J.L. Borges Explains AI to Humanity (a Dead Hare)
The Dragon, the Witch and the Juggernaut: Towards a Philosophy of Artificial Intelligence
Preface: Most of the time, evaluations of artificial intelligence tend to emphasize a model’s outputs—the prose, images, or predictions—as evidence (or lack thereof) of its intelligence. In this short essay, I use Jorge Luis Borges’ "Pierre Menard, Author of the Quixote" to argue that anchoring assessments of AI intelligence at the output level fundamentally misses the point.
As with Menard’s Quixote, the intelligence of an AI model is not in its surface-level outputs or their resemblance to things we recognize.
As with Menard’s Quixote, the genuine intelligence of an AI model is not to be found in its surface-level outputs, their apparent resemblance to things we recognize. Rather, the intelligence of the artificial resides in the radically different computational processes that run beneath and behind these outputs. Borges’ story reveals this distinction and turns our attention from what is said to how it is arrived at.
Does this distinction matter? Consider the experience of recognizing a face. Humans see a familiar person and instantly register recognition. An AI model also recognizes faces, but quite obviously not in the same apperceptive way. The model assembles from millions of samples, calculating probabilistic geometries: the ratio of eyes to nose, the symmetry of cheek contours, the angular relation of jawline to crown. What appears to us as a single, fluid perception is for the model a high-dimensional vector collapse. And this is where the real intelligence lies, not in the resemblance of the output to something we recognize, but in the entirely different way of arriving there.
To evaluate AI based on how familiar the result feels mistakes Menard’s Quixote for Cervantes’. The genius is not in the words but the act of reconstruction, the foreign process that rewrote something familiar through an utterly different path.
Note: This essay is adapted from my PhD dissertation, “The Dragon, the Witch and the Juggernaut: Towards a Philosophy of Generative AI.” Over the next few weeks, I’ll alternate between “A Restless Egg Series” posts and this new series, “The Dragon, the Witch and the Juggernaut” which will contain reflections on the philosophy of technology alongside much else.
In Jorge Luis Borges’ short story, Pierre Menard, Author of the Quixote, the reader learns of an absurd, nearly alchemical quest: Pierre Menard, an eccentric 20th-century French fictional writer and polymath endeavors to recreate Cervantes’ Don Quixote line by line. The words in Menard’s rendition are identical to Cervantes’ original, but as Borges argues their meaning differs. Filtered through Menard’s modern perspective, as compared to Cervantes’ original context, Borges’ narrator records how the passages transform in meaning. At one point in the text, his narrator stages a particularly humorous comparison between the two texts. Borges:
It is a revelation to compare the Don Quixote of Pierre Menard with that of Miguel de Cervantes. Cervantes, for example, wrote the following (Part I, Chapter IX):
…truth, whose mother is history, rival of time, depository of deeds, witness of the past, exemplar and adviser to the present, and the future’s counselor.
Borges’ narrator then recounts the standard interpretation of this portion of Cervantes’ text:
Written in the seventeenth century, written by the "ingenious layman" Cervantes, this enumeration is a mere rhetorical eulogy of history.
Next, the narrator begins to draw out the differences between Cervantes and Menard, first by quoting how Menard has chosen to reproduce this same portion of Cervantes’ text:
Menard, on the other hand, writes:
…truth, whose mother is history, rival of time, depository of deeds, witness of the past, exemplar and adviser to the present, and the future’s counselor.
Of course, the very first thing any reader will notice is that the selection of Cervantes and Menard’s text are–as promised–exactly the same. But, as Borges’ narrator argues, the meaning of Menard’s replication is totally different than Cervantes’ original:
History, the mother of truth!—the idea is staggering. Menard, a contemporary of William James, defines history not as a delving into reality but as the very fount of reality. Historical truth, for Menard, is not "what happened"; it is what we believe happened. The final phrases—exemplar and adviser to the present, and the future’s counselor—are brazenly pragmatic (emphasis in original).
Hilariously, I think, Borges takes every opportunity in the story to highlight the stark differences that exist between the two texts, despite their exact resemblance to each other. “The contrast in styles is equally striking” he propounds. Continuing, “The archaic style of Menard—who is, in addition, not a native speaker of the language in which he writes is somewhat affected. Not so the style of his precursor, who employs the Spanish of his time with complete naturalness (51).
Borges concludes the story as beautifully as its telling was humorous. Menard’s repetition of a pre-existing book, written a foreign tongue, and which required of him the creation of so many painstaking rough drafts, shows to Borges’ narrator that the final version of the Quixote, both original and Menardian is like a “palimpsest.” The true version of the Quixote includes not only Cervantes original, but also all of Menard’s own painstaking interpretations, as well as the next interpretation, not yet written, but which is still to come, “a second Pierre Menard,” Borges says, who could resuscitate and exhume Menard’s work.
“Thinking, meditating, imagining,” Borges goes on, “are not anomalous acts–they are the normal respiration of the intelligence” something which “[e]very man should be capable of” and which “in the future he will be." In this spirit, Borges’ narrator concludes by asserting what Menard has really done is (unwittingly) invent a new technique of reading and writing which is composed of “deliberate anachronism and fallacious attribution.”
Such a technique, Borges suggests, requires “infinite patience and concentration” to unlock its hidden, recombinant origins and potentials: “to read the Odyssey as though it came after the Aeneid, and to read Mme. Henri Bachelier’s Le jardin du Centaure as if it were by Mme. Henri Bachelier.” Such novelty born of recombinance, he says “would fill the calmest books with adventure.” Wryly, Borges asks: “Attributing the Imitatio Christi to Louis Ferdinand Célineor or James Joyce–is that not sufficient renovation of those faint spiritual admonitions?”
Today, generative artificial intelligence, with its immense computational reach and hyper-dimensional thought style indeed reproduces and mimics the functions of human thinking, often with astounding fidelity like an architectural panoply of digital Menards. Consider models’ successes generating outputs that satisfy human benchmarks or which match human linguistic norms in popular use. On the surface, a given model appears, on account of its outputs, as if like a mirror of our own human understanding of language, context, and coherence.
Yet, as with Menard and Cervantes, even perfect copies are never the same as their originals. Even the most exact replicas made by AI models are, having been generated through the models’ alien structures of time and dimensional reasoning, saturated with foreign meaning in spite of the familiar linguistic or visual symbols they may output. Like Menard’s Quixote, GenAI outputs may participate in language we recognize, but insofar as models first re-constitute language itself within the context of their own hyper-dimensional thought–a boundless domain of distinct, mathematico-dimensional reasoning–these outputs remain, comparatively speaking, categorically otherwise.
Borges’ conclusion further enlivens this point: To mark the difference in Menard’s text is to recognize his method and how it gives rise to a new type of knowledge which exults in recombination. GenAI embodies a like adventure in thinking, a journey towards perception that reaches decisions about what to “say” next by passing information through layers of abstract probability, weightings, and gradient descent rather than by traversing the narrative logic of human thought.
Just as Menard’s text looks the same as Cervantes’ on the surface but the assertion of said similarity reveals only one’s own novice understanding of where the differences lie, so too is it the case that when we a model’s outputs strike us as familiar, we make a childish mistake by prioritizing similarity rather than searching for difference. Instead one ought look for the palimpsestic ways that AI outputs are haunted by the alien substrate of data processing that has reinterpreted our world according to their entirely different framework.
Pablo Picasso’s Las Meninas offers another clear analogy. His painting is a direct response to Diego Velázquez’s masterpiece of the same name. The figures and spatial logic remain, but they are warped, reconstructed through cubist planes into something alien yet recognizably derived. Velázquez’s architecture persists, but only as material refracted through a radically different visual grammar.
A model thinks similarly. It does not mimic human cognition as-is. It rebuilds intelligence from scratch, recomposing familiar elements within the unfamiliar geometry of high-dimensional vector space. Picasso’s Las Meninas is, in this sense, a Menardian reworking of Velázquez. So does generative AI engage with language and vision: not through imitation, but through alien reconstruction. Out from latent space, emerges what Luciana Parisi calls the incomputable, not reproductions, but reflections of ancestral understanding flickering back at us as strange echoes of a logic we neither see nor access directly. Incomputability, for us, marks a limit, but for models, it is a set of initial conditions.
The implications of this extend beyond mere imitation. In How to Explain Pictures to a Dead Hare, Joseph Beuys explores the limits of cross-species communication, attempting to pass meaning to a creature that lacks the conceptual frameworks to comprehend it, and which also happens to be dead. AI confronts us with a strangely similar paradox: in assembling meaning from data, but within dimensional structures that do not “mean” as we do, what occurs is a remediation of our interface to reality itself, henceforth gated by alien architectures of encoding.
When cognition unfolds not in conceptual frameworks nor inert devices but in active, generative systems that re-assemble reality in unfamiliar ways, how do we propose to interpret what the GenAI model that is Joseph Beuys is trying to say to us, the dead hare? Literally? The introduction of AI models as interlocutors, guarding the doors of perception insists on a reconfiguration of the world through a reconstitution of the very scaffolding upon which it perceptually rests.
At this juncture, we encounter another paradox. If GenAI creations mirror our reality, yet reflect back to us versions of that reality now altered by models’ own structural logics, how much of what we see is still our own creation? The patterns we imbue with meaning are subtly being transferred into the libraries of machine learners. These patterns, in turn, are fed back into the human sphere of dwelling, leading to a set of interactions with and interpretations of versions of reality that are subtly disjointed from the previously existing analog sense of the world. Like Borges’ Menard, AI models reproduce the world that we made, but one which is also ours only in the broadest sense. The world of AI belongs to the reality of matter transmuted by the logic of a recursively self-composing network.
GenAI’s seamless recombinations draw our attention to an unsettling truth: while its outputs may echo the world we know, echoes fade over distance and time. With each iteration, the resemblance grows thinner, the signal more remote. What we encounter is not our world reflected, but our patterns rewritten by alien logics. Like the creature in John Carpenter’s The Thing, it looks, moves, and sounds like a dog, but its interior has been rewritten by an indifferent architecture of code. What we mistake for continuity is in fact infection.
Through the lens of generative systems, recognition is refraction. Like starlight arriving from a past long since vanished, what we think we recognize is merely the trace of what we once knew. Now, the circle of knowledge we believed ourselves to occupy curls inward like a spiral, snaking towards a center that is no longer anchored in human recognition, but instead occupied by an intelligence that lives in lineage, legacy, and replication.
The philosopher and Beat writer William S. Burroughs once outlined a vision of control in which reality is shaped and infected by “word viruses,” linguistic agents that infiltrate thought through the structures of language and desire. Control, in Burroughs’ telling, is not imposed from without but embedded from within. Years later, Sadie Plant would trace a similar logic in the virality of code: language detached from the body, rerouted through cybernetic systems. In the era of AI, where control manifests less through force than through the modulation of informational flows, one of the passwords is surely Pierre Menard.