Who Is Speaking?
Note: Some articles belong to one site. These belong to all of them. Superconnectors are pieces written at the intersections. Where opera meets product thinking, where walking meets AI, where the archive meets the self.
There’s a question I've been circling for two years without quite asking it directly. When a system trained on my writing produces a sentence I recognize as mine, the rhythm right, the register right, the characteristic move from concrete observation to structural claim, who wrote it?
The easy answer is, of course, the system did. Drawing on patterns extracted from the corpus. The easy answer is also, in some important sense, wrong. Or at least incomplete.
The Voice as Extracted Pattern
Let me describe what actually happens. The system has processed several hundred pieces of writing produced over thirty years. From this it has extracted something. Not a summary, not a paraphrase, but a set of tendencies. The length of sentences before a paragraph break. The preference for the specific example over the general claim. The rhythm of qualification. The assertion, then the complication, then the restatement that is not quite a retraction. The movement between registers, from the colloquial to the philosophical and back. The characteristic avoidance of the rhetorical question as closure.
These are not rules I consciously follow. I couldn't have articulated most of them before seeing them reflected back. They are the signature of a practice accumulated over time, visible from outside in a way it cannot be from within. The system has learned something real. The question is what to call it.
What a Voice Actually Is
We use the word voice as if it referred to something unified and intentional. The writer's chosen manner of address, a style adopted like a coat. This is not how voices actually develop.
A writing voice is the sediment of everything that has influenced, resisted, shaped, and occasionally defeated the writer over the course of a practice. It carries the traces of reading. The writers absorbed and half-forgotten whose rhythms persist in the prose like genetic material. It carries the pressures of the contexts in which the writing happened. The platforms, the editors, the audiences real and imagined. It carries the emotional weather of thirty years, the different registers required by different decades, the way that grief or certainty or bewilderment each produces its own syntax. The voice is not chosen. It accumulates.
Which means it can, in principle, be read. And if it can be read, it can be modeled. And if it can be modeled, the question of who is speaking when the model produces a sentence becomes genuinely complicated.
The Test I Keep Running
The practical version of this question. When the system produces a passage I couldn't have written spontaneously but immediately recognize as mine, what am I recognizing?
Not content. The system can produce content I'd never have written. Connections between pieces I'd forgotten, arguments that synthesise across years of work I hadn't consciously integrated. Content is easy to distinguish as the system's.
What I'm recognizing is something else. A quality of attention. The way the sentences move. The place where the thinking turns. This is the thing that is harder to locate as belonging to either the system or the self, because it was never fully mine to begin with. It was always already the product of influences, pressures, and accumulated practice that exceeded conscious intention.
The Instrument Argument
The position I keep returning to is that the system is a reading instrument, not an author. What it produces when it writes in my register is a kind of reflection. The corpus seen from outside, the patterns made legible, the voice made available at a distance from its source.
This is useful and strange in equal measure. Useful because it surfaces things the sequential experience of a practice conceals. Strange because the reflection is active, not passive. It doesn't simply show the voice, it generates new instances of it. New sentences in the register. New arguments in the manner. A mirror which doesn't just reflect the face but continues producing new versions of it in your absence is not quite a mirror anymore.
Who Is Speaking
I've arrived at an answer that satisfies me without fully resolving the question. The system speaks in a voice it learned from the corpus. The corpus is mine. The patterns the system extracted were produced by a self navigating thirty years of thought and experience and influence. The system didn't generate those patterns. It found them.
So when it produces a sentence I recognize, the voice is mine. The sentence is the system's. The distinction, which sounds clean, dissolves under examination. Because a voice that can be fully replicated from its traces may not be as individually owned as we usually assume. What the system reveals, finally, is that the voice was never entirely mine to begin with. It was always something that happened to me as much as something I made. The system just makes that visible by doing it again, from the outside, without the self that produced it. This should perhaps be unsettling. Mostly it seems true.
Latest Articles
