Jonathan Swift is remembered as a satirist, a moralist, a literary provocateur. He is not typically remembered as a prophet of artificial intelligence. Yet a close reading of Gulliver’s Travels, published in 1726, reveals something that should give pause to anyone who assumes that large language models are a purely modern invention of thought.
The argument is not metaphorical. It is structural.
The Machine in the Academy of Lagado
The third voyage of Gulliver’s Travels is the least celebrated of the four. While the Lilliputians and the Brobdingnagians have entered popular culture, the visit to the Academy of Lagado in the country of Balnibarbi remains largely overlooked. This is where Swift’s anticipation of AI becomes most explicit.
In the Academy, Gulliver encounters a professor overseeing a large mechanical frame, roughly 20 ft by 20 ft, operated by 40 assistants. The device consists of small wooden blocks, each covered with a piece of paper bearing a word from the local language. The blocks are connected by wires and fitted with 40 handles or cranks. When the assistants turn the cranks, the arrangement of words shifts randomly. Scribes then record any intelligible combinations that emerge. The process runs six hours a day.
The professor’s ambition is explicit: this machine would allow “the most ignorant person” to produce works in philosophy, poetry, politics, law, mathematics, and theology, without any study or native genius. The volumes already produced consist of broken sentences, which the professor plans to assemble into what he calls “a compleat Body of all Arts and Sciences.” He asks only for adequate funding and 500 additional frames.
Strip away the 18th-century mechanics and the satirical framing, and the architecture is recognizable. A system operates mechanically on language. It produces outputs by recombining existing words. It promises to democratize intellectual production. It requires significant resources to scale. The human role is reduced to turning handles and filtering results.
What Swift Got Right, and What He Missed
The parallel to large language models is not perfect, and the source is honest about where it breaks down. Swift’s machine lacks the capacity to respond to a specific question from an individual user. That interactive dimension, the prompt-and-response loop that defines the ChatGPT experience, is absent from the Lagado device.
But that absence is less significant than it might appear. A substantial portion of what AI systems do today is not question-answering. It is content generation: producing text from a corpus of accumulated language, applying statistical patterns to generate new material. That is precisely what the Lagado machine does, in its cumbersome, hand-cranked way.
What Swift captures with particular precision is the underlying logic of the operation. The machine does the work. The operator needs no expertise. The output is mechanically derived from language rather than from understanding. And the system produces a great deal of noise alongside whatever signal it manages to generate. In Lagado, scribes filter out the gibberish. This is what most coverage of AI’s limitations misses: Swift identified the hallucination problem three centuries before the term existed. His scribes are the quality-control layer that modern AI deployment often lacks.
Swift’s real target, according to the analysis grounded in the text, was the Royal Society of London and its culture of ambitious, sometimes absurd experimental projects. The satire is aimed at the gap between the grandeur of the promise and the poverty of the actual output. That gap has not closed.
Why a 300-Year-Old Satire Still Lands
The reason Swift’s Lagado episode resonates is not that he predicted the technology. It is that he identified the ideology behind it: the belief that mechanical operations on language can substitute for understanding, and that access to such a machine confers intellectual capability on anyone who uses it, regardless of their knowledge or judgment.
This is the passive model of intelligence that the source identifies as the common thread between the Lagado machine and contemporary LLMs. The user does not need to think. The system thinks for them, or appears to. The promise of “remarkable gains in knowledge and productivity,” as the source puts it, rests on this assumption.
Swift was skeptical of that assumption in 1726. The skepticism is worth inheriting. AI systems genuinely augment human capacity in ways that matter: they handle volume, surface patterns, accelerate drafting, and free up cognitive bandwidth for higher-order judgment. But the Lagado critique points to what gets lost when the human in the loop stops filtering, stops questioning, stops reading the output with the critical attention of a scribe.
The professor in Lagado was proud of his volumes of broken sentences. He had a plan to make them cohere. He just needed more funding and more frames.
In Short
Jonathan Swift’s Academy of Lagado, from Gulliver’s Travels (1726), describes a mechanical device that recombines words to generate texts on any subject, operated by 40 assistants working six hours a day, promising universal access to knowledge without study or expertise. The structural parallel to large language models is precise enough to be instructive. What Swift adds, beyond the technical sketch, is a critique: the passive model of intelligence that underpins such systems produces noise alongside signal, and without active human filtering, the noise wins. Three centuries later, that warning has not been answered.
Based on reporting from The Conversation.