Copyright Considerations In Artificial Intelligence

This was confirmed by the decision in Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018). This is the case in which a seven-year-old crested macaque monkey named Naruto allegedly took several photographs of himself with a camera owned, and left unattended, by wildlife photographer David Slater. The photographs were published in a book by Slater and Wildlife Personalities Ltd. created through Blurb Inc.'s website. In 2015, People for the Ethical Treatment of Animals (PETA) and Dr. Antje Engelhardt filed a complaint for copyright infringement against Slater, Wildlife and Blurb on behalf of Naruto. The U.S. Court of Appeals for the Ninth Circuit held that Naruto lacked standing to sue under the Copyright Act because animals cannot sue for infringement. The appeals court elaborately distinguished between human and non-human authors of a work. Specifically, the court reasoned that because other sections of the Copyright Act refer to "children" and "widow," for example, the statutory construction of the term "author" requires that the author must be a human being.

Under the Naruto rationale, AI software itself could not own the rights to an AI-created work. This interpretation would also align with a recently published petition decision by the U.S. Patent and Trademark Office (USPTO) that, under current law, only natural persons may be named as an inventor in a patent application. If AI systems cannot be credited as an inventor in a patent, then current policy dictates it cannot be credited as an author in a copyright. Therefore, AI-created work may be ineligible for copyright protection and enter the public domain immediately upon creation.

Alternatively, the copyright may be held by an individual who created the AI program. This conclusion can also be derived from the ruling in Naruto. In particular, in its decision the Ninth Circuit noted that copyright ownership rights of corporations were predicated on the fact that corporations are considered "persons" under precedent established by the U.S. Supreme Court because corporations "are formed and owned by humans." Therefore, copyrights for works created AI software, also formed by humans, could be owned by the author of the software. This would also be consistent with the field of AI.

In general, AI software requires some initial input from the author of the software code. As such, any subsequent work would be the fruit of that initial work, making the programmer the original author. Without additional guidance from Congress or the U.S. Copyright Office, creators of AI systems should apply to register copyrights for AI-created work as the "author" of that work. Or else, they would risk losing the ability to protect the work and exercise ownership rights against potential infringers.

Copyright law has not yet defined a clear line between infringement and fair use for AI-manipulated works. The interplay between copyrighted works being used for machine learning and reproducing a work in a culpable manner is highlighted analogously through Authors Guild v. Google, 804 F.3d 202 (2d Cir. 2015), in which copyright owners sued Google for scanning digital copies of works, creating searchable text and providing free "snippets" of the texts containing searched terms.

In Authors Guild, major libraries contracted with Google to make digital scans of books that could be machine-readable. As the U.S. Court of Appeals for the Second Circuit noted, Google "retain[ed] the original scanned image of each book" and "indexed more than 20 million books" for its Google Books project. Google did not display advertising to users who searched for books using search terms and did not "receive payment by reason of the searcher's use of Google's link to purchase the book."

In upholding the district court's grant of summary judgment of fair use under 17 U.S.C. §107, the Second Circuit found that creating the searched indexes was transformative, with a purpose "to make available significant information about those books" and to provide readers the necessary information to educate them on whether or not they needed to purchase the books without providing so much information that the full context of the book was no longer necessary. Transitioning to a similar example in AI contexts, it is clear that the exact output of the AI would have to be weighed against the input material to determine how the AI treats the work and the overall transformative nature of the AI-created content.

Second, the appeals court noted that the §107 fair use factor as to "the nature of the copyrighted work[,] … rarely play[s] a significant role in the determination of a fair use dispute." With respect to AI, this factor weighs in favor of fair use for factual information. But to the extent that the AI relies on expressive works for its machine learning algorithms, this factor could weigh against fair use. The second factor highlights the murky common-law and statutory framework AI is currently operating in, and the need for practitioners to stay up-to-date on any upcoming guidance from the U.S. Copyright Office (as well as USPTO).

Third, the Authors Guild court looked to "the amount and substantiality of the portion used in relation to the copyrighted work as a whole" under §107. Noting that Google had copied the entire copyrighted work, the Second Circuit found that "what matters … is not so much the amount and substantiality of the portion used in making a copy, but rather the amount and substantiality of what is thereby made accessible to a public for which it may serve as a competing substitute." The court also found that there is no categorical rule precluding fair use from works that are taken in their entirety. Although AI may take entire works to learn, this factor highlights the uncertainty between the ability to claim AI's reproduction of copyrighted materials for machine learning as fair use, without a detailed review of an AI's purpose, input and output.

In evaluating the effect on the market in Authors Guild, the appeals court found that Google must create a "meaningful or significant effect 'upon the potential market for or value of the copyrighted work,'' citing 17 U.S.C. §107(4). Depending on an AI algorithm's output, the effect of AI-generated content may make the underlying content obsolete, complement it or leave such content's market completely undisturbed.

Clearly, under the current statutory framework and common law doctrines, fair use can absolve AI of infringing copyrighted works. See also, Sega Enters. Ltd. V. Accolade Inc., 977 F.2d 1510 (9th Cir. 1992) (holding that a copyrighted video game code may be copied to gain understanding of unprotected elements under fair use); Sony Computer Entertainment Inc. v. Conntectix, 203 F.3d 596 (9^th Cir. 2000) (finding the reverse engineering and intermediate copying of Sony PlayStation's copyrighted BIOS software for a transformative use constituted fair use); and Perfect 10 Inc. v. Amazon.com Inc., 508 F.3d 1146 (9th Cir. 2007) (finding thumbnails of copyrighted work to be sufficiently transformative to qualify to support a finding of fair use to survive summary judgment).

An AI training model may use a copyrighted image for an entirely different purpose than that of the copyrighted image but may also use a copyrighted image for a purpose that is not sufficiently transformative. While current legal frameworks support that artificial intelligence may include fair use, the same framework makes it unclear as to the applicability to AI generally, rather than having to ask courts to provide a case-by-case analysis of when AI crosses the line from fair use to infringement.

It was at the end of 2019 that the USPTO asked for public comment on how developments in AI should impact U.S. IP law and on April 27, 2020, that the USPTO published the decision in which it found that inventors are limited to natural persons for purposes of patents. While no official guidance or precedent has been set with respect to AI's implication on copyright laws, any official USPTO guidance could impact the applicability of AI in creating works capable of copyright protection.

Additionally, while there is analogous law on how AI may be said to infringe or qualify for fair use of copyrighted works, there has yet to be an on-point decision by any court that squarely decides the issue. Practitioners should advise clients of the importance of considering how their AI manipulates copyrighted works to determine if AI does indeed infringe the copyright of another work.

*****

Shaleen J. Patel is an associate in the Intellectual Property Department of Pepper Hamilton LLP, based in the firm's Boston office. He can be reached at [email protected]. Sushmitha Rajeevan is a Pepper Hamilton associate also based in the firm's Boston office. The views expressed in this article are those of the authors and not necessarily the views of their clients or other attorneys in the firm. This article was previously published in The Intellectual Property Strategist, a Law Journal Newsletters sibling of Entertainment Law & Finance.