“A critical policy issue to explore is the obligation, ethical or otherwise, to identify works that a computer has created…. Future regulations will need to explore the obligations, if any, that are placed on authors and computer users when seeking protection over generative works.”
If you follow Reese Witherspoon on Twitter, you may notice she has been tweeting about non-fungible tokens, or NFTs, a lot. She currently features an NFT as her Twitter profile picture (abbreviated “pfp” for those in the know). In October 2021, Witherspoon became a partner in an NFT art collection called World of Women, or WoW, which was created and illustrated by the artist Yam Karkai. Through an auction-style bidding process, the WoW collection is currently available on OpenSea, one of the largest NFT marketplaces. As of publication, an individual WoW NFT auction starts at around 7 Ethereum (ETH), the cryptocurrency used to purchase on OpenSea, which currently equates to approximately USD 20,000.
Understanding the NFT Collection
There are many types of non-fungible tokens. NFTs are most commonly made available using the Ethereum blockchain network. A non-fungible asset can be one unique item or multiple unique items, and the token can provide a range of rights or permissions to the owner of the token. NFTs are not limited to artwork and can include virtual and real-world benefits to an owner, such as the Bud Light NFT launched during the 2022 Super Bowl.
An NFT collection is a digital art collection released by an artist, or group of artists, containing a limited number of individual NFTs. There are usually several stages of a release, after which time the NFTs trade on the open market. It’s most common to see cartoon, or vector, style NFT collections of 10,000 pieces that can be used as an avatar or pfp. One may be left wondering how an artist, or even a group of 1,000 artists, could generate 10,000 individual, unique NFTs so quickly. In a way, they don’t. That’s where computer software, sometimes with an underlying artificial intelligence or machine learning component, comes into the mix.
Is It Generation or Creation of NFT Collections?
NFT Collections start with an artist (or artists) creating a base frame or outline of a character’s head for their NFT collection, such as an ape, punk, or cat. The artist then creates a set of unique attributes in the form of easily manipulated digital image layers. For example, an artist may build out various options for backgrounds, facial expressions, hair, eyes, accessories, colors, outfits, and more, referred to as attributes. The layers are then combined by software to generate the larger collection of works autonomously. Furthermore, artists can also incorporate traits, such as defining a matching top and bottom attribute, to further direct the software used in creating these generative works of art. A quick search on YouTube will reveal hundreds of instructional videos explaining how to create an NFT collection, with many promising results in under an hour.
The rarity of the computer-generated combinations can cause the price of individual NFTs to skyrocket. Online tools, liked rarity.tools, offer resources to explore the various rarity rankings of the NFTs within a collection, using a unique calculation depending on how the NFT collection is valued by its community (such as a limited number of attribute and trait combinations). In addition, through a range of dashboards and charts, most NFT collections self-publish information regarding the outcomes of the software combinations, such as how many of a particular trait or attribute combination are available.
There are varying levels of sophistication in the software used by artists creating NFT collections. Some software simply runs through the mathematical process of generating every possible combination of the attributes and traits. The software can be further refined to follow sets of rules. For example, if a character has a red hat, there will be a 25% chance of having a nose ring. This rules-based process is often referred to as generative coding and can assist with more complex NFT collection creations. Other software utilizes artificial intelligence or machine learning to output more sophisticated and autonomously curated combinations, such as the Eponym NFT project.
The concept of generative art, or computer art, is nothing new. The 2009 paper “What is generative art?” by Margaret A. Boden & Ernest A. Edmonds and published in the peer-reviewed journal Digital Creativity, provides a deep exploration into the use of computers to generate works of art both entirely autonomously or with minimal human involvement. Adobe Sensei technology provides AI-powered tools to do everything from basic tasks such as removing a background or an object from an image or video to more complex tasks such as the delivery of real-time, intelligent analytics from data. Autonomous creations are even taking place within some metaverses.
Copyright Protection for Works Generated Using Artificial Intelligence or Machine Learning Software
Presently in the United States, there is no federally recognized copyright protection afforded to the output of artificial intelligence software, or animal selfies, for that matter. It’s worth noting that copyright protections extend to works created by humans with the assistance of computers (such as AI-powered word processors or templatized creation of works, such as drag-and-drop graphics editors); however, such protections stop when a work is the expression of a computer alone or the result of a random or process-driven expression. For further discussion supporting copyright laws recognizing computer users initiating the creation of computer-generated works, see “Ex Machina: Copyright Protection for Computer-Generated Works and Session Two of the February 2020 USCO and WIPO event, Copyright in the Age of Artificial Intelligence.
Section 306 of the Compendium of U.S. Copyright Office Practices, Third Edition, (Compendium) explains that works by non-human authors will not be registered. The Compendium entry references two cases from the late 1800s: Trade-Mark Cases, 100 U.S. 82, 94 (1879) and Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884). However, a recent court decision involving a monkey selfie and other ongoing debates about artificial intelligence have further solidified the human authorship rule, of which the Compendium includes two specific examples for works created by animals: “a photograph taken by a monkey” and “a mural painted by an elephant.”
Furthermore, the Compendium makes the human author requirement abundantly clear in Section 313.2 when referencing the U.S. Copyright Office, Report to the Librarian of Congress by the Register of Copyrights 1965:
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 a machine.”
Note that the Compendium in its current form incorrectly references and links to the 1966 Report, but the discussion of works generated by a computer is found in the 1965 Report.
As evidenced in the Supreme Court’s decision in Feist Publications, Inc., v. Rural Telephone Service Co., 499 U.S. 340 (1991), the U.S. Copyright Office will not register works generated or otherwise created by non-humans because they often fail to meet the creativity requirement. This means machines, algorithms, artificial intelligence, and related technological advancements have faced issues with respect to registration for federal copyright protections, and general recognition as being capable of producing copyrightable works without some level of human intervention.
A similar, but arguably distinct, issue arises within the context of database registrations. The threshold authorship question is whether the database “incorporate[s] authorship in the programmer’s expression of original ideas, as distinguished from the ideas themselves.” (H.R. REP. NO. 94-1476, at 54 (1976), reprinted in 1976 U.S.C.C.A.N. at 5667.) To be registered with the U.S. Copyright Office, the claimant must own the copyright in the material.
Could there be an argument that the materials underlying the NFT collection rise to the level of creativity necessary for copyright protection? Or, would those fail under an analysis similar to the useful articles issue in Star Athletica v. Varsity Brands and qualify as a protectable work separate from the useful article?
Authorship may be identified through the selection, coordination, and/or arrangement of the materials within the database, according to Section 727.3(D) of the Compendium. Consider when software is used to create an NFT collection. In that case, it’s arguable that in the case of NFT collection creation, the software itself performs the actions of selection, coordination, and arrangement, removing any supporting argument of human authorship.
The Typical NFT Collection Contract
Putting aside the issue of whether there are any rights to be transferred, precisely what rights do owners of an NFT receive? In a true lawyer answer, it depends. Every NFT project can be structured however the artist, or the technical developers supporting the artist, decide. A corporate entity, or in some cases a more complex multi-corporation structure, is typically established that is used to release an NFT project. On paper, this provides for a liability shield to the artists when it comes to regulatory oversight (current or future), personal liability, and other risks not-yet-known in the world of cryptocurrencies, blockchain, and NFTs.
There are considerable gaps in most NFT collection license agreements. There typically will not be any representations or warranties made by the artists that they own or have the rights necessary to grant the permissions to the NFTs. Arguably, most of these licenses fall short of a quitclaim license structure – missing only the express language essential to protecting the artists. The overall lack of representations and warranties could be a good thing for artists, given the argument that copyright protection doesn’t extend to these types of works. However, there are also no liability caps to protect the artists, or their corporate entities, from claims that might come up due to the NFT collection project. In addition, there is no indemnification language to cover either the artist or the NFT owner. The artist should include indemnification coverage for any bad actions taken by the NFT owner. Likewise, the NFT owner should seek indemnification coverage, or at least an obligation that the artist pursues (or permits the NFT holder to pursue) infringement claims by third parties, to ensure they have uninterrupted, or quiet enjoyment, of title to their artwork.
It’s worth noting that blockchain-based NFT transactions are, by default, partially anonymous due to the inherently decentralized, and therefore easily anonymized, manner in which cryptocurrencies operate. As such, it would be challenging, if not impossible, to uncover the true identity of a licensee that has breached the license agreement without having some external registration system for NFT holders.
Are any of these terms or transactional nuances ultimately considered deal breakers? Not for the tens of thousands of transactions logged every month on the Ethereum blockchain. However, it seems odd that individuals spend tens of thousands to millions of dollars on NFTs without giving much thought to the underlying contract terms that govern.
Crafting a Policy for the Future
If software is used to autonomously create 10,000 musical compositions based on a set of four cords, should those compositions be eligible for copyright protection? Does it make a difference if an AI or machine learning model was used? What if the model was trained using the top songs from the Billboard charts to mimic popular music styles? Musicians have already been experiencing an unprecedented amount of copyright infringement lawsuits. The chances of potentially infringing on an existing work would substantially increase, along with the risks of releasing music for fear of infringement claims.
Historically, U.S. congressional action drives regulatory changes to copyright laws in the face of technological advancements. The USCO is tasked with implementing any legislative changes to the law. The Office of Registration Policy and Practice within the USCO is currently exploring the potential ramifications of AI on copyright law, including during the February 2020 USCO and WIPO event Copyright in the Age of Artificial Intelligence.
During the event, then-Acting Register of the U.S. Copyright Office Maria Strong shared the Supreme Court’s thoughts from Burrow-Giles, “There is no question a copyright law allows for the use of tools in creation, but at what point does setting something into motion mean authorship?” Strong commented, “this is something photographs and AI have in common: where is the line of authorship? What about ownership issues?”
In February 2022, the Review Board of the United States Copyright Office (CRB) issued an opinion affirming a 2019 decision to refuse registration of Stephen Thaler’s work titled “A Recent Entrance to Paradise,” which he noted “was autonomously created by a computer algorithm running on a machine.” This was the second time Thaler requested reconsideration. In his registration application, Thaler listed the author of the work as “Creativity Machine” and included a transfer statement “ownership of the machine.” The CRB refused to accept either of Thaler’s two arguments: (1) denial of registration is unconstitutional and unsupported by case law; and (2) a machine is capable of creating a work made for hire.
Globally, the issue of ownership over machine creations or generative art is unsettled and varies significantly from jurisdiction to jurisdiction. In the UK, computer-generated works are eligible for copyright protection, but the concept is under review by the Intellectual Property Office. Under French law, which is partially applicable to WoW by governing law of the license agreement, the laws recognizing ownership are not entirely clear. For patent law, exploration of a similar issue regarding the use of AI to either assist with or completely support the invention process is underway. Jurisdictions around the globe continue to explore AI’s impact on both ownership and inventorship.
A critical policy issue to explore is the obligation, ethical or otherwise, to identify works that a computer has created. There are few means of readily identifying works created entirely by a computer. A fully computer-generated work could likely be registered if an application to register a work with the USCO is submitted. Future regulations will need to explore the obligations, if any, that are placed on authors and computer users when seeking protection over generative works.
Artists within some NFT communities are beginning to explore the implications of copyright law on their creations. It seems that many NFT collections will continue to be released by artists, and purchased by willing buyers, despite the lack of clear ownership rights within established intellectual property laws. Is this wrong? Or, is it a new form of digital ownership under a concept of ersatz ownership, or ownership established by contract. We’ll find out at some point in the future once regulatory regimes catch up to technology, and, hopefully, everyone will retain some form of ownership over their newly minted profile pictures. Otherwise, we may all be free to use Reese Witherspoon’s pfp however we please.