WORK

The Library of Babel

Borges's 1941 story — the infinite library containing every possible book, every truth and every gibberish — the cautionary extreme that Calvino's multiplicity requires a shaping consciousness to avoid.

Jorge Luis Borges's 1941 story imagines a library of infinite hexagonal rooms, each containing shelves of identical books. Between them, the books contain every possible combination of letters and therefore every possible text. Every masterpiece. Every error. Every almost-masterpiece differing from a masterpiece by a single misplaced letter. Every catalog of the library. Every false catalog. The totality is useless. The infinite, without selection, is indistinguishable from nothing, because the ratio of signal to noise approaches zero as the collection approaches completeness. The librarians who inhabit the library spend their lives searching for the book that would explain the library — and cannot, because the book exists but so do all its counterfeits, and the means to distinguish them do not. The story is Calvino's cautionary example in the memo on multiplicity, and the structural parallel to large language model training data.

In the AI Story

Hedcut illustration for The Library of Babel — *The Library of Babel*

Borges wrote the story during his tenure at the Miguel Cané municipal library in Buenos Aires, where the bureaucratic tedium of cataloging provided the material for the fable. The library of the story is a thought experiment taken to its logical conclusion: if language is combinatorial and the combinations are finite, then the set of all possible books is finite but astronomically vast, and a library containing all of them would contain every truth alongside every falsehood, every meaning alongside every noise.

The philosophical structure of the thought experiment is precise. The library is comprehensive — every possible text exists within it. But comprehensiveness is not usefulness. To find the true catalog of the library, one would need to distinguish it from the countless false catalogs; to find a true book on any subject, one would need to distinguish it from the countless books on that subject that are false in subtly different ways. The means of distinction must come from outside the library, because the library itself contains every possible claim about what is inside it. The librarians, confined to the library, cannot escape the problem.

A language model's training data is a practical Library of Babel. It contains the written record of human civilization — the accumulated output of millions of minds across thousands of years, in hundreds of languages, on every conceivable subject. The range is extraordinary. The multiplicity is genuine. And the selection problem is identical to the one Borges diagnosed: the model can access everything, which means that the act of choosing what to access, what to foreground, what to connect — the act that transforms multiplicity from noise into meaning — must come from elsewhere.

Calvino's invocation of the library in his memo on multiplicity is precise. He celebrates the encyclopedic ambition of Gadda, Borges, Musil, Perec. He also insists that multiplicity requires a shaping intelligence, because unshaped multiplicity is the Library of Babel — the totality that contains everything and therefore means nothing. The shaping intelligence selects. It chooses which thread to follow, which connection to pursue, which possibility to realize. The selection is the authorship. The selection is what transforms the library from a repository of potential meaning into an actual work.

The argument extends to human-AI collaboration. The model provides the library. The human provides the selection. The human's judgment — what to ask, what to accept, what to reject, what to follow — is what transforms the model's comprehensive generativity into something that bears the shape of a mind. Without the selection, the model's output is the library: everything and therefore nothing. With the selection, the output becomes a world, in the specific sense Calvino meant — a bounded arrangement that bears the imprint of the consciousness that shaped it.

Origin

Jorge Luis Borges wrote the story in 1941 while working at the Miguel Cané municipal library in Buenos Aires. It was first published that year in the collection El Jardín de senderos que se bifurcan (The Garden of Forking Paths) and later included in Ficciones (1944), which established Borges's international reputation.

Key Ideas

Infinite combinations, finite meaning. Every possible text exists in the library, which means the true texts are indistinguishable from their false counterparts.

The comprehensiveness problem. Completeness is not usefulness; the totality without means of selection is indistinguishable from nothing.

The librarians' predicament. Those who inhabit the library cannot escape its logic, because the means of distinction must come from outside it.

The training-data parallel. Language models have access to a practical Library of Babel; the selection problem Borges diagnosed is the problem of AI-era meaning-making at scale.

Calvino's prescription. Multiplicity requires shaping; the shaping intelligence — the selecting consciousness — is what transforms library into world.

Debates & Critiques

Whether the Library of Babel is a pessimistic or optimistic image is contested. Pessimistic readings emphasize the impossibility of finding meaning in the totality. Optimistic readings emphasize that the totality contains every truth — the problem is selection, which is soluble. The Calvino volume's reading is operational: the library is neither optimistic nor pessimistic but diagnostic, and it names the specific task that falls to the human in the age of AI.

Appears in the Orange Pill Cycle

Italo Calvino — On AI