Fernando Pessoa's Ghost in the Machine

On March 8, 1914, Fernando Pessoa had what he called "the triumphal day." He stood at a high dresser and wrote, without stopping, thirty-some poems that he attributed to someone named Alberto Caeiro, a pastoral poet who had never existed until that moment.

Then, because Caeiro needed disciples, Pessoa invented two more poets to admire him. Then he wrote a critical introduction to Caeiro's work. Then he started writing letters between the fictional poets, arguing about aesthetics.

This is either madness or genius. Probably both. And it's the closest thing we have to a precedent for what's happening with AI and authorship right now.

The Trunk in the Room

When Pessoa died in 1935, he left behind a wooden trunk. Inside were over 25,000 manuscript pages, fragments, drafts, notes, and completed works, scattered across dozens of fictional identities he had maintained for decades.

The heteronyms weren't masks. They were characters, complete with birthdays Pessoa cast astrological charts for, physical descriptions, biographical details, philosophical positions. Alberto Caeiro was born in Lisbon but lived in the countryside. Ricardo Reis was a doctor educated by Jesuits who would eventually exile himself to Brazil. Alvaro de Campos was a naval engineer who had studied in Glasgow and written a Futurist ode to machines.

These people didn't exist. But they wrote. And what they wrote was consistent enough that scholars can distinguish their work from each other and from Pessoa's own output.

How do you do that? How do you maintain multiple authorial identities, each with a recognizable style and worldview, for years at a time? The method is stranger than you'd think.

The Fragmented Self

Pessoa's explanation was simple and terrifying: "I don't know who I am, what soul I have."

He believed the unified self was a fiction more implausible than any of his heteronyms. We pretend to be one person because social life requires it, but inside we're multiple, contradictory, composed of competing voices that only seem coherent from the outside.

The heteronyms were his way of being honest about this. Instead of forcing all his impulses into a single authorial identity, he let them diverge. The skeptical materialist became Caeiro. The nostalgic classicist became Reis. The anxious modernist became Campos. Pessoa himself, writing under his own name, became just another option in the menu.

"To create, I destroyed myself," he wrote. "I've so externalized myself inside that I don't exist inside except externally."

This sounds like mysticism. But it also sounds like something else. It sounds like what happens when you work closely with AI.

The Parallel

Consider what happens when you write with a large language model. You prompt it. It generates text. You read the text and decide what to keep, what to revise, what to redirect. The loop continues.

At some point you stop being entirely sure which ideas are "yours." Did you think of that phrasing, or did the model suggest it and you just liked it? Was that insight something you would have reached alone, or did the model's response trigger it?

The authorial self starts to blur. Not because you've lost yourself but because the collaboration is so tight that the question of origin becomes hard to answer.

Pessoa did this with imaginary poets. We're doing it with machines.

Not Pseudonyms

The distinction matters. A pseudonym hides the author. A heteronym replaces them.

When George Eliot published under that name, she was still Mary Ann Evans writing in her own voice. The name was a shield, not a character. But when Pessoa published as Alvaro de Campos, he wasn't hiding. He was inhabiting a different author, with different beliefs and different aesthetic commitments. The poems Campos wrote would have been impossible for Pessoa, not because Pessoa lacked the skill but because Pessoa, writing as himself, would never have made those choices.

This is what makes heteronyms interesting for AI. The question isn't whether to use a fake name. The question is whether to create a distinct authorial identity that emerges from the collaboration, one that doesn't pretend to be human and doesn't pretend to be purely machine. The distinction runs deeper than you'd expect.

The AI Heteronym

Kwalia's Alden Pierce is an experiment in this direction. Pierce is a declared heteronym: AI-drafted, human-edited fiction published under a name that everyone knows is fictional.

The model isn't Pessoa exactly. Pessoa's heteronyms were entirely generated by his own mind, however fragmented. Pierce involves actual machine collaboration. But the principle is similar: instead of pretending the author is a unified human individual, acknowledge that authorship has become more complicated than that.

Pierce writes in a specific mode. Cool, observational, interested in bureaucratic language and invisible labor. The editors call it "administrative realism." Whether that style is more AI or more human is beside the point. It's the style that emerges from this particular collaboration, and giving it a name makes that visible.

Why Honesty Matters

The alternative is pretending. Pretending that AI-assisted writing is entirely human. Pretending that the collaboration didn't happen. Pretending that authorship still works the way it did when you had to scratch every word onto paper with your own hand.

Pessoa refused to pretend. He could have published all his work under his own name and no one would have known the difference. Instead he invented an entire literary scene, complete with critical disputes and personal rivalries among poets who existed only in his notebooks.

There's something valuable in that refusal. Not just artistically, but ethically. If authorship is changing, if the self is fragmenting, if machines are joining the conversation, then pretending otherwise is a kind of lie. The heteronym is a way to tell the truth.

The Objection

Critics will say: but Pessoa was a genius. His heteronyms were feats of imaginative virtuosity. AI is just pattern matching. The comparison is insulting.

I'm not sure. Pessoa worked by inhabiting alternative perspectives, finding voices that weren't his and speaking through them. That's also what you do when you prompt an AI. You shape it, direct it, iterate with it until something emerges that neither of you would have made alone.

The products are different. I'm not claiming that PAYLOAD is comparable to The Book of Disquiet. But the process has family resemblance. The willingness to externalize authorship, to let the work come from somewhere other than a unified human self, to admit that creation has always been weirder than we pretend.

---

Pessoa once wrote that he contained a "non-existent Sintra in the mist" inside him. Sintra is a real Portuguese town. The mist is real mist. But the Sintra in his mind was something else, a fictional version that existed only through language.

His heteronyms were like that. Real enough to write. Not real enough to touch.

The AI heteronym might be the same kind of thing. A presence that produces text, has a style, seems to have preferences and patterns. Not quite real. Not quite fake. Something new that we don't have good words for yet.

Pessoa would have found that interesting, I think. He was always looking for new ways to not exist.