The Non-Copyrightability Doctrine for Generative AI
This past February, the U.S. Copyright Office rescinded parts of the copyright registration it had initially extended to Kristina Kashtanova for the comic book, Zarya of the Dawn. The Office had learned that the imagery was generated by the AI system Midjourney, and was therefore “not the product of human authorship,” though it allowed copyright for the text and layout, which were the author’s. I believe the Copyright Office made the right decision in enforcing what could be called the “non-copyrightability doctrine” of generative AI. Understanding why can guide future AI regulation.
As a form of intellectual property, copyright is governed by laws and customs that, at least in much of the modern world, trace their underpinnings to John Locke’s theory of property rights. Locke asked, If natural resources are a God-given bounty that start off “in common,” how can individuals legitimately claim to own specific things?
In what is perhaps his most quoted excerpt, Locke’s answer was this: “The labour of [a person’s] body, and the work of his hands, we may say, are properly his. Whatsoever then he removes out of the [commonly owned] state that nature hath provided, and left it in, he hath mixed his labour with, and joined to it something that is his own, and thereby makes it his property.” So, investing labor — one’s effort — is what justifies initial ownership, after which property can be used, destroyed, traded, or given away as the owner wishes. In the classic example, if you made the effort to explore unowned land, staked out a claim, and then farmed it, then you should own the land and the harvest.
Arguably, Locke was merely articulating an intuition embedded deep in Western conceptions of property, and which has since gone global by way of the European Enlightenment. Our conceptions of who “deserves” material benefits are deeply tied to our sense of their proportional effort. And property, in turn, incentivizes effort. By giving creators ownership of their effort, intellectual property is believed to foster innovation.
The crux with generative AI is that there isn’t much labor involved, at least by the human user of it, because the current generation of AI tools enables the effortless creation of text, imagery, music, and video. If someone didn’t have to put in much effort to create content, do they deserve ownership of it?
In fact, arguments for either side of the issue rest on the question of effort. In her response to the Copyright Office, Kashtanova appealed to the fact that she went through a process involving “hundreds or thousands of descriptive prompts” to Midjourney. A similar claim was made by Jason M. Allen, who won a blue-ribbon prize at the Colorado State Fair last year for a Midjourney-generated artwork; he said he tried over 900 prompts over 80 hours before hitting on the one that resulted in his prize-winning work. In both cases, a significant effort was expended.
But, was it the right kind of effort? As the Copyright Office and many professional artists have argued, their efforts were not of a kind consistent with authorship of visual imagery. The Copyright Office’s reasoning is subtle and rests on multiple points, but a key rationale was that there is a great distance between the input text prompts and the output imagery of a generative AI like Midjourney. Text prompts alone don’t make the person “the inventive or master mind” behind a graphical work. The Office formalized what feels like common sense: If the AI drew it for you, then you’re really not the artist, even if you guided the AI at some high level, or Photoshopped it a bit afterward.
Arguments about this decision have raged online, with some insisting that generative AI is just another tool, in the same way that pre-AI Photoshop was a tool — and people should own what they make with mere tools. Some draw an analogy to photography, which was initially not accepted as art, because it only took a button click to make an image. Over time, it was acknowledged that there was something to composition, to the choice of camera settings, to being at the right place at the right time — all requiring skilled effort. Maybe AI prompting will go that way, too, as we discover that some people have a special ability to elicit the best from generative AI.
But, I wouldn’t bet on it. Generative AI will only get better and better, and the realm of what it can’t do will only shrink. We’re in the midst of an AI gold rush, after all, and tech companies are chasing after machines that will deliver the best for the asking, no matter who the user is. In effect, generative AI results in whole new bounties of creative content that arguably should start “in common” — collectively owned by all of humanity — until, as Locke might suggest, someone mixes their labor with it. The converse results in a ridiculous world in which anyone with access to a generative AI system could churn out creative masterpieces by the microsecond, and claim to own them all.
A wonderful side-benefit of the non-copyrightability doctrine is some measure of economic protection for human creators. If graphics produced by generative AI can’t be copyrighted, that leaves room for a market of copyrightable work by human illustrators. Content that can’t be copyrighted would be much less profitable for publishers and distributors because they can’t prevent rampant copying. Publishers would continue to focus on human-authored work that they could hope to profit from. Hollywood studios wouldn’t bother releasing movies that didn’t involve human writers and human actors, because content produced by machine could be legally “pirated.” Of course, with the cheap availability of AI-generated content perfectly tailored to one’s tastes — copyrightable or not — consumers might never again have to purchase creative content at all, so human creators might lose their jobs, anyway… but, that’s a question for another day.
