By Sean Carlin
@redlineproject
AI disclosure: Generative AI was used to outline and create an annotated bibliography for purposes of organization, along with creation of related media. Models utilized for organization, editing and summarization include Chat-GPT 4o and Grammarly. Models utilized for artwork were Chat-GPT 4o for prompt generation, MidJourney for image creation, and Dream Machine for video creation.
Generative AI creates content using extensive datasets, often including copyrighted materials, without permission or compensation to the original creators. Current U.S. copyright law does not recognize AI-generated content as copyrightable because it requires human authorship, an outdated standard that does not consider AI’s modern capabilities to synthesize ideas from past works. This exclusion raises questions about who should own or control AI-generated work — the AI user or the company.
Economically, AI challenges traditional compensation models, as its use often undermines the value of human-created work. Some suggest adapting copyright laws to allow artists fair compensation for their contributions to AI training or transferring rights to the users or companies leveraging AI. Ultimately, the U.S. copyright system needs reform to address the ethical and economic implications of AI, ensuring fair protection for creators while accommodating AI’s transformative role.
AI has been widely adopted within the last decade, though its origins can be traced to the 1940s. It has been in the public consciousness as a concept for even longer due to various fictional explorations of machinery mimicking or surpassing human intelligence. Generative AI, however, has been a more recent development that is a manifestation of those long-seeded ideas.
Furthermore, Generative AI allows for text, audio, speech, images, or videos to be created without much human intervention beyond a prompt provided by an individual user. These AI models are often trained on vast swaths of data and information regarding their specific use case, often including copyrighted works like essays, blogs, photos and videos. This usage is rarely known by the artists whose work is used to train these models, making most of their usage a copyright violation without adequate compensation. To further complicate the copyright usage and claimant, the current United States Copyright Office does not recognize AI-generated works as work that can be copyrighted.
With these complications, a dilemma emerges regarding how copyright and compensation should be handled within the US copyright system. Within the laws governing personal property and works of art, this usage without attribution, compensation, and/or licensure constitutes what’s known as a copyright violation. The U.S. Copyright Office states that “copyright infringement occurs when a copyrighted work is reproduced, distributed, performed, publicly displayed, or made into a derivative work without the permission of the copyright owner.” Artificial intelligence falls into a grey area where the works are not directly reproduced unless achieved through skilled prompt engineering by users of the AI models. Instead, AI text, image, and video creators often reproduce the creative work necessary to bring a new idea to fruition. Instead of taking the works directly, AI is trained to spin the works into a new idea.
With these points under consideration, the following will be an exploration of why the US copyright system in its current form is ill-equipped to handle the influx of new Generative AI works and the compensation issues that arise from training data. With the wide adoption of generative AI models like ChatGPT, Midjourney, and Dall-E, this paper seeks to provide answers and explore the possibilities of how the US Copyright system can be adapted for the best interest of all parties involved with the Generative AI boom.
AI and the Current Copyright Framework
The current framework surrounding the copyright of Generative AI works is currently outlined by the US Copyright Office’s Human Authorship Requirement. The Human Authorship Requirement plainly states that “copyright can protect only material that is the product of human creativity. Most fundamentally, the term ‘author,’ which is used in both the Constitution and the Copyright Act, excludes non-humans.” This policy is not only explicitly exclusive of any work that AI may have generated but also fails to consider in-between cases where the ratio of human work outnumbers the ratio of Generative AI work. To further complicate matters, the framework upon which this ruling is laid out is outdated and doesn’t reflect modern knowledge of computational ability or computers as a whole. One of the primary legal benchmarks used by the US Copyright Office is two legal cases from 1884 and 1879, making both precedents over a century old to modern standards.
The crucial point of contention amongst these points is that the precedent fails to consider when creation can be a resynthesis of old ideas and concepts rather than pure inspiration of the human mind. Generative AI, when broken down, mimics the human creative process through reference points in training data, similar to how lived experiences and memories influence the human creative process. However, with the current markup of US Copyright law, the expression of an idea is protected, rather than the idea itself. On this principle, no copyright will ever be administered to Generative AI works as they’re only synthesizing the idea rather than the expression of a specific artist. However, this does not account for when artists seek to use Generative AI for artistic expression, and they are denied the ability to claim that expression as their own.
One of the main tests for the US Copyright Office is that it is an author’s original work, and contributions are beyond merely trivial and recognizably one’s own work. Generative AI has a hard bar to pass with this idea as it cannot mimic true inspiration and originality within a work. Generative AI models are often built off large quantities of training data that cannot adequately reflect the ideas of human inspiration. Instead, these data sets utilize previously created, usually already copyrighted works.
Lastly, there is much industry disagreement about how Generative AI can be best attributed to the US Copyright system. In a study by Hristov, an AI scholar at the University of Science and Technology of China, found that in a survey of fifty-seven industry experts, scholars, and copyright specialists, there was wide disagreement about how AI should be attributed to modern copyright law. This divided nature within industry professionals results in a more prominent call for reform within the contemporary understanding on copyright attribution.
Challenges of Authorship and Ownership in AI
Challenges arise when the source of inspiration and re-creation are questioned when, as mentioned previously, Generative AI models are trained on previously created human work. These data sets are often encompassing millions of photos, videos, and text samples being utilized to train generative AI on what a desired output may look like. This frequently starves creatives of income they otherwise may be entitled to, as usage of their work without credit, attribution, or compensation is in violation of current US Copyright standards.
With current rules and regulations, one of the few copyright attributions available to Generative AI is the code base for the AI itself. This allows companies creating Generative AI to copyright their code as a literary work, as code is an acceptable format for copyright under US law. This allows for AI as a concept to be protected and for the creation of such programs, their variations, and concepts to be copyrighted with full rights held to the company responsible for their creation. However, given the randomness seen throughout Generative AI results because of the code necessary to build them, current copyright law sees work created by Generative AI fall within the public domain.
Another primary consideration if copyright were to be adapted to reflect more modern sensibilities, would the works of Generative AI be attributed to the individual user that prompted the AI for the work, or would it be attributed to the AI algorithm itself? As stated earlier, the copyright applies to the expression of the idea rather than the idea itself, which would natively mean that the attribution should go to the individual prompting the Generative AI for its creative output. Accounting for the idea that most Generative AI programs are repurposing old creative works to generate new ones, it also calls into question how an AI can hold copyright if the expression behind the piece has no lived experience or influence behind it. Plainly stated, given that Generative AI has no lived experience and, therefore, is not human, it will never receive copyright protection under US law unless it has a human intervening on its behalf to do most, if not all, of the work. With an AI unable to hold copyright protection for its work due to lack of humanity, the next reasonable rights holder for Generative AI works is the users utilizing the tools given by Generative AI.
The Economic and Ethical Implications of AI in Copyright
One of the primary concerns that has always been raised about the usage of Generative AI is its reliance on pre-made copyrighted works generated by humans. These works vary in nature but are still utilized nonetheless to make new features for Generative AI to utilize. A primary example of these concepts is in the creation of AI-generated music, where the basis of these works is the efforts of teams of individuals who have created pieces of music through hours of personal labor and expensive amounts of equipment. Generative AI can sidestep those ideas by creating a quick piece of music in the tone, genre, and mood the prompt user desires and have a finalized product within moments of submitting the prompt to the given AI.
There is pushback against this idea as people in entire industries have specialized knowledge of the equipment and skills necessary to create music to fit specific uses. Their argument lies on the idea that their effort and creativity in creation are undermined through the usage of these Generative AI models, and that argument extends beyond just the music industry and can be applied to nearly any creative industry where the personal touch is needed for the creation of a piece. Plainly stated, the higher the skill necessary in the work, the greater the compensation received, but what AI can do with ease greatly undermines the effort put forth by artists.
Additionally, it is worth mentioning that the music industry doesn’t follow the typical standards of copyright as other industries. The music industry specifically follows the Recording Industry Association of America’s (RIAA) policies on licensing and distribution. While in a similar manner to the US Copyright Office, they handle most of the licensing necessary for individual artists so their work can be adequately registered. One popular myth surrounding musical copyright is that the song Happy Birthday still holds its copyright to this day due to defense by RIAA and Warner Chappell Music, even after the artist’s death in 1946. However, that is no longer the case, as the song was deemed in the public domain in 2015. The public domain is the body of works that anyone can use freely without attribution or compensation to the original artist, as their original copyright claim can no longer be held. In the context of training Generative AI models, these works are free to use without issue or concern.
With the usage of copyrighted pieces to train Generative AI, there is often a backlash in ideas that the usage of previously copyrighted works goes uncompensated. However, a primary issue with trying to seek compensation from these models and their creators is the fact that the data and information taken are often in such wide swaths that it’s hard to directly pay a given user for their contribution to the training of the generative AI model. At times, a specific artist’s style or creation can be recreated, with it being used for good or nefarious reasons. Bad actors can utilize generative AI for misinformation but would also often corrupt an original artist’s work in the process. While not in direct conflict with the copyright of an original artist’s work, it can still detriment an artist’s work if they know their creations are feeding into the training of potentially hostile algorithms. However, even if consenting to its usage, an artist wouldn’t be able to withdraw their work’s usage within a Generative AI model as their work is un-removable memory within the basis of a Generative AI model.
Future of AI and Copyright
Copyright policy has been changed before and can be changed again to bring about this new era of Generative AI. In 1974, a commission was formed by the US Congress that was known as the Commission of New Technological Uses of Copyrighted Works (CONTU). In its consideration, it concluded that there were no major alterations needed to change US Copyright law as computers were tools to assist human creators. However, a more modern set of recommendations should be made regarding the new iteration of technology we face with Generative AI. AI Scholar Kalin Hristov has another set of recommendations that better encompass new ideas, like acknowledging the new creative capabilities of computers and keeping authorship strictly to humans but introducing new interpretations of employer and employee relationships within the legal framework. These allow for ownership rights to be transferred, rather than the idea of pure authorship, where the Generative AI model can be “employed” for its given task and have copyright transferred to the employer at the end of its assigned work.
Additionally, it is recommended that AI models be created to distribute royalties or compensation for the usage of copyrighted works within a Generative AI model’s training data. These models would be able to sort through the vast quantities of data used better than having an individual try to backtrack and assess royalties owed to a respective artist out of millions. When properly compensated, concerns about attribution and copyright usage are mitigated as the copyright system protects the intellectual property of an individual artist or corporation rather than a Generative AI model.
In conclusion, the current system of copyright within the US is in need of drastic changes to address AI concerns regarding creativity and attribution of work adequately. While these changes may not be adopted overnight, the rate of change with AI is currently outpacing the legal system set up to safeguard it and its users. With adequate consideration and change, an equitable system can be adopted where Generative AI companies can protect their models and their usage while adequately compensating the artists whose work was used to help in their development.
Editor’s note: Students in Mike Reilley’s AI Journalism (COMM 294) Fall 2024 undergraduate course experimented with AI storytelling tools to create these stories, following the AI use guidelines on The Red Line Project’s Principles page.