Last week, several major record companies filed copyright infringement lawsuits against the producers of two generative AI music applications, Suno and Udio, in U.S. courts. The record companies claim that the AI companies have copied numerous sound recordings owned by the record companies and produced outputs that are very similar to these recordings.
The record companies are demanding $150,000 in damages for each song allegedly infringed.
The lawsuits allege that Udio produced outputs with “striking similarities” to songs such as ABBA’s “Dancing Queen” and Mariah Carey’s “All I Want For Christmas Is You,” while Suno is accused of producing songs similar to James Brown’s “I Got You (I Feel Good)” and Chuck Berry’s “Johnny B. Goode.”
These lawsuits are not the first to challenge the rapidly growing generative AI industry. Visual artists have sued the producers of image-generating systems, and various newspapers have filed similar claims against OpenAI, the owner of ChatGPT. The outcome of these lawsuits may determine the future validity of generative AI products.
How do music-generating AIs work?
For those unfamiliar with such products, here’s how they operate: You write a text command, for example, “compose a female jazz song to overcome Monday morning blues.” You can also provide your own lyrics if you wish.
The application then generates an MP3 song consisting of a combination of vocals and instrumentation, which can be downloaded by the user.
To create the song, the AI is trained on a vast dataset. The lawsuits claim that this data consists of pre-existing sound recordings owned by various record companies and was copied without permission. These sound recordings are at the heart of the issue.
The outcome of the case is likely to depend on whether Suno and Udio’s use of these sound recordings falls under “fair use.”
In the U.S., fair use is a defense mechanism against copyright infringement. In Australia, there is a narrower “fair use” doctrine, which pertains to specific uses such as research and review.
How Will the Court Decide?
The court will evaluate the use of the record companies’ songs by Suno and Udio based on four factors:
- Purpose and Character of the Use
- Nature of the Original Copyrighted Work
- Amount and Substantiality of the Portion Used
- Effect of the Use on the Market Value
The most contentious factor is the purpose and character of the use. This involves examining whether the generative AI music is sufficiently “transformative,” meaning whether it provides new meaning, expression, or value to the original work.
Suno and Udio argue that their technologies are sufficiently transformative. They assert that their AIs generate new and original outputs rather than copying or reproducing pre-existing songs.
The court will assess the quantity and significance of the copied songs. Specifically, it will evaluate how the allegedly copied songs were used in the AI training process and output production.
The significance may be qualitative rather than quantitative. This means that, in addition to the amount copied, the court may consider whether a substantial portion of the song was copied.
Additionally, the effect of generative AI on the market value of the original sound recording will be considered. A usage that occupies the market position of the original song may be more significant. This point can be argued both ways.
What is Sound Cloning?
A major concern for the music industry is sound cloning. This refers to the ability of other generative AI music applications (not Suno or Udio) to clone the voice of a famous singer onto any song.
Suno stated on X that sound cloning is not possible with their applications because they do not allow users to reference specific singers. This issue may be discussed in court.
What Does the Future Hold?
Predicting the outcome is difficult. There may be a settlement before the trials, or new licensing arrangements could be developed, similar to OpenAI’s recent collaboration with News Corp.
One certainty is that new AI voice cloning innovations targeting monetization and licensing are being developed through start-up companies. An example is Hooky, an AI voice modeling licensing platform that offers artists control over the use of their voices.
If the record companies’ lawsuits proceed, U.S. courts will have the opportunity to clarify whether the training activities and outputs of generative AI music applications fall under fair use. This decision may set a precedent for other types of generative AI applications.
