fill the gaps.” The machine reproduces structure; the reader produces meaning (Pipkin 2016).

The outputs of a language model are stochastic, sampled from probability distributions with an element of engineered “randomness.”² But this is not the unpredictability of a decision, it is the unpredictability of cast lots. The model does not deliberate over branching possibilities; a token is sampled, and we supply the meaning after the fact.

The late author David Foster Wallace, while an undergraduate, showed that the philosophical argument for fatalism could be defused through careful attention to the modalities embedded in language—the difference between what didn’t happen and what couldn’t have happened (Wallace, Cahn, and Eckert 2011).³ A language model cannot make this distinction. It is optimized for plausibility, not truth; it has no model of possible states of affairs against which to check its output, and so cannot represent counterfactuals—cannot hold in view the paths not taken at the moment of generation (Pearl and Mackenzie 2018). An interpreter can say “this could have gone otherwise”; a prediction engine can only say “this is what came next.” A culture that mistakes the latter for the former—that takes the stochastic output of a prediction engine for the speech of an oracle—has not escaped myth. It has reinstalled it at the center of its technical apparatus.

This writing is, in part, about authority, but it does not speak with authority. The subject of the underlying nature of language and how to model it is one of the most intense debates in science and philosophy. As a former art student I have some understanding of semiotics—the use of signs and their relationship to what is signified. As a hobby programmer who uses code and language in my art projects, I have a beginner’s understanding of a statistical model of language, calculating the likelihood of one letter, word, or group of words to follow another. [^15] I’m also aware from my research of the dogged projects to distill language into sets of syntactic rules or structural forms. There is also a metaphysical, even mystical way of speaking about language, a consideration of the ‘magic’ uses of reading, writing, and speaking, and imaginings of ‘pure’ or ‘original’ languages. While I cannot speak with authority about any of these modes of inquiry, I am an authority on my own experience of communication and agency. The self may be an illusion, but I feel solitude. Free will may be an illusion, but I feel anxiety of action and guilt. Language may be some primordial force, a product of evolution, math, structure, an alien life form, something I am not qualified to speak about, but I speak. I read and listen, and I construe meaning from what I apprehend as language.