SUBSCRIBE NOW!

Vol. 41 No. 9 — Nov/Dec 2024

LET'S GET STRATEGIC

The statements and opinions of the experts in this article are theirs alone and do not represent any company.

You ask ChatGPT (or one of a growing number of similar automated content generators) to write copy for a blog post that you publish to your company’s website. The copy was entirely written by the AI tool. Later, you discover that another company has lifted the post, verbatim, for use on its own site. What happens if you sue the company for copyright infringement? That’s a difficult question to answer.

For creators, the issue is pretty straightforward, notes Jonathan Bailey, a copyright/plagiarism consultant who runs Plagiarism Today and CopyByte. “Nearly all AI systems were trained on large amounts of unlicensed but publicly available content. Some of the content was even pirated. It was then used to train AI systems, which are being used to make similar works. Many are not happy about this, and there have been dozens of lawsuits against AI firms over this issue,” he says. Here’s the rub: If a generative AI (gen AI) tool created the content, who is liable for copyright infringement? The emergence and rapid adoption of gen AI tools to create content of all kinds have led to some interesting—and perplexing—questions related to the ownership of that content, such as: Who owns the copyright to gen AI-created content? At this point, the answer is nobody.

OWNERSHIP ISSUES

The U.S. Copyright Office has made it clear that AI-generated content can’t be registered under a copyright. It states, “In the Office’s view, it is well-established that copyright can protect only material that is the product of human creativity. Most fundamentally, the term ‘author,’ which is used in both the Constitution and the Copyright Act, excludes non-humans.” But the Copyright Office also acknowledges the need for additional guidance on the matter and has “launched an agency-wide initiative to delve into a wide range of these issues,” indicating that it is “seeking public input on additional legal and policy topics, including how the law should apply to the use of copyrighted works in AI training and the resulting treatment of outputs.”

There’s much debate on the topic, which is anything but clear cut. Beth Cooper, VP of sales and marketing at KNB Communications and an attorney who practiced contract and copyright law, disagrees with the notion that AI-generated content shouldn’t be subject to copyright. She makes an analogy to using cameras to create images: “There are some photos that do not warrant copyright protection, and there are some that do. It depends on the creative vision of the photographer. In AI-generated works, the creativity is in the prompt writing.” What’s more problematic is the potential for creators’ own work to be used in datasets to train gen AI models, Cooper says, adding, “Many creators are finding that their work has been included in training without their permission or compensation and, unsurprisingly, they by and large don’t love that. Sure, humans can copycat their art, but AI can do it at an incredible scale. It has the potential to make a giant impact on not only their livelihood, but also their legacy.”

CREATOR ISSUES

“The recent advancements in generative AI technology have raised a wide range of issues concerning copyright ownership, infringement, and the unauthorized use of copyrighted materials,” says Craig Smith, an IP litigator and partner at Lando & Anastasi. While Smith points out that tech companies such as Amazon, Google, Meta, Microsoft, and OpenAI have committed to helping ensure the “safe and transparent development of AI technology,” there is currently no comprehensive AI legislation in the U.S. In October 2023, the Biden administration issued the Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, which requires federal agencies to establish standards and guidance for AI use, safety, and security. “These standards do not have the same force as actual legislation, but they will help companies think more critically about how they use and implement AI technology,” he asserts.

Content creators, like writers, “complain that both the input and output of AI systems violate copyright laws,” Smith notes. They’re in direct conflict, though, with companies such as OpenAI. “[A]uthors argue that the use of their works to train AI models constitutes copyright infringement. Not surprisingly, AI companies disagree. OpenAI has argued that using copyrighted material to train AI models is fair use,” he says.

TEXT, AUDIO, AND VIDEO CONTENT

Oindrila Mandal, a senior game product manager at Electronic Arts, points out that there is a continuum in terms of the maturity of AI across text, audio, and video content. “Gen AI is very mature in the text space,” she says. “Not only can gen AI generate quality text content, but it can also significantly reduce editing efforts. And gen AI text identification solutions are also quite mature, so it is possible to distinguish original human-written content from AI-generated text.”

For video, though, Mandal notes, the solutions are not very mature yet: “While content creators can use it for editing purposes, it is significantly harder and more expensive to generate high-quality gen AI video from scratch. Hence currently its usage is limited, except when it is used to modify existing video media content.” AI-generated audio such as voices or music is in a gray area between these two worlds, according to Mandal. “AI audio generation is relatively ubiquitous and has wide applications in entertainment, gaming, social media content, ebooks, and more,” she says, but there are legal issues that need to be resolved between content creators and the owners of the audio, especially with regard to acceptable use policies.

INFRINGEMENT RISKS

A similar continuum applies in terms of copyright risks when using AI-generated text, audio, or video, Cooper notes. “Each medium brings its own set of complexities when AI is thrown into the mix.” With text, the subtleties of language have led many linguists to “believe one’s choice of words and syntax are so unique as to be akin to a fingerprint.” She adds this: “AI has the ability to generate text that mimics the style and structure of existing authors, making it incredibly hard to distinguish between original content and something that’s derivative.” A writer’s style, though, isn’t subject to copyright. Copyright protects expression, not ideas or methods. “That said, if someone generates a work that is so close as to be considered a derivative work, or generates work using unique characters or worlds, there is a potential for copyright infringement claims,” Cooper states.

There’s more danger of infringement when it comes to audio, especially music. Here, Cooper says, there’s risk “of AI-generated compositions sampling or replicating existing songs, melodies, or chords.” Even a snippet of a song can have copyright protection. “Big record labels have advanced software and crackerjack legal teams to track down their owned music on the internet and make sure no one else makes a dime off of it.” Cooper explains that the same intellectual property considerations exist with video, but that video presents another serious consideration—the potential for deepfakes. Deepfake videos, she says, “make it appear that someone said or did something they never did. It’s an ethical and legal minefield.”

Modern copyright law treats all types of creative content the same from a legal perspective, says Bailey. “The biggest difference is that certain works, by their nature, have different rules out of practicality. For example, there are only so many notes and so many chords in music. So, reusing common notes, chords, etc., is less likely to be infringing than repeating sentences verbatim,” he explains.

Content creators need to be aware of their ownership rights. “Creators run the risk of losing ownership to their content if they either use gen AI to create their content or gen AI is used to modify their content by either authorized or unauthorized third parties,” Mandal cautions. To mitigate these risks, she says, it’s important for content creators to be aware of gen AI copyright laws and to make sure contracts explicitly state who owns any content being created. In terms of U.S. guidance, in July 2024, the Copyright Office released Part 1 of its report on copyright and AI, addressing the topic of digital replicas. Additional guidance is anticipated. These reports are a must-read for anyone creating content of any kind—and any companies using gen AI tools to generate copy. That’s pretty much everyone these days.