Generative Artificial Intelligence: A Tool Or A Crutch? Examining The Authorship Requirement For Copyright Protection In The Context Of AI-Assisted Works
Introduction
In a graphic novel titled Zarya of the Dawn, Kristina Kashtanova used a generative artificial intelligence (generative AI) tool, Midjourney, to assist her in authoring her novel.[1] The original conception and inspiration for the novel were based on Kashtanova’s personal lived experiences.[2] After Kashtanova lost her grandmother, she sought to explore her grief by creating a story of a girl named Zarya who wakes up in an abandoned world with no memories of the world around her.[3] Kashtanova personally wrote all the text in the novel.[4] Kashtanova provided Midjourney with hundreds of text prompts for each final image, which the program then used to create the images in the novel.[5] In February 2023, the United States Copyright Office (USCO) granted limited copyright protection to Kashtanova’s novel.[6] The USCO granted Kashtanova copyright protection for the text in the novel and for the selection, coordination, and arrangement of the novel’s visual elements; however, the actual images were not granted copyright protection as the USCO asserted they were not the product of human authorship.[7]
With the rise of generative AI technology, many artists, like Kashtanova, have elected to use generative AI tools to assist them in making their works (AI-assisted works).[8] To use a generative AI tool, a user must first submit a text prompt directing the AI program to generate an output consisting of text, audio, or an image.[9] Once the AI program generates an initial output, the user can submit further prompts directing the AI tool to make changes or create variations of the initial output.[10] A user can repeat this process an infinite number of times to generate a final output that meets their expectations and needs.[11] As with any expressive works, artists like Kashtanova seek intellectual property protection, specifically copyright protection, to protect their AI-assisted works.[12]
The authorship requirement for copyright protection poses the main challenge for artists to obtain protection for their AI-assisted works.[13] With the rise of generative AI technology, there are questions as to how copyright law should approach the authorship requirement as it pertains to AI-assisted works.[14] In March 2023, the USCO published a copyright registration guidance for works containing material generated by AI (USCO AI Guidance).[15] While the USCO AI Guidance stated that the USCO would allow artists to protect elements of AI-assisted works if they contain sufficient human authorship, the USCO stated that the copyright would only extend to the human-authored aspects of the work.[16] However, the current legal framework regarding authorship as developed by courts is sufficiently equipped to extend copyright protection to the artists behind AI-assisted works.[17] Artists like Kashtanova demonstrate significant human authorship in the creation of their AI-assisted works and should be awarded copyright protection for the final product that the generative AI tool produces.[18] When artists have sufficiently demonstrated that they contributed human-authored elements through extensive iterative prompting of a generative AI tool, artists should be regarded as the sole authors of the final AI-assisted works.[19]
Part I discusses the rise of generative AI technology, examines how generative AI tools work, and discusses how artists have used generative AI tools.[20] Part II considers the existing copyright framework for authorship, including the constitutional and statutory basis for copyright protection and the copyright doctrines that courts have established pertaining to authorship today.[21] Part III examines the legal challenges that artists have faced in attempting to protect their AI-assisted works and outlines arguments made by scholars who oppose copyright protection for AI-assisted works.[22] Part IV argues that the current copyright doctrines regarding authorship can extend to AI-assisted works and that providing copyright protection for AI-assisted works is consistent with public policy rationales offered for copyright protection.[23]
I. What is Generative AI Technology & How Is It Used by Artists?
Generative AI models are complex technologies that can be used in a number of contexts, including personal and commercial applications.[24] Artists use generative AI technology in various ways ranging from complete content creation to simple content enhancement of already created works.[25] To appropriately examine whether artists can display sufficient human authorship when creating AI-assisted works to satisfy the authorship requirement of copyright protection, it is important to understand how the technology works.[26] Generative AI tools differ from traditional AI tools.[27] Traditional AI tools are predictive systems that are programmed with specific sets of rules and make predictions by interpreting the data they were fed.[28] While generative AI tools create new content when prompted, traditional AI models predict future outcomes.[29]
A. How Does Generative AI Technology Work?
Generative AI technologies are models that take publicly available raw data and generate probable outputs when prompted.[30] Such models can generate text, images, videos, sounds, and other content based on the data they were trained on.[31] Prior to training, software developers prepare large datasets, which may be composed of publicly available data.[32] Developers must preprocess the data and transform it into suitable formats to ensure that the data is consistently formatted before feeding the data to the generative AI model.[33] The training process begins with developers feeding the generative AI model data.[34] Depending on the generative AI model, developers will feed the model using different types of data—for example, if the generative AI model generates images, developers will train the model with images, and if the generative AI model generates text, developers will train the model with text samples.[35] During the training process, the generative AI model analyzes the data and begins to learn patterns and relationships within the dataset.[36]After training, developers will typically refine the hyperparameters of the generative AI model or expand the training dataset to continue improving the model.[37]
For the generative AI model to generate an output, an individual must first input a prompt.[38] The model analyzes the input and generates an output based on the data it was trained on.[39] The generative AI tool produces higher-quality output when the prompt is more descriptive and specific.[40] Furthermore, the user can alter or build on a prior prompt to guide the generative AI tool toward the output the user is seeking.[41] While the output generative AI tools produce is a representation of the data they were trained on, generative AI tools produce a distinct output that is not identical to the original data.[42]
B. Rise and Development of Generative AI Technology
Generative AI technology was first introduced in the 1960s in the form of chatbots.[43] The Eliza chatbot is one of the earliest examples of generative AI.[44] This chatbot was composed of simple software that used a rule-based approach to generate text responses to inputs.[45] Since then, the technology has evolved drastically and now works with other types of data, including images, speech, and videos.[46] The first class of generative AI models to generate images and speech was introduced in 2013.[47] The most popular publicly available generative AI tools now include text-based tools, such as ChatGPT and Google Bard, and image-based tools, such as DALL-E and Midjourney.[48]
In recent years, generative AI tools have become extremely popular and are now widely used by the public.[49]Within two months of OpenAI’s release of ChatGPT in 2022, the free-to-use generative AI tool ChatGPT became the fastest-growing consumer application in history.[50] OpenAI is a company that aims to develop and provide access to generative AI technology to the public.[51] While large technology companies such as IBM, Microsoft, and Google used similar generative AI technology before 2022, the AI industry garnered the attention of regular consumers with the release of ChatGPT.[52] Additionally, generative AI companies are regularly improving their technology and continuously releasing new capabilities to their existing technology.[53]
C. Use of Generative AI Technology by Artists
Artists have used generative AI tools in several contexts in the creation of their AI-assisted works, including employing generative AI technology for content creation and content enhancement.[54] For visual artists, there are a number of generative AI tools that they can use to create content.[55] The most popular generative AI tools for visual artists include the text-to-image models Midjourney, DALLE-3, and Stable Diffusion.[56]
1. Various Applications of Generative AI Tools by Visual Artists
Visual artists can use generative AI technology to create or enhance images in two different ways.[57] First, artists can prompt a generative AI tool to create an image and then modify it within the generative AI platform.[58] Many artists also elect to download the image created by a generative AI tool and make manual modifications to it using other computer software such as Photoshop or Adobe Illustrator.[59] Second, artists can upload their own images and prompt the generative AI tool to enhance them in specific ways.[60]
The two most popular generative AI tools used by artists to instruct and generate images are DALLE-3 and Midjourney.[61] DALL-E 3 is an advanced image generator created by OpenAI, the same company that created ChatGPT, which launched in October 2023.[62] When a user inputs a text prompt, DALL-E 3 will generate an image.[63]If a user is not satisfied with the image, the user can input further text prompts to tweak the image.[64] On OpenAI’s website, the company makes clear that the images created by DALL-E 3 can be used freely by customers for commercial purposes without express permission from the company.[65]
Similar to DALL-E 3, Midjourney requires a user to first submit a text prompt to generate an image.[66]Midjourney then creates four image options based on the initial prompt—these images are similar to one another but have unique differences, such as variations in backgrounds and styles.[67] The user then has three options.[68] First, the user can select one of the four images, download it, and begin editing the image on the Midjourney platform.[69] Second, the user can request Midjourney to re-process the text prompt to generate another four new images.[70] Third, a user can select an image from which Midjourney will generate a series of four images that a user can edit.[71] With this last option, the user can make edits to the image on the Midjourney platform, such as zooming out to extend the original boundaries of the image or panning out in a specific direction to expand the image in a specific direction.[72] Midjourney allows the user to download the final image when the user is satisfied with the final output.[73]
2. Kashtanova’s Use of Midjourney in Creating Zarya of the Dawn
All the images in Kashtanova’s Zarya of the Dawn were generated by Midjourney.[74] To develop each image in the novel, Kashtanova undertook an extensive iterative prompting process wherein she fed Midjourney with hundreds of iterative text-based prompts.[75] Although Kashtanova has not disclosed all the prompts she inputted into Midjourney, in her letter to the USCO, she illustrated the iterative prompting process for one of the most important images in the novel.[76] For the image Zarya Holding a Postcard, Kashtanova aimed to portray the main character in the novel, Zarya, holding a postcard with an abandoned world behind her.[77] Kashtanova prompted Midjourney hundreds of times before Midjourney generated the final image that Kashtanova envisioned and was satisfied with.[78] The first prompt instructed Midjourney to depict hands holding a photograph and provided specific details on the appearance of the hands, including being dark-skinned and elderly.[79] Kashtanova also instructed Midjourney on the aspect ratio of the image.[80] While the first image Midjourney generated accurately depicted Kashtanova’s prompt, it was much less developed than the final image.[81] The first version of Zarya Holding a Postcard solely featured hands holding a blank postcard, whereas the final image portrayed Zarya with closed eyes holding a postcard containing text against the backdrop of a city skyline surrounded by overgrown trees.[82]
Kashtanova’s iterative prompts directed Midjourney on every element that appeared in the final version of Zarya Holding a Postcard.[83] Some of the images generated by Midjourney only depicted the content that Kashtanova intended to appear inside the postcard held by Zarya.[84] Other images only portrayed the city skyline that Kashtanova intended to be in the image's backdrop.[85] The intermediate versions of the images generated by Midjourney included all the independent elements that Kashtanova intended to appear in the final image.[86] The final version of Zarya Holding a Postcard that appeared in the graphic novel incorporated the independent elements from the intermediate versions into one cohesive image.[87] In addition to iteratively prompting Midjourney, Kashtanova further refined some of the images by digitally cropping or framing them herself and using Photoshop to enhance the depth and details in the images.[88]
3. Distinction Between AI-Generated and AI-Assisted Works
Since artists use generative AI technology in a multitude of ways, it can be difficult to make a distinction between AI-generated works and AI-assisted works.[89] However, the distinction between the two is important because it plays a part in how the authorship of such works should be evaluated.[90] For example, the USCO has stated that where a work is created entirely by a generative AI tool through a prompting process, the work’s elements will lack human authorship.[91]Alternatively, if an artist selects, arranges, or modifies material generated by AI, those human-authored aspects of the work may constitute sufficient human authorship.[92]
The World Intellectual Property Organization (WIPO) published a paper discussing the issues raised by AI in the intellectual property system.[93] Specifically, WIPO noted a distinction between AI-generated and AI-assisted works.[94]WIPO defined AI-generated content as outputs created by AI without any human intervention.[95] In contrast, it defined AI-assisted works as outputs generated with human intervention and direction.[96]
Others define AI-assisted works differently.[97] For example, in its policy regarding AI content, Amazon provided its own distinction between using AI to create a work and using AI in the process of creating a work.[98] Amazon stated that AI-generated work is content wholly created by an AI tool.[99] This definition of AI-generated work applies even if an individual made changes to the work after the AI technology generated the content.[100] Amazon defined AI-assisted work as content created by a human who used AI tools to edit, improve, or refine the content of the work.[101] It additionally applies to content created by a human author who used AI tools to brainstorm or generate ideas.[102]
4. Market Value of AI-Assisted Works
Over the past several years, AI-assisted works have gained notoriety and have demonstrated to have market value.[103] Creators of AI-assisted works have been selling their works at auction houses across the nation.[104] In October 2018, the AI artwork Portrait of Edmond Belamy sold for $432,500 at the world-renowned auction house Christie’s.[105]An AI model, Botto, has generated over $3 million in revenue from AI art generated by prompts that were automatically fed by the model itself.[106] These sales demonstrate that a market exists for AI-assisted works, and such a market is likely to continue to grow with the continued advancement of generative AI technology.[107] One report predicts that the market size for generative AI art is expected to grow to $5.84 billion by 2032.[108]
II. Existing Framework for Copyright Protection
The basis of copyright protection is grounded in the United States Constitution.[109] The main purpose of copyright law is to promote progress and enrich society through the arts.[110] To advance this purpose, authors of artistic works are granted specific rights to their works.[111] The Copyright Act of 1976 grants authors exclusive rights to their expressive works.[112]
A. Constitutional and Statutory Basis for Copyright Protection
Article 1, Section 8, Clause 8 of the United States Constitution authorizes Congress to enact legislation securing rights for authors of expressive works (Copyright Clause).[113] The policy considerations for awarding inventors and authors intellectual property protection are centered around promoting progress and innovation in the fields of science and the arts.[114] Public policy rationales based on the promotion of progress are supported by the history behind why the framers of the Constitution included the Copyright Clause in the Constitution.[115]
1. Rationales for Awarding Copyright Holders Exclusive Rights
The copyright framework in the United States reflects a utilitarian approach.[116] The Supreme Court has affirmed that the purpose of copyright protection is to encourage the production of expressive works that will benefit society as a whole.[117] Awarding copyright holders exclusive rights to their works is based upon the balancing of two interests: (1) encouraging and rewarding authors’ creative works by granting exclusive rights to copyright holders to induce authors to continue producing creative works, and (2) ensuring public access to information that may require limiting the grant of exclusive rights.[118] Thus, artists would not expend time, energy, and costs to create expressive works if their work is unprotected and susceptible to copying.[119] Still, society would not benefit if artists were given exclusive rights too broadly because access and circulation of works would be limited—this is known as the incentive-access paradigm.[120]
The incentive-access paradigm in copyright protection recognizes that the creation of expressive works requires skill, time, and money, while the copying of expressive works requires few costs.[121] While copyright holders may choose to price the cost of copies at marginal costs to disincentivize copying, they would not be able to recoup the costs of producing the work.[122] Therefore, copyright protection serves to strike a balance between the costs of limiting access to expressive works and providing incentives to authors to create the original work.[123] If the creator of a copyright is unable to exclude others from copying the work, the market price of the work will decrease to the marginal cost of copying it.[124] In such a situation, creators of expressive works will not be incentivized to create the works in the first place because they would be unable to recoup the costs of creating the expressive works.[125]
However, if exclusive rights are granted too broadly to copyright holders, other individuals may be blocked from using the work.[126] Creating new expressive works typically involves adapting or borrowing material from prior works and adding new original expressions.[127] The broader copyright protection is, the less a creator can quote, reference, or adapt previous works.[128] Thus, society would lose out on potential new expressive works that would contribute to public discourse and valuable educational, scientific, or cultural forms of progress.[129] The Copyright Act of 1976 aimed to strike a balance between granting authors exclusive rights and ensuring public access to those works.[130]
2. The Copyright Act of 1976
The history of the Copyright Act demonstrates that Congress’s understanding of copyright protection is dynamic.[131] The first Copyright Act was enacted in 1790 and was amended several times.[132] The currently governing Copyright Act is the Copyright Act of 1976, where Congress awarded authors longer terms of protection and introduced the fair use doctrine, which allows authors to use elements of copyrighted works in certain contexts without liability for infringement.[133]
For the USCO to grant a copyright to a particular artist, the Copyright Act of 1976 requires several elements to be met.[134] Although registration with the USCO is voluntary, the issuance of a certificate constitutes prima facie evidence of the validity of a copyright.[135] There are three general requirements for an author to receive copyright protection for a particular work.[136] The work must be fixed in a tangible medium of expression.[137] Second, the work must meet one of the statutory categories of a work capable of being copyrighted.[138] Additionally, the work must demonstrate authorship and be original.[139]
B. Authorship Requirement of Copyright Protection
Although copyright protection is awarded to authors of expressive works, neither the Constitution nor the Copyright Act defines authorship.[140] Thus, courts have been left to define the authorship requirement of copyright protection.[141] Authorship is an important element of copyright protection, considering the Copyright Act grants an author of a work legal ownership of a copyright.[142] Subject to some limitations, such as works made for hire, the Copyright Act treats the owner of a copyright as the author-in-fact of the work.[143]
There are several theories of authorship.[144] A sole author is one individual who is the creator of the expression.[145] A joint author is an individual who has created the work with at least one other individual.[146] The Copyright Act defines a joint work as a work prepared by two or more authors who intend to merge their work together into a single work.[147] In a work made for hire, the employer or person who ordered the work to be prepared is considered the author of the work for purposes of copyright protection.[148] There are statutory requirements that must be met in order for the work-for-hire doctrine to be applied to a specific work—the work must either be created by an employee within their scope of employment, or the parties must have a written agreement stating that the work will be treated as a work made for hire.[149]
The authorship requirement is inextricably linked with the originality requirement.[150] The Supreme Court has enumerated that an author is the originator of a work.[151] The originality standard requires that the work be independently created by the author rather than copied from other sources and that the work has some minimal level of creativity.[152] The level of creativity required to satisfy the originality standard is extremely low; as long as a work possesses some creative spark, the originality standard will be met.[153] The policy justification for setting a low threshold for creativity is to prevent the judiciary from making subject judgments about what qualifies as art.[154]
In Burrow-Giles Lithographic Co. v. Sarony, one of the earliest copyright cases heard by the United States Supreme Court, the Court defined an author as the person who originates a work of science or literature.[155] In the case, the plaintiff took a photograph of Oscar Wilde, which is the work at issue.[156] The plaintiff posed Wilde in front of the camera, selected Wilde’s costume and the accessories in the backdrop of the photograph, and arranged the lighting in the photograph.[157] The Court held that the plaintiff demonstrated sufficient authorship for copyright protection.[158] While a photograph may be described as a mere mechanical reproduction of physical objects, the Court found that the artist’s selection, arrangement, and representation of elements in a photograph constituted sufficient artistic expression to make the photographs protectable by copyright.[159] Thus, the Court defined an author as the mastermind who had control over the expressive elements in a work.[160]
In Aalmuhammed v. Lee, the Ninth Circuit gave effect to the Burrow-Giles definition of authorship in the context of a claim to joint authorship of a work.[161] Warner Brothers hired defendant Spike Lee to produce and direct a movie about Malcolm X.[162] Lee was responsible for much of the movie’s production; Lee co-wrote the screenplay and directed and co-produced the movie.[163] Aalmuhammed consulted the actor playing Malcolm X since he possessed extensive knowledge about Malcolm X and Islam.[164] Aalmuhammed contributed to the film in several ways—he suggested script revisions, some of which were included in the final version of the film, he directed actors on set, created two complete scenes, and edited portions of the movie.[165] The end credits of the film credited Aalmuhammed only as an Islamic Technical Consultant.[166] Aalmuhammed filed a complaint, seeking a judgment granting him a co-owner of the copyright, arguing that he co-authored the movie.[167] The court held that he was not a co-author of the film.[168]The court stated that Aalmuhammed was not the mastermind behind the movie; although Aalmuhammed made suggested changes to the screenplay, it was Lee’s authority whether or not to accept them.[169] The court also reasoned that finding Aalmuhammed as a co-author would be contrary to public policy because copyright protection exists to progress science and the arts, and progress would be hampered if authors could not confer with others to the benefit of their work.[170]
The Third Circuit has held that a person who creates a work by using mechanical means to translate an idea into a tangible expression is an author and entitled to copyright protection.[171] In Andrien v. Southern Ocean County Chamber of Commerce, the plaintiff created a map by compiling preexisting maps and other publicly available geographical information.[172] Additionally, the plaintiff conducted a personal survey of the area and noted certain streets and landmarks that were unmarked in preexisting maps.[173] The plaintiff worked with a printing company to prepare the map, and the printing company assigned a specific employee to print the plaintiff’s map.[174] The printing employee was provided with the collection of preexisting maps, and the plaintiff directed the employee to coordinate the scales and label street names.[175] The plaintiff spent approximately one hour at the print shop daily over three weeks.[176] The court held that the plaintiff was the author of the work because individuals could be entitled to copyright protection even if they did not perform the work themselves.[177]
A court has held that an individual who directs the filming of a work but does not physically perform the act of filming can be considered the author of such footage.[178] In Lindsay v. Wrecked & Abandoned Vessel R.M.S. Titanic, the plaintiff was a director who directed a documentary film about a salvage expedition of the Titanic.[179] The director sailed with a salvage expedition team to the wreck site and remained at sea with the crew for one month.[180] Prior to filming, the director created storyboards for the film detailing specific camera angles and sequences to be used.[181] The director also designed and constructed the underwater light towers that were used to emit light for the filming of the wreckage.[182] While on board, the director conducted daily meetings with the expedition crew and provided the film crew with specific instructions on how to use the light towers during filming.[183] The court rejected the defendants’ argument that the director did not have a right to the footage since he did not personally photograph the wreckage himself.[184] The court stated that because the director created the storyboard for the film and provided specific directions to the film crew, the film was the product of the director’s personal and original intellectual expressions.[185] In holding that the director was the author of the film for purposes of copyright protection, the court reasoned that he exercised a significant degree of control over the entirety of the filming operation, including the camera angles, lighting, and other artistic elements in the film.[186]
Under the work-for-hire doctrine, the employer or person who ordered the work to be prepared is considered the author of the work for purposes of copyright protection—rather than the author-in-fact.[187] In Community For Creative Non-Violence v. Reid, the Supreme Court stated that, under the work-for-hire doctrine, to determine whether a work was created by an employee or an independent contractor, a court must apply common law agency principles.[188] A nonprofit association, Community For Creative Non-Violence (CCNV), commissioned an artist to produce a sculpture to display the plight of the homeless.[189] CCNV had developed the idea for the display as well as the material it was to be cast in.[190] The artist agreed to make the sculpture, and CCNV agreed to make the steam gate and pedestal for the statue.[191] The artist suggested an alternative material that would be more cost-efficient and easier to create within the specified time period, and CCNV agreed.[192] The parties did not sign a written agreement or discuss copyright ownership.[193] The Supreme Court held that the sculpture was not a work made for hire, reasoning that the parties did not have a written agreement and that CCNV and the artist did not have an employer-employee relationship.[194]
While courts seem to agree that an author does not have to physically undertake the labor of creating the work, courts have exhibited a reluctance to award non-human authors copyright protection.[195] In Naruto v. Slater, a wildlife photographer, Slater, left his camera in a reserve largely occupied by monkeys.[196] One monkey took multiple photographs of himself (selfies) with Slater’s camera.[197] Two animal advocacy groups brought a copyright infringement claim on behalf of the monkey after Slater reproduced and distributed copies of the monkey’s selfies.[198]The court held that the monkey cannot be considered an author under the Copyright Act.[199] The court reasoned the USCO does not recognize authorship for non-humans.[200]
Although the USCO is merely a regulatory body, the USCO interprets the authorship requirement based on judicial opinions.[201] Drawing on such judicial opinions, the USCO asserted that authorship is limited to only human-created works.[202] In its compendium, the USCO states that a machine or mechanical process cannot be considered to be an author.[203] Thus, the USCO does not register works wholly created by a machine that does not require human input.[204] The USCO provided that, in the context of a work created partly by a machine, it will evaluate whether the expressive elements in the work were created and executed by a machine or a human.[205]
III. Challenges In Copyrighting AI-Generated and AI-Assisted Works
Artists like Kashtanova have faced challenges in attempting to copyright works created with the help of generativeAI technology.[206] The main challenge authors have faced concerns the authorship requirement for copyright protection.[207] Although some courts have held that an artist’s use of digital tools like Photoshop does not preclude them from being considered the authors of their works, the USCO remains skeptical of generative AI technology.[208] Even when artists have inputted hundreds of original prompts to appropriately guide the generative AI tool to reach the output that they envision, the USCO rejects that the final output constitutes human authorship.[209]
One artist garnered national attention when he won a prize at the Colorado State Fair’s annual art competition for an AI-generated work.[210] The artist, Allen, used Midjourney to create an artwork titled Théâtre D’opéra Spatial.[211]The artist input more than 600 variations of text prompts into Midjourney to generate the initial image.[212] Allen attempted to register the work for copyright protection but the USCO rejected his requests for copyright registration.[213]The USCO stated that Théâtre D’opéra Spatial was not a product of human authorship because the work was generated by Midjourney.[214] Because Allen refused to disclaim the elements of the work created by Midjourney, the USCO rejected his application for copyright registration.[215]
One court has ruled that the user of an AI tool cannot receive copyright protection for an AI-generated work.[216]In Thaler v. Perlmutter, the plaintiff used an AI program that he developed to create an image.[217] Although it is unclear what type of AI program the plaintiff developed, it does not appear to have been a generative AI program.[218] After the program created the work, the plaintiff tried to register it with the USCO.[219] The plaintiff listed the author of the work as the AI program but attempted to claim the copyright of the work for himself, arguing that the transfer of ownership from the AI program to him constituted a work for hire.[220] After the USCO twice denied the application on the basis that the work did not contain human authorship, the plaintiff filed a claim to challenge the decision.[221] The District Court of the District of Columbia held that the work could not be protected by copyright because it lacked human authorship.[222] The court relied on precedent that declined to extend copyright protection to works created without human involvement.[223] The plaintiff argued that the work was a work made for hire because the plaintiff specifically ordered the machine to create the work.[224] The court rejected the plaintiff’s argument and stated that the work-for-hire doctrine could not apply to the work because the doctrine applies only when the entity that creates the work could be treated as a legal owner of the work and that no property rights existed for the AI program.[225]
Some scholars have expressed reservations about awarding copyright protection for AI-assisted works.[226]Scholars have opined that the protection of AI-assisted works is contrary to the public policy rationales that exist for copyright protection.[227] These arguments center around the notion that once a generative AI tool is developed, it can create an infinite number of works for little to no cost.[228] They assert that because AI-assisted works are easy and inexpensive to produce, the traditional incentives that copyright law grants artists to create new works are reduced.[229]They argue that if copyright protection does not extend to AI-assisted works, no market failures would occur because artists would continue to create AI-assisted works due to their inexpensive nature—and copyright protection largely exists to cure market failures.[230] Under this approach, protecting AI-assisted works does not serve the traditional notions of why copyright protection exists under the utilitarian framework of copyright protection in the United States.[231]
Additionally, some scholars argue that the benefits of providing copyright protection for AI-assisted works do not outweigh the risks of blocking access to the public, who would likely want to use materials from copyrighted elements in AI-assisted works in the creation of their new works.[232] These scholars argue that awarding copyright protection to AI-assisted works would create an impediment to individuals who seek to use content from those AI-assisted works.[233]Therefore, these scholars contend that because little incentives are required for generative AI users to create new works and the risks of preventing access to the public from using materials from AI-assisted works are high, copyright protection should not subsist in AI-assisted works because there is no utilitarian justification for doing so.[234]
The USCO expresses concerns about awarding copyright protection for AI-assisted works for other reasons.[235]The USCO has asserted that AI-assisted works would likely not meet the authorship requirement because users do not exercise control over the outputs that generative AI tools produce.[236] In its AI Guidance, the USCO concluded that the act of AI users prompting a generative AI technology could not alone constitute human authorship because generative AI tools ultimately determine how prompts are implemented and executed.[237] As a result, generative AI tools are in control of the expressive elements of the output, and such an output cannot be considered to be a creation of human authorship.[238]
IV. Analyzing the Authorship Requirement for Copyright Protection of AI-Assisted Works
It is likely that many artists will encounter issues in trying to protect their AI-assisted works because the USCO has asserted that copyright protection cannot extend non-human authored elements of AI-assisted works.[239] The USCO stated that it applies and relies upon the current interpretations of authorship as created by courts.[240] However, the current doctrines of authorship as established by courts can sufficiently extend to AI-assisted works.[241] Copyright law should recognize that AI users, like Kashtanova, are the sole authors of their AI-assisted works because they can be deemed the masterminds behind the outputs generated by generative AI tools.[242]
A. Framing the Distinction Between AI-Generated Works and AI-Assisted Works
Providing a distinction between AI-generated works and AI-assisted works is essential to evaluate the authorship of works created by artists who utilize generative AI technology.[243] For example, under WIPO’s definitions of AI-generated and AI-assisted works, Kashtanova’s images in Zarya of the Dawn, which were generated by Midjourney, would constitute AI-assisted works because she provided Midjourney with a large number of prompts to direct the tool in the creation of the final images.[244] However, under other definitions, Kashtanova’s images would be considered AI-generated works regardless of her extensive prompting of Midjourney because the images were ultimately generated by Midjourney and were a result of its output.[245] Conversely, under both of these differing definitions, the plaintiff’s work in Thaler would be regarded as an AI-generated work because the AI program created the image without any guidance or prompting from the plaintiff.[246]
AI-generated works should be defined as works that are generated with little to no human prompting, and AI-assisted works should be defined as works that are generated with the necessary level of prompting required to demonstrate that the AI user guided the generative AI technology to produce the output it did.[247] The way that generative AI technologies currently exist, human intervention is required if an author has a specific idea of the output they are seeking.[248] Therefore, a work should be regarded as an AI-assisted work if the AI user prompts the generative AI tool to the extent that is required to ensure that the expressive elements generated by the generative AI tool are a product of the prompts the user provided.[249]
Applying the proposed definition of AI-assisted works demonstrates that both Kashtanova’s and Allen’s works constitute AI-assisted works, whereas the work of the plaintiff in Thaler constitutes AI-generated work.[250] Kashtanova had to input hundreds of prompts to reach the images she desired.[251] Similarly, Allen inputted 600 variations of text prompts into Midjourney to generate the initial image he was seeking.[252] Thus, it logically follows that such outputs that were created by a generative AI technology were created because a human was guiding the technology to do so, and thus, should be defined as AI-assisted works.[253] On the contrary, the plaintiff in Thaler did not guide the AI technology in the form of prompts to generate his work—rather, the AI tool generated the image autonomously.[254] The work in Thaler should be classified as AI-generated work because the AI technology was wholly unguided in the generation of the work.[255]
AI-generated works should not meet the authorship requirement because authorship requires some level of intellectual conception attributable to a human.[256] Although the court in Thaler applied current precedent regarding human authorship, the court’s decision does not implicate AI-assisted works because the plaintiff’s work was AI-generated work, which requires substantially less human involvement than AI-assisted work.[257] In fact, in Thaler, the court relied upon Burrow-Giles to argue that, unlike the photographer in Burrow-Giles, who demonstrated intellectual conception through the selection and arrangement of the expressive elements in the photograph, the plaintiff in Thaler did not demonstrate any intellectual conception in the making of the AI-Generated Work.[258] Had the plaintiff exercised greater intervention in the creation of the work, as the photographer did in Burrow-Giles, by prompting the AI tool, the district court may have decided the case differently.[259]
B. Analyzing Authorship Cases Defining an Author as the Mastermind
Examination of cases attempting to define who constitutes an author of a work reveals that authorship is centered around who exercises control over the expressions in a work.[260] Reading Burrow-Giles through the lens of AI-assisted works reveals that the Court’s opinion can easily encompass the type of activities undertaken by artists of AI-assisted works.[261] The plaintiff chose the costume Wilde wore and picked the accessories in the backdrop of the photograph.[262] Prior to capturing the image, the plaintiff had a mental conception of how he wanted the elements in the photograph to appear.[263] Because the plaintiff was responsible for controlling what appeared in the photograph and how it appeared in the photograph, the Court deemed him as the originator of the work.[264]
For similar reasons, the Aalmuhammed court rejected Aalmuhammed’s claim to co-authorship of the Malcolm X movie.[265] Although Aalmuhammed arguably made some important contributions to the film, such as suggesting script revisions, directing actors on set, and creating two scenes of the film, the court declined to consider him a co-author of the movie because he did not superintend the work and exercise control over the movie.[266] Ultimately, Lee was the creative mastermind behind the work because it was his decision whether or not to include the revisions suggested by Aalmuhammed or the scenes created by Aalmuhammed.[267]
Both Andrien and Lindsay demonstrate that an individual need not carry out the physical labor of creating a work to be considered an author; rather, an author can be an individual who directs another to translate his ideas into tangible expressions.[268] In Andrien, the plaintiff possessed the original idea to create a map by compiling preexisting maps and including information collected by him through personal surveys.[269] Even though the plaintiff directed an employee at a printing shop to physically compile the information requested by him, and the employee carried out that labor himself, the court held that the plaintiff was the author of the final compiled map because he had the original conception of the work.[270] Similarly, in Lindsay, the director of the documentary film about the Titanic was held to be the author of the film even though the film crew filmed all the scenes underwater themselves.[271] The court gave importance to the fact that the director had created detailed storyboards and provided the crew with specific instructions on how to film the wreckage.[272] The court effectively ruled that because the director was the mastermind behind the film and had exclusive authority over the final product, he was the author of the final work.[273] Thus, examination of court precedent demonstrates that an author is regarded as the mastermind behind the work and the individual who exercises control over the work.[274]
C. How Should the Authorship Requirement Be Satisfied in Cases of AI-Assisted Works?
Since copyright law establishes that authorship is centered around who exercises control over a work, AI users (artists) of AI-assisted works should be deemed as the sole authors of the final works when they demonstrate that the output the generative AI technology produced was a result of their control.[275] Although the USCO has argued that generative AI tools are unpredictable and humans do not exercise creative control over the final product, this argument is not compelling.[276] Artists like Kashtanova amend their inputs hundreds of times to direct the generative AI tool appropriately.[277] Such artists demonstrate creative control in the creation of their AI-assisted works because, with each iterative prompt, they tweak their inputs to guide the generative AI tool closer to generating the desired output.[278]
1. Establishing the Appropriate Standard for Authorship for AI-Assisted Works
The issue of how much prompting is sufficient to constitute authorship of an AI-assisted work is undoubtedly a difficult question.[279] However, the appropriate standard should be based on both the quantity and the quality of the prompts given to the generative AI tool.[280] The quantity of prompts alone cannot demonstrate that an artist was directing the generative AI tool to carry out the artist’s mental conception of the work because authorship is based upon the intellectual motivations of the artist rather than a numerical number of affirmative acts undertaken by the artist.[281]Conversely, the quality of the prompts alone cannot demonstrate that an artist directed the generative AI tool to carry out the artist’s mental conception because generative AI technologies in their current form require iterative prompting to generate a final product that depicts an artist’s original mental conception for the work.[282]
Examination of whether an artist who uses generative AI technology to create an AI-assisted work can qualify as an author would ultimately need to be determined through the artist’s disclosure of the prompts used in the making of the work.[283] The artist’s prompts can help reveal whether the expressive elements in the AI-assisted work are reflective of the artist’s mental conception behind the work or whether the expressive elements in the AI-assisted work were autonomously generated by the generative AI tool without specific instructions or prompting by the artist.[284] Thus, an artist should submit the prompts used in the creation of the work alongside the copyright application to the USCO when seeking copyright registration to allow the USCO to evaluate whether the expressive elements in the AI-assisted work are attributable to the artist’s prompts to the generative AI tool.[285]
The expressive elements in an AI-assisted work should be attributable to the artist when the artist has demonstrated that they provided the generative AI tool with iterative prompts that are deliberate, detailed, and descriptive.[286] Additionally, it should be apparent, when comparing the prompts written by the artist and the output generated by the generative AI tool, that the generative AI tool generated an output that captured the detailed instructions in the prompts.[287] When an artist demonstrates through iterative prompting that the final output the generative AI tool produced was based upon the artist’s mental conception of the work, which can develop throughout the iterative prompting process, the artist should be considered the author of the final AI-assisted Work.[288]
Since the authorship requirement is closely linked with the originality standard for copyright protection, an artist claiming authorship over an AI-assisted work must demonstrate that they meet the originality standard.[289] The originality standard only requires that an author creates the work independently with some minimal level of creativity.[290] Artists like Kashtanova should meet the originality standard when the iterative prompts they input into the generative AI tool are independently created rather than copied from other artists or sources, and the prompts contain some minimal level of creativity such as setting the scene, mood, or subjects for the desired image.[291] Most artists should not face serious impediments in claiming that their AI-assisted works meet the originality standard, considering the originality standard is extremely low.[292]
2. Applying the Cases Defining Authorship to AI-Assisted Works
Applying the cases defining an author as the mastermind of a work demonstrates that artists such as Kashtanova should be considered the authors of their final AI-assisted works.[293] Some cases, like Kashtanova’s, will be easier to evaluate than others.[294] Using Kashtanova as an example, Kashtanova should be considered the sole author of the images in Zarya of the Dawn.[295] The concept for the graphic novel itself was born out of Kashtanova’s grief after she experienced the loss of her grandmother and best friend.[296] She sought to depict this experience through a story of a girl named Zarya who is transported to uninhabited worlds with no memories of her previous life, left only with one postcard in her pocket.[297] The images in the novel were meant to illustrate this story.[298] Just as the plaintiff in Burrow-Giles had a mental conception of how he wanted the elements in the photograph to appear, Kashtanova had a clear image in mind for every element in the images in the graphic novel.[299]
Examining one of the images in Zarya of the Dawn illustrates how Kashtanova is the mastermind behind the image due to her iterative prompting process.[300] For the most important image in the novel, Zarya Holding a Postcard, Kashtanova undertook an extensive iterative prompting process that involved hundreds of versions of the image.[301]Kashtanova aimed to portray Zarya holding a postcard with an abandoned world behind her.[302]
Figure 1. Final Version of Zarya Holding a Postcard.[303]
As shown in Figure 1, the final image that appeared in the graphic novel depicted Zarya with closed eyes, holding a postcard containing text against the backdrop of a city skyline surrounded by overgrown trees.[304]
Comparing the first version of Zarya Holding a Postcard generated by Midjourney with the final version of the image that appeared in the novel illustrates how Kashtanova’s iterative prompts guided Midjourney to develop the final version of the image.[305] Kashtanova’s first prompt aimed to direct Midjourney on only how the hands should appear.[306] Kashtanova instructed Midjourney to generate an image of dark-skinned and old hands holding a photograph and specified the aspect ratio of the image.[307]
Figure 2. First Version of Zarya Holding a Postcard Generated by Midjourney.[308]
Figure 2 portrays the first version of Zarya Holding a Postcard generated by Midjourney, which included four image options that accurately reflected Kashtanova’s prompt by depicting dark-skinned hands holding a photograph.[309] The first version of the image is significantly less developed and refined than the final version of the image, a result attributed to the iterative prompts that Kashtanova provided to Midjourney.[310]
Similar to how the photographer inBurrow-Gilesexercised control over what appeared in the photograph and how it appeared in the photograph, Kashtanova’s iterative prompts directed Midjourney on every element that appeared in the final version ofZarya Holding a Postcard.[311]Some of the intermediate versions of the image only depicted independent elements that Kashtanova intended to appear in the final image, indicating that Kashtanova prompted Midjourney on each independent element individually.[312]
Figure 3. First Example of the Intermediate Versions of Zarya Holding a Postcard.[313]
Figure 4. Second Example of the Intermediate Versions of Zarya Holding a Postcard.[314]
Some of the intermediate images, as illustrated in Figures 3 and 4, portrayed only the content Kashtanova intended to appear inside the postcard held by Zarya.[315] Other intermediate images, as shown in Figure 4, depicted only the city skyline that Kashtanova intended to appear in the backdrop of the final image.[316] Comparing the intermediate versions of Zarya Holding a Postcard with the final version of the image demonstrates how the final version of the image incorporated the independent elements from the intermediate versions generated by Midjourney.[317] Similar to the photographer in Burrow-Giles, who determined the subject’s appearance and selected the subject’s costume, accessories, and backdrop, Kashtanova determined the elements that appeared in the final image, including Zarya’s appearance (with closed eyes), the content inside the postcard Zarya held, and the backdrop of the image.[318]
While Kashtanova did not personally execute the creation of the images in her novel, that should not preclude her from being considered the author of the final images.[319] In both Andrien and Lindsay, the courts held that an author can encompass an individual who directs another to translate his mental conception of the work into tangible expressions.[320] Although the plaintiffs in Andrien and Lindsay directed other individuals to undertake the physical labor of creating the works rather than computer programs, their holdings can sufficiently extend to cases like Kashtanova’s because their reasonings were centered around the fact that the plaintiffs had the original conceptions behind the works.[321]Additionally, courts have held that artists can be considered authors even when they utilize digital tools like Photoshop in the creation of their works.[322]
3. What Theory of Authorship Should Be Adopted for Artists of AI-Assisted Works?
The artists prompting generative AI tools should be considered the sole authors of their final AI-assisted works.[323] The main three theories of authorship include sole authorship, joint authorship, and works made-for-hire.[324]Because artists like Kashtanova are the sole individuals who have the mental conception of how the final work is to be portrayed, they should be considered the sole authors of the expressions generated by generative AI tools.[325] In Burrow-Giles, Andrien, and Lindsay, the courts held that the plaintiffs were the sole authors of the final works because they were the sole individuals who exercised ultimate control over the expressive elements in the work.[326] Similarly, artists like Kashtanova are the only individuals who exercise control over the generative AI tools’ execution of the expressive elements in their AI-assisted works.[327]
The theory of joint authorship is not preferable in the context of AI-assisted works.[328] As asserted by the court in Aalmuhammed, even if an individual makes suggestions as to the expressive elements in the work, if another party has the ultimate authority over whether or not to accept them, the individual cannot be considered a joint author of the work.[329] Applying the court’s reasoning to cases like Kashtanova’s reveals that generative AI tools are similar to the plaintiff in Aalmuhammed—generative AI tools can make suggestions as to expressive elements in a work through intermediate versions of images, but it is ultimately the artist’s discretion whether or not to accept them.[330]
The work-for-hire doctrine is also not appropriate in the context of AI-assisted works because artists would not be able to satisfy the statutory requirements under the Copyright Act.[331] The Court in Community For Creative Non-Violence explained that the Copyright Act requires that the work be created in some context of an employer-employee relationship, and the relationship between the artist and a generative AI tool cannot be regarded as an employee–employer relationship.[332] Additionally, as the court in Thaler discussed, the work-for-hire doctrine necessarily requires the employee to have a legally cognizable right to the work.[333] A generative AI tool does not have a legally cognizable right to a work because courts do not recognize authorship for non-human entities.[334] Therefore, artists should be regarded as the sole authors of their AI-assisted works because alternative theories of authorship are not preferable, especially when they would necessitate the restructuring of fundamental doctrines of copyright law.[335]
E. Why Providing Copyright Protection for AI-Assisted Works Is Consistent with Public Policy Rationales
Although some scholars dubiously question the protection of AI-assisted works, these critics mistakenly assert that protecting AI-assisted works is inconsistent with the utilitarian justifications offered for copyright protection.[336] These scholars argue that AI-Assisted Works should not receive copyright protection because the creation of AI-Assisted Works do not require considerable time and money, and therefore, the benefits of granting monopolies to artists of AI-Assisted Works do not outweigh the costs of blocking the public’s access to the expressive elements in those works.[337] However, artists like Kashtanova do spend considerable time and effort in the creation of their AI-assisted works.[338] Kashtanova spent over a year creating Zarya of the Dawn, and for each image in the novel, she provided Midjourney with hundreds of prompts.[339] Additionally, the Copyright Act contains safeguards in place, such as fair use, which ensures that the public can access elements in copyrighted works in certain contexts.[340]
The utilitarian justifications offered for copyright protection in the United States compel the finding that artists of AI-assisted works should be considered authors of such works.[341] The Constitutional rationale centered behind granting monopolies to copyright holders is to induce authors to continue producing creative works.[342] If artists are not awarded copyright protection for their AI-assisted works, artists like Kashtanova would cease to spend the considerable time and energy it takes to create their AI-assisted works.[343] Patrons of the arts have demonstrated an interest in AI art, and the market value for AI-assisted works is likely to increase significantly over the next decade.[344] If copyright protection does not extend to AI-assisted works, society would be disadvantaged by the lack of availability of artworks that fall within the intersection of art and technology.[345]
Conclusion
The Copyright Clause of the Constitution demonstrates that the promotion and progress of the arts is an essential public policy interest in the United States.[346] In cases where artists use substantial human insights in the creation of AI-assisted works, artists should be awarded copyright protection.[347] Although the USCO has stated that artists of AI-assisted works cannot be considered authors under copyright law, the current legal framework regarding authorship as developed by courts is sufficiently equipped to extend authorship to artists like Kashtanova.[348] Granting copyright protection to artists like Kashtanova for their AI-assisted works aligns with the public policy rationales supporting copyright protection because it ensures that artists are incentivized to invest the considerable time and energy required to create such works, which ultimately promotes the progress and development of the arts.[349]
[1] See Letter from Van Lindberg, Taylor English Duma LLP, to Robert J. Kasunic, U.S. Copyright Off. (Nov. 21, 2022) [hereinafter Kashtanova’s Letter], https://www.copyright.gov/docs/zarya-of-the-dawn.pdf, 1 (discussing how Kashtanova used Midjourney to create Zarya of the Dawn).
[2] See id. at 3 (explaining Kashtanova’s original conception of the novel).
[3] See id. at 3–4 (detailing both the inspiration for Kashtanova’s novel and the main plot of the novel).
[4] See id. at 2 (“[T]he text of the Work was written entirely by Kashtanova without the help of any other source or tool, including any generative AI program.”).
[5] See id. at 9 (“[E]ach iteration of each image is the result of a unique set of inputs composed by Kashtanova”).
[6] See Letter from Robert J. Kasunic, U.S. Copyright Off., to Van Lindberg, Taylor English Duma LLP (Feb. 21, 2023) https://www.copyright.gov/docs/zarya-of-the-dawn.pdf, 1 (affirming copyright protection for some elements in Zarya of the Dawn but failing to extend protection to the images in the novel generated by Midjourney).
[7] See id. (finding that “the images in the Work that were generated by the Midjourney technology are not the product of human authorship”).
[8] See Kevin Roose, A.I.-Generated Art is Already Transforming Creative Work, N.Y. Times (Oct. 21, 2022), https://www.nytimes.com/2022/10/21/technology/ai-generated-art-jobs-dall-e-2.html (discussing how a game designer, interior designer, filmmaker, and service designer used generative AI programs in the creation of their artworks).
[9] See, e.g., Midjourney, https://docs.midjourney.com/docs/quick-start (last visited Nov. 3, 2023) (detailing how individuals can use Midjourney to generate an image).
[10] See id. (explaining how a user can create variations of the initial output and outlining how a user can enhance or modify the output by using various functions such as zooming in and out of the image or expanding the canvas of the image).
[11] See, e.g., Kashtanova’s Letter, supra note 1, at 10 (explaining how Kashtanova “painstakingly shaped each set of inputs and prompts over hundreds of iterations to create as perfect a rendition of her vision as possible”).
[12] See id. at 13 (requesting the USCO to grant Kashtanova copyright protection for the images generated by Midjourney in Zarya of the Dawn).
[13] See, e.g., Thaler v. Perlmutter, No. 22-1564, 2023 U.S. Dist. LEXIS 145823, at *20–21 (D.C. Dist. Ct. 2023) (upholding the USCO’s denial of copyright registration for an AI-assisted work because it lacked human authorship).
[14] See Carys Craig & Ian Kerr, The Death of the AI Author, 52 Ottawa L. Rev. 31, 31 (2021) (discussing how generative AI tools are prompting scholars to confront “the question of how to treat seemingly original works of expression that are not the product of ‘authorship’ in the traditional sense”).
[15] See Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 37 C.F.R. pt. 202 (2023) [hereinafter USCO AI Guidance] (setting out a guidance for how the USCO intends to examine applications for copyright protection of works created by generative AI technology).
[16] See id. at 4 (“If a work’s traditional elements of authorship were produced by a machine, the work lacks human authorship and the Office will not register it.”).
[17] See, e.g., Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884) (defining that an author is the individual “originating, making, producing, as the inventive or mastermind, the thing which is to be protected”); see also Aalmuhammed v. Lee, 202 F.3d 1227, 1234 (9th Cir. 2000) (applying Burrow-Giles in the context of a joint work and stating that, the author is the one who “‘superintends’ the work by exercising control”).
[18] See generally Kashtanova’s Letter, supra note 1 (arguing that Kashtanova should be awarded copyright protection for Zarya of the Dawn as she exhibited human authorship in the form of extensive iterative prompting).
[19] See, e.g., id. at 8 (discussing how for one of the images in Zarya of the Dawn, Kashtanova inputted the following prompt, “a holographic elderly white woman named Raya, raya is having curly hair and she is inside a spaceship,” as well as mood and style-related directions, such as “cinematic,” “Raya as a hologram,” and “unreal engine”).
[20] See infra Part I (discussing what generative AI technology is and examines how they work in the context of visual art).
[21] See infra Part II (examining the existing copyright framework as it pertains to authorship and the utilitarian justifications offered for copyright protection in the United States).
[22] See infra Part III (discussing the legal challenges that artists have faced in attempting to copyright their AI-assisted works).
[23] See infra Part IV (analyzing whether the current copyright doctrines as they pertain to authorship can be applied to the protection of AI-assisted works and discussing whether copyright protection of AI-assisted works is consistent with the policy justifications offered for copyright protection).
[24] See Top Generative AI Industry Applications of 2023: An In-Depth Look, Turing, https://www.turing.com/resources/generative-ai-applications (last visited Nov. 16, 2023) (“[G]enerative AI applications and tools are empowering both organizations and individuals to automate tedious tasks, make better decisions, and streamline operations for maximum efficiency.”).
[25] See Nantheera Anantrasirichai & David Bull, Artificial Intelligence in the Creative Industries: A Review, 55 A.I. Rev. 589, 601 (2021) (analyzing the various ways in which artists use generative AI technology).
[26] See, e.g., Kashtanova’s Letter, supra note 1, at 9 (arguing that because Kashtanova used the generative AI tool Midjourney, which requires prompts to appropriately guide the tool, she displayed original authorship to satisfy the authorship requirement for copyright protection).
[27] See Bernard Marr, The Difference Between Generative AI and Traditional AI: An Easy Explanation For Anyone, Forbes (July 24, 2023), https://www.forbes.com/sites/bernardmarr/2023/07/24/the-difference-between-generative-ai-and-traditional-ai-an-easy-explanation-for-anyone (outlining the differences between generative AI tools and traditional AI technologies).
[28] See id. (explaining that traditional AI tools are not the same as generative AI tools because they are rule-based systems that make simple predictions based on the data fed to them).
[29] See id. (discussing how generative AI tools can generate new content, while traditional AI tools simply predict information).
[30] See Kim Martineau, What is Generative AI?, IBM (Apr. 20, 2023), https://research.ibm.com/blog/what-is-generative-AI (explaining how generative AI tools generate outputs from a technical standpoint).
[31] See Danica Lo, What is Generative AI? Your Questions Answered, Fast Co. (Dec. 3, 2023), https://www.fastcompany.com/90826178/generative-ai (providing examples of different AI models that generate images, videos, and sound).
[32] See Greg Pavlik, What Is Generative AI? How Does It Work?, Oracle (Sept. 15, 2023), https://www.oracle.com/artificial-intelligence/generative-ai/what-is-generative-ai) (“[G]enerative AI models are trained on large, diverse data sets . . . [and] must always be trained, at least initially, using unsupervised learning (where data is unlabeled, and the AI software is given no explicit guidance).”); see also Lauren Leffer, Your Personal Information Is Probably Being Used to Train Generative AI Models, Sci. Am. (Oct. 19, 2023), https://www.scientificamerican.com/article/your-personal-information-is-probably-being-used-to-train-generative-ai-models/ (explaining how training data sets for generative AI models include data from the internet and books).
[33] See How to Train Generative AI Models, Deepchecks, https://deepchecks.com/how-to-train-generative-ai-models/ (last visited Nov. 14, 2023) (stating that preprocessing could include tasks such as “resizing and standardizing images to a consistent resolution,” “normalizing audio data to ensure consistent volume levels,” or “converting text data into a standardized format, removing special characters or stopwords”).
[34] See id. (describing that the training process consists of developers presenting data to the model, from which the model learns the data and refines its ability to generate content).
[35] See id. (explaining how during the training process, the model is exposed to training data; “For image generation it would be a dataset of real images, while text generation, it could be a corpus of text samples”).
[36] See Google Cloud, https://cloud.google.com/use-cases/generative-ai#section-3 (last visited Nov. 15, 2023) (describing how generative AI models are typically trained through supervised learning where developers give the model human-created content with corresponding labels).
[37] See How to Train Generative AI Models, supra note 33.
[38] See Getting Started with Prompts for Text-Based Generative AI Tools, Harv. Univ. Info. Tech. (Aug. 30, 2023), https://huit.harvard.edu/news/ai-prompts (explaining how to appropriately prompt a generative AI tool).
[39] See id. (“After you enter a prompt, the AI model analyzes your input and generates a response based on the patterns it has learned through its training.”).
[40] See id. (“The information, sentences, or questions that you enter into a Generative AI tool . . . are a big influence on the quality of outputs you receive.”).
[41] See id. (“Change the wording or tone or add more context and specificity to guide the AI toward the output you’re looking for.”).
[42] See Martineau, supra note 30 (stating that generative AI tools “encode a simplified representation of their training data and draw from it to create a new work that’s similar, but not identical, to the original data”).
[43] See George Lawton, What is Generative AI? Everything You Need to Know, TechTarget, https://www.techtarget.com/searchenterpriseai/definition/generative-AI (last visited Nov. 15, 2023) (discussing the earliest examples of generative AI technology and how they worked); see also Oshan Jarow, How the First Chatbot Predicted the Dangers of AI More than 50 Years Ago, Vox (Mar. 5, 2023), https://www.vox.com/future-perfect/23617185/ai-chatbots-eliza-chatgpt-bing-sydney-artificial-intelligence-history (explaining how the Eliza chatbot was used in the field of psychotherapy and asked users questions about their mental health conditions).
[44] See Jarow, supra note 43 (explaining that Joseph Weizenbaum created the Eliza chatbot in the 1960s).
[45] See id. (stating that “early implementations [of generative AI] used a rules-based approach that broke easily due to a limited vocabulary, lack of context and overreliance on patterns, among other shortcomings”); Ben Tarnoff, Weizenbaum’s Nightmares: How the Inventor of the First Chatbox Turned Against AI, The Guardian (July 25, 2023), https://www.theguardian.com/technology/2023/jul/25/joseph-weizenbaum-inventor-eliza-chatbot-turned-against-artificial-intelligence-ai (discussing how the Eliza chatbox “looked at the user input and applied a set of rules to generate a plausible response”).
[46] See Martineau, supra note 30 (discussing variational autoencoders, the first types of generative AI models that analyze more complex types of data such as images and videos).
[47] See id. (discussing how variational autoencoders were first introduced in 2013 and “were the first deep-learning models to be widely used for generating realistic images and speech”).
[48] See Explore Generative AI Tools, Harv. Univ. Info. Tech., https://huit.harvard.edu/ai/tools (last visited Nov. 16, 2023) (covering some of the various publicly available and free generative AI tools).
[49] See Sofia Bliss-Carrascosa & Jeff Cercone, What is Generative AI and Why is it Suddenly Everywhere? Here’s How Tools Like ChatGPT and Dall-E Work, Poynter (June 22, 2023), https://www.poynter.org/fact-checking/2023/what-is-generative-ai-and-why-is-it-suddenly-everywhere-heres-how-tools-like-chatgpt-and-dall-e-work/ (explaining how consumers of generative AI technology include regular consumers and individuals from different types of industries like education and private corporations).
[50] See id. (explaining that the general public began to use generative AI technology when OpenAI released ChatGPT and made it available for individuals to use for free); see also Paul Smith-Goodson, The Extraordinary Uniquity of Generative AI And How Major Companies Are Using It, Forbes (July 21, 2023), https://www.forbes.com/sites/moorinsights/2023/07/21/the-extraordinary-ubiquity-of-generative-ai-and-how-major-companies-are-using-it/?sh=568edc332124 (discussing the emergence of Generative AI and outlining the technology companies that have been using the technology).
[51] See Cameron Hashemi-Pour, Open AI, TechTarget, https://www.techtarget.com/searchenterpriseai/definition/OpenAI (last visited Nov. 16, 2023) (discussing OpenAI’s membership, origin, and goals).
[52] See id. (discussing how OpenAI garnered the attention of regular consumers through the release of ChatGPT in 2022).
[53] See, e.g., Nitasha Tiku, OpenAI Unveils Better Image Generator, DALL-E 3, As AI Arms Race Deepens, Wash. Post (Sept. 20, 2023), https://www.washingtonpost.com/technology/2023/09/20/openai-dall-e-image-generator/ (discussing how OpenAI will release a new version of its generative AI tool to include capabilities for the technology to create images from written prompts).
[54] See Anantrasirichai & Bull, supra note 25 (discussing the creative applications of AI technology, namely “content creation, information analysis, content enhancement and postproduction workflows, information extraction and enhancement, and data compression”).
[55] See Kishor K., How Generative AI is Revolutionizing Drawing and Art, Medium (June 17, 2023), https://medium.com/@Nontechpreneur/how-generative-ai-is-revolutionizing-drawing-and-art-6c4e99e67ba9 (providing examples of different generative AI tools that artists are utilizing).
[56] See Harry Guinness, The Top AI Art Generators in 2024, Zapier (July 9, 2023), https://zapier.com/blog/ai-art-generator/ (discussing the most popular AI art generators like Midjourney, DALLE-2, DALLE-3, and Stable Diffusion).
[57] Compare Midjourney, supra note 9 (providing that a user can prompt the generative AI tool to generate an image), with Adi Robertson, OpenAI’s DALL-E AI Image Generator Can Now Edit Pictures, Too, The Verge (Apr. 6, 2022), https://www.theverge.com/2022/4/6/23012123/openai-clip-dalle-2-ai-text-to-image-generator-testing (explaining how an AI user can upload their own image and prompt the generative AI tool to make modifications to their image).
[58] See, e.g., Midjourney, supra note 9 (discussing how a user can make modifications to images generated by Midjourney); Andrew, How to Use Stable Diffusion, Stable Diffusion Art, https://stable-diffusion-art.com/beginners-guide/ (last updated Sept. 30, 2023) (discussing how a user can “ask Stable Diffusion to roughly follow an input image when generating a new one”).
[59] See, e.g., Kashtanova’s Letter, supra note 1, at 11–12 (discussing how Kashtanova used Photoshop after downloading some of the images generated by Midjourney and editing them in various ways such as “smoothing of gradients and modifications of lines and shapes”).
[60] See Robertson, supra note 57 (explaining that users of DALL-E 2 can upload images to the generative AI tool, select a specific area of the image, and instruct the model to edit it; “You can block out a painting on a living room wall and replace it with a different picture, for instance, or add a vase of flowers on a coffee table”).
[61] See Sabrina Ortiz, The Best AI Image Generators to Try Right Now, ZDNet (Feb. 5, 2024), https://www.zdnet.com/article/best-ai-image-generator/ (discussing how OpenAI’s DALL-E tool is the original text-to-image based tool that produces accurate depictions of text prompts and how Midjourney produces the highest quality photos).
[62] See OpenAI, https://openai.com/dall-e-3 (last visited Nov. 15, 2023) (demonstrating how DALL-E 3 is powered by ChatGPT and built on the same technology).
[63] See id. (“When prompted with an idea, ChatGPT will automatically generate tailored, detailed prompts for DALL-E 3 that bring your idea to life.”).
[64] See id. (“If you like a particular image, but it’s not quite right, you can ask ChatGPT to make tweaks with just a few words.”).
[65] See id. (“[T]he images you create with DALL-E 3 are yours to use and you don’t need our permission to reprint, sell or merchandise them.”).
[66] See Midjourney, supra note 9 (detailing that a user must first submit a text prompt for Midjourney to generate a set of images).
[67] See id. (“[T]he Midjourney Bot processes your request, creating four unique image options within a minute.”).
[68] See id. (providing the different options a user has once Midjourney generates the four images, such as prompting further variations of a selected image or re-running the job based on the original prompt).
[69] See id. (discussing the image selection process, namely selecting an image and giving access to a user to “download” the image and give access to “additional editing and generation tools”).
[70] See id. (stating that a user can re-run a job, which would “re-run the original prompt producing a new grid of images”).
[71] See id. (discussing that a user can select one of the four images from which Midjourney will generate “a new image grid that maintains the general style and composition of the selected image”).
[72] See id. (illustrating that a user can elect to zoom out of an image, prompting Midjourney to include additional background artwork).
[73] See id. (explaining that the user can save the final image and download a copy of it).
[74] See Kashtanova’s Letter, supra note 1, at 1–2 (explaining that the images in Zarya of the Dawn were created through Midjourney’s image generation service).
[75] See id. at 3, 5 (describing how Kashtanova went through an extensive process of iterative prompting to develop every image in the novel and how the development of “each individual image took hours”).
[76] See id. at 5 (detailing the iterative prompting process for the most important image in the novel and stating that “each and every image in the Work was created in a similar fashion”).
[77] See id. at 5 (explaining how the most important image in the novel sought to “establish Zarya’s character and the setting of the story,” was refined through an extensive iterative process).
[78] See id. (explaining that to create Zarya Holding a Postcard, Kashtanova “went through an extensive iterative process involving hundreds of versions of images”).
[79] See id. at 8 (“Kashtanova specified a subject (‘hands’), an object (‘a … photograph’), and descriptive context (‘dark skin,’ ‘holding,’ and ‘old’).”).
[80] See id. (describing how Kashtanova directed Midjourney to constrain the image to a 16:9 aspect ratio).
[81] See id. at 5–6 (including the first version of Zarya of the Dawn generated by Midjourney and providing the intermediate versions of the images, demonstrating how the iterative prompting process guided Midjourney to develop the image closer to Kashtanova’s mental conception for the work with each intermediate version).
[82] See id. at 5 (comparing the first version of Zarya Holding a Postcard, which only depicted dark-skinned hands holding a blank holding a blank postcard, with the final version that included was ultimately included in the book, which included considerably more details).
[83] See id. at 5–6 (depicting the hundreds of the intermediate versions of Zarya Holding a Postcard generated by Midjourney, and illustrating how some of the versions only depicted specific elements for the final image).
[84] See id. at 5 (illustrating in “Screenshot #2 of intermediate versions of ‘Zarya Holding a Postcard’” that some of the images generated by Midjourney only portrayed the details that were to appear in the postcard held by Zarya).
[85] See id. (illustrating in “Screenshot #2 of intermediate versions of ‘Zarya Holding a Postcard’” and “Screenshot #3 of intermediate versions of ‘Zarya Holding a Postcard’” that some of the images generated by Midjourney only depicted the backdrop of the city skyline that was to appear in the backdrop of the image).
[86] See id. at 5–6 (portraying over 90 versions of Zarya Holding a Postcard and demonstrating that the intermediate versions of the images included only specific elements that later appeared in the final version of the image).
[87] See id. at 7 (“Different elements of the final image are created, developed, refined, and relocated [in the intermediate versions]. The final image includes multiple elements from different generations of intermediate images all brought together into a cohesive whole.”).
[88] See id. at 10, 12 (discussing how the images in the final work were “cropped, framed, and placed to better convey the story and feel Kashtanova had in mind” and how Kashtanova used Photoshop to add more details such as brushstrokes into the final images).
[89] Compare Revised Issues Paper on Intellectual Property Policy and Artificial Intelligence, WIPO/IP/AI/2/GE/20/1 REV. (May 21, 2020) (defining AI-assisted works as outputs generated with human intervention and direction), with Content Guidelines, Kindle Direct Publ’g, https://kdp.amazon.com/en_US/help/topic/G200672390 (last visited Sept. 22, 2023) (defining AI-assisted works as works created by a human originally who later uses AI to modify or edit the works).
[90] See USCO AI Guidance, supra note 15 (“[T]he Office will consider whether the AI contributions are the result of ‘mechanical reproduction’ or instead of an author’s ‘own original mental conception, to which [the author] gave visible form.’”).
[91] See id. at 4 (“[W]hen an AI technology receives solely a prompt from a human and produces complex written, visual, or musical words in response, the ‘traditional elements of authorship’ are determined and executed by the technology–not the human user.”).
[92] See id. (“In other cases . . . a work containing AI-generated material will also contain sufficient human authorship to support a copyright claim.”).
[93] See WIPO, supra note 89 (discussing aspects of AI that are specific to intellectual property issues such as whether works generated by generative AI tools can qualify for copyright protection).
[94] See id. (defining what “AI-generated” and “AI-assisted” means in the context of intellectual property).
[95] See id. (“‘AI-generated’ and ‘generated autonomously by AI’ are terms that are used interchangeably and refer to the generation of an output by AI without human intervention.”).
[96] See id. (“‘AI-assisted’ outputs . . . are generated with material human intervention and/or direction.”).
[97] See Content Guidelines, supra note 89 (providing that AI-generated content is content originally created by AI, while AI-assisted content is originally created by a human who later used AI to modify the content).
[98] See id. (illustrating the difference between AI-generated and AI-assisted content for purposes of Amazon’s copyright policies).
[99] See id. (“We define AI-generated content as text, images, or translations created by an AI-based tool. If you used an AI-based tool to create the actual content (whether text, images, or translations).”).
[100] See id. (“[I]t is considered ‘AI-generated,’ even if you applied substantial edits afterwards.”).
[101] See id. (“If you created the content yourself, and used AI-based tools to edit, refine, error-check, or otherwise improve that content (whether text or images), then it is considered ‘AI-assisted’ and not ‘AI-generated.’”).
[102] See id. (“[I]f you used an AI-based tool to brainstorm and generate ideas, but ultimately created the text or images yourself, this is also considered ‘AI-assisted’ and not ‘AI-generated.’”).
[103] See, e.g., Is Artificial Intelligence Set to Become Art’s Next Medium?, Christie’s (Dec. 11, 2018), https://www.christies.com/en/stories/a-collaboration-between-two-artists-one-human-one-a-machine-0cd01f4e232f4279a525a446d60d4cd1 (explaining how an AI artwork sold for 45 times its high estimate at Christie’s art auction house).
[104] See Sam Gaskin, When Art Created by Artificial Intelligence Sells, Who Gets Paid?, Artsy (Sept. 17, 2018), https://www.artsy.net/article/artsy-editorial-art-created-artificial-intelligence-sells-paid (discussing how a Turkish artist sold an AI artwork for $8,000 at a charity auction house in San Francisco and how directors of art galleries have been pricing AI art “quite aggressively”).
[105] See Is Artificial Intelligence Set to Become Art’s Next Medium?, supra note 103 (describing that the sale of Portrait of Edmond Belamy signaled “the arrival of AI art on the world auction stage”); see also Gabe Cohn, AI Art at Christie’s Sells for $432,500, N.Y. Times (Oct. 25, 2018), https://www.nytimes.com/2018/10/25/arts/design/ai-art-sold-christies.html (explaining how Portrait of Edmond Belamy sold for “well over double the price” of an Andy Warhol print and a Roy Lichtenstein work auctioned alongside it).
[106] See Harsh Notariya, How This AI Artist Made $3 Million as NFT Markets Fell, BeinCrypto (Apr. 25, 2023), https://beincrypto.com/ai-artist-made-3-million-selling-nfts/ (discussing how the BottoDAO Image Generator has made $3 million in revenue on non-fungible token (NFT) marketplaces).
[107] See Gaskin, supra note 104 (explaining that a director of an auction house stated that AI art could grow the art market and extend beyond traditional patrons of artworks; one artist stated, “I’ve seen many atypical art collectors purchasing my work–including scientists, video game makers, and researchers in computer vision and AI”).
[108] See MarketResearch.BIZ, https://marketresearch.biz/report/generative-ai-in-art-market/ (last visited Feb. 10, 2024) (“Generative AI in Art Market size is expected to be worth around USD 5,840 Mn by 2032 from USD 212 Mn in 2022, growing at a CAGR of 40.5% during the forecast period from 2023 to 2032.”).
[109] See U.S. Const. art. I, § 8, cl. 8 (providing the constitutional basis for copyright protection and granting Congress authority to protect expressive works).
[110] See Wheaton v. Peters, 33 U.S. 591, 661 (1834) (“The sole interest of the United States and the primary object in conferring the monopoly . . . lie in the general benefits derived by the public from the labors of authors.”).
[111] See Twentieth Century Music Corp. v. Aiken, 422 U.S. 151, 156 (1975) (expressing that the monopoly granted by copyright protection “is intended to motivate the creative activity of authors . . . and to allow the public access to the products of this genius”).
[112] See 17 U.S.C. §§ 101–1511 (setting out the federal regime governing copyright protection and providing authors of expressive works limited monopolies to exploit their expressive works).
[113] See U.S. Const. art. I, § 8, cl. 8 (granting Congress the power “[t]o promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries”).
[114] See id.; see also Sean O’Connor, The Overlooked French Influence on the Intellectual Property Clause, 82 U. Chi. L. Rev. 733, 803 (2015) (arguing that the Copyright Clause aimed to encourage the promotion of the arts and sciences because at the time, “‘to promote’ was ‘synonymous with the words ‘to stimulate,’ ‘to encourage,’ or ‘to induce’”).
[115] See L. Ray Patterson & Craig Joyce, Copyright In 1791: An Essay Concerning the Founders’ View of the Copyright Power Granted to Congress in Article I, Section 8, Clause 8 of the U.S. Constitution, 52 Emory L.J. 909, 945 (2003) (explaining how the framers of the Constitution aimed to “implement the three copyright policies inherited from the Statute of Anne—the promotion of learning, the right of public access, and the public domain”).
[117] See Washingtonian Pub. Co. v. Pearson, 306 U.S. 30, 36 (1939) (stating that copyright serves “to afford greater encouragement to the production of literary works of lasting benefits to the world”).
[118] See Twentieth Century Music Corp. v. Aiken, 422 U.S. 151, 156 (1975) (explaining that awarding copyright holders limited monopolies reflects a balance between how “creative work is to be encouraged and rewarded, but private motivation must ultimately serve the cause of promoting broad public availability of literature, music, and other arts”).
[119] See id. (“The immediate effect of our copyright law is to secure a fair return for an ‘author's' creative labor. But the ultimate aim is, by this incentive, to stimulate artistic creativity for the general public good.”).
[120] See id.; see also Fox Film Corp. v. Doyal, 286 U.S. 123, 127 (1932) (“The sole interest of the United States and the primary object in conferring the monopoly lie in the general benefits derived by the public from the labors of authors.”).
[121] See William M. Landes & Richard A. Posner, An Economic Analysis of Copyright Law, 18 J. Legal Stud. 325, 326 (1989) (“While the cost of creating a work subject to copyright protection . . . is often high, the cost of reproducing the work, whether by the creator or by those to whom he has made it available, is often low.”).
[122] See id. (explaining how “once copies are available to others, it is often inexpensive for these users to make additional copies,” thus, the creator can choose to price copies “at or close to marginal cost, but the creator’s total revenues may not be sufficient to cover the cost of creating the work”).
[123] See Richard A. Posner, Intellectual Property: The Law and Economics Approach, 19 J. Econ. Persp. 57, 57 (2005) (“The traditional focus of economic analysis of intellectual property has been on reconciling incentives for producing such property with concerns about restricting access to it by granting exclusive rights in intellectual goods . . . enabling the owner to charge a price for access that exceeds marginal cost.”).
[124] See id. at 58–59 (arguing that while the creator of a copyrighted work can be compensated for the cost of the creation of the work, if the creator “has no right to exclude others from access to the property, the price is forced down to marginal cost by competition from copiers”).
[125] See Landes & Posner, supra note 121, at 328–29 (explaining that creators of copyrightable works would be uncertain about the return costs of creating works if their works could not be protected, which would in turn lead to “an additional disincentive to create works in the absence of copyright protection”).
[126] See id. at 332 (discussing how at a certain point, copyright protection would affect the creation of new works because individuals would be excluded from using materials from prior works).
[127] See id. (“A new work of fiction, for example, will contain the author’s expressive contribution but also characters, situations, plot details, and so on, invented by previous authors. Similarly, a new work of music may borrow tempo changes and chord progressions from earlier works.”).
[128] See id. at 332–33 (explaining how a later author of a work would prefer minimum protection for works created by others because he would be able to borrow from previous works without infringing upon their copyright).
[129] See Mark S. Nadel, How Current Copyright Law Discourages Creative Output: The Overlooked Impact of Marketing, 19 Berkeley Tech. L.J.785, 855 (arguing that overbroad copyright protection “diminishes net new creations” and undermines the public policy rationales to encourage the creation of new expressive works).
[130] See e.g., 17 U.S.C. § 103 (explaining that individuals can receive limited copyright protection for derivative works and compilations provided that the material has been used lawfully by the author); see also 17 U.S.C. § 107 (providing limitations on exclusive rights granted to authors and allowing authors to use portions of copyrighted works in certain contexts without being at risk of liability for infringement).
[131] See Shyamkrishna Balganesh, Copyright As Legal Process: The Transformation of American Copyright Law, 168 U. Pa. L. Rev. 1101, 1127 (2020) (providing for a history of copyright law in the United States).
[132] See id. (discussing the early history of federal copyright law).
[133] See 17 U.S.C. §§ 101–1511 (setting out the current legislative regime for copyright protection, expanding the scope of copyright protection by awarding copyright owners longer monopolies over their rights, and introducing the fair use regime which allows authors to use material from already copyrighted works without liability of infringement).
[134] See 17 U.S.C. § 102 (describing the requirements for copyright protection, specifically that the work must be original, fixed in a tangible medium of expression, and belongs to one of the enumerated categories of works of authorship).
[135] See U.S. Copyright Office, Compendium of U.S. Copyright Office Practices § 306 (2021) (stating that an issuance of certificate constitutes prima facie evidence of the validity of a copyright in any judicial proceedings).
[136] See 17 U.S.C. § 102 (providing that “copyright protection subsists . . . in [1] original [2] works of authorship [3] fixed in any tangible medium of expression”).
[137]See id. (providing that a work must be fixed in “any tangible medium of expression, now known or later developed, from which they can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device”).
[138] See id. (providing the eight categories of works that are capable of receiving copyright protection: “(1) literary works; (2) musical works, including any accompanying words; (3) dramatic works, including any accompanying music; (4) pantomimes and choreographic works; (5) pictorial, graphic, and sculptural works; (6) motion pictures and other audiovisual works; (7) sound recordings; and (8) architectural works”).
[139] See id. (stating that copyright protection requires “original works of authorship” but not providing a definition for the terms).
[140] See U.S. Const. art. I, § 8, cl. 8 (failing to provide a definition for who constitutes “Authors”); see also 17 U.S.C. § 102 (providing that copyright protection subsists in “original works of authorship” but failing to provide a definition for “authorship”).
[141] See, e.g., Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 57–58 (1884) (providing an explanation that an author is the individual who originates an expressive work).
[142] See 17 U.S.C. § 201 (“Copyright in a work protected under this title vests initially in the author or authors of the work.”).
[143] See id. (providing that in a work made-for-hire, “the employer or other person for whom the work was prepared is considered the author”).
[144] See David Nimmer & Melville B. Nimmer, Nimmer on Copyright § 1.06 (rev. ed. 2018) (describing how the theories of authorship include sole authorship, joint authorship, and authorship under the work made-for-hire doctrine).
[145] See id. (finding the plaintiff as the sole author of the work because he personally selected and arranged all the elements in the copyrightable work).
[146] See 17 U.S.C. § 101 (defining a joint work as a work prepared by more than one person).
[147] See id. (“A ‘joint work’ is a work prepared by two or more authors with the intention that their contributions be merged into inseparable or interdependent parts of a unitary whole.”).
[148] See 17 U.S.C. § 201 (providing that the employer or person for whom the work was prepared is considered the author of such work); see also Nimmer & Nimmer, supra note 144 (illustrating that while the constitutional sense of an author is the original creator of a work, an employer may also be considered an author under the work made for hire doctrine).
[149] See 17 U.S.C. § 101 (describing that a work made for hire is a work prepared by an employee within the scope of his employment or a work that falls within a specific class of works that is specially commissioned by another under an agreement between the parties that the work shall be considered a work for hire).
[150] See Nimmer & Nimmer, supra note 144 (describing how “the one indispensable element of authorship is originality” and that an individual is an author of a work “if the resulting work is the product of one’s own independent efforts”); see also Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., Inc., 499 U.S. 340, 345 (1991) (“The sine qua non of copyright is originality. To qualify for copyright protection, a work must be original to the author.”).
[151] See Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 57–58 (1884) (holding that the author of the work is the individual who is the originator of the work and one who makes the expressive elements in a work original).
[152] See Feist Publ’ns, 499 U.S. at 345 (providing that originality “means only that the work was independently created by the author (as opposed to copied from other works), and that it possesses at least some minimal degree of creativity”); see also Nimmer & Nimmer, supra note 144, at § 2.01 (“[A] work is original and may command copyright protection even if it is completely identical with a prior work, provided it was not copied from that prior work but is instead a product of the independent efforts of its author.”).
[153] See Feist Publ’ns, 499 U.S. at 345 (explaining how the originality requirement only requires a very minimal level of creativity and stating that “the vast majority of works make the grade quite easily, as they possess some creative spark”).
[154] See Bleistein v. Donaldson Lithographing Co., 188 U.S. 239, 251–52 (1903) (“It would be a dangerous undertaking for persons trained only to the law to constitute themselves final judges of the worth of pictorial illustrations, outside of the narrowest and most obvious limits. At the one extreme some works of genius would be sure to miss appreciation. . . . At the other end, copyright would be denied to pictures which appealed to a public less educated than the judge.”).
[155] See Burrow-Giles, 111 U.S. at 57–58 (“An author . . . is ‘he to whom anything owes its origin; originator; maker; one who completes a work of science or literature.’”).
[156] See id. at 55 (explaining how the plaintiff produced the picture).
[157] See id. (describing the plaintiff’s affirmative actions in selecting and arranging the expressive elements in the photograph, including posing the subject and selecting and arranging all the elements in the backdrop of the photograph).
[158] See id. at 60 (holding that the photograph was an original work of art and “the product of plaintiff’s intellectual invention” because the plaintiff created the photograph “entirely from his own mental conception, to which he gave visible form” by posing the subject and selecting and arranging the costume and accessories in the photograph).
[159] See id. at 59–60 (describing that the plaintiff demonstrated authorship “by posing the said Oscar Wilde in front of the camera, selecting and arranging the costume, draperies, and other various accessories in said photograph, arranging the subject so as to present graceful outlines, arranging and disposing the light and shade, suggesting and evoking the desired expression”).
[160] See id. at 61 (defining that an author includes “originating, making, producing, as the inventive or mastermind, the thing which is to be protected”).
[161] See Aalmuhammed v. Lee, 202 F.3d 1227, 1234 (9th Cir. 2000) (applying Burrow-Giles in the context of a joint work and stating that, in the absence of a contract, the author is the one who “‘superintends’ the work by exercising control”).
[162] See id. at 1229 (explaining how Warner Brothers contracted with Lee and his production companies to create the movie Malcolm X which was to be based on a book).
[163] See id. (discussing Lee’s role in the movie, including how he was a director, co-writer of the screenplay, and a co-producer).
[164] See id. (describing how Denzel Washington, who played the role of Malcolm X asked Aalmuhammed to “assist him in his preparation for the starring role” due to his detailed knowledge about Malcolm X’s life and Islam and because he previously wrote, directed, and produced a documentary film about Malcolm X).
[165] See id. at 1229–30 (discussing that Aalmuhammed’s contributions to the film were “to ensure the religious and historical accuracy” of the movie and stating that Washington testified that “Aalmuhammed’s contribution to the movie was ‘great’ because he ‘helped to rewrite, to make more authentic’”).
[166] See id. at 1230 (describing that Aalmuhammed requested writing credit as co-writer of the film but was turned down and credited “only as an ‘Islamic Technical Consultant,’ far down the list”).
[167] See id. (recounting that Aalmuhammed applied for a copyright with the USCO claiming he was a co-author of the movie and that he subsequently filed suit against Spike Lee, his production companies, and Warner Brothers, seeking declaratory relief and accounting under the Copyright Act).
[168] See id. at 1236 (holding that the district court correctly granted summary judgment in favor of defendant because the record before it “established no genuine issue of fact as to Aalmuhammed’s co-authorship of Malcolm X”).
[169] See id. at 1234–35 (discussing how Warner Brothers and Spike Lee superintended the work and “control in many cases will be the most important factor” in determining whether an individual is an author).
[170] See id. (discussing that the Constitution gave authors copyrights to promote the process of science and the arts and “[p]rogress would be retarded rather than promoted, if an author could not consult with others and adopt their useful suggestions without sacrificing sole ownership of the work”).
[171] See Andrien v. S. Ocean Cnty. Chamber of Com., 927 F.2d 132, 134 (3d Cir. 1991) (holding that a plaintiff who compiled preexisting maps to create a new map could be considered the author for purposes of copyright protection).
[172] See id. at 133 (“[The plaintiff” collected available maps of local taxing bodies and a divers’ map locating shipwrecks in the area.”).
[173] See id. (describing how the plaintiff conducted a personal survey to determine the information he sought to include in the maps).
[174] See id. (describing how the plaintiff took a collection of preexisting maps and the information he collected through personal surveys to a printing shop to prepare the map).
[175] See id. (explaining how during his deposition, the plaintiff testified that the employee performed all the assignments at his direction, “‘with me at her elbow practically’”).
[176] See id. (“Almost daily he spent about an hour at the print shop over a three week period.”).
[177] See id. at 135 (describing how “[p]oets, essayists, novelists, and the like may have copyrights even if they do not run the printing presses or process the photographic plates necessary to fix the writings into book form”).
[178] See Lindsay v. Wrecked & Abandoned Vessel R.M.S. Titanic, No. 97 Civ. 9248, 1999 U.S. Dist. LEXIS 15837, at *15–16 (S.D.N.Y. Oct. 12, 1999) (holding that a plaintiff who directed the filming of a documentary film but did not physically perform the filming could be considered the author of the film).
[179] See id. at *4–5 (describing the plaintiff’s role in the making of the documentary film).
[180] See id. at *5 (explaining how the plaintiff sailed with the expedition crew to the wreck site and “remained at sea for approximately one month”).
[181] See id. at *6 (“[T]he plaintiff created various storyboards for the film, a series of drawings which incorporated images of the Titanic by identifying specific camera angles and shooting sequences.”).
[182] See id. (describing how the plaintiff “personally constructed the light towers”).
[183] See id. at *7 (detailing how the plaintiff directed the filming “from on board the salvage vessel . . . after leading daily planning sessions with the crew”).
[184] See id. at *12–13 (rejecting the defendants’ argument that the director was not the author of the footage because he did not personally photograph the wreckage and stating that the defendants’ argument “does not hold water”).
[185] See id. at *14 (“[The plaintiff’s] alleged storyboards and the specific directions he provided to the film crew regarding the use of the light-towers and the angles from which to shoot the wreck all indicate that the final footage would indeed be the product of [the plaintiff’s] ‘original intellectual conceptions.’”).
[186] See id. at *15–16 (finding that when a plaintiff “alleges that he exercised such a high degree of control over a film operation,” and the final product “duplicates his conceptions and visions of what the film should look like,” the plaintiff can be the “author” of the work).
[187] See 17 U.S.C. § 201 (providing that the employer or person for whom the work was prepared is considered the author of such work).
[188] See Cmty. For Creative Non-Violence v. Reid, 490 U.S. 730, 740 (1989) (holding that an artist who produced a sculpture in his own time and with his own resources was not an employee under the work for hire doctrine, and thus, was entitled to copyright protection of the sculpture).
[189] See id. at 733 (describing how CCNV reached out to Reid who agreed to sculpt the sculpture which would consist of three human figures).
[190] See id. (explaining how CCNV “conceived the idea” for the sculpture, including the outline and size of the figures in the sculpture, and a steam grate, which would rest on a base under the sculpture).
[191] See id. (“[The artist] agreed to sculpt the three human figures. CCNV agreed to make the steam grate and pedestal for the statue.”)
[192] See id. at 734 (describing how the artist proposed that the work be cast in bronze to ensure a lower cost for the sculpture and faster production time).
[193] See id. (“The parties did not sign a written agreement. Neither party mentioned copyright.”).
[194] See id. at 752 (holding that the work was not a work made for hire because there was no agreement between CCNV and the artist expressing that the work was to constitute a work made for hire, and finding the artist was an independent contractor because did not have a relationship akin to employment with CCNV).
[195] See e.g., Naruto v. Slater, No. 15-cv-04324-WHO, 2016 U.S. Dist. LEXIS 11041, at *10 (N.D. Cal. Jan. 28, 2016) (holding that a monkey cannot bring a suit of copyright infringement because non-human authors do not satisfy the authorship requirement under the Copyright Act), aff’d, 888 F.3d 418, 426 (9th Cir. 2018).
[196] See id. at *2 (describing how Slater visited a reserve in Indonesia occupied by monkeys and left his camera in the reserve, allowing one monkey to take an image of himself).
[197] See id. (explaining that the monkey was accustomed to seeing cameras in the reserve and took an image of himself by “‘independent, autonomous action’ in examining and manipulating Slater’s unattended camera and ‘purposely pushing’ the shutter release multiple times, ‘understanding the cause-and-effect relationship between pressing the shutter release, the noise of the shutter, and the change to his reflection in the camera lens.’”).
[198] See id. at *2–3 (providing that the People for Ethical Treatment of Animals and Next Friends brought a copyright infringement suit on behalf of the monkey after Slater displayed, advertised, reproduced, distributed, and sold copies of the monkey’s selfies).
[199] See id. at *9–10 (holding that the monkey was not an author within the meaning of the Copyright Act because authorship cannot extend to non-human entities).
[200] See id. (reasoning that the USCO “agrees that works created by animals are not entitled to copyright protection” by relying upon to the Compendium of U.S. Copyright Office practices issued by the USCO, which provides that authorship requires that the work is created by a human being).
[201] See USCO AI Guidance, supra note 15 (“The Office’s registration policies and regulations reflect . . . judicial guidance on [the issue of human authorship].”).
[202] See U.S. Copyright Office, supra note 135 (stating that only a human author can register a copyright).
[203] See id. at § 313.2 (stating that the USCO “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.”).
[204] See id. (stating that the USCO will not consider machines to be authors).
[205] See id. (“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.’” (quoting U.S. Copyright Office, Report to the Librarian of Congress By the Register of Copyrights 5 (1996)).
[206] See Letter from Robert J. Kasunic to Van Lindberg, supra note 6 (stating that the images generated by Midjourney in Zarya of the Dawn do not constitute human authorship).
[207] See Thaler v. Perlmutter, No. 22-1564, 2023 U.S. Dist. LEXIS 145823, at *20–21 (D.C. Dist. Ct. Aug. 18, 2023) (holding that an AI-generated work does not demonstrate human authorship).
[208] See, e.g., Etrailer Corp. v. Onyx Enters., Int’l Corp., No. 4:17-CV-01284-AGF, 2018 U.S. Dist. LEXIS 19916, at *3, *8 (E.D. Mo. Feb. 7, 2018) (holding that the plaintiff was the author of the photographs in dispute even considering that he used Photoshop to edit the photographs).
[209] See Letter from Robert J. Kasunic to Van Lindberg, supra note 6 (maintaining that despite Kashtanova’s hundreds of prompts, the USCO did not find that the elements generated by Midjourney constituted human authorship).
[210] See Kevin Roose, An A.I.-Generated Picture Won an Art Prize. Artists Aren’t Happy, N.Y. Times (Sept. 2, 2022), https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html (discussing how artist, Allen, created the AI-Generated Work using Midjourney, which won him a prize at the Colorado State Fair’s annual art competition).
[211] See id. (stating how Allen prompted Midjourney to create the artwork but failing to provide the prompt used by Allen because Allen refused to disclose it).
[212] See Letter from Suzanne Wilson, U.S. Copyright Off., to Tamara Pester, Tamara S. Pester, LLC (Sept. 5, 2022), https://fingfx.thomsonreuters.com/gfx/legaldocs/byprrqkqxpe/AI%20COPYRIGHT%20REGISTRATION%20decision.pdf, 2 [hereinafter Allen’s Letter] (discussing how Allen “input numerous revisions and text prompts at least 624 times to arrive at the initial version of the image”).
[213] See id. at 1 (affirming the USCO’s initial decision to reject copyright protection for the work because the work was generated by an AI tool and Allen refused to disclaim the elements in the work that were generated by the AI tool).
[214] See id. at 5 (providing that the artwork contains “more than a de minimus amount of content generated by artificial intelligence” and thus must be rejected).
[215] See id. at 3 (“Because Mr. Allen has refused to disclaim the material produced by AI, the Work cannot be registered as submitted.”).
[216] See Thaler v. Perlmutter, No. 22-1564, 2023 U.S. Dist. LEXIS 145823, at *20–21 (D.C. Dist. Ct. Aug. 18, 2023) (holding that an AI-generated work is not protectable by copyright because it does not satisfy the authorship requirement under the Copyright Act).
[217] See id. at *2 (describing how the plaintiff created the work using an AI system called the “Creativity Machine” that he developed himself).
[218] See id. *2–3 (failing to enumerate the type of AI program the plaintiff created but stating that the work was autonomously and fully created by the program); see also Marr, supra note 27 (explaining the differences between a generative AI program and a traditional AI program, namely that a generative AI program requires a user to first enter an input to prompt the generative AI program to generate content).
[219] See id. at *3 (describing the plaintiff’s attempt to register the image with the USCO).
[220] See id. (outlining the plaintiff’s argument that because he created the AI program, the output generated by the AI program constituted a work-for-hire, which in the traditional employer-employee relationship recognizes the employer as the author of the work created by the employee).
[221] See id. at *4–5 (stating that the plaintiff sought a court judgment after his application for copyright registration was denied by the USCO twice).
[222] See id. at *11 (emphasizing that “human authorship is a bedrock requirement of copyright”).
[223] See id. at *15 (discussing how in Burrow-Giles, the copyrightability of the photograph rested on the fact that the human creator conceived of and used the camera to capture the image).
[224] See id. at *3 (describing the plaintiff’s argument that the AI program created the artwork under his direction, and thus, the machine’s work constitutes a work-for-hire).
[225] See id. at *20 (“Given that the work at issue did not give rise to a valid copyright upon its creation, plaintiff's myriad theories for how ownership of such a copyright could have passed to him need not be further addressed.”).
[226] See, e.g., Patrick Zurth, Artificial Creativity? A Case Against Copyright Protection for AI-Generated Works, 25 UCLA J.L. & Tech. 1, 17 (arguing that copyright protection should not extend to AI-assisted works because existing incentives for companies exist to employ and invest in AI technologies).
[227] See Ryan Abbott & Elizabeth Rothman, Disrupting Creativity: Copyright Law in the Age of Generative Artificial Intelligence, 75 Fla. L. Rev.1141, 1187 (2023) (outlining the copyright skeptics’ arguments that the benefits of protecting AI works are reduced because an AI tool does not need any incentives to create new works).
[228] See id. (“Once a creative AI like DALLE 2 exists, the marginal cost of having it create additional works may be close to zero.”).
[229] See id. (explaining that under these arguments, “once an AI is fully operational, it can simply continue to make a practically limitless number of works”).
[230] See id. (“As a result, the scale tips against protection because many or most AI-generated works would be created in the absence of copyright protection.”).
[231] See id. (explaining the argument that copyright protection is not necessary to ensure that AI-assisted works remain to be created); Landes & Posner, supra note 121 (discussing the rationale behind the incentive-access framework which balances the interests of awarding copyright owners exclusive rights and ensuring access to expressive works to future creators); see also Xiao Wang, AI Output: A Human Condition that Should Not Be Protected Now, or Maybe Ever, 20 Chi.-Kent J. Intell. Prop. 136, 136 (2021) (arguing that “the protection of [AI-assisted works] is neither useful nor necessary for achieving the copyright system’s goal of promoting the public interest and other social benefits”).
[232] See Wang, supra note 232, at 154, 161 (explaining that the copyright system promotes public interest by “accelerating the emergence of new cultural products on the market” and “reserving space for subsequent use that will be free from the liability of infringement,” and arguing that granting protection to AI-assisted works does not promote public interest); see also Nadel, supra note 129 (discussing the risks of overbroad copyright protection, specifically that new authors entering the market will be blocked from using materials from works that are already copyrighted); Landes & Posner, supra note 121, at 332 (explaining that authors may be required to borrow materials from already existing copyrighted works to create new works).
[233] See Martin Senftleben & Laurens Buijtelaar, Robot Creativity: An Incentive-Based Neighboring Rights Approach, 42 Eur. Intell. Prop. Rev. 797, 810 (2022) (“Therefore, the described benefits accruing from the grant of protection—an incentive for the full development and dissemination of robot works—must be weighed against potential detriments, such as an impediment of creative remix and reuse of AI-generated creations, including their use as training material for other robots.”).
[234] See Wang, supra note 232, at 167 (arguing that incentives for creating new AI-assisted works would not be supported by copyright protection, and upon “weighing the possible benefits and expenses in protecting AI outputs . . . we should not grant copyright protection at least under the current technological level”).
[235] See USCO AI Guidance, supra note 15 (outlining that the USCO’s concerns regarding copyright protection of AI-assisted works are centered around the issue of human authorship, and specifically, stating that “copyright can protect only material that is the product of human creativity”).
[236] See id. (“Based on the Office’s understanding of the generative AI technologies currently available, users do not exercise ultimate creative control over how such systems interpret prompts and generate material.”).
[237] See id. (“For example, if a user instructs a text-generating technology to ‘write a poem about copyright law in the style of William Shakespeare,’ she can expect the system to generate text that is recognizable as a poem, mentions copyright, and resembles Shakespeare’s style. But the technology will decide the rhyming pattern, the words in each line, and the structure of the text.”).
[238] See id. (“When an AI technology determines the expressive elements of its output, the generated material is not the product of human authorship.”).
[239] See id. (stating that copyright protection will not extend to outputs of generative AI tools when that output is generated as a result of users’ prompts alone).
[240] See id. (“The Office’s registration policies and regulations reflect . . . judicial guidance on [the issue of human authorship].”).
[241] See, e.g., Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884) (defining that an author is the individual “originating, making, producing, as the inventive or mastermind, the thing which is to be protected”); see also Aalmuhammed v. Lee, 202 F.3d 1227, 1234 (9th Cir. 2000) (applying Burrow-Giles in the context of a joint work and stating that, the author is the one who “‘superintends’ the work by exercising control”).
[242] See, e.g., Burrow-Giles, 111 U.S. at 61 (1884) (defining an author as the master mind or originator behind the copyrightable work).
[243] See USCO AI Guidance, supra note 15 (“In the case of works containing AI-generated material, the Office will consider whether the AI contributions are the result of ‘mechanical reproduction’ or instead of an author’s ‘own original mental conception, to which [the author] gave visible form.’”).
[244] See Kashtanova’s Letter, supra note 1, at 10 (explaining how Kashtanova “painstakingly shaped each set of inputs and prompts over hundreds of iterations to create as perfect a rendition of her vision as possible”); WIPO, supra note 89 (stating that an AI-generated work is an output that involved no human intervention, whereas an AI-assisted work is an output that involved some level of human intervention).
[245] See Kashtanova’s Letter, supra note 1, at 11–12 (discussing how Kashtanova used Photoshop to modify some of the images after they were generated by Midjourney); see also Content Guidelines, supra note 89 (providing that an AI-generated work includes works that are generated by AI but then edited by a user afterward).
[246] See Thaler v. Perlmutter, No. 22-1564, 2023 U.S. Dist. LEXIS 145823, at *19–20 (D.C. Dist. Ct. Aug. 18, 2023) (explaining that the plaintiff did not submit any evidence to demonstrate that he guided the AI technology to create the final image).
[247] See, e.g., WIPO, supra note 89 (stating that an AI-generated work is an output that involved no human intervention, whereas an AI-assisted work is an output that involves some level of human intervention).
[248] See Kashtanova’s Letter, supra note 1, at 10 (explaining how Kashtanova’s prompts guided Midjourney to generate the final images she was satisfied with).
[249] But see USCO AI Guidance, supra note 15 (arguing that the output created by a generative AI tool cannot be attributable to the AI tool because the generative AI tool ultimately determines how to implement the prompts in its output).
[250] Compare Kashtanova’s Letter, supra note 1, at 9–10 (providing that each image in Zarya of the Dawn was created as a result of hundreds of iterative prompts written and created by Kashtanova), and Allen’s Letter, supra note 213 (describing how Allen used iterative prompts to guide the generative AI technology to produce the final image), with Thaler, 2023 U.S. Dist. LEXIS 145823, at *19–20 (discussing that there was no human intervention on the part of the plaintiff to guide the generative AI tool to produce the output it generated).
[251] See Kashtanova’s Letter, supra note 1, at 10 (describing how Kashtanova inputted hundreds of prompts into Midjourney to reach the final images that she was satisfied with in Zarya of the Dawn).
[252] See Allen’s Letter, supra note 213 (discussing how Allen “input numerous revisions and text prompts at least 624 times to arrive at the initial version of the image”).
[253] See Midjourney, supra note 9 (explaining that users can input prompts to provide further modifications to the output that Midjourney generates).
[254] See Thaler, 2023 U.S. Dist. LEXIS 145823, at *19–20 (explaining that no human intervention was used to create the output generated by the AI technology).
[255] See id. (discussing how the plaintiff’s image was fully generated by the AI technology).
[256] See Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884) (providing that authorship is exemplified by an individual’s intellectual conception).
[257] See Thaler, 2023 U.S. Dist. LEXIS 145823, at *2, *14 (analyzing precedent defining authorship as human created expression and providing that the plaintiff did not display any human intervention in the making of the work).
[258] See id. at *15 (discussing how in Burrow-Giles, the copyrightability of the photograph rested on the fact that the human creator conceived of the image in his head prior to taking the photograph); see also Burrow-Giles, 111 U.S. at 60 (holding that the plaintiff was the author of the photographs because he produced the photograph of his mental conception of how he wanted the costume, accessories, and lighting in the photograph to appear).
[259] See Thaler, 2023 U.S. Dist. LEXIS 145823, at *15, *20 (comparing how, unlike the plaintiff in Burrow-Giles, the plaintiff in the case did not demonstrate any human involvement or conception in the AI-generated work).
[260] See Burrow-Giles, 111 U.S. at 58–59 (describing how authorship involves “intellectual conception . . . in which there is novelty, invention, originality, and therefore comes within the purpose of the Constitution in securing its exclusive use or sale to its author”); see also Aalmuhammed v. Lee, 202 F.3d 1227, 1236 (9th Cir. 2000) (holding that an author is an individual who exercises control over the final product of the work).
[261] See Burrow-Giles, 111 U.S. at 59–60 (describing the ways in which the plaintiff created the elements in the photograph from his own mental conception).
[262] See id. at 55 (explaining the plaintiff’s affirmative actions in selecting and arranging the elements in the photograph).
[263] See id. at 59–60 (explaining that the bedrock of authorship is centered around the mental conception of the individual creating the work).
[264] See id. at 61 (providing that the plaintiff was the author who was the inventive or mastermind of the photograph to be protected).
[265] See Aalmuhammed, 202 F.3d at 1236 (holding that the district court correctly granted summary judgment in favor of defendant because Aalmuhammed did not demonstrate that he exercised control over the final version of the movie).
[266] See id. at 1229–30, 1236 (discussing the various ways in which Aalmuhammed contributed to the movie, but ultimately holding that he was not a co-author of the movie because he was not the “mastermind” behind the work).
[267] See id. at 1234–35 (discussing how Spike Lee superintended the work and “control in many cases will be the most important factor” in determining whether an individual is an author).
[268] See Andrien v. S. Ocean Cnty. Chamber of Com., 927 F.2d 132, 134 (3d Cir. 1991) (holding that a plaintiff who directed a printing employee to compile preexisting maps into one map was the author of the final work); see also Lindsay v. Wrecked & Abandoned Vessel R.M.S. Titanic, No. No. 97 Civ. 9248, 1999 U.S. Dist. LEXIS 15837, at *15–16 (S.D.N.Y. Oct. 12, 1999) (holding that a plaintiff who directed the filming of a documentary film but did not physically perform the filming could be considered the author of the film).
[269] See Andrien, 927 F.2d at 133 (providing that the plaintiff sought to create a map by compiling information from preexisting maps and including information collected by him through personal surveys of the areas to be included in the map).
[270] See id. at 133–34 (explaining how although the plaintiff directed the printing employee and asked him to “do the ‘art work,’” the plaintiff was the author of the work).
[271] See Lindsay, 1999 U.S. Dist. LEXIS 15837, at *15–16 (holding the director of the film was the author of the film for purposes of copyright protection even though the film conducted the physical act of filming the shipwreck footage).
[272] See id. at *6 (“[T]he plaintiff created various storyboards for the film, a series of drawings which incorporated images of the Titanic by identifying specific camera angles and shooting sequences.”); see also id. at *7 (detailing how the plaintiff directed the filming “from on board the salvage vessel . . . after leading daily planning sessions with the crew”).
[273] See id. at *15–16 (finding that when a plaintiff “alleges that he exercised such a high degree of control over a film operation,” and the final product “duplicates his conceptions and visions of what the film should look like,” the plaintiff can be the “author” of the work).
[274] See Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884) (defining that an author is the individual “originating, making, producing, as the inventive or mastermind, the thing which is to be protected”); see also Aalmuhammed v. Lee, 202 F.3d 1227, 1234 (9th Cir. 2000) (applying Burrow-Giles in the context of a joint work and stating that, the author is the one who “‘superintends’ the work by exercising control”).
[275] See Aalmuhammed, 202 F.3d at 1234–35 (holding that Aalmuhammed was not a co-author of the movie because he did not exercise creative control over the production of the movie).
[276] See USCO AI Guidance, supra note 15 (“Based on the Office’s understanding of the generative AI technologies currently available, users do not exercise ultimate creative control over how such systems interpret prompts and generate material. Instead, these prompts function more like instructions to a commissioned artist—they identify what the prompter wishes to have depicted, but the machine determines how those instructions are implemented in its output.”).
[277] See Kashtanova’s Letter, supra note 1, at 10 (explaining how Kashtanova’s prompts guided Midjourney to generate the final images she was satisfied with).
[278] See, e.g., Allen’s Letter, supra note 213 (discussing how Allen “input numerous revisions and text prompts at least 624 times to arrive at the initial version of the image”).
[279] See, e.g., id. (inquiring whether an AI-assisted work is copyrightable depends on examining whether a human, or a machine, conceived the traditional elements of authorship).
[280] See, e.g., Kashtanova’s Letter, supra note 1, at 10 (arguing that Kashtanova should be considered the author of the images in Zarya of the Dawn because she provided Midjourney hundreds of high-quality descriptive prompts).
[281] See Andrien v. S. Ocean Cnty. Chamber of Com., 927 F.2d 132, 134 (3d Cir. 1991) (holding that the plaintiff was the author of the maps even though another individual executed the compiling of the final maps because the plaintiff had the intellectual conception behind the work and directed the individual on how to execute the expressive elements in the final work).
[283] See USCO AI Guidance, supra note 15 (discussing that the inquiry as to whether an AI-assisted work can qualify for copyright production is to be determined on a case-by-case basis, and providing that applicants must disclose the use of generative AI technology in their application for copyright protection).
[284] See id. (enumerating that the critical question as to whether AI-assisted works can be protected is whether the computer is used merely as an assisting instrument or whether the “traditional elements of authorship in the work . . . were actually conceived . . . by a machine”).
[285] See id. (requiring artists submitting copyright registration applications for works involving the use of generative AI technology to disclose the ways in which they used the generative AI tool in the creation of the works).
[286] See, e.g., Kashtanova’s Letter, supra note 1, at 8 (discussing how for one of the images in Zarya of the Dawn, Kashtanova inputted the following prompt, “a holographic elderly white woman named Raya, raya is having curly hair and she is inside a spaceship,” as well as mood and style-related directions, such as “cinematic,” “Raya as a hologram,” and “unreal engine”).
[287] See, e.g., id. at 7–8 (specifying that for one of the images, Kashtanova inputted a prompt to Midjourney that contained the words “dark skin hands holding an old photograph,” and the image Midjourney produced reflected the prompts inputted by Kashtanova and contained dark-skinned hands holding a photograph).
[288] See generally USCO AI Guidance, supra note 15 (discussing that in a case-by-case inquiry of whether an AI-assisted work can be protected, the USCO will consider whether the authorship elements of the work could be attributed to the artist).
[289] See Nimmer & Nimmer, supra note 144 (explaining that “one indispensable element of authorship is originality”); see also Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 57–58 (1884) (holding that the author of the work is the originator of the work and person responsible for making the expressive elements in a work original).
[290] See Feist Publ'ns, Inc. v. Rural Tel. Serv. Co., Inc., 499 U.S. 340, 345 (1991) (providing that the two requirements for originality include that the work was created independently by the author and a minimal level of creativity).
[291] See, e.g., Kashtanova’s Letter, supra note 1, at 9 (disclosing one of Kashtanova’s prompts, which included language to set the scene for the image she desired, “sci-fi scene future empty New York . . . no people, tall trees, New York Skyline forest punk”).
[292] See Feist Publ'ns, 499 U.S. at 345 (describing how the originality standard is extremely low and that “the vast majority of works make the grade quite easily”); see also Bleistein v. Donaldson Lithographing Co., 188 U.S. 239, 251–52 (1903) (explaining that the originality standard is intentionally low to ensure that the judiciary does not pass judgments on what it believes art should be).
[293]See Burrow-Giles, 111 U.S. at 59 (defining an author as the original conceiver of an idea and the “mastermind” behind the idea); see also Kashtanova’s Letter, supra note 1, at 10 (explaining how Kashtanova’s “painstakingly” prompted Midjourney through hundreds of prompts to generate the final images she was satisfied with).
[294] Compare Kashtanova’s Letter, supra note 1, at 10 (demonstrating that Kashtanova disclosed examples of the prompts she submitted to Midjourney), with Allen’s Letter, supra note 213, at 6 n.8 (explaining that Allen “declined to disclose any specific prompt” to the USCO).
[295] See Kashtanova’s Letter, supra note 1, at 8 (illustrating the detailed and extensive prompting Kashtanova undertook to guide Midjourney to create the final image).
[296] See id. at 3 (explaining how Kashtanova came up with the idea of Zarya of the Dawn due to the loss of her grandmother and best friend).
[297] See id. at 3–4 (describing how, in the story, Zarya travelled through new worlds and discovered how she could “adjust to any of them without any additional equipment”).
[298] See id. at 4 (describing that the work is “designed to communicate–through words and pictures,” the experience of Zarya traveling through desolated worlds).
[299] See id. at 5 (explaining that each image in the novel went through an iterative prompting process that began with Kashtanova’s clear intent and vision for the images); see also Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884) (describing that the plaintiff chose the costume Wilde wore and picked the accessories in the backdrop of the photograph).
[300] See Kashtanova’s Letter, supra note 1, at 5–6 (illustrating how the most important image in the novel, “Zarya Holding a Postcard,” which sought to “establish Zarya’s character and the setting of the story,” was refined through an extensive iterative process).
[301] See id. at 5 (explaining that to create Zarya Holding a Postcard, Kashtanova “went through an extensive iterative process involving hundreds of versions of images”).
[302] See id. (describing how Kashtanova intended Zarya Holding a Postcard to “establish Zarya’s character and the setting for the story”).
[303] See id. (including the final version of the image of Zarya Holding a Postcard which appeared in Kashtanova’s graphic novel).
[304] See id. (providing the final version of the Zarya Holding a Postcard image, which includes several detailed elements including Zarya as the subject of the image holding a postcard in her hands with her eyes closed and a scenic lake and city skyline in the backdrop of the image).
[305] See id. (comparing the first version of Zarya Holding a Postcard, which only depicted dark-skinned hands holding a blank holding a blank postcard, with the final version that was ultimately included in the book, which included considerably more details).
[306] See id. at 8 (describing how the first input “while simple, contains the core creative input that went into this initial version of the ‘Zarya Holding a Postcard’ image”).
[307] See id. (explaining how “Kashtanova specified a subject (‘hands’), an object (‘a … photograph’), and descriptive context (‘dark skin,’ ‘holding,’ and ‘old.’)” and directed Midjourney to constrain the image to a 16:9 aspect ratio).
[308] See id. at 5 (providing the first version of Zarya Holding a Postcard generated by Midjourney which accurately depicted Kashtanova’s prompts directing it to include dark-skinned, old hands holding a photograph).
[309] See id. at 7–8 (illustrating how the first version of the Zarya Holding a Postcard that was generated by Midjourney was responsive to her inputs and how Midjourney generated four output images based upon Kashtanova’s inputs); see also Midjourney, supra note 9 (describing how Midjourney generates four image options based on the prompt provided by the user).
[310] See Kashtanova’s Letter, supra note 1, at 5 (comparing the first version of Zarya Holding a Postcard, which only depicted dark-skinned hands holding a blank holding a blank postcard, with the final version that included was ultimately included in the book, which included considerably more details).
[311] See Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884) (discussing how the plaintiff selected and arranged all the elements in the photograph, including the subject, accessories, and backdrop in the image); Kashtanova’s Letter, supra note 1, at 5–6 (portraying through ninety intermediate versions of Zarya Holding a Postcard that Kashtanova directed Midjourney to generate individual images for the individual elements that were to be incorporated in the final version of the image).
[312] See Kashtanova’s Letter, supra note 1, at 5–6 (illustrating that some of the intermediate versions of the Zarya Holding a Postcard image included only specific elements that later appeared in the final version of the image).
[313] See id. at 5 (providing the image in the figure, titling it “Screenshot #1 of intermediate versions of ‘Zarya Holding a Postcard’”).
[314] See id. at 6 (providing the image in the figure, titling it “Screenshot #2 of intermediate versions of ‘Zarya Holding a Postcard’”).
[315] See id. at 5–6 (providing the images in Figures 3 and 4, which depict versions of Zarya Holding a Postcard that only included the content inside the postcard that Kashtanova intended to incorporate in the final image).
[316] See id. at 6 (providing the image in Figure 4, which depicts versions of Zarya Holding a Postcard that included only the desolate city skyline surrounded by overgrown trees that Kashtanova intended to be in the backdrop of the final version of the image).
[317] See id. at 7 (“Different elements of the final image are created, developed, refined, and relocated [in the intermediate versions]. The final image includes multiple elements from different generations of intermediate images all brought together into a cohesive whole.”).
[318] See Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 55 (1884) (describing the plaintiff’s affirmative actions in selecting and arranging the elements in the photograph, including posing the subject and selecting and arranging all the elements in the backdrop of the photograph); see alsoKashtanova’s Letter, supra note 1, at 5–6 (illustrating how Kashtanova directed Midjourney to generate images that depicted Zarya’s appearance, the contents inside the postcard she held, and the city skyline that appeared in the background of the final image).
[319] See Andrien v. S. Ocean Cnty. Chamber of Com., 927 F.2d 132, 134 (3d Cir. 1991) (holding that an individual need not carry out the physical labor to be considered the author of the work).
[320] See id. (holding that a plaintiff who directed a printing employee to compile preexisting maps into one map was the author of the final work); see also Lindsay v. Wrecked & Abandoned Vessel R.M.S. Titanic, No. 97 Civ. 9248, 1999 U.S. Dist. LEXIS 15837, at *15–16 (S.D.N.Y. Oct. 12, 1999) (holding that a plaintiff who directed the filming of a documentary film but did not physically perform the filming could be considered the author of the film).
[321] See Andrien, 927 F.2d at 135 (describing that plaintiff had the mental conception of how he wanted the elements in the maps to appear and stating that “[p]oets, essayists, novelists, and the like may have copyrights even if they do not run the printing presses or process the photographic plates necessary to fix the writings into book form”); Lindsay, 1999 U.S. Dist. LEXIS 15837, at *15–16 (finding that when a plaintiff “alleges that he exercised such a high degree of control over a film operation,” and the final product “duplicates his conceptions and visions of what the film should look like,” the plaintiff can be the “author” of the work).
[322] See, e.g., Etrailer Corp. v. Onyx Enters., Int’l Corp., No. 4:17-CV-01284-AGF, 2018 U.S. Dist. LEXIS 19916, at *3, *8 (E.D. Mo. Feb. 7, 2018) (holding that the plaintiff was the author of the photographs in dispute even though he used Photoshop to create the images).
[323] See Burrow-Giles, 111 U.S. at 57–58 (holding that the photographer was the sole author of the photograph because authorship depends upon who conceives of the original idea for the expressive work).
[324] See Nimmer & Nimmer, supra note 144 (describing how the theories of authorship include sole authorship, joint authorship, and authorship under the work made for hire doctrine).
[325] See Kashtanova’s Letter, supra note 1, at 2–3 (arguing that Kashtanova should be considered the author of the images because in her iterative process, which “included multiple rounds of composition, selection, arrangement, cropping, and editing for each image in the Work,” was undertaken solely by her).
[326] See Burrow-Giles, 111 U.S. at 59–60 (holding that sole authorship is appropriate when one individual had the mental conception behind the work; Andrien, 927 F.2d at 134–45 (holding that the plaintiff was the sole author of the final maps even though he did not physically undertake the work himself because he exercised control over the elements that appeared in the work); Lindsay, 1999 U.S. Dist. LEXIS 15837, at *15–16 (holding that the director was the sole author of the work because the work was conceived by him alone and the individuals who exercised the work did not have a claim of authorship in the work).
[327] See Kashtanova’s Letter, supra note 1, at 3, 5–6 (illustrating how Kashtanova directed Midjourney to generate every element in the final image of Zarya Holding a Postcard and Midjourney functioned only an “assisting instrument” to execute her creative expressions).
[328] See 17 U.S.C. § 101 (defining a joint work as a work “prepared by two or more authors with the intention that their contributions be merged into inseparable or interdependent parts of a unitary whole”).
[329] See Aalmuhammed v. Lee, 202 F.3d 1227, 1234–35 (9th Cir. 2000) (holding that Aalmuhammed was not a co-author of the film because Warner Brothers and Spike Lee superintended the work and “control in many cases will be the most important factor” in determining whether an individual is an author).
[330] See id. (describing how Aalmuhammed was not a co-author because Lee had the authority to accept them); see also Kashtanova’s Letter, supra note 1, at 5–6 (providing the intermediate images generated by Midjourney for the Zarya Holding a Postcard image and illustrating that Kashtanova picked the individual elements that she sought to include in the final version of the image).
[331] See 17 U.S.C. § 101 (describing that a work made for hire is a work prepared by an employee within the scope of his employment or a work that falls within a specific class of works that is specially commissioned by another under an agreement between the parties that the work shall be considered a work for hire).
[332] See Cmty. For Creative Non-Violence v. Reid, 490 U.S. 730, 752 (1989) (holding that the plaintiff’s work was not a work made for hire because the plaintiff was an independent contractor with no relationship with CCNV akin to an employer-employee relationship).
[333] See Thaler v. Perlmutter, No. 22-1564, 2023 U.S. Dist. LEXIS 145823, at *20 (D.C. Dist. Ct. Aug. 18, 2023) (“Common law doctrines of property transfer cannot be implicated where no property right exists to transfer in the first instance. The work-for-hire provisions of the Copyright Act, too, presuppose that an interest exists to be claimed.”).
[334] See Naruto v. Slater, No. 15-cv-04324-WHO, 2016 U.S. Dist. LEXIS 11041, at *10 (N.D. Cal. Jan. 28, 2016), aff’d, 888 F.3d 418, 426 (9th Cir. 2018) (holding that a monkey cannot bring a suit of copyright infringement because non-human authors do not satisfy the authorship requirement under the Copyright Act); see also U.S. Copyright Off., supra note 203 (stating that copyright protection is only available to humans because a non-human entity cannot be considered an author).
[335] See Thaler, 2023 U.S. Dist. LEXIS 145823, at *15 (describing that courts are compelled to continue applying established doctrines of authorship).
[336] See Abbott & Rothman, supra note 228 (discussing the overprotection argument that the benefits of protecting AI works are reduced because no incentive exists for AI to create new works).
[337] See Wang, supra note 232, at 154, 161 (arguing that copyright protection should not be granted to AI-assisted works is not in the public interest because the traditional justifications offered for copyright protection do not apply in the same way to AI-assisted works).
[338] See e.g., Kashtanova’s Letter, supra note 1 (explaining the time-consuming and “painstaking” process Kashtanova undertook to create the images in Zarya of the Dawn).
[339] See id. at 3 (explaining how Zarya of the Dawn “took over a year from conception to creation” and that Kashtanova’s development of the images “took hours; finalizing each individual page took a day or more”).
[340] See 17 U.S.C. § 107 (providing limitations on exclusive rights granted to authors and allowing authors to use portions of copyrighted works in certain contexts without being at risk of liability for infringement); see also 17 U.S.C. § 103 (explaining that individuals can receive limited copyright protection for derivative works and compilations provided that the material has been used lawfully by the author).
[341] See, e.g., Washingtonian Pub. Co. v. Pearson, 306 U.S. 30, 36 (stating that copyright serves “to afford greater encouragement to the production of literary works of lasting benefits to the world”).
[342] See Twentieth Century Music Corp. v. Aiken, 422 U.S. 151, 156 (1975) (explaining that awarding copyright holders limited monopolies reflects a balance between how “creative work is to be encouraged and rewarded, but private motivation must ultimately serve the cause of promoting broad public availability of literature, music, and other arts”); Landes & Posner, supra note 121 (explaining that copyright protection helps ensure that authors continue to produce creative works because “[w]hile the cost of creating a work subject to copyright protection . . . is often high, the cost of reproducing the work, whether by the creator or by those to whom he has made it available, is often low”).
[343] See id. (“The immediate effect of our copyright law is to secure a fair return for an ‘author's' creative labor. But the ultimate aim is, by this incentive, to stimulate artistic creativity for the general public good.”); see, e.g., Kashtanova’s Letter, supra note 1, at 3 (explaining the lengthy process Kashtanova undertook to create the images in Zarya of the Dawn and describing how each image took Kashtanova several hours and hundreds of prompts to develop).
[344] See Is Artificial Intelligence Set to Become Art’s Next Medium?, supra note 103 (describing how the world-renowned auction house, Christie’s, sold an AI-assisted work for over $432,500); see also MarketResearch.BIZ, supra note 108 (describing how the art market size for AI-assisted works “is expected to be worth around USD 5,840 Mn by 2032 from USD 212 Mn in 2022, growing at a CAGR of 40.5% during the forecast period from 2023 to 2032.”).
[345] See Fox Film Corp. v. Doyal, 286 U.S. 123, 127 (1932) (explaining that the Constitutional rationale behind awarding authors copyright protection is to allow society to benefit through the enrichment of the arts).
[346] See U.S. Const. art. I, § 8, cl. 8 (providing that the intent behind the Copyright Clause is to promote progress in science and the arts).
[347] See Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 58 (1884) (providing that authorship is exemplified by an individual’s intellectual conception); see also Aalmuhammed v. Lee, 202 F.3d 1227, 1234 (9th Cir. 2000) (applying Burrow-Giles in the context of a joint work and stating that, the author is the one who “‘superintends’ the work by exercising control”).
[348] See USCO AI Guidance, supra note 15 (explaining that the USCO will only protect elements in AI-assisted works that were not generated by generative AI tools because “[i]f a work’s traditional elements of authorship were produced by a machine, the work lacks human authorship”). But see Burrow-Giles v. Sarony, 111 U.S. 53, 58 (1884) (defining that an author is the individual with the mental conception behind the work); Andrien v. S. Ocean Cnty. Chamber of Commerce, 927 F.2d 132, 134 (3d Cir. 1991) (holding that an individual need not carry out the physical labor of executing the work to be considered the author of the work); Lindsay v. Wrecked & Abandoned Vessel R.M.S. Titanic, No. No. 97 Civ. 9248, 1999 U.S. Dist. LEXIS 15837, at *15–16 (S.D.N.Y. 1999) (holding that a plaintiff who directed the filming of a documentary film but did not physically perform the filming could be considered the author of the film).
[349] See Senftleben & Buijtelaar, supra note 234 (arguing that analyzing whether AI-assisted works should be capable of receiving copyright protection requires weighing the benefits provided by the grant of protection, specifically “an incentive for the full development and dissemination of robot works” against any potential detriments); see also Twentieth Century Music Corp. v. Aiken, 422 U.S. 151, 156 (1975) (describing how awarding artists monopolies for their works serves the ultimate aim of copyright protection, which is to promote the “broad public availability of literature, music, and other arts”).