Generative content is having a moment in the form of non-fungible tokens (NFTs) powered by blockchain platforms. Continued investment and growth in the NFT market is bringing into the mainstream legal edge cases concerning the use of digital technology to create intellectual property. Recently, the U.S. Copyright Office’s Review Board issued a decision addressing whether an artificial intelligence “Creativity Machine” can meet the statutory requirements of an author for copyright purposes. See Decision dated Feb. 14, 2022, available here . The Board held that the Creativity Machine does not meet the statutory requirements of an author, consistent with the Office’s position that an author must be a human being.
This article explores what the Board’s decision means more broadly for generative content, including generative NFTs, and offers some practical guidance for those looking to secure copyright protection.
The Human Authorship Requirement for Copyrighted Works in the United States
Statutorily, a copyright in a work “vests initially in the author or authors of the work.” 17 U.S.C. § 201(a). While “author” is not explicitly defined, the U.S. Copyright Office has taken the position that the author must be a human being:
The copyright law only protects “the fruits of intellectual labor” that “are founded in the creative powers of the mind.” Trade-Mark Cases, 100 U.S. 82, 94 (1879). Because copyright law is limited to “original intellectual conceptions of the author,” the Office will refuse to register a claim if it determines that a human being did not create the work. Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884).
Compendium of U.S. Copyright Office Practices, § 306. The Copyright Office further highlights that the “crucial question is ‘whether the ‘work’ is basically one of human authorship, with the computer [or other device] merely being an assisting instrument, or whether the traditional elements of authorship in the work (literary, artistic, or musical expression or elements of selection, arrangement, etc.) were actually conceived and executed not by man but by machine.” Id., § 313.2.
U.S. Courts have taken a consistent view in interpreting the statute as requiring human authorship. See, e.g., Cmty. For Creative Non-Violence v. Reid, 490 U.S. 730, 737 (1989) (“As a general rule, the author is the party who actually creates the work, that is, the person who translates an idea into a fixed, tangible expression entitled to copyright protection”). Human authorship came into focus in the case of Naruto v. Slater, No. 15-CV-04324-WHO, 2016 WL 362231, at *3 (N.D. Cal. Jan. 28, 2016), aff’d, 888 F.3d 418 (9th Cir. 2018), aka the “Monkey selfie” case. There, photographer David J. Slater was in Indonesia to take pictures of wildlife, when a six-year-old macaque named Naruto picked up his camera and took several images of himself. The People for the Ethical Treatment of Animals (PETA) attempted to obtain authorship status for Naruto; however, the Court ruled that Naruto, as a non-human, could not be an author.
In view of the foregoing precedent, the U.S. Copyright Office’s Review Board reaffirmed the requirement for human authorship in refusing to register a two-dimensional artwork “autonomously generated by a computer algorithm running on a machine” as a work-for-hire. A “Creativity Machine,” consisting of deep artificial neural networks, generated the artwork shown above.
The Copyright Office explained that the Supreme Court “has continued to articulate the nexus between the human mind and creative expression as a prerequisite for copyright protection,” citing Mazer v. Stein, 347 U.S. 201, 214 (1954) and Goldstein v. California, 412 U.S. 546, 561 (1973). Notably, however, the Board’s decision leaves open the key issue of “under what circumstances human involvement in the creation of machine-generated works would meet the statutory criteria for copyright protection.”
The Meaning of “Generative” in Generative Content
Generative content encompasses digital artwork such as images, video, text, and graphics created by computer programs that use some degree of autonomy. Typically, an artist will instruct a computer program to create content using certain algorithms, possibly within certain boundaries.
Many popular generative NFTs rely on some form of algorithmic assistance to randomly select variables within artistic boundaries established by a person. For example, the CryptoPunks NFT collection includes stylized pixel art characters consisting of “24×24 pixel art images, generated algorithmically.” See https://www.larvalabs.com/cryptopunks, accessed Feb. 22, 2022.Similarly, the Bored Apes collection includes cartoon apes, each of which is “unique and programmatically generated from over 170 possible traits, including expression, headwear, clothing, and more.” See https://boredapeyachtclub.com/#/home, accessed Feb. 22, 2022. Some platforms, such as Art Blocks, allow a user to select a collection, and in exchange for a cryptocurrency payment, the platform will algorithmically generate a new and unique version of content from the selected collection on-demand.
By contrast, certain advanced machine learning (ML) techniques, including generative modeling, give computer programs much more autonomy in creating digital content. Generative modeling uses as input an initial set of data, the program learns patterns and features about that data, and the program creates as output new data that could have been taken from the original set of data. There are several techniques for building and training generative model applications using neural networks, and many today use Generative Adversarial Networks (GAN). GANs involve the use of two models: (1) a generator that creates new examples, and (2) a discriminator that classifies the new examples as from the original set of data (real) or not (fake).
Generative models are used create new information with varied properties, such as predicting the next word in a sequence, synthesizing training data, generating video predictions, synthesizing speech or text, among many others. For example, GANs gained some notoriety for their use in creating “deepfake” content—including images, audio, and video content that looks real but is not.
Is Generative Content Subject to Copyright?
The Board’s decision leaves open the question of what minimum level of human input is required in a generative content creation process to meet the statutory requirements of an author. At its heart, the question will likely come down to whether there is an identifiable nexus between the potentially creative elements supplied by the human artist and the output from the computer program. The Compendium of U.S. Copyright Office Practices, § 313.2 provides a few guideposts for where “the Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author:”
- Reducing or enlarging the size of a preexisting work of authorship.
- Making changes to a preexisting work of authorship that are dictated by manufacturing or materials requirements.
- Converting a work from analog to digital format, such as transferring a motion picture from VHS to DVD.
- Declicking or reducing the noise in a preexisting sound recording or converting a sound recording from monaural to stereo sound.
- Transposing a song from B major to C major.
- Medical imaging produced by x-rays, ultrasounds, magnetic resonance imaging, or other diagnostic equipment.
- A claim based on a mechanical weaving process that randomly produces irregular shapes in the fabric without any discernible pattern.
For many generative NFT collections, there is arguably an identifiable nexus between any potentially creative elements supplied by the human artist and output of the program. Namely, a human programs artistic boundaries as to the composition of the digital work, and the artist uses the program as a tool to create content within those artistic boundaries. By contrast, the Creativity Machine at issue in the Board’s recent decision uses significantly more autonomous generative AI modeling techniques to create digital art. With such advanced technology, any potential nexus between the human input and the machine output arguably becomes more tenuous.
The precise legal boundaries of human creativity and machine autonomy continue to evolve with technology, and the Board noted “it is not aware of a United States court that has considered whether artificial intelligence can be the author for copyright purposes.” Critically, it is significant to note that the copyright applicant did not attempt to claim human authorship. Applicants looking to secure copyright protection over generative content would be well advised to name a human author and identify—and preferably document—the contributions that human made in defining the creative output generated by the computer program.