Imagining a New Form of Intellectual Property for AI-Generated Works


By Lauren Mack.  With the rise of generative artificial intelligence (AI) tools becoming widely available for public use—such as ChatGPT, DALL·E 2, and Midjourney—the topic of whether and how AI-generated output should be protectable intellectual property in the United States is becoming increasingly critical. The primary debate has centered on how we define a human-generated original work and whether using the works of human creators to train AI requires a license from those creators. While global policymakers weigh their long-term strategies for encouraging innovation while protecting creators in this new landscape of emerging technology, some may determine that a new kind of intellectual property right is needed to appropriately govern content generated by AI technologies.

Recent Cases in the AI Copyright Debate

In 2018, computer scientist Steven Thaler attempted to register a copyright the work “A Recent Entrance to Paradise”, which Thaler identified as having been “autonomously created by a computer algorithm running on a machine”. He listed the author of the work as the creative machine system he built, known as DABUS, under the theory that his machine had created it for him as a work made for hire. The Copyright Office ultimately denied the claim, writing that the work “lacks the human authorship necessary to support a copyright claim”. This was consistent with the Copyright Office’s stance during a previous lawsuit that an Indonesian monkey named Naruto who took some “selfies” using a nature photographer’s camera did not own the rights to its own photographs.

Earlier this year, the Office found that a short graphic novel, titled “Zarya of the Dawn”, which was generated by Kristina Kashtanova in part with help from Midjourney, did constitute a copyrightable work—though with a significant caveat. While the copyright extends to the creative elements of the story expressly generated by Kashtanova, as well as her compilation of the images created using Midjourney, the images themselves that were generated by Midjourney were not copyrightable as they “are not the product of human authorship”.

Intellectual property questions not only surround AI output, but also the input. Several lawsuits have already been filed by artists and authors, alleging that generative AI tools committed copyright infringement by using their works as training fodder without permission. While what materials the AI tool was trained on may not always be obvious unless the style of a certain author is requested by the user, Getty claims to have caught Stable Diffusion red-handed when its output included warped versions of Getty’s watermark, prompting an additional trademark infringement claim from the art licensing giant.

The U.S. Copyright Office’s Position

In response to a growing number of applications to register AI-created works, the Copyright Office has issued a new registration guidance that requires applicants to disclose whether AI has been used to any measurable degree in the creation of their content. According to the Copyright Office, “[i]f a work’s traditional elements of authorship were produced by a machine, the work lacks human authorship and the Office will not register it…. Based on the Office’s understanding of the generative AI technologies currently available, users do not exercise ultimate creative control over how such systems interpret prompts and generate material. Instead, these prompts function more like instructions to a commissioned artist—they identify what the prompter wishes to have depicted, but the machine determines how those instructions are implemented in its output”.

In further clarifications provided by the Copyright Office, it explained that if any element of work contains an “appreciable” contribution from a human—meaning that the human’s contribution standing alone could be copyrightable—that portion of the work is eligible for copyright protection. However, any work that is created solely by AI, or with only minor contribution(s) from a human, such as changing a single word or the color of someone’s eyes, would not be copyrightable since the human-added element could not be copyrighted on its own.

Evaluating AI Lawmaking

U.S. policymakers will no doubt be looking to see how other countries around the world are navigating the sticky issue of whether and how intellectual property laws govern AI technologies. Earlier this month, the European Parliament passed a draft law of its AI Act, first proposed in 2021. While the focus of the regulatory framework is expansive, the AI Act would require generative AI models like ChatGPT to disclose key information about how their technologies work, including requiring summaries of the copyrighted data that were used in their training.

At the same time, China is similarly speeding ahead with plans to regulate generative AI technologies. Its cyberspace regulator in April released draft rules to govern generative AI tools, and China’s rules would not only apply to how these technologies are trained, but would extend even to the content they’re permitted to create. It’s widely anticipated that the EU or China could be among the first governments to issue sweeping legal frameworks for generative AI.

The U.S. Copyright Office acknowledges that its hurdles to finding an easy workaround to copyright protections for generative AI are multifaceted. In its registration guidance, the Office writes that “the term ‘author,’ which is used in both the Constitution and the Copyright Act, excludes non-humans, adding that its “registration policies and regulations reflect statutory and judicial guidance on this issue”.

In the Age of Generative AI, We Need a New Form of Intellectual Property

Given the current stance by the U.S. Copyright Office, Congress may consider looking beyond traditional copyright protections to a unique type of intellectual property right, such as database rights in the U.K. or the U.S.’s own mask work framework, a type of intellectual property protection that specifically applies only to semiconductor chips. Mask works are not protected by copyright law in the U.S., and the requirements for this sui generis protection are distinguished “in terms of eligibility, ownership rights, registration procedures, term, and remedies for rights violations”.

These types of frameworks may offer a solution that’s more nuanced and tailored than current copyright protections. If I were to propose a unique property right for AI output, my suggestion would be to grant the right to reproduce, distribute, and otherwise exploit the output to the creator of the AI tool. Works eligible for protection should be completely original and not commonplace, and registration would be required for protection. Any infringement claim would be required to show that the alleged infringer had access to work and created an exact copy. Further, these works should have a significantly shorter term of protection, such as three to five years, to account for the lack of human creativity and meteoric evolution of AI technologies.

As an additional consideration, we might consider a requirement for the receipt of intellectual property protection a disclosure that such works were created by AI. In the same way that publications disclose the source of their imagery from databases like Getty or influencers disclose compensation for a review, rules for using AI imagery might similarly require language that makes it clear the content was not a unique work captured or created entirely, or at all, by a human being (which could, additionally, help mitigate the spread of spoofed or intentionally misleading imagery across the media landscape).

Further, granting a right in AI output does not make sense unless the authors of the works used to train the AI tool are also compensated. While end result of the current lawsuits by copyright owners regarding the use of their works remains to be seen, it’s entirely possible we’ll start to see creators willingly opting into their work training these generative AI systems as a source of licensing income. In the meantime, ensuring that the work of human creators, and that of their AI-friendly human collaborators, are protected and able to exploit their works in a manner intended to “promote the progress of sciences and useful arts” as mandated by the U.S. Constitution is paramount for all stakeholders.