The production of creative works by artificial intelligence (AI) provokes many responses—philosophical, cultural, economic, and legal. I have already argued against copyright protection for works created by AI, supporting the longstanding doctrine that copyright rights can only attach to works of human authorship. But one paragraph in a recent article by attorney Adam Adler raises a potentially difficult question as to whether human prompts directing an AI to produce work could ever constitute authorship of the resulting work? Adler writes:
… proponents of AI art don’t have to look very hard to find the required creative contribution. The most prominent AI works are generated through trial and error using specially crafted word prompts. For example, Jason Allen, the winner of the Colorado State Fair, spent 80 hours crafting the prompts he used to generate the art and tested over 900 different prompts before settling on the winner. Given the sensitivity of AI art generators, one could argue that the selection and refinement of prompts (at least as they are used today) involves significant creative work, analogous to placing a camera or framing a shot. And because a human’s prompt selection informs the creation of the entire work, there would not be any obvious way to disentangle the creative and non-creative elements of the work
Whether Adler endorses the view that the prompts in this example should vest James Allen with the rights of authorship in the resulting image, he is probably correct that advocates for copyrightability of AI works will advance this argument. But is the position valid? If I ask a friend to paint a picture of a weeping willow by a brook, my broad description does not constitute even joint authorship in the painting, and that example is arguably no different than my recent playing around with DALL-E 2 writing prompts with an existing painting in mind—Henry Wallis’s “Chatterton” (1856).
Although the results were nothing like the original work (and I am admittedly a novice propter), the image on the left could, eerily enough, be passed off by a would-be forger as an early sketch in the development of the Wallis painting, and it was admittedly astounding to watch these, and other variations appear in a matter of seconds.
But returning to the theme of this post, I maintain that I did not author these images or the other variations output by DALL-E 2. Authorship flows from the creative choices made by a human, and there would need to be a colorable nexus between my prompt writing and the selection and arrangement of the choices made in the work—in this case a visual work—upon its fixation.
Prompts by themselves may be protectable “literary works” under copyright law—not unlike computer code, which can be sufficiently creative while also serving a utilitarian function as a set of instructions. But the potential copyrightability of prompts themselves does not necessarily extend to protection of the resulting work—not even in the case of Allen writing complex prompts into the app Midjourney to produce the visual work he called “Théâtre D’opéra Spatial.”
Neither the 80 hours Allen spent nor the 900 different prompts he tested has any bearing on a potential claim of authorship in the resulting image because copyright does not protect “sweat of the brow.” Copyright also does not protect ideas or concepts; and the incident of copyrightability is agnostic with regard to the author’s intent, message, or methodology. Unless there is a lack (i.e., less than a modicum) of originality in the work, copyright attaches upon fixation, and without consideration as to how or why the work was made. But the Allen example implies a potential difficulty in the doctrine to which Adler alludes in his description.
Given the time and energy Allen spent on the prompts, we can assume he developed a somewhat complex set of instructions, and it is conceivable that there may be a point at which a creative arrangement of prompts could approach a defensible claim of human authorship in the resulting work output by the AI. But it’s tricky, and I am skeptical.
Ordinarily, the moment a work is fixed in a tangible medium, the human’s creative choices may be inferred, credited solely to the human, and the choices need not be explained. In the photojournalist’s image of the factual event, it is longstanding doctrine that the existence of the image itself is evidence that she made sufficient choices (even in a second or two) to meet the “modicum of originality” threshold, and copyright rights are vested in her without asking her to describe or defend the choices made.
AI production may frustrate this doctrine in the near future by providing a reason to ask how a vast amount of work has been produced and about the nature of the human involvement in its production. Then, even if prompts may be protectable literary works on their own, the consideration as to whether this confers authorship to the human in, for instance, a resulting visual work or music work output by the AI implies a case-by-case consideration of copyrightability that would be administrative chaos for the Copyright Office.
It is hard to imagine a generally applicable doctrine that would harmonize the human authorship requirement with a definable nexus between prompt writing and the resulting work, and this suggests that the law must hold that copyright does not attach where an AI has made any (and likely most) of the creative “choices” in the work being claimed.
This view is consistent with the purpose of art and the purpose of copyright—both of which are profoundly human constructs. Neither copyright’s utilitarian origins (i.e., the author must earn a living), nor its civil rights origins (i.e., the product of the author’s mind is naturally his property) has any meaning whatsoever to a machine, just as machine made “art,” in my view, will ultimately mean nothing to humans.
Art without human creators may be decorative, interesting to a point, useful, entertaining, or even conducive to computer science in other contexts, but the products themselves are bloodless in every sense. Making art and engaging with art is one of the most human of all activities—transcendent and spiritual for many—and I have no idea why it would ever be outsourced to computers. One might as well suggest that the Buddhist set his mobile device in front of the alter to chant for him while he does something else with his body, mind, and hands.
When Henry Wallis revealed “Chatterton” at the Royal Academy in 1856, it caused a stir—both because it was considered a masterwork made early in the painter’s career, and because its romantic yet grim subject matter was viewed by many as a comment on the poor treatment of artists. Chatterton’s suicide by arsenic at the age of seventeen was believed to be at least partly the result of his abject poverty due to the failure of publishers to pay him for his writing. His “Last Verses,” found with his lifeless body, contain an 18th century version of the artist who was supposed to live on “exposure” rather than compensation.
Farewell Bristolia's dingy piles of brick, Lovers of Mammon, worshippers of trick! You spurned the boy who gave you antique lays, And paid for learning with your empty praise.
Two and a half centuries later, “Lovers of Mammon” have invested billions in technologies and business models designed to devalue creative work and infringe copyright rights at massive scale. And now, we enter the next phase, when machines are being “trained” on volumes of human-authored works to potentially replace humans in the production of literary works, visual arts, music, and perhaps eventually, performing arts. And, as usual, the technology is advancing apace without any consideration as to whether it can reasonably be called progress.
Illustration by: zdeneksasek
