For most of the history of copyright law, the question of who counted as a work's author answered itself: whoever held the pen, the brush, or the camera did. The law occasionally quarrelled over edge cases -- whether a photograph showed enough deliberate choice of angle and light to count as authored rather than merely mechanically recorded -- but the basic assumption survived intact for two centuries: a human being made a specific set of choices, and the resulting work reflected those choices closely enough that crediting the human as its author was not a fiction. Generative software complicates this assumption in a way the old edge cases never quite did, because it introduces a genuine gap between what a person asked for and what the machine actually produced.

Consider two people who each end up with a striking photograph of a sunset. The first walks to a particular hillside, waits for a particular hour, and adjusts a particular exposure setting to capture a particular arrangement of cloud and colour; a different choice at any of those steps would have produced a visibly different photograph, and the final image is, in a meaningful sense, the accumulated residue of those choices. The second types a nine-word description into an image generator and receives, among the first handful of results, an image striking enough to keep. Change the nine words only slightly and an entirely different image might have resulted -- not a variation on a theme the second person controlled, but something the system arrived at through a process the second person neither performed nor, in any detailed sense, directed. Both people end up pleased with an image. Only one of them can point to a specific chain of choices that explains why the image looks exactly as it does rather than some other way.

Defenders of treating such generated images as ordinary authored works reach quickly for the tool analogy: a camera is also a machine that does most of the actual work, and no one doubts a photograph is authored by the photographer rather than by the camera. The analogy has real force but a real limit. A camera's unpredictability is narrow and bounded -- light behaves according to physics the photographer already understands well enough to anticipate, and a competent photographer can explain, after the fact, exactly which choice produced which visible effect. A generative system's unpredictability is neither narrow nor fully explicable even to the engineers who built it; the same short prompt, run twice, can return two images with little in common, and no one, including the system's own designers, can specify in advance exactly which pixel arrangement a given prompt will return. The tool analogy quietly assumes a controllable tool, and the entire question is whether that assumption still holds.

Courts and copyright offices, faced with this gap, have mostly declined to resolve the deeper philosophical question of whether a machine can be genuinely creative, and have instead drawn a narrower, more administrable line: copyright protection attaches only to the portions of a work that reflect meaningful human creative control, and a work produced by typing a prompt and accepting the first output that comes back typically fails that test, while a work built by generating many outputs, selecting among them, and combining or substantially editing the results may pass it, because the human's curatorial choices, not the machine's generative process, become the thing the law is actually crediting. This does not require anyone to decide whether the machine was creative in some deeper sense; it only requires deciding whether enough specifically human judgment entered the final result to justify treating a human as its author.

What this manoeuvre reveals is that authorship, legally speaking, was never really a question about where creativity metaphysically resides. It was always a question about whom the law should treat as responsible for a work, and whom it should reward with the resulting incentive. A rule granting no protection to purely machine-generated output does not depend on any settled answer about whether the machine possesses genuine creative agency; it depends only on the practical judgment that extending exclusive rights to whoever merely presses a button, with no meaningful choice behind the pressing, would reward too little human contribution to serve the purpose the right was built for in the first place. The old, comfortable assumption -- that authorship and human choice were simply the same fact under two names -- has not survived. What has survived, adapted rather than abandoned, is the practical question the assumption was always standing in for.