AI and Copyright
By John T. Pacconi, Posted September 26, 2023
Technology is advancing exponentially and changing the legal landscape of copyright law in ways that Congress could not have foreseen when codifying the Constitution and IP statutes years before artificial intelligence (AI) came into existence. For decades, humans have used technology to assist their development of copyrightable material. For example, authors use Word to type up their latest novel, and photographers use digital cameras and Photoshop to take and edit photos. These technologies are clearly tools that assist the authors in carrying out their creative vision. No one would dare assert that Word or Photoshop is the author of their respective works; clearly, the human is the author. With the advent of AI, however, the line between whether technology is a tool that assists the author or the author itself is blurring. This article will analyze the current state of AI and copyright law, arguing that AI-assisted works of art can and should receive copyright protection.
Copyright protection, specifically for AI-assisted works of art, is important to ensure that development of innovative AI continues to increase. Having the intellectual property of an AI-assisted work of art allows the owner of the intellectual property to reap the benefits of the AI’s creative output through licensing agreements and royalties. However, if the U.S. Copyright Office refuses to extend intellectual property to the authors, the work of art is placed in the public domain where anyone can access and use the material. Placing intellectual property in the public domain allows anyone to copy the design without the original author earning profits. This public alternative to copyright slashes incentives to innovate AI further. After all, why would anyone funnel money into a technology that they cannot profit from. Affording copyright protection to AI-assisted works of art, on the other hand, would cause a significant uptick in investment for innovative AI and be a boon for the AI industry and the economy broadly because AI works much more efficiently than humans can. There needs to be an incentive to innovate, and copyright protection provides that monetary incentive.
When applying for copyright protection, there are 6 requirements that you need to fulfill: fixation, human authorship, copyrightable subject matter, originality, independent creation, and creativity. The most important of which, in the realm of AI, is human authorship because AI is clearly not human. Section 313.2 of the Compendium of U.S. Copyright Office Practices states, “To qualify as a work of ‘authorship’ a work must be created by a human being,” and further states, “the Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author.” This human authorship requirement has been the principle guiding most court decisions regarding the extension of copyright protection to AI-assisted works of art. There is a whole body of case law that argues that AI-generated content is not copyrightable. If you take an algorithm and hit a button telling it to create something, the user of the AI has not demonstrated enough creative influence to receive copyright protection of the AI-generated work of art. Instead, the AI-generated work is placed into the public domain where anyone can access it and the user of the creative AI is unable to receive compensation for licensing.
Just this past week, the District Court of D.C. reaffirmed this opinion in Stephen Thaler v. Shira Perlmutter. In this case, Thaler’s AI, the “Creativity Machine,” produced a painting called “A Recent Entrance to Paradise” and filed for copyright protection. The Court, however, denied the request, arguing that an “artificial system absent human involvement should [not] be eligible for copyright” because it does not fulfill the human authorship requirement. The problem with this case and many other cases for AI-assisted intellectual property is that there was not a human in the loop during the production of the copyrightable material. Indeed, Thaler did not contest this claim, “confirming that the work ‘was autonomously generated by an AI’ and ‘lack[ed] traditional human authorship.’” It is our opinion, however, that, if the applicant can prove that there is a human author who has significant creative control over the development of the work of art, copyright protection should be extended.
Before discussing how AI-assisted works of art can receive copyright protection, it is important to investigate the current state of artificial intelligence as it relates to creative design. Currently, AI is considered “narrow” or focused on a single, specific task that it is programmed to complete. Programmers and developers code algorithms that are applied to a single use. AI is not yet “general” or fully autonomous, meaning that a human must tell an AI what to do in order for it to complete that task. Artificial General Intelligence (AGI) refers to AI that can learn, think, and act the way humans do. AGI would not require developers to program the algorithm to do a narrow function because it can perform a wide array of tasks, like a human can. While many AI experts believe AGI will be developed within the coming decades, in the meantime, AI will remain fixed to a narrow application. While AGI is not an issue that this article will address, it is important to note that, if AI becomes fully autonomous, the human authorship question will be further muddled. The fact that AI is currently narrow is important to the discussion of copyright protection because it influences the way the human authorship requirement should be litigated. AI is not yet capable of conceiving of and creating original works of art without human input and intervention, meaning that if the person filing for copyright protection can demonstrate that they had an outsized influence on the creative vision of the AI-assisted work of art, they should be able to receive rights to the intellectual property.
Narrow AI should be viewed as a tool assisting the human author rather than the author itself. The ways that AI is used to create works of art are similar to other forms of technology. No one would dare assert that the author of the Mona Lisa is Leonardo da Vinci’s paintbrush because da Vinci clearly had the creative vision and an outsized influence on the creative output. Then why should we assume that AI is the author of a work of art that a human has outsized influence on? Sure, AI is much more advanced than a paintbrush, but, like a paintbrush, AI is unable to conceive of and create works of art without human influence because it is not yet fully autonomous. A user of AI is like a painter using a paintbrush: AI, like a paintbrush, is a tool that allows the programmer to produce an original work of art. Now, imagine that that paintbrush was attached to a machine, and the painter told that machine what colors to use, where to place them, and what the painting should be of. This scenario is analogous to how many users of AI generate original works of art. There is no doubt that AI significantly minimizes the role the human has in actually creating the work of art, but, if there is no human user, there is no creative vision. Therefore, as long as there is a human in the loop who exercises sufficient creative influence, the human should be afforded copyright protection.
There are multiple ways that an applicant can demonstrate that a human author had “creative input or intervention.” The first of which is by programming the algorithm that creates the work of art. There is a clear difference between someone telling ChatGPT to write a horror thriller and someone coding an algorithm that uses big data and deep data analysis to scan every Stephen King novel, learning information about how horror stories are written, and subsequently asking the algorithm to use machine learning to write a horror story involving certain people, in a certain place, at a certain time, doing certain things and improve upon prior writings until a final product is reached. Obviously, the latter example involves a human who has a much larger creative vision and influence on what the horror story looks like than the former. If you write the code for a program that creates a work of art, that person should qualify as the author because they have a large influence in the work of art; they are not simply typing a few words into a chatbot that spews out a unique, unpredictable story that the user of the AI could have never predicted.
A second way that an applicant could qualify as an author of an AI-assisted work of art is to have a significant influence on the inputs that go into the algorithm as the user of the AI. Even if the user did not program the algorithm that generates the work of art, they could potentially be considered an author of the AI-generated work of art by controlling the creative vision of the algorithm. If the user of the algorithm told the AI to analyze H.P. Lovecraft or Bram Stoker instead of Stephen King, the algorithm would shoot out a different style of novel. Similarly, if the user told the algorithm to use a different set of characters in a different place and time, the results of the algorithm would be different. While controlling one of these variables is unlikely to be enough to qualify for authorship, the more of these factors that the user controls, the larger the case can be made that the user qualifies as an author. After all, what goes into the algorithm effects what comes out.
A third way that an applicant could qualify as an author of an AI-assisted work of art is to alter a work of art that is generated by the algorithm after the fact. This is known as a derivative work—a work of art that includes major copyrightable elements of a previously created work of art, but major revisions have been made to make it distinct. Imagine a user of an algorithm used an AI to create a painting, and, after the fact, made major revisions to the painting, such as adding additional elements not included in the original AI-generated work of art or changing the colors of the painting in a major way. This derivate work could be afforded copyright protection. This is no different than The Notorious B.I.G. sampling Diana Ross’s “I’m Coming Out” in his hit song, “Mo Money Mo Problems.” Even though Biggie sampled the beat of a copyrighted Diana Ross song, “Mo Money Mo Problems” was still able to receive copyright protection because Biggie made substantial changes to the original work of art. In the same way, if a human makes meaningful changes to an AI-generated work of art, the human who made the final version of the work of art could be able to receive copyright protection.
While there may be ways that AI-assisted works of art could receive copyright protections, there are a few valid criticism of this strategy. Notably, any attempt to draw a line on how much human influence is necessary for that human to be considered in the loop and thus the author of the work of art will involve a degree of arbitrariness. If AI takes on too much of the load, the work of art will be placed in the public domain, so it is necessary to have a clear bright line on what is sufficient to constitute authorship. A good rule of thumb is that the greater the influence that a person has on as many of the different steps of the creative process, the stronger the case that that person should be considered the author of the work of art because the degree to which the AI is responsible for the work of art is substantially reduced. Arbitrariness can also result when multiple people are involved in the creative process. Should a programmer receive authorship or should a user? The arbitrariness criticism can be easily resolved through a carefully crafted legal regime. Clear licensing and contract agreements could establish that the person who has the largest creative control over the development of the work of art should be the sole or primary beneficiary of royalties and licensing payments.
