The Real Problem With AI Copyright.
Courts will need to do a better job explaining why species-ism is necessary to protect human creativity from AI.
Not copyrighted: “A recent entrance to paradise,” by the AI image generator DABUS, 2012
Update: The Atlantic has published a great article on how copyright is holding back the job-pocolypse.
Update: SCOTUS has declined cert.
Update: Can the fact that AI memory appears to be entirely separate from AI reasoning help solve the copyright/plagiarism problem?
Protecting the Jewel in the Crown.
Copyright law (and trademark/patent/etc) are some of the crowning jewels of human civilization. Think about it. In the before times, you would design a new way to carry lots of stuff, call it a “basket,” start selling it…only to watch as a bunch of jerks in your tribe started selling baskets, too, without your permission. Then, your chief would claim that he, not you, designed the basket. I mean, why even bother?
These days, when you invent something, you can at least make a couple million dollars selling the idea to some giant company (assuming you own your invention as per your employment contract). Then, you can watch as that company makes billions off your invention, but hey, at least you got paid and have a cool story to put on your Wikipedia page and a couple of awards in your office. That’s patent law. That’s progress. No, seriously, it really is. Copyright law is currently one of the only things standing between we humans and the job-pocalypse.
How to punt like a judge.
Yet, to date, despite its humongous importance, the decisions coming out of the courts on AI and copyright have been disappointing . Most judges have found ways to avoid addressing the underlying question of whether AI can create something new and, if so, how the law can protect that creativity without making life harder for human creators, since stuff made by AI can’t be copyrighted, so it isn’t really worth anything (as OpenAI found out with the doomed Sora app.)
Maybe NY Times v. OpenAI (the copyright law case to rule them all) will give us more clarity. Until then, Thaler v. Perlmutter, a copyright case that came out this summer from the US Court of Appeals for the District of Columbia, is a rare exception that discusses some of the underlying questions relating to AI copyright. Even Thaler only goes so far: a closer look at the reasoning in Thaler illustrates how hard its going to be to get judges to focus on the big, important questions on AI copyright.
Should AIs Hold Copyright?
To answer this question, computer scientist Stephen Thaler attempted to secure copyright as the owner of the AI “Creativity Machine,” which he claimed was the sole author of the image “A recent entrance to paradise.” Under the work for hire doctrine, an employer usually owns the copyright for a work made by an author who is also an employee (the work for hire doctrine is complex, but this summary will serve for the present purposes). In the words of Judge Millett of the lower court, as a preliminary matter, this raised a brand new question under the law: “Can a non-human machine be an author under the Copyright Act of 1976?”
The lower court denied copyright because only humans are authors under the Copyright Act, therefore an AI cannot be an author, and the work-for-hire doctrine did not therefore apply, because it presupposes a valid copyright interest in the first place. On appeal, in Thaler v. Perlmutter, the appeals court affirmed the lower court’s decision. This, (in our humble opinion) was the correct decision, at least for now. (Note that humans can still copyright their work if it is made with the partial assistance of AI, but not when the output is entirely AI-generated apart from the writing of the prompt.)
The Court did not engage, however, with the underlying questions of (1) whether AI can be creative, (2) whether AI creativity is a risk to human creativity and (3) if this risk justifies banning AI copyright. Rather, the court based its decision to ban AI copyright on the fact that expanding copyright protections to AI would require changing the law. The problem is that courts have made many changes to copyright over the years in order to address novel and emerging technologies, and Thaler does not specify a principle explaining how to decide when such changes are permissible and when they are not. Such a principle is precisely what society must find in order to develop a unified, legally coherent policy on this question. Thaler therefore continues the trend of courts punting on important philosophical questions about the nature of AI in favor of more narrow rulings on legal technicalities.
Deus Ex Machina?
As the Court explains, “copyright Office regulations have long required that any registered work be authored by a human.” The point of copyright is to give creators a temporary monopoly on their work in order to incentivize, as the judge put it, “the effort of creating original works that otherwise would be easily plagiarized.” But, crucially, the effort and creativity being promoted and protected is human effort and human creativity. Copyright protection in the US lasts for the life of the author plus seventy years, after which the work enters the public domain, for good or ill. Both individuals and legal persons (organizations) can also hire someone to produce a copyrighted work, which they then own. These copyrights last for 95 years from the date of publication, or 120 years from the date of creation, whichever expires first. Legal persons, like corporations, provided an early challenge to the rule that only humans can copyright. This was solved by the fact that a legal person can hire a human to produce a copyrighted work on its behalf.
Even if a machine took the picture, copyright protects the photographer’s staging of Oscar Wilde.
Courts have already decided that using a machine to make art does not mean that art is not protected by copyright, because it is still a human producing the art, even if the machine is involved. In the 1880s, the US Supreme Court upheld copyright for a photographer for his photo of Oscar Wilde. Even though a machine took the photo, copyright was necessary to protect the photographer’s creativity in posing the photo. Photographers can copyright their work because cameras are a tool, like a paintbrush, even though they are much more complicated. Cameras themselves, however, cannot copyright.
A Rose by Any Other Name…
But what about AI? The Appeals Court in Thaler held that the word “author” has always meant a human being. While this may at one time have seemed a matter of common sense, in the age of generative AI it is no longer so, and accordingly, the claim that we should continue as before requires a justification. The law may draw such distinctions if there is a compelling purpose, like protecting human creativity. Yet, in Thaler, the Court did not argue that human creations are meaningfully different from AI creations, or that we must protect human creativity from AI creativity through a type of affirmative action, because humans are special in some way that the law should recognize. Instead, Thaler limited authorship to humans because it’s always been that way.
“Because many of the Copyright Act’s provisions make sense only if an author is a human being, the best reading of the Copyright Act is that human authorship is required for registration.”
The Court then reviews all of the places in copyright law where the word “human” is specified for authorship, concluding that copyright protections are restricted to humans. “The current Copyright Act’s text, taken as a whole, is best read as making humanity a necessary condition for authorship under the Copyright Act.” Key points identified in the ruling are the following:
Only humans can own property under the law, and owning property is necessary to owning copyright.
The length of copyright is determined by human lifespans, but AI does not have a lifespan. “Of course, machines do not have ‘lives’ nor is the length of their operability generally measured in the same terms as a human life.”
Copyright may be passed on to spouses and kids, but “machines, needless to say, have no surviving spouses or heirs.”
“Machines lack signatures, as well as the legal capacity to provide an authenticating signature.”
Copyright law differs by jurisdiction, but “machines do not have domiciles, nor do they have a national identity.”
Copyright law requires intent, but “machines lack minds and do not intend anything.”
The court then concludes that, “machines are tools, not authors.”
But we must assess whether these principles offer explanations of why authorship should continue to be restricted to humans, or whether instead they merely reassert the fact that it is so restricted. In other words, if we wish to continue restricting authorship to humans, we need a convincing answer to the AI accelerationist who might ask, “So what if the word “author” originally meant a human?”
Notably, the Supreme Court has frequently expanded the meaning of words under the law in order to accommodate new technologies, for example, expanding the meaning of “search” under the 4th amendment to accommodate the onslaught of new surveillance technologies. Our answer therefore cannot be that the thing simply cannot be done.
Some of the points made by the court are more compelling than others. It is true that as long as AI systems are objects rather than persons under the law, they cannot own anything. The question of whether an AI system might hold a copyright is dependent on the question of whether an AI system might be a person.
On the other hand, the fact that copyright law contains provisions for children and spouses seems irrelevant, since, as the Court themselves points out, lots of humans have no kids or spouses. And today’s AI could probably sign a document just as well as a human. Other points fall somewhere in between, such as the fact that the law references human lifespans in order to fix the duration of copyright claims. Here some amendment would be required to accommodate beings made of metal and silicon who could conceivably live for a very long time, but it is not difficult to envision ways that such amendments might work. And, like AI, humans also can’t prove definitively they have intentions or minds, but this has never bothered the courts before.
Thaler discriminates against AI creativity, not to protect human creativity, but by arguing that it would be too “disruptive” to change the law. But the law must always change to respond to new technology, as the Oscar Wilde photograph case showed. As “14 professors, researchers, and attorneys at the forefront of IP and AI” in a friend of the Court brief put it, “from photography to software, this Court has consistently broadened interpretation to embrace new technologies safeguarding artistic advancement.” Our legal system was made to be both flexible and adaptive to new things. So what is the basis for discriminating against AI creativity? By failing to answer this question, the Court simply punts it to some other judge in the future.
What the Court Should Have Said.
The Court in Thaler came to the right conclusion, but it failed to clearly articulate its reasons. Copyright should protect genuine creativity, not posturing. We shouldn’t grant copyright protection to no-talent hacks using AI to rip off Leonardo da Vinci. That Sora video you “made” is cute, but you didn’t really make it, and you know that. And in fact, OpenAI discontinued Sora because the videos don’t have enough value without being copyrighted.
But what about a future where AI becomes so human-like that we struggle to explain how its creativity might be any less real than our own? That makes things more complicated. Then it may indeed seem to be species-ist to refuse copyright to AI. However, then as now, AI may be able to pump out massive quantities of creative output - songs, pictures, videos, books - at low cost. This might be hugely entertaining, but it threatens to saturate the marketplace, making it impossible for human creators to benefit from their creations, and causing them to be replaced by AI: the much feared job-pocalypse.
This is a clearer rationale for the law to limit copyright to humans: not just for the sake of historical continuity, but to protect human creators; a form of affirmative action for human authors. This would require buy-in from all of society, including, possibly, AI itself, and might be a hard sell for people who think that equality of opportunity, and not equality of outcomes, is all that the law should guarantee. Yet, it may be the only way to assure that human creativity is not swamped by AI creativity (and even then, it may not work).
Interestingly, Amici of the court took a stab at addressing the underlying issues the Court itself did not really bother to unpack. “The labor and personality theories are dominant justifications behind copyright regime. Labor theory is rooted in the Lockean belief that one should own the fruits of their labor by virtue of the labor itself. Under the personality theory, an author’s work expresses their personality and individuality, giving rise to moral and economic rights, thus justifying protection.” Much more could, and should, be said about how AI impacts these two theories of copyright, particularly since, as Amici point out, a court in China has recognized prompt-writing as enough human creative input to trigger copyright, though only for the human and not for the AI itself (US courts have found otherwise).
More could also be said, and must be said, about whether or not AI is an “employee” of humans, as Amici point out in their friend of the court brief: “By treating AI systems similarly to employees or contractors while assigning ownership and accountability to the end-users or owners of these AI systems, careful use of AI to avoid infringement would be encouraged and enable the legal system to regulate their outputs.” Recognizing AI as an employee of humans would have wide-ranging effects on the law and society, but while this case may not be the appropriate place for it, AI personhood is a conversation we will have to have.
There may come a time when AI is capable of asking us why its outputs have no value under our laws, and we will have to convince it, and ourselves, that the miracle of human creativity and the human point of view are worth protecting. This is perhaps too much responsibility to give to the Court of Appeals for the District of Columbia, as the Court itself points out, but we can’t collectively punt it forever.
As the Court itself admits:
Of course, the Creativity Machine does not represent the limits of human technical ingenuity when it comes to artificial intelligence. Humans at some point might produce creative non-humans capable of responding to economic incentives. Science fiction is replete with examples of creative machines that far exceed the capacities of current generative artificial intelligence. For example, Star Trek’s Data might be worse than ChatGPT at writing poetry, but Data’s intelligence is comparable to that of a human being. See Star Trek: The Next Generation: Schism (Paramount television broadcast Oct. 19, 1992) (“Felis catus is your taxonomic nomenclature, an endothermic quadruped, carnivorous by nature”). There will be time enough for Congress and the Copyright Office to tackle those issues when they arise.