The brave new world of AI-generated intellectual property disputes

Imagine a digital artist prompting an AI art generator with the following: "Abraham Lincoln in the style of Andy Warhol's Marilyn series." The AI produces silk screen-like images that could pass for the pop artist's work. Or imagine a computer programmer typing four words into a code editor. The AI generates an almost-verbatim reproduction of copyrighted code. These are now standard capabilities in some commercially available software - with significant implications for intellectual property law and practice.

When artificial intelligence (AI) trains on publicly available copyrighted works, what rights can the copyright holders enforce? Who owns the generated content? A series of pending cases may generate jurisprudence to help attorneys counsel clients in light of technological advances.

The Copilot Case

On Nov. 3, a putative class action lawsuit was filed on behalf of software copyright holders against GitHub, Microsoft, and OpenAI in the U.S. District Court for the Northern District of California. The complaint alleges that GitHub Copilot, an AI code generator, infringes copyright by ignoring copyright holder license terms. J. Doe, et al. v. GitHub, Inc, et al., Case No. 3:22-cv-06823 (N.D. Cal., Filed Nov. 3, 2022).

Microsoft owns GitHub and Copilot is GitHub's commercial AI code generator. According to the Copilot FAQ, Copilot "has been trained on natural language text and source code from publicly available sources, including code in public [OSS] repositories on GitHub." Open Source Software (OSS) licensing can impose obligations on those who use OSS in their own projects, such as requiring attribution.

The GitHub complaint quotes from a GitHub customer-support message as supplying the alleged justification for its use of copyrighted codes, "Training machine learning models on publicly available data is considered fair use across the machine learning community" Complaint, ¶ 84. By contrast, the GitHub complaint argues: "[R]egardless of this concept's level of acceptance in 'the machine learning community,' under Federal law, it is illegal." Id. The complaint claims that no court has considered the question of whether fair use includes training machine learning systems with publicly available data. Id. at ¶ 85.

The Warhol Case

The area between a "transformative" fair use (a defense to copyright infringement) and a derivative work (a copyrighted work that comes from another copyrighted work, requiring a license) is murky. The Supreme Court of the United States recently held that a verbatim copy of certain computer code can be transformative fair use. In Google LLC v. Oracle Am., Inc., 141 S. Ct. 1183 (2021), Google copied Sun Java's application programming interface (API), but not its implementation. In other words, Google's plug looks just like Java's plug, but the wiring behind the walls is different.

The court determined the purpose behind Google's copying of the API was "to expand the use and usefulness of Android-based smartphones," thereby creating a "new product [that] offers programmers a highly creative and innovative tool." The "purpose and character" of Google's use was therefore transformative. The Google court acknowledged that a copyright on software also protects its derivative works, but it did not discuss the issue further.

Unlike Google v. Oracle, Copilot not only recreates APIs (plugs), but it also generates implementation (wiring) and synthesizes and sometimes recreates almost-verbatim copyrighted code. But AI-generated code's reliance on copyrighted material, and its resemblance to the IP upon which it relies, is not unique.

The U.S. Supreme Court recently heard oral argument in another fair use case, Andy Warhol Foundation for Visual Arts, Inc. v. Goldsmith, 11 F.4th 26 (2nd Cir. 2021), involving Andy Warhol's unlicensed use of copyrighted photos of the musician Prince to create his "Prince Series" of silkscreen prints. The Second Circuit Court of Appeals rejected the Warhol Foundation's fair use defense and discussed the "inherent tension in the Copyright Act between derivative works reserved to the copyright holder ... and 'transformative' fair uses of the copyrighted work by others."

The court's Warhol Foundation decision will address human-generated art, but it's easy to see how its reasoning could apply to AI-generated works. If the court accepts Warhol's fair use defense, would it see things any differently for an AI model trained on copyrighted works? If Warhol's Prince Series is transformative, would an AI-generated artwork be the same, even if it modified or resembled existing works? And, if so, who would own the copyright?

Copyright in AI-Generated Works

Whether AI-generated art and Copilot-generated code have their own copyright protections is not directly relevant to whether those outputs are infringing someone else's copyright, but it has implications for further commercialization, and it could implicate underlying license terms, including based on whether the outputs are derivative works.

There has never been a case in the United States that has decided the protectability of an AI-generated work. However, the U.S. Copyright Office has a "Human Authorship Requirement" under which it refuses to register copyright in AI-generated works. That requirement is being challenged in a case pending in U.S. District Court for the District of Columbia. Thaler v. Perlmutter, et al., Case No. 1:2022cv01564, (D.D.C. filed June 2, 2022). Different jurisdictions take different approaches to AI-generated works and copyright. For instance, the United Kingdom's Copyright, Designs and Patents Act 1988 (CDPA), explicitly provides for copyright protection of these works. CDPA Section 9(3).

Alternative Dispute Resolution for Intellectual Property Disputes

The head-spinning pace of technological evolution and the litigation these disputes generate beg the question: Are the courts the ideal forum? Although some cases are valuable for their precedential effect, others might benefit from a more flexible and streamlined process. (As an aside, by the time the U.S. Supreme Court decided Google v. Oracle, Android development had moved on from the Java APIs at issue.) Mediation and arbitration tend to expedite conflict resolution, decrease costs, protect confidentiality and lead to a broader range of commercially oriented outcomes. Ryan Abbott, Jeremy Lack, and David Perkins, Managing Disputes in the Life Sciences, 36 Nature Biotechnology 697 (2018). Perhaps one of the greatest benefits in the context of IP-related disputes is that parties have the option to select their neutrals, which can be particularly valuable for disputes involving specialized areas of law and technical complexity.

Disclaimer: The content is intended for general informational purposes only and should not be construed as legal advice. If you require legal or professional advice, please contact an attorney.