At long last, I have posted a preprint draft of my article Prompting Progress: Authorship in the Age of AI, 76 Fla. L. Rev. (forthcoming 2024). You can download it from SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4609687
The Article explains why the Copyright Office’s recent guidance and position denying the copyrightability of AI generated works is wrong. It follows my Washington Post Op-Ed, A terrible decision on AI-made images hurts creators, published earlier this year.
TL;DR version
- The Copyright Office’s current approach to AI generated works is wrong.
- The Office and the federal courts should return to the first principles of authorship under the Copyright Clause and reject the imposition of any artificial requirements—such as the exercise of sufficient control, avoidance of random elements in the creative process, prediction of the final work ahead of time, or dictation of specific results—that restrict authorship or progress in the United States. Nothing in the text of the Copyright Clause restricts authorship to such a rigid task.
- The proper test of authorship examines whether the person contributes a minimal level of creativity in the origination of the work, which may be satisfied simply by an individual’s selection or arrangement of elements in the work. The requisite level is, as the Supreme Court recognized, “extremely low.” There is no constitutional restriction imposed on the process of authorship, much less any expectation of a set path that all authors must take in lockstep. Given this low threshold, or the bare minimum, Kashtanova’s selection of elements in the images, including the aspects of their non-binary character, Zarya, consistently depicted in 24 images involving different scenes, easily satisfies the test. But the scope of copyright for works whose originality lies in its selection is thin, only protecting against identical copies. Under this approach, (i) works created with a minimal level of creativity of the creator, such as in the selection or arrangement of elements, are eligible for copyright, while (ii) works created autonomously and randomly, without human input in the creation, are uncopyrightable.
- The Office’s rigid view of authorship is not evidence based or supported by any empirical studies showing such a rigid view describes what authors do in practice. In fact, studies show the importance of serendipity, randomness, and lack of control in the creative process.
- The Office’s rigid view of authorship gives insufficient attention to the Supreme Court’s recognition of a principle of judicial avoidance of making case by case determinations of whether a work qualifies for copyright outside the narrowest and most obvious limits. Judicial avoidance serves the First Amendment in preventing the government from making content-based discrimination that may result in subjective or arbitrary decisions.
- Photography provides an important historical example in which critics argued that photographs didn’t involve authorship, but instead, were merely mechanical creations of cameras, similar to the objection against AI generated works today. But, starting with Burrow-Giles, the federal courts correctly rejected that narrow view and recognized that virtually all photographs involve authorship, at least based on a selection or arrangement of elements. That same test should apply to prompt-engineered images.
Author’s note
For the record, I have the utmost respect for the U.S. Copyright Office. They do incredibly important work for our copyright system. I have praised their outstanding work on authors’ resale royalties by then Register Maria Pallante (and am probably in the minority among academics in doing so). On this issue, I think they are headed down the wrong path. My article and comments are offered in the spirit of healthy critique and deliberation to get the correct legal approach. Their notice of inquiry suggests they are open to revising their view. And, if they don’t, the federal courts will be reviewing it in due time.