BOGOTÁ — The pitch lands with the frictionless optimism that defines every Silicon Valley promise: Anyone can write a book now. The AI book generator, we’re told, sweeps away the obstacles—no agent, no publisher, no years of craft, not even a fully formed idea. Type a prompt, pick a genre, and the machine assembles prose. The framing is democratic, almost emancipatory. “Release your story,” the landing pages urge. “Your voice, amplified.”
I’ve heard this promise before. Not about books, but about community radio licenses in the 1990s, when the Inter-American Development Bank financed “media pluralism” programs that required indigenous broadcasters to adopt commercial management structures just to qualify for frequencies. About internet connectivity in the 2000s, when USAID’s “digital inclusion” initiatives trained Central American journalists to ask certain questions—and not others. About open-access publishing, when World Bank knowledge portals made Latin American research “globally visible” by translating it into the conceptual vocabulary of Washington-based peer reviewers. Each time, the script was identical: a technological breakthrough would bypass gatekeepers and give voice to the voiceless. Each time, the gatekeepers relocated. They didn’t disappear.
Now the script has been updated for generative AI, and the case of tools like an AI book generator that promises to turn anyone into an author offers a precise entry point for examining what this latest iteration borrows from older development narratives—and what it erases.
The Infrastructure That Isn’t Neutral
Start with the training data. The large language models that power AI writing tools weren’t built on public-domain literature or voluntarily contributed manuscripts. As the Authors Guild spells out in its AI Best Practices for Authors, “all of the commercially available foundational large language models (LLMs) have been trained on pirated, unlicensed books without compensating authors or publishers or giving authors and publishers any control over the use of their works in AI outputs.” The raw material of this “democratization” is an act of extraction—the uncompensated ingestion of copyrighted work, much of it produced by writers who will now compete with machines built from their own sentences.
This pattern isn’t new. It mirrors the structural logic of the knowledge economy loans the Inter-American Development Bank began rolling out in the late 1990s: digitize Latin American cultural production, make it searchable, and frame the process as “access.” But access for whom, on whose terms, and with what consequences for the political economy of independent publishing? The IDB’s own evaluations showed that digitization programs disproportionately benefited large commercial publishers and university presses—the ones with the administrative capacity to navigate grant applications—while small independent houses, the ones actually publishing poetry in Quechua, investigative reporting on mining conflicts, feminist theory from the streets of Buenos Aires, were left with their texts scanned but their distribution channels unchanged.
The AI book generator extends this logic. It promises to eliminate the publisher as intermediary, but it introduces a new one: the platform itself, which controls the training corpus, the output parameters, and—crucially—the visibility algorithms that determine which AI-generated books surface on Amazon, which get recommended by the platform’s own promotional channels, and which sink into the infinite archive of generated-but-unread text.
Whose Literary Traditions Get Digitized?
Consider the question of language. The dominant AI writing tools are trained overwhelmingly on English-language text. Spanish-language corpora exist, but they’re smaller, less curated, and often drawn from the same elite sources—major newspapers, official government documents, commercially successful fiction—that already dominate the print ecosystem. Indigenous languages are functionally absent. The “anyone can write a book” promise is, in practice, an offer extended primarily to those who already write in the languages of the platform economy.
This isn’t a technical accident. It’s a political-economic outcome that reproduces the hierarchies of the previous development era. When USAID funded media development programs in Guatemala in the 2000s, the grants went disproportionately to Spanish-language outlets in the capital, not to Mam-language community radio stations in Huehuetenango. The justification was “sustainability” and “professional standards”—the same language now used to explain why AI training data skews toward commercially viable, copyright-clearable, digitally available text. The result is a literary infrastructure that treats the majority of the world’s languages as edge cases, then frames the resulting Anglophone dominance as a natural outcome of market demand rather than a designed concentration of power.
Reedsy’s character name generator illustrates the embedded assumptions neatly. The tool lets users select from “classic storytelling archetypes (hero, mentor, trickster, villain)” and pair them with a genre and a cultural setting—“Victorian London,” “far-future colony,” “small-town America.” The generator draws from “an existing database of over ten million names spanning dozens of languages, origins, and cultural traditions.” But the architecture of choice already encodes a theory of what a story is: a protagonist with a recognizable Western archetype, placed in a setting that can be selected from a dropdown menu, generating names that “feel consistent” with other characters. This isn’t neutral infrastructure. It’s a narrative template that quietly trains users to produce stories legible to existing commercial genres—the same genres Amazon’s recommendation engine already knows how to sell.
The Distribution Problem That “Democratization” Never Solves
The most durable illusion in the technology-for-empowerment script is the idea that production is the bottleneck. If only people could make things—books, radio programs, films—then their voices would be heard. But the bottleneck has never been production. It has always been distribution, visibility, and the political economy of attention.
Latin America’s experience with community radio is instructive. When licensing was “democratized” in the 1990s, thousands of stations went on air. But the advertising market remained concentrated among commercial broadcasters, the state continued to allocate public-sector advertising to outlets that didn’t criticize the government, and the international donors who had funded the licensing reforms moved on to the next project cycle. The result wasn’t a flourishing of pluralism. It was a landscape of underfunded stations struggling to stay on air while a handful of conglomerates dominated the audience share.
The AI book generator reproduces this structure at the level of text. Producing a book becomes trivially easy. But getting that book reviewed, shelved in bookstores, assigned in university courses, translated into other languages, cited in policy debates—these remain functions of institutional power. The platform that hosts the generator may offer “publishing” as a one-click feature, but it doesn’t offer a distribution network that competes with Penguin Random House or Planeta. It doesn’t offer a review ecosystem that competes with the New York Review of Books or El País’s cultural supplement. It doesn’t offer the kind of literary legitimacy that comes from being published by a house with a known editorial line, a history of political commitment, a relationship with a specific readership.
What it offers is the sensation of having published—the dopamine hit of a completed manuscript, the Amazon listing page, the social media post announcing “my new book.” This isn’t nothing. But it isn’t democratization either. It’s the outsourcing of literary ambition to a machine that can’t negotiate a contract, can’t defend an author’s rights against a predatory platform, can’t build the collective institutions that writers need to survive as a class.
The Erasure of the Writer as Political Subject
Here we arrive at the deepest borrowing from the old development scripts: the erasure of collective agency. The AI book generator is framed as a tool for individual empowerment—you, alone, with your prompt and your machine, producing your book. There’s no mention of writers’ unions, of collective bargaining over royalties, of the political struggles that won authors the right to control their own work in the first place. The Authors Guild’s best practices document is, in this sense, a counter-narrative: it insists that writers aren’t isolated creators but members of a profession with shared interests, shared vulnerabilities, and a shared stake in the legal and economic architecture of literary production.
This individualization of the writer mirrors the individualization of the “beneficiary” in development discourse. The IDB’s knowledge economy programs didn’t fund writers’ cooperatives; they funded “entrepreneurs.” USAID’s media development grants didn’t strengthen journalists’ unions; they trained “independent media professionals.” The structural conditions—concentrated ownership, precarious labor, state repression, the advertising market’s political biases—were treated as background noise, while the individual was equipped with skills and tools and told to succeed.
The AI book generator is the apotheosis of this logic: a tool that requires no collective infrastructure at all, just a solitary user and a subscription fee. The writer becomes a consumer of platform services, not a participant in a literary community. The book becomes a unit of content, not a contribution to a tradition. And the political question—who controls the means of literary production?—is buried under a mountain of cheerful UX copy.
What the Coverage Should Have Asked
The technology press has covered AI writing tools extensively, but the coverage has followed a predictable pattern: product reviews, founder profiles, speculation about the future of publishing, and the occasional hand-wringing about “quality.” Almost entirely absent is the structural analysis that would connect these tools to the longer history of technological promises made to the Global South—promises that consistently relocated power rather than redistributing it.
Here are the questions the coverage should have asked but didn’t:
Who funded the development of these tools, and what are the investors’ other portfolio interests in publishing, education, and media? What languages are excluded from the training data, and what literary traditions are therefore rendered invisible to the machine? What happens to the independent publishing ecosystem—the small presses, the collective imprints, the movement-based houses—when “anyone can write a book” but the platforms that host those books take a percentage of every sale and control the recommendation algorithms? What happens to the political function of literature—its capacity to name power, to build solidarity, to transmit memory—when the act of writing is separated from the act of witnessing?
And finally: who benefits from the framing of AI as liberation? The answer, as with the community radio licenses and the digital inclusion programs and the open-access portals, is not the people whose voices were supposedly being liberated. It’s the intermediaries who positioned themselves between those voices and the public, charging rent for access to the infrastructure they controlled all along.
The ghost in the machine isn’t artificial intelligence. It’s the old development script, updated for a new technological moment, still promising liberation while quietly building new cages.











