Following the dawn of generative AI writers such as ChatGPT, additional developments in AI are making headway. Meta’s AI music generator, MusicGen, has struck a chord with audiences worldwide and people are tuning in to the advances of generative AI.
The speed of AI innovation has stirred up legal issues that call for a re-examination of our current legal frameworks. In the last year alone, several copyright-related cases involving generative AI have surfaced. In this article, I will discuss key legal aspects surrounding the application of copyright law to generative AI.
Understanding How an AI Music Generator Works
Before we dive into the legalities, let’s get a grasp on the technology we are discussing. An AI music generator uses machine learning algorithms to create original music compositions.
An AI music generator learns patterns, rhythms, harmonies, and melodies by analyzing a dataset of existing musical compositions. According to Meta’s GitHub page, MusicGen has been trained with 20,000 hours of music composed of 10,000 tracks and data from music libraries at ShutterStock and Pond5.
MusicGen utilizes a language model that requires a user to input text to generate music. By writing a description of music to be generated, its algorithm produces a 15-second clip of a song based on the text. You can demo the capabilities of MusicGen here.
AI Music Generators and Copyright Laws
Whether AI music generators infringe on copyrights remains unanswered by any U.S. court. A review of a few basic copyright laws will help identify and evaluate potential issues.
U.S. Copyright Laws
Under 17 U.S.C. § 106, copyright owners possess exclusive rights to their original work. These rights extend to reproducing the work, creating derivative works, distributing copies, and public performance, among others. If someone else uses the copyrighted work without permission, they could potentially face a copyright infringement claim.
In the context of an AI music generator, potential copyright infringement could arise if the AI produces a song that is substantially similar to a copyrighted song used in its training without permission. Recent legal cases, such as comedian Sarah Silverman’s class action lawsuit against Meta and OpenAI, highlight this issue. While this case is centered on language models, the same principles apply to an AI music generator that may use copyrighted music recordings for training.
Potential Copyright Claims Involving AI Music Generators
When an AI music generator creates a song that is substantially similar to a copyrighted piece used in its training, a copyright infringement claim might be plausible if the AI creator didn’t secure the copyright holder’s permission.
To illustrate, a recent class action lawsuit filed by comedian Silver Silverman against Meta and OpenAI involves that issue. Silverman alleges that the companies used her book, “The Bedwetter,” without permission to train their language models, ChatGPT and LLaMA.
According to the lawsuit, Silverman’s book and others were downloaded from “shadow libraries,” which are online repositories that make books available without the authors’ consent. The complaint further alleges that these AI language models generate derivative works that infringe upon the original copyrighted books.
While Silverman’s lawsuit pertains to text rather than music, the principles remain the same. An AI music generator, like its text-generating counterparts, requires pre-existing data for training that potentially includes copyrighted music.
Meta claims MusicGen was trained using licensed music, suggesting they have permission to use the tracks. However, there are no details as to the permissions granted by the licenses.
There’s a question as to whether Meta’s licenses merely allow them to possess copies of each song for private use, enabling the training of MusicGen. It’s worth noting that buying a song or album for private use doesn’t automatically include a license to create derivative works. So, if Meta’s licenses don’t allow for the creation of such works, there’s a chance they could infringe on music copyrights if their AI music generator creates songs remarkably similar to copyrighted work.
Defenses to Copyright Infringement by an AI Music Generator
Two possible defenses against copyright infringement in the context of AI music generators are fair use and independent creation.
Fair Use
The fair use defense is evaluated using the four following factors:
- The purpose and character of your use, including whether the use is of a commercial nature or is for nonprofit educational purposes or if the alleged infringing work is transformative in nature;
- The nature of the copyrighted work;
- The amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
- The effect of the use upon the potential market for or value of the copyrighted work.
If an AI-generated song is deemed transformative—meaning it presents a new expression, meaning, or message—it might be entitled to a fair use defense. The applicability of the defense will depend on the characteristics of the generated music. In comparison to the original copyrighted song, the AI generated music is transformative if it can be considered a new expression, meaning, or message to the original work.
Independent Creation
Another potential defense that may be raised in a copyright infringement claim against AI generated music is independent creation. To successfully establish the defense, the alleged copyright infringer must demonstrate the infringing work was created independently and without any knowledge or access to the copyrighted material.
The independent creation defense asserts that any similarities between works are coincidental, not the result of copying. However, proving independent creation may be challenging if the AI was trained on the copyrighted material.
Who Owns the Music Created by an AI Music Generator?
Ownership of copyrights in the context work created by generative AI was addressed this year. The U.S. Copyright Office canceled a copyright registration for AI-created digital art based on Section 313.2 of its Compendium of Compendium of U.S. Copyright Office Practices that copyrightable work “must be created by a human being” and “[w]orks that do not satisfy this requirement are not copyrightable.”
Another denial of copyright registration for AI-generated work was challenged in federal court by Stephen Thayler, an AI system inventor. However, the federal trial court upheld the Copyright Office’s decision to deny copyright registration, and the federal appellate court affirmed the lower court’s ruling. As a final effort, Thayler sought review by the United States Supreme Court, which denied Thayler’s petition on April 24, 2023.
The disputes involving copyright registration of AI-generated work led the U.S. Copyright Office to offer guidance on AI-generated work.
Based on the United States Supreme Court’s refusal to consider Stephen Thayler’s petition, the U.S. Copyright Office has the right to deny copyright registration for work generated solely by AI. Unless Congress decides to change U.S Copyright laws, work created by an AI music generator, and any other generative AI, is not entitled to copyright protection.
Conclusion
The issue involving copyright ownership of AI-generated work was resolved this year. Based on the current state of U.S. copyright laws, no one is entitled to a copyright for music created by an AI music generator. Whether an AI music generator or use of the technology constitutes copyright infringement has yet to be determined by any U.S. Courts. As the AI technology advances, the updates to the law will follow.