Navigating copyrights and authorship in age of generative AI

“A Recent Entrance to Paradise” (2012) is an artwork autonomously generated by computer scientist Stephen Thaler's AI system referred to as the “Creativity Machine.” Courtesy of Stephen Thaler

“Undoubtedly, we are approaching new frontiers in copyright as artists put AI in their toolbox,” wrote U.S. district judge Beryl A. Howell in her ruling on one particular case that grabbed international headlines in August.

This lawsuit marked computer scientist Stephen Thaler’s latest bid to secure copyright registration for a 2012 artwork. What then made it a high-profile case? The work in question was entirely generated by Thaler’s brainchild – an AI system known as the “Creativity Machine.”

The federal judge ultimately rejected his request citing the absence of “human authorship” – a bedrock prerequisite for a valid copyright claim. The case, nevertheless, underscores the growing conundrum concerning the nature of intellectual property and creativity within the rapidly evolving landscape of generative AI.

And Thaler’s case is just one example of the legal disputes emerging as AI continues to encroach upon the creative realm once believed to be the unique purview of human minds. In the U.S., a string of copyright battles over AI-generated works has already begun.

Korea is experiencing its own share of legal and ethical dilemmas.

In July 2022, the Korea Music Copyright Association (KOMCA) abruptly announced a decision to discontinue royalty payments for six tracks composed by one of its members, EvoM (Evolutionary Music), after belatedly learning that "she" was an AI-driven music composition software and not a human songwriter.

“For a long time, AI was a theoretical problem in copyright,” Andrew Foglia, deputy director of policy and international affairs for the U.S. Copyright Office (USCO), remarked during this year’s Seoul Copyright Forum, Oct. 26. “The last few years have changed that.”

In addition to the debates about the copyrightability of AI-generated content, there has been a wave of lawsuits filed by rights holders – including Getty Images and “Game of Thrones” author George R. R. Martin – against leading AI firms this year.

These plaintiffs are accusing the companies behind ChatGPT and Stability Diffusion of using their copyright-protected materials without permission to train generative AI models.

The outcomes of these litigations are likely to carry far-reaching implications that will shape the future of AI copyrights worldwide, including Korea.

In June, a coalition of webtoon creators launched an online boycott of Naver Webtoon, a digital comic platform operated by Korea's largest portal, Naver, to protest the potential use of their own uploaded works as training data for the portal's AI-integrated services. Earlier, the platform had faced criticism for a comic series believed to have been created with the aid of AI engines, evident in its awkward compositions and uncanny illustrations.

During the subsequent months, the Korean Association of Newspapers and the Korea Online Newspaper Association issued statements regarding generative AI tools, claiming that scraping news content for AI training purposes without proper authorization and a legal basis infringes on the rights of media organizations as database producers.

A coalition of webtoon creators initiated an online boycott of Naver Webtoon this summer, protesting the potential use of their uploaded works as training data for the portal's AI-integrated services. Screen capture from Naver Webtoon

Can AI-generated content gain copyright status?

As of today, most countries, including Korea, lack legislative provisions dealing explicitly with AI creations.

Korea’s Copyright Act, however, defines a “copyrighted work” as “a creative production that expresses human thoughts and emotions.” Similarly, the Compendium of U.S. Copyright Office Practices specifies that “to qualify as a work of ‘authorship’ a work must be created by a human being.”

Human authorship, therefore, remains a foundational requirement for determining the eligibility for copyright protection. It effectively rules out any attempt to register works produced solely by non-human entities, including AI engines.

But questions arise when we look at images and texts born from a “collaboration” between a human user and AI.

According to the USCO’s latest copyright registration guidance, if an individual makes sufficiently creative modifications, edits and/or compilations using the initial AI-generated output, they can seek limited copyright protection extending only to the human-authored portions.

This still leaves a grey area in defining what constitutes a sufficient human contribution.

But one important thing to note is that merely inputting text prompts into a generative AI model does not qualify as a copyrightable, human-authored act. That is because after an AI engine receives initial commands from human users, it makes a series of autonomous, interpretive choices unbeknownst to us to produce the final result. This problem is known as the “black box.”

“The idea is that authorial decisions are happening inside the model, opaque to the user and sometimes even to the AI developer itself,” USCO's Foglia said. “With DALL-E 3, for example, the system will even invisibly rewrite the users’ prompts sometimes to engineer a better output.”

In Korea, the Copyright Commission is working on guidelines to address these issues, including stipulations regarding AI-assisted creations, with a draft already outlined and the final version slated for publication by the end of the year.

“Zarya of the Dawn," a graphic novel that combined text written by human author, Kristina Kashtanova, with images generated by AI-powered Midjourney, was granted limited copyright protection by the USCO. Courtesy of Kristina Kashtanova

One intriguing case in the U.S. sheds light on how copyrightability may be determined for a “joint” work involving a human and AI in the future in Korea.

In February, the USCO granted limited copyright protection to the graphic novel, “Zarya of the Dawn.” This 18-page novel combined text written by human author, Kristina Kashtanova, with images generated by AI-powered Midjourney.

So, which portion became eligible for copyright? Interestingly, it covered the book’s human-authored text, as well as the work as a compilation (specifically, the overall “selection, coordination and arrangement of the work’s written and visual elements”) – even though the AI-generated images themselves were still not copyrightable.

From a legal perspective, if an individual were ever to seek copyright for an entire work produced with the aid of AI, he or she would need to make “transformative” changes to the initial output “so much so that it imparts a sense of aesthetics entirely different from the original and expresses genuinely creative authorship,” according to Yang Jin-young, a partner lawyer at Minwho Law Group, a law firm specializing in information technology and AI.

In such cases, there would no longer be “substantial similarity” between the initial, AI-generated output and the final, human-authored work, she told The Korea Times.

“But of course, assessing substantial similarity falls within the realm of subjective judgment, so legal disputes are likely to arise."

Is using copyrighted works to train AI models an infringement?

Another pressing issue that demands attention is the question of copyright infringement in the training of generative AI systems – and whether “fair use” reasoning or other legal exceptions can be applied.

In the U.S., a series of class-action lawsuits filed by high-profile names – including comedian Sarah Silverman, Pulitzer Prize-winning Michael Chabon and Getty Images – are currently pending against firms like OpenAI, Stability AI and Meta.

While no formal legal action has yet been initiated in Korea, numerous creators and rights holders in the music, media and webtoon industries have voiced their concerns in response to AI’s march into the creative sphere.

KOMCA, for instance, established an AI Response Task Force in March, arguing that its human musicians and composers are filled with great apprehension about the possibility of being replaced by their AI counterparts.

"Can we realistically compete with AI that can generate all the data humanity has ever created up to this point in the blink of an eye?" Hwang Seon-cheol, KOMCA's director general of licensing bureau 2, noted at the Seoul Copyright Forum.

Content creators and rights holders are accusing AI developers like OpenAI of using their copyright-protected materials without permission to train generative AI models. AP-Yonhap

These rights holders are accusing AI developers of performing “web scraping” and “text and data mining (TDM)” to train their models, which result in the uncompensated and unauthorized use of copyright-protected materials available online.

It remains debatable whether “fair use” – the legal doctrine that allows limited use of copyrighted materials without acquiring permission for “transformative” purposes of research, etc. – can work in AI companies’ defense.

“There’s a fair degree of unpredictability when it comes to fair use,” Mira Sundara Rajan, a professor at UC Davis School of Law, noted at the forum. “Whether economic harm is generated to authors and artists will need to be weighed against ‘transformative use’ (for fair use to justify AI’s scraping).”

Another legal provision that could work in favor of AI companies is TDM exception, which allows text and data mining of copyrighted works for limited, usually non-commercial, purposes. It has already been adopted in countries such as the United Kingdom, Japan, the European Union and Singapore. And since 2021, several bills have been proposed to Korea's National Assembly to include this exception as part of the efforts to revise the Copyright Act.

To this, KOMCA's Hwang stated, "It's not only the AI companies that matter in Korea. The country has a thriving culture sector, with its music, films and webtoons enjoying international success and contributing to 'hallyu.' Is it necessary to push for TDM exception so aggressively to the point of potentially disrupting our creative ecosystem?"

What does the future hold for AI copyrights?

Amid ongoing legal uncertainties surrounding AI copyrights, experts stress that a critical task lies ahead when it comes to devising relevant legislative guardrails – striking a careful balance between protecting the rights of human creators and facilitating AI firms’ model training to foster a virtuous cycle of creativity and technological advancement.

In fact, generative AI technology can only progress when the creative industries supplying the original data are preserved.

“Unless there is cooperation, you can only do so much to cannibalize your own data source,” Ben Zhao, a professor of computer science at the University of Chicago, told Built In, a U.S.-based online platform for tech companies and startups.

Ko Han-gyu, a senior research engineer at LG Electronics’ AI Lab, echoed this sentiment at the Seoul forum: “The quality and quantity of human-generated data used to train highly proficient AI models are just as important as the technology itself.”

“In the end, for AI engines, the absence of data is comparable to lacking fuel."