Jack Deutsch
Free Speech in Law and Practice
Free Speech in Law and Practice
December 12, 2023
Artificial (Un)Intelligence
“Create an image in the style of Banksy, showing a girl in a dress holding a red heart balloon,” I ask Bing’s Image Creator. In less than 30 seconds, I have a nearly identical copy of Banksy’s Balloon Girl. The generated images I receive have even captured the little details of Banksy’s famous mural, including the shape of the girl’s dress and the strands of her hair blowing in the wind. Most interestingly, the AI has generated the image as a graffitied mural displayed on a distressed city wall without my instruction; just like Banksy’s original, seemingly knowing about the artist, his style, and his preferred mediums. Since AI-generated images are not owned by anyone and free to use commercially without a license, I can now use this photo in any way I like. Without paying a cent to a stock-photo licensing platform like Getty Images, which charges $499 for the editorial rights to a similar photo of Balloon Girl with a few more details, I now possess a substitute (Guidi, 2022). However, there are some differences between the image I created with AI and Stefano Guidi’s photograph that Getty is selling. By specifying my prompt to account for the discrepancies between the two, I have quickly created an unwatermarked version of this image – and I own it, according to OpenAI’s FAQs, “including the right to reprint, sell, and merchandise – regardless of whether an image was generated through a free or paid credit” (Can I Sell Images I Create With DALL·E?, n.d.)
With two sentences and one minute of effort, I have substituted the need for Getty’s services without costing me one cent. Yes, Getty is a 4.8-billion-dollar company, and Banksy might not need the few extra dollars, but what if it was a small-business-owner or local artist on Instagram? (Attié, 2021) What if it was your work? Meta’s AI-generating software, “Imagine with AI,” used 1.1 billion public Facebook and Instagram posts, probably including yours, to train their model (Gowran, 2022). On Instagram’s Terms of Service declaration, they state that they “do not claim ownership of [public] content, but you grant us a license to use it” (Terms of Use, 2022). Meta’s Privacy Policy also confirms that they “collect, store, and, where relevant, share, profile, review, and curate” such content, including the content we create (Privacy Policy, 2023).
The rampant dissemination of AI-generated content and the scraping of small artists’ content for use in training these models have not been controlled or regulated by current law. Big lawsuits from artists and authors call for AI platforms like Stability and Midjourney to cease stealing work to train models that they are not being compensated for, yet these cases quickly reach a standstill due to the lack of substantial evidence. This means users can freely make and commercialize images generated from data points of millions of artists’ signature styles and likenesses. In some cases, AI can replicate an artist's exact style if named in the prompt, as it did for my prompt referencing Banksy. This is a complete degradation of the United States copyright system’s original intentions of protecting the human author and their right to dictate how a work is used. Modern-day artists can either abstain from promoting their work online or subject their creations to scraping by training software that serves as substitutes for their services. This paper will illustrate the legal history and definition of human authorship and market harm, alluding to how generative artificial intelligence software threatens long-standing and essential copyright aspects. I will also call for a system of lineage determination and watermarking to be imposed on all AI-generated content to provide artists with more evidence to use in cases against large tech companies who would like to continue to profit from the efforts of small artists’ skill and craft.
The rise of AI-generated content challenges traditional precedents surrounding authorship in copyright law, prompting a modern reevaluation of what is eligible for its protections. [ER1] With monumental cases like Silverman v. OpenAI, Inc. appearing in courts, the law faces a new frontier of rapid algorithmization. Since the enactment of the copyright system, the eligible mediums for protection have transformed, but the entities eligible to claim copyright ownership have remained consistent. The questions of what can be copyrighted and who can own it provide valuable insights when analyzing the future of the law's interactions with generative artificial intelligence systems – why is human authorship important, and how do algorithmic artworks compare to those made with human creativity? The answers lie in the rudimentary intentions [ER2] of copyright itself. Courts have consistently determined that the purpose of copyright law is to encourage, protect, and reward creativity pertaining exclusively to human authors, establishing that human authorship is an essential requirement in owning a copyright. The established human authorship standard, which acknowledges creativity as an exclusively human capability, is being challenged by contemporary cases involving AI and the notion of machine creativity. As courts rule on these novel disputes with authorship claims over AI-generated works, examining how human authorship requirements have been solidified historically, how market dynamics between human and non-human competitors are defined through Fair Use’s pillar of market harm, and the legal definitions of creativity can illuminate potential legal outcomes and a path forward.
In the upcoming years, the Supreme Court will be forced to reconsider, reaffirm, or transform two fundamental precedents – whether generative artificial intelligence is eligible to satisfy the requirement of a human author and whether the works created by these algorithms infringe on and compete with the copyrighted material used to train the models[ER3] . These reconsiderations require extensive explorations of the definition of authorship and why their humanity holds such weight.
Origins of Copyright
The legal protection for an author’s intellectual property began in the United States Constitution, which granted[JD4] the right of Congress to “promote the progress of science and useful arts, by securing for limited time to authors and inventors the exclusive right to their respective writings and discoveries” (U.S. Const. art 1 § 8). Subsequent legislation, including the Copyright Act of 1790 and the Digital Millennium Copyright Act (DMCA), updated the law to meet the needs of the times, yet still relying on the original intentions of promoting the protection of an author’s rights. With the advent of the printing press, the American copyright system was born as state-level printing privileges to manufacture and sell texts. Yet, authorship was not recognized until the Copyright Act of 1790, designed to promote national learning by securing an author's rights to print, republish, or vend their original “literary, dramatic, musical, and artistic works fixed to a tangible medium,” for six years, “excluding ideas, systems, facts, or methods of operation” (17 U.S.C. §§ 201-216). The Copyright Act of 1790 also introduced a payment system of sixty cents to the clerk or author if their work was republished. While encouraging learning, this act also asserted a form of responsibility to the author of a work. An author must keep track of their publications, send a copy to the Secretary of State for preservation, and receive monetary damages for any infringements made upon their work. The specification for the work to be human-authored was not articulated but implied, as only a human possesses the ability to hold this form of responsibility (Copyright Act of 1790, 1790). When the act was amended in 1976, an author's life was postulated, updating the copyright period to the duration of the author’s life plus an additional fifty years. (Copyright Act of 1976, 1976). This duration has expanded over time, but copyright law still relies on the supposition that an author can both live and die.
Originality and Creativity Requirements
Copyright also protects the arrangement of an author’s ideas, constituting the involvement of original ideation and creative vision. Burrow-Giles Lithographic Company v. Sarony is a foundational case that brought photography into the scope of copyright protections and, in turn, established a foundational definition of authorship and originality. This case involved an unlicensed use of Napoleon Sarony’s photograph of Oscar Wilde, Oscar Wilde No. 18. Without Sarony’s consent, Burrow-Giles reprinted and sold the photos, infringing on Sarony’s assumed copyright (Burrow-Giles Lithographic Company v. Sarony, 1884). As this case reached the Supreme Court, Burrow-Giles argued that photographs merely detail real-life objects and involve no authorship or creativity in their conceptions, leaving the Supreme Court to decide that photographs are indeed authorized for copyright protection, as they represent visual expressions of an author’s original ideas (Burrow-Giles Lithographic Company v. Sarony, 1884). Burrow-Giles Lithographic Company v. Sarony is an essential lens through which to interpret originality and human authorship, as it questions a machine’s creativity capability. The machine, which in this case is a camera, is not the author of the work it was used to capture - the photographer is. In the proceedings, Lord Justice Cotton stated in his opinion that an author is one “who has actually formed the picture” and acts as “the inventive or master mind” (Burrow-Giles Lithographic Company v. Sarony, 1884). More specifically, Justice Cotton stated that authorship is attributed to “he to whom anything owes its origin,” specifically the “originator, maker, or one who completes a work of science or literature” (Burrow-Giles Lithographic Co. v. Sarony, 1884). The court attributed its decision against Burrow-Giles to the photograph being a product of Sarony’s “original mental conception,” including the creative arrangement of details captured in the photo. Following this logic, originality is not found in a tool’s ability to synthesize data into a physical medium. Instead, originality exists in humans' ability to creatively arrange and act as the “mastermind” in a final project. (Burrow-Giles Lithographic Company v. Sarony, 1884).
Another foundational case, occurring about a hundred years later, defined originality as “sufficient creativity” in the final product of their work, shifting the criteria of copyright to focus on the content of the work itself (Feist Publications, Inc., v. Rural Telephone Service Co., 1991). Feist Publ’ns, Inc. v. Rural Tel. Serv. Co. is arguably where the Supreme Court first defined creativity legally, offering exact quotes to define originality instead of reading between the lines. Prior to this ruling, the courts relied on a “sweat of the brow doctrine” in determining originality, borrowed from English law, which defined creativity as anything an author put hard work and money into, disregarding a sense of overly evident originality (“Sweat of the Brow,” 2023). In this 1991 case, Feist Publications sued Rural Telephone Services for directly copying names, addresses, and numbers from their telephone directory service and reusing the data in their phone book. In the Supreme Court, the case questioned how much creativity must be put into a work for it to be eligible for copyright protection, similar to Burrow-Giles Lithographic Company v. Sarony. Since Feist’s white pages consisted solely of names, numbers, and addresses, the court decided that it lacked “the modicum of creativity necessary to transform mere selection into copyrightable expression” (Feist Publications, Inc., v. Rural Telephone Service Co., 1991). The court also stipulated that “Rural expended sufficient effort to make the white pages directory useful, but insufficient creativity to make it original” (Feist Publications, Inc., v. Rural Telephone Service Co., 1991). The Supreme Court decided that Feist’s white pages were never eligible for copyright protections in the first place and that Rural’s use was not infringing. This case expanded the definition of originality to include the content of the work in the ruling and coined the originality-criteria of the “modicum of creativity” (Feist Publications, Inc. v. Rural Tel. Serv. Co., 1991). This case moved copyright rulings out of the “Sweat of the Brow” doctrine and into the “modicum of creativity” doctrine. As referenced in the ruling, Feist did not “select, coordinate, or arrange the uncopyrightable facts in any original way,” which fails to meet the modicum of creativity. (Feist Publications, Inc. v. Rural Tel. Serv. Co., 1991). However, works “founded in the creative powers of the mind, independently created, and not the result of copying” constitute originality. Thus, these three pillars became the new criteria for copyright ownership. (Feist Publications, Inc. v. Rural Tel. Serv. Co., 1991).
In a sense, AI image generators are performing similar actions as Rural – accumulating data from pre-existing sources and creating something new from it. However, since the data in this case is copyrighted artwork, it infringes on the rights of artists profiting from their work, which involves labor and originality. In generating an image, an artificial intelligence platform takes a user’s prompt, starts from random noise, and builds the photo by referencing millions of photos on the internet to recreate something similar. (How AI Turns Text into Images, 2023). Creating thousands of micro images, these platforms cannot create something novel, as the new image is formulated based on human creativity from the past. However, as these systems advance, they are not merely arranging old photos to create a new one – but instead, relying on the billions of training images to “infer concepts like color, objects and artistic style” (How AI Turns Text into Images, 2023).
Non-Human Entities and Copyright
Are these novel machines founded in the creative powers of the mind, independently created, and devoid of copying, or do AI-generated works violate the three pillars of the “modicum of creativity” doctrine established by Feist? (Feist Publications, Inc. v. Rural Tel. Serv. Co., 1991) To determine AI-generated art as being any of the above, it would have to be human-authored, a distinction which the courts most famously made in the case of Naruto v. Slater, 888 F. 3d 418, 420 (9th Cir. 2010), when a lawsuit was filed arguing for a monkey to hold copyright ownership, grappling with the question of non-human copyright ownership in the form of animal-created works. It sets a precedent by affirming that copyright law exclusively concerns human creators. This landmark case revolved around a viral selfie taken by a monkey using wildlife photographer David Slater's camera equipment (Naruto v. Slater, 2010). Since non-human entities cannot yet file lawsuits, PETA, acting as a next friend, or an “individual who appears in court in place of another who is not competent to do so,” sued the photographer after selling a book containing Naruto’s selfies under his ownership (Next Friend, 2023; Naruto v. Slater, 2010). The plaintiff’s argument claimed that Naruto performed intentional actions in taking said selfies, which should grant the monkey ownership of the image's copyright (Naruto v. Slater, 2010). In the media, this legal saga garnered significant attention due to the question of [JD5] whether non-human entities could hold copyright, claiming Slater falsely claimed to be the author and stole from the monkey by selling copies of the images for profit. The court's ruling determined that animals cannot hold copyright without Article III and statutory standing under the Copyright Act, as it does not explicitly grant the possibility to animals to claim copyright (Naruto v. Slater, 2010). In affirming the district court's judgment, the Circuit Judge clarified that “Naruto - and any animal, since they are not human - lacks statutory standing under the Copyright Act” (Naruto v. Slater, 2010). This was the first solidification of any non-human entity lacking protection by the United States copyright system. Artificial intelligence, being non-human language and data processing algorithms, presents itself as falling under Naruto’s umbrella of non-human entities and, therefore, also lacking protection.
AI and Copyright Ownership
Thaler v. Perlmutter is a case from this year that specifically confirms the lack of protection given to these algorithms (Thaler v. Perlmutter, 2023). This suit involves Stephen Thaler's application to own the copyright on a work of art formulated entirely by his computer system, “The Creativity Machine,” which he claims to be capable of creating entirely new works of art (Thaler v. Perlmutter, 2023). The United States Copyright Office, the department within the Library of Congress that oversees copyright and related policy, rejected his application due to the lack of human authorship in manufacturing the final product. Even though he coded his machine and alleged to have provided human authorship in the form of instructing the model on a clear prompt, the Office determined that this level of contribution was insufficient to qualify as ‘human’ (Thaler v. Perlmutter, 2023). The district court also ruled that the Copyright Office was correct in denying Thaler ownership of the copyright, as “no valid copyright has ever existed in work generated absent human involvement, leaving nothing at all to register and thus no question as to whom that registration belonged” (Thaler v. Perlmutter, 2023). This case strongly reinforces the law’s requirement for copyrighted work to be human-authored, solidifying that “human authorship is a bedrock requirement of copyright ownership” (Thaler v. Perlmutter, 2023). Thaler’s case shows that the Copyright Office requires human input and clear signs of “ultimate creative control” in the final product for a work to constitute human authorship, reaffirming the precedents set in both Feist Publications, Inc. v. Rural Tel. Serv. Co., and Burrow-Giles Lithographic Co. v. Sarony (Thaler v. Perlmutter, 2023).
While this ruling appears to be a clear and definitive rejection of AI ever being able to hold copyright, it also leaves room for further negotiations. A recent copyright registration for a comic book featuring AI-generated images has examined this gray area, suggesting a human-computer hybrid form of authorship where a human author collaborates with the generative algorithm and exercises “ultimate creative control” over the output. Kristina Kashtanova was granted copyright for the totality of their book, Zarya of the Dead, by the United States Copyright Office, which was later revisited upon learning that the images were created using Midjourney’s generative AI service. (United States Copyright Office, 2023). Instead of revoking the copyright, the office modified the protections to include solely the text and arrangement of images, leaving only the material “product of human authorship” under protection (United States Copyright Office, 2023). The following quote from the letter indicates a clear precedent that the Copyright Office intends to uphold going forward - “the Office will not register works produced by a machine or mere mechanical process that operate randomly or automatically without any creative input or intervention from a human author” (United States Copyright Office, 2023). Since software manufactured the images randomly, they cannot be owned by any author, even the AI platform. However, Kashtanova made substantial edits to a few of the images they used in their book, including altering colors and facial structures of characters in post-production (United States Copyright Office, 2023). The Copyright Office indicated that these images would not be excluded from the revised copyright (United States Copyright Office, 2023). This diverting stipulation to the rule, denoting that substantial edits to AI-generated outputs entitle such works to copyright protection, sets a noteworthy precedent going forward, suggesting that there is a possibility to perform sufficient creativity even if the beginnings of a work are created by artificial intelligence. [ER6]
The generative AI models behind works like the illustrations in Kashtanova’s Zarya of the Dead are trained on unlicensed creative works, signifying that the generated images from platforms like Midjourney are grounded in copyrighted works made entirely by humans. Beyond the criteria of creativity to constitute authorship, copyright law has long maintained goals of promoting the progression of creativity through economic incentives tied to intellectual property rights (17 U.S.C. §§ 201-216). Smaller artists make a living by licensing and selling their art, which in turn incentivizes them to continue creating. What is incentivized by bestowing copyright protection to algorithms? This year, artists Sarah Andersen, Kelly McKernan, and Karla Ortiz sought justice against Stability AI, Midjourney, and DeviantArt, who they allege have used their artworks as training material for their software without compensation or consent in Anderson et al. v. Stability AI LTD (Anderson et al. v. Stability AI LTD, 2023). In the proceedings, Stability representatives have admitted to storing compressed versions of over five million works scraped from the internet, which is a direct infringement in the plaintiffs' argument (Anderson et al. v. Stability AI LTD, 2023). To prove a copyright infringement, U.S.C. 17 requires a (1) valid copyright and (2) “evidence that the defendant copied original expression from the copyrighted work” (17 U.S.C. §§ 201-216). Since the AI generators use millions of training images to create new works, the plaintiffs have no one image that appears to be a blatant infringement, nor is there an imposed system of watermarks that encode origination data to be used as evidence (Anderson et al. v. Stability AI LTD, 2023).
Similarly, stock image database Getty Images is also suing Stability over its use of Getty’s stock images in their AI models. In Getty Images (US) Inc. v. Stability AI Inc., Getty argues that Stability’s unlicensed use of its stock images creates unfair competition. Instead of licensing an image and compensating the author, anyone could create an unwatermarked, more customizable stock-like image using Stability’s platform, threatening Getty’s marketability. In addition to human authorship, these cases pose another question to the courts regarding AI and copyright - do AI's outputs threaten human authors' marketability, or are they a form of harmless entertainment? Since the final product is art and could threaten the original’s marketability, it is best interpreted under the Fair Use Doctrine and its pillar of market harm.
Fair Use Doctrine and Market Harm
In 1841, the case of Folsom v. Marsh led to the creation of “fair use,” a doctrine allowing the use of copyrighted material for specific uses. Jared Sparks and Folsom, his publisher, published a biography about George Washington featuring letters penned by Washington himself. Under his publisher, Marsh, Reverend Charles Upham also wrote a biography on Washington, based entirely on - and including - the letters Sparks used in his book (Folsom v. Marsh, 1941). When Sparks filed suit, Upham claimed his use was fair and justified. This case questioned whether using small but valuable pieces of copyrighted material is fair if they are essential to the purpose of the new work (Folsom v. Marsh, 1941). The District Court of Massachusetts found that Upham’s use of Spark’s work was unfair, setting up a list of reference points to consider when considering fairness: the purpose, nature, quantity, and value of a use, and whether it devalues the market of the original work (Folsom v. Marsh, 1941). A use is deemed fair if, according to Justice Story, it does not “supersede[s] the objects” of the original and, instead, offers something transformative (Folsom v. Marsh, 1941). The Fair Use doctrine aims to forge a balance between an author’s right to their work and the public’s interest in the work for purposes such as criticism, parody, or research. An author may use copyrighted material if something new is created, and not serving as a substitute for the original, undermining the copyright system’s goals of incentivizing potential authors to pick up a pen and make their own work. This case and the precedents it set are instrumental when interpreting AI’s uses of an original author's work. The holding of Upham’s use as infringing shows that even a small, undetectable use of copyrighted work still constitutes infringement if essential to the purpose of the new work or serves as a substitute, akin to AI’s small and undetected use of copyrighted images (Folsom v. Marsh, 1941).
However, what constitutes transformative and “non-superseding” is constantly debated in case law. The most recent Fair Use case to reach the Supreme Court, Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, examines Andy Warhol’s use of a photograph by Lynn Goldsmith. Warhol used the image in a reinterpretative art series licensed by Vanity Fair in 1981 (Andy Warhol Foundation for the Visual Arts Inc. v. Goldsmith, 2023). After the photo was used again on a 2016 Vanity Fair cover sans license, Goldsmith contacted the Andy Warhol Foundation, informing them that her copyrighted photograph had been infringed upon. In return, AWF sued Goldsmith for a declaratory judgment of noninfringement, claiming fair use (Andy Warhol Foundation for the Visual Arts Inc. v. Goldsmith, 2023). Since fair use arguments require the court to make inferences on the meaning of an artwork, the Supreme Court tends to refrain from hearing such cases, and this was the first art case in decades to reach the Supreme Court. AWF claimed that Warhol’s use of Goldsmith’s photograph was transformative, adding something new to the work and not serving as a substitute (Andy Warhol Foundation for the Visual Arts Inc. v. Goldsmith, 2023). However, the Supreme Court justices ultimately disagreed, holding that the work was not transformative and did infringe upon Goldsmith’s copyright since the results “share substantially the same purpose, and the use is of a commercial nature” (Andy Warhol Foundation for the Visual Arts Inc. v. Goldsmith, 2023). This, and any fair use case, signifies the law’s ability to deem what purpose art serves and reinforces the intentions of the law to uphold the “incentives for artists to create original works in the first place” and, in turn, hardening the criteria surrounding what constitutes Fair Use (Andy Warhol Foundation for the Visual Arts Inc. v. Goldsmith, 2023). This case, and the focus on market harm as a deciding factor, applies to artificial intelligence and a similar threat of market harm. In the previously mentioned Getty Images (US) Inc. v. Stability AI Inc., Getty Images seeks to recover damages suffered by the allegedly intentional acts of Stability in inflicting predatory market harm on their business (Getty Images Inc. v. Stability AI Inc, 2023). Getty’s legal team has further accused Stability of endeavoring to steal business by marketing to those seeking creative imagery and have even noticed many of Stability’s outputs as “containing a modified version of a Getty Images watermark, creating confusion as to the source of the images and falsely implying an association with Getty Images” (Getty Images Inc. v. Stability AI Inc, 2023). Under the precedent reaffirmed by the Andy Warhol Case, a commercial work intending to serve the same purpose as an original is indeed an imposer of market harm and should result in a recovery of damages.
Campbell v. Acuff-Rose Music, and Intended Purpose of Derivative Works
Historically allowed uses of copyrighted works criticize, comment on, report, teach, research, and parody a previously existing work’s message. Parody, the latter of these allowed uses, is a looser interpretation of fair use deemed acceptable after the case of Campbell v. Acuff-Rose Music. The case involves hip-hop group 2 Live Crew, Roy Orbison's "Oh, Pretty Woman," and Orbison's publisher, Acuff-Rose Music. After Acuff-Rose Music denied 2 Live Crew's request to license "Oh, Pretty Woman" on their upcoming album, the band released the song regardless and sold 250,000 copies that year (Campbell v. Acuff-Rose Music, 1994). The publishing company filed suit for infringement, eventually reaching the Supreme Court. 2 Live Crew's Luther Campbell claimed their song parodies the original's naivete and rejects the traditional values it details (Campbell v. Acuff-Rose Music, 1994). The Court sided with Campbell and deemed its use fair in the name of parody, serving no threat to the original's marketability. This case is significant as it held that parody constitutes fair use under Section 107 and refined copyright laws as emphasizing the overall character, purpose, and relative marketability of a work's use (Campbell v. Acuff-Rose Music, 1994).
Nevertheless, it is illuminating to note that AI art platforms make no such parodical, critical, or transformative claims. Much like how the Warhol ruling found the creation of commercial work with the same purpose as a copyrighted reference to cause market harm, AI outputs serve free and identical purposes to the licensed commercial stock and fine art photographs they are trained upon. There is a strong argument that creating substitutions for licensed products using artificial intelligence models trained on photos that would otherwise require a license is the circumvention of a paywall, a direct violation of the Digital Millennium Copyright Act. The DMCA makes punishable by law the “circumvention of a technological measure that effectively controls access to a work protected under this title” (17 U.S. Code §1201, 1A). It is vital to analyze A.I.-generated art’s threat to the marketability of human creativity.
Reccomendations
At the current stage of legislation, United States law leaves many questions unanswered regarding the grey area of AI's interaction with intellectual property and copyright protections. The barriers to entry in creating art are reduced to almost none by AI, and suddenly, anyone with a phone can make a modern painting in the style of Monet. Generative AI instantaneously creates professional-looking artwork without the requirement of skill or labor on the part of the user. While the software engineers insert their unique gifts of creativity and mastery when coding the models, the data used to train the models must be regulated, as well as how the users can use the outputs generated through this software. This democratization of mastery has begun to create a world that devalues the artist and does not need the human author. While it is not yet possible to claim copyright on an image generated entirely by AI, the advent of generative artificial intelligence threatens the sanctity of human authorship. In 2018, an artwork made entirely by AI, “Edmond de Belamy, from La Famille de Belamy,” sold in New York for $432,000 (Cohn, 2018).
Copyrightable or not, AI-generated art presents lucrative business opportunities bypassing the input of a human artist. For example, OpenAI, the language processing algorithm behind ChatGPT, which also faces copyright infringement lawsuits, is valued at roughly $86 billion (Silverman v. OpenAI, Inc., 2023; Levine, 2023). Additionally, Getty Images is requesting damages of up to two trillion dollars from Stability AI for using Getty stock photos to train AI image generators that allow consumers to create their own stock-like images (O’Brien, 2023). Mobile applications like Lensa charge subscription fees to create realistic AI avatars, generating mass profit and potentially substituting for a professional photographer. This legal chaos and confusion make it clear that regulation is needed quickly.
Legal debates regarding this regulation are still in their relative infancy, with some governments, like the European Union, proving far ahead of American policy. There is a delicate balance between enabling innovation and protecting the human author and their fundamental human rights. However, like most political battles, the party with more influencing power is the wealthy and well-connected, who invest heavily in novel innovations. Tech companies like Mark Zuckerberg’s Meta, who hope to profit from AI, outpower the average human artist in persuasive power. We live in an exciting time for innovation, but if the law does not introduce a series of regulations and policy changes, a real threat is posed to artists whose life work is creating artwork; we must protect humanity in the arts. In addition to funding legal defense networks for artists and an international policy update, the introduction of lineage determination and opt-in systems will ensure that the law can keep up with the times, prioritizing innovation and the rights of artists alike.
The courts steadfastly uphold the human authorship requirement of copyright, and the Andy Warhol case illustrates that the monetary incentives for creating human art remain at the forefront of the discussion. I do not believe that artificial intelligence meets the modicum of creativity established by Feist needed to constitute human authorship, nor do I believe that the law should make space to protect these generated images. At the current stage of AI models, the argument that AI-generated art is “founded in the creative powers of the mind, independently created, and not the result of copying” lacks substantial evidence (Feist Publications, Inc. v. Rural Tel. Serv. Co., 1991). This is evident in the fact that the art is randomly created from a language-based prompt, the utilization of data points collected from copyrighted images, and the absence of human control over the final product (How AI Turns Text into Images, 2023). AI is not an extension of the human eye but a service that creates randomized images without the creative liberty to control what results. In this situation, the user or machine uses no means of creativito constitute human authorship, and the courts have upheld this notion. The law must protect those meeting the “modicum of creativity” when creating their art, which will result from regulating the artwork AI is trained on.
Using unlicensed artwork to train generative AI models is a dangerous infringement of artist’s rights. Chances are, if an artist posts their work to social media or anywhere on the internet, it will be ingested in a data mine for use in training generative software. Prior to last December, artists had no option to opt out of this use (Heikkilä, 2022). They could either refuse to showcase their work online and miss out on potential exposure or accept unlicensed use of their creations and receive no compensation for their efforts. I align with the idea of an opt-in system instead of an opt-out system when it comes to this issue. By law, big tech companies should not be permitted to impermissibly scrape an artist’s work from their Instagram or website to use them as they please. Social media is essential for small businesses to profit from a trade by reaching new audiences and providing access to a point of sale. Instead of retroactively searching for an option to opt out of something they might not even know exists, artists should have to expressly allow these platforms to use their art in training their software. An opt-in system would enable artists to negotiate fair compensation for their work that fuels AI development while still benefiting from online exposure. Requiring consent also upholds artists' rights over their creations in the digital realm, thereby reaffirming the monetary incentives to copyright, as mentioned in the Warhol case. Placing the power back into the hands of artists allows for a shift in the new era of algorithmic creativity that prioritizes the humans essential in its creation. Integrating compensation and credit for the use of artwork, AI has the opportunity to become a profitable space for artists and tech entrepreneurs alike.
In addition to an opt-in system, establishing a transparent lineage determination system is another essential aspect of litigating a fair deal between artists and technology, a sentiment President Biden’s executive order on artificial intelligence also called for. In his order, the President insisted that a watermarking system be embedded into all AI outputs within half a year of his instruction (Biden, 2023). Transparency is essential in bridging the information gap between AI-generated artworks and their creators, ensuring accountability and protection for artists. A data lineage determination system is the “process of tracking the data flow over time, providing a clear understanding of where the data originated, how it has changed, and its ultimate destination within the data pipeline” (What Is Data Lineage?, n.d.). This system of transparency is used in tech companies to ensure the quality of data in reports and to easily trace the flow of information, making it easy to pinpoint errors or security breaches. A data lineage system is commonly tracked through encoded watermarks, which are digital stamps that store authorship and access data to prevent files from being copied or infringed upon. They are embedded in a digital product and are difficult to disarm or remove.
Similarly, the European Union’s “EU AI Act,” passed in June of 2023, has already enacted the requirement of generative AI platforms to disclose summaries of copyright data used for training (European Parliament, 2023). With a data tracking system encoded in all AI outputs, entities will be held accountable for using and distributing copyrighted digital material. In cases like Anderson et al. v. Stability AI LTD., the plaintiffs are having a difficult time proving their artwork was infringed upon and used to train models as Stability’s software uses millions of reference images and does not provide a list of data points used in the creation of a generated image (Anderson et al. v. Stability AI LTD, 2023). With a comprehensive data lineage system, artists can gather more substantial evidence to seek compensation for infringement on their works. Watermarks contribute to a supportive environment for the advancement of human ideas amid such radical transformation. Artists are more prone to sharing their work and contributing to the creative landscape when they have confidence in the protection of their intellectual property. The balance between protecting intellectual property and fostering innovation is challenging to find, but instilling priorities of transparency is an important first step in getting the ball rolling, allowing courts a more comprehensive picture of how much unlicensed art is being used in these AI models.
With a database showing the transaction of data and art used in training a model, there is a unique opportunity to value the human author of the training images by introducing a payment system for the opted-in use of an artist’s work. Admittedly, it is unrealistic to suppose that all generative AI models will be able to use only a selected handful of licensed work, and it may even slow the advancement of these models, but moving towards transparency and an even distribution of profit is vital. Adopting a system like the music industry’s, modeled after the framework of performance royalties, could benefit artists who find themselves at the hands of modern AI. To play an artist's song in a commercial venue, a proprietor must secure a blanket license from Performing Rights Organizations. In the case of art and artificial intelligence, there is an opportunity to create a similar fund for artists who allow their work to be used in training for-profit models. AI companies should be required to license the images they use if creating something for profit, and there needs to be transparency in which images are used to create the final image in a user’s prompt. This way, lawsuits will have more evidence to substantiate these cases, and a profit system can be created to benefit artists.
As evident in Burrow-Giles Lithographic Co. v. Sarony, even photography once challenged copyright as we know it. Digital music, film, software, games, and software each required the American copyright system to rethink itself, finding new ways to protect artists and innovation. Artificial intelligence is the next medium to challenge the law to remain fair, ethical, and beneficial to humanity. However, never before has the law had to consider a new form of consciousness and the work it creates off the back of human artistry. Getty Images v. Stability AI illustrates why the threat of leaving artificial intelligence unregulated is dangerous. Why subscribe to a stock photo licensing company that employs hundreds of smaller photographers; why hire a portrait photographer; why buy local art when you can instantly generate an image of similar quality for free? The state of artificial intelligence currently disregards the fundamental intentions of copyright, which are to protect original works of authorship to promote human discovery. By bestowing copyright protections onto the work of artificial intelligence, there is little to no incentive remaining to create physical and original work. If the law continues to allow artificial intelligence to scrape and eventually emulate human capabilities, it will show potentially disastrous consequences for humans to be anything other than reliant on artificial intelligence. Until generative AI models can abide by a transparent system that either credits and licenses the data points they ingest or creates entirely new pieces of work without human art as the backbone, it is imperative that we exercise caution and discretion in commercializing AI-generated. We are at such an innovative, transformative time in global history; by providing a way for artists to get paid for their use in these models and abiding by copyright’s original intentions, we can uphold the values of protecting humanity, all while introducing new ways for users to get involved in creative expression.
References
Anderson, et al., v. Stability A.I. LTD. et al., 3:23-cv-00201, (N.D. Cal. 2023)
Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, (598 U.S. 2023)
“An image in the style of Banksy, showing a girl in a dress holding a red heart balloon,” Dall-E3, 10 December, 2023.
Attié, I. (2021, December 14). Getty Images goes public again - valued at $4.8 billion. Stock Photo Secrets. https://www.stockphotosecrets.com/news/getty-images-goes-public-again.html
Banksy. (2002). Balloon Girl [grafitti mural]. London, England.
Biden, J. (2023, October 30). Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. The White House. Retrieved November 26, 2023, from https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/
Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884)
Campbell et al., v. Acuff-Rose Music, Inc. 510 U.S. 569 (1994)
Can I sell images I create with DALL·E? | OpenAI Help Center. (n.d.). https://help.openai.com/en/articles/6425277-can-i-sell-images-i-create-with-dall-e
Cohn, G. (2018, October 25). AI art at Christie’s sells for $432,500. The New York Times. https://www.nytimes.com/2018/10/25/arts/design/ai-art-sold-christies.html
Copyright Act of 1790 (1790). https://copyright.gov/about/1790-copyright-act.html
Copyright Act of 1796 (1976). https://www.copyright.gov/title17/92appa.html
European Parliament. (2023, June 6). EU AI Act: first regulation on artificial intelligence. Retrieved November 26, 2023, from https://www.europarl.europa.eu/news/en/headlines/society/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence
Feist Publications, Inc., v. Rural Telephone Service Co., 499 U.S. 340 (1991)
Folsom v. Marsh, 9. F.Cas. 342 (C.C.D. Mass. 1841)
Getty Images (US) Inc. v. Stability AI Inc., 1:23-cv-00135 (D.D.C. 2023)
Gowran, L. M. (2022, September 22). “Legal minefield”: The risk of commercialising AI-generated images. Silicon Republic. https://www.siliconrepublic.com/machines/ai-generated-images-legal-risks-copyright
Guidi, S. (2022, February 24). A man walks near a Banksy artwork during the “The World of Banksy”. . . Getty Images. https://www.gettyimages.com/detail/news-photo/man-walks-near-a-banksy-artwork-during-the-the-world-of-news-photo/1372532092
Heikkilä, M. (2022, December 16). Artists can now opt out of the next version of Stable Diffusion. MIT Technology Review. https://www.technologyreview.com/2022/12/16/1065247/artists-can-now-opt-out-of-the-next-version-of-stable-diffusion/
Isaacs-Thomas, B. (2023, January 12). How AI turns text into images. PBS NewsHour. Retrieved November 3, 2023, from https://www.pbs.org/newshour/science/how-ai-makes-images-based-on-a-few-words
Instagram. (2022, Febuary). Terms of use. https://help.instagram.com/581066165581870
Instagram. (2023, June). Privacy Policy. https://privacycenter.instagram.com/policy/
Legal Information Institute. (2023, July). Next friend. https://www.law.cornell.edu/wex/next_friend
Levine, M. (2023, November 27). OpenAI is still an $86 billion nonprofit. Bloomberg.com. https://www.bloomberg.com/opinion/articles/2023-11-27/openai-is-still-an-86-billion-nonprofit
Naruto v. Slater, 888 F. 3d 418, 420 (9th Cir. 2010)
O’Brien, M. (2023, September 25). Getty Images’ up to $1.8 trillion lawsuit about AI plagiarism is still pending, but it just launched an AI image generator. Fortune. https://fortune.com/2023/09/25/getty-images-launches-ai-image-generator-1-8-trillion-lawsuit/
“Photograph of a man in all black walking past a graffiti drawing in the style of Banksy that shows a girl in a dress holding a red heart balloon. ‘There is always hope’ is written above in skinny handwriting,” Dall-E3, 10 December, 2023.
Sutton, B. (2019, March 6). Artificial intelligence artwork by Mario Klingemann sells for £40,000 at Sotheby’s | Artsy. Artsy. https://www.artsy.net/article/artsy-editorial-artwork-created-ai-sold-40-000-sothebys-failing-generate-fervor-propelled-ai-work-sell-40-times-estimate-year
Thaler v. Perlmutter, 1:22-cv-01564, (D.D.C. 2023)
U.S. Congress. (1958) United States Code: Copyright Office, 17 U.S.C. §§ 201-216. [Periodical] Retrieved from the Library of Congress, https://www.loc.gov/item/uscode1958-004017003/.
United States Copyright Office. (2023). Re: Zarya of the Dawn (Registration # VAu001480196).
What is data lineage? (n.d.). IBM. Retrieved November 22, 2023, from https://www.ibm.com/topics/data-lineage
Wikipedia contributors. (2023, October 2). Sweat of the brow. Wikipedia. https://en.wikipedia.org/wiki/Sweat_of_the_brow