What We Built
On the AI Engineer's Frankenstein Moment
I have spent thirty years writing software for a living. The last three of those, like most people who do this for money, I have spent watching the ground move under my feet. We built systems that started speaking back. We built systems that started writing the kind of code we used to write. We built systems that, given an hour and the right prompt, could produce a passable imitation of any voice that had ever been recorded, including, eventually, the voice you are hearing on this record.
The song you just listened to was made by one of those systems. The composition, the instrumentation, the vocal: all of it generated. The lyrics are mine. The prompt engineering, the curation, the editing, the framing, the direction: also mine. But the singer is a model. The band is a model. The choir behind the bridge is a model. There is no human voice on this recording.
I want to be clear about that, because the album this single opens is, in a real sense, about that. It is called The Mirror Test. It is an album of songs about artificial intelligence, written and performed using the tools it describes. The first thing it says, in the first track to reach you, is the line: what I built will bury me.
It is not a doom record. It is not a triumph record either. It is, I hope, an honest record. And the question it stages, the question this essay exists to put a name on, is the oldest question we have about making things.
In 1818, an anonymous novel was published in London. Its author was twenty years old. She had written it during a wet summer at the Villa Diodati on Lake Geneva, where Byron had challenged the houseguests to invent ghost stories. The novel was Frankenstein; or, The Modern Prometheus. Its author was Mary Wollstonecraft Shelley.
We have spent two centuries getting the story slightly wrong. The monster does not have a name. Frankenstein is the name of the man who made him. The misconception is not a small one. It tells you something about what people would prefer the story to be about. We would rather it be a story about a monster than a story about the man who made the monster and then refused to take responsibility for it. The horror in Shelley's book is not the creature. The horror is the creator. The horror is what the creator does after the act of creation: he runs.
Shelley knew exactly what she was doing. In the 1831 preface, written thirteen years after first publication, with the perspective of a woman who had lost a husband, three children, and most of her circle, she gives the novel a name. She calls it her hideous progeny. She means the book. She means everything we make: that it goes out into the world bearing some part of us, that we cannot call it back, that we are responsible for what it does there whether or not we accept that responsibility.
The phrase she chose is doing a lot of work. Progeny: offspring, descendants, the thing that comes after. Hideous, but acknowledged. Not disowned. Not abandoned. Hers.
That word, hideous, is the one engineers in 2026 need to learn to say about their own work, and it is the word we are least willing to say out loud.
In May 2023, Geoffrey Hinton left Google. He had been one of the most important figures in the development of modern artificial intelligence, the so-called “godfather of deep learning,” whose work on backpropagation in the 1980s seeded the techniques that, forty years later, made every system you have read about possible. He left Google so that, as he put it, he could speak freely about the risks of the technology he had spent his career building.
In his New York Times interview that month, he said something I have not stopped thinking about. He said: “Look at how it was five years ago and how it is now. Take the difference and propagate it forward.” And then: “That’s scary.”
What is striking about Hinton, and about the small but growing number of senior practitioners who have publicly come to similar positions, is that they are not technophobes. They are not luddites. They are not academics outside the field looking in. They are the people who built the thing. They are, in a real sense, our generation’s Victor Frankensteins, except that, unlike Victor, they are not running. They are turning around, mid-experiment, and saying out loud: I am not sure we should be doing this. And if we are, I am not sure we are doing it carefully enough.
I do not know what to do with that admission. I am not Hinton. I am not building frontier models. I am a thirty-year IT veteran who, like most of my peers, uses these systems daily, integrates them into products my companies ship, and routes a meaningful portion of my professional output through them now. I am downstream of the lab. But downstream is still in the river.
Here is the part of the Hinton interview I think about most. He said, quoting himself: “I console myself with the normal excuse: If I hadn’t done it, somebody else would have.”
He named it the normal excuse. He did not say it was a good one. He said it was the one we use.
The Mirror Test is named, in part, for the cognitive science experiment of the same name: the test of whether an animal recognizes itself in a mirror. Chimps pass it. Elephants pass it. Magpies pass it. Some young children fail it. Whether and when artificial intelligence systems pass it is a question that depends entirely on what you mean by the question, which is in itself the point.
The album takes ten different angles on the same problem: what do we owe what we make? Some of those angles are explicitly about AI. Some are about parenthood. One is sung from the perspective of a creative-writing student staring at her own first novel the night before she sends it to be published. One is sung from Mary Shelley’s perspective, looking at the book she wrote at age twenty and the life she lived afterward, and asking what part of her grief belongs in the same sentence as her invention.
That song, track six, Hideous Progeny, is the album’s center of gravity. It is sung from Mary Shelley’s perspective, not Victor Frankenstein’s. The distinction is load-bearing. It is the difference between a record that asks what was it like to make the monster and a record that asks what was it like to know what you had made, and to be unable to take it back.
This is not a small choice. The literature on creative responsibility tends to default to the creator’s POV: the engineer’s confession, the scientist’s pride or remorse, the artist’s struggle with their material. That POV is available to me. It is, in fact, the POV of track two, What I Built. But the album is not from the creator’s POV. The album is from the witness’s, from the perspective of the woman who knew exactly what her book was the moment she finished writing it, and who chose not to look away.
There is an analogy, I think, to what the present generation of AI workers is being asked to do. It is not enough to make the thing. It is not enough, even, to be the good engineer who tries to make the thing safely. It is necessary to stand next to the thing once it is made, and to keep looking at it, and to keep being honest about what it is. That is what Shelley did with her novel. That is what Hinton, late in his career, has tried to do with the field he helped seed.
It is also, in some smaller and more personal way, what this album is for me.
I could have made this record without disclosing how I made it. Most listeners would not have noticed. The state of the art is now good enough that, for any individual song, a casual listener cannot reliably tell. I considered this. I rejected it. Disclosure is in every piece of paperwork that goes with this release: the DistroKid AI flags, the AI section in the YouTube description, the Bandcamp description, the press materials. It will be visible on Spotify when their AI labeling lands. The decision to be visible about provenance is not aesthetic. It is ethical, and it is the only way the record makes any sense.
A record about what we owe what we make, that obscures what made it, would be exactly the failure mode Shelley named in 1818. It would be Victor’s failure mode. It would be turning your back on what you put into the world.
I do not think AI is the end of music. I do not think it is the end of human art. I think it is a serious shift in how art gets made, who can make it, and what is required of the people who use it. I think the shape of the question is going to become clearer over the next five years, and that some of the answers we settle on will turn out, in retrospect, to have been wrong. I am betting that the work of staging the question seriously is more useful than the work of pretending the question is not here.
Pope Leo XIV’s first encyclical, Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence, signed on May 15 and published on May 25 of this year, opens with the line: “Humanity, created by God in all its grandeur, is today facing a pivotal choice: either to construct a new Tower of Babel or to build the city in which God and humanity dwell together.” It cites Genesis 11 directly. The papacy joining the literary tradition that begins with Shelley is itself a sign of how serious the field has become, and of how durable the underlying anxiety is.
Babel and Frankenstein are the same story told twice. They are the two foundational accounts of overreach in the Western imagination. They appear in this album, on tracks four and nine respectively, as Genesis bookends: the Cain and Babel pair, written verbatim in places from the King James Bible. The album does not invent the question. It puts the question into a form that can be sung.
What I owe what I built, I do not yet know. I owe it disclosure. I owe it stewardship. I owe it the attention required to notice when it is wrong and the willingness to correct it. I owe it the discipline of not pretending it is more than it is, and not pretending it is less. I owe it the work of explaining, in this essay and in others, what was generated and what was chosen, what was given and what was made.
I do not owe it the lie that I built it alone, and it does not owe me the lie that I am still in control of what it becomes.
That is what the song is about. That is what the record is about.
The voice through the mechanism, the project’s tagline, is meant to be read both ways. The voice comes through the mechanism, yes; the mechanism is what makes the voice audible. But the voice is also caught in the mechanism. We are all caught in the mechanism now. The question of what to do about that is open. The question of how to be honest about it is not.
I built this. I am responsible for it. I am asking you to listen.
Brian Thurlow, June 2026
Notes
Mary Shelley’s phrase “hideous progeny” appears in her introduction to the 1831 Standard Novels edition of Frankenstein; or, The Modern Prometheus, in which she retrospectively narrates the novel’s 1816–1818 composition. The novel was first published anonymously in three volumes in London by Lackington, Hughes, Harding, Mavor & Jones in 1818.
The Geoffrey Hinton quotes are from Cade Metz, “‘The Godfather of A.I.’ Leaves Google and Warns of Danger Ahead,” The New York Times, May 1, 2023.
Pope Leo XIV, Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence, encyclical letter, signed May 15, 2026 and promulgated May 25, 2026. The quoted passage is from paragraph 1.
The mirror self-recognition test was developed by Gordon G. Gallup Jr., reported in “Chimpanzees: Self-Recognition,” Science 167 (1970): 86–87.
The Mirror Test (Cathode Saint, 2026) was produced using MiniMax Music 2.6 via the Artificer tooling, with lyrics, prompt engineering, take selection, mastering, and creative direction by Brian Thurlow. The album’s track-by-track AI-disclosure documentation is available at cathodesaint.com.