By Madeline McCabe
One of the significant issues related to the usage of artificial intelligence (AI) in the realm of copyrights is how companies train their generative AI models. Various AI-related lawsuits are occurring due to companies allegedly using copyrighted material to train AI models.
Unsurprisingly, potential copyright infringement in the training of AI models is a pressing issue in the music industry. In June 2024, 11 record companies, including Sony Music Entertainment, Universal Music Group Recordings, and Warner Records, filed two complaints: one against AI music generator Suno and the other against Udio.
The plaintiffs in both complaints allege that Suno’s and Udio’s AI programs were trained through copying and ingesting sound recordings into the AI models to generate outputs similar to the original sound recordings owned by the plaintiffs. According to the complaints, Udio and Suno did not deny or proffer any facts to undermine plaintiff’s direct accusations of copying the plaintiff’s sound recordings in pre-trial litigation. However, neither company has disclosed how the AI models are trained, with Udio claiming that the training data is “competitively sensitive” and contains trade secrets.
To help showcase that Udio and Suno used copyrighted works in training the AI models, the plaintiffs used a series of prompts that pinpoint a particular sound recording by referencing specific subject matter, genre, artist, instruments, vocal style. Both Suno’s and Udio’s services generated outputs that closely matched the targeted copyrighted recordings, reflecting the machine learning phenomenon known as “overfitting.” An AI model is overfitted when the model is too closely adapted to the trained data, creating difficulty for the model to generalize new data sets. One symptom of overfitting is when a model replicates portions of its training data.
For instance, one example provided in the Suno complaint described how plaintiffs generated 29 different outputs extremely similar to Chuck Berry’s “Johnny B. Goode” in Suno by using prompts like “1950s rock and roll, rhythm & blues, 12 bar blues, rockabilly, energetic male vocalist, singer guitarist,” along with snippets of the song’s lyrics. One generated song, “Deep down in Louisiana close to New Orle” almost exactly replicated the “Go, Johnny, go, go” chorus.
In the Udio lawsuit, the plaintiffs generated a dozen outputs similar to Mariah Carey’s “All I Want for Christmas Is You.” One of the outputs, also titled “All I Want For Christmas is You,” was generated with Mariah Carey lyrics and the prompt “mariahcarey, contemporary r&b, holiday, Grammy Award-winning American singer/songwriter, remarkable vocal range.” The output contains chords and a melody virtually identical to the original on every line except for “I don’t care about the presents underneath the Christmas tree.” Here is a link to listen to other AI outputs alleged to be trained on copyrighted recordings.
According to the complaints, both Udio and Suno have claimed that they are protected under the fair use defense, but the defendants themselves have not yet filed a response. Nonetheless, the plaintiffs argue that AI companies can’t claim the fair use defense. The plaintiffs state that the doctrine of fair use promotes human expression by permitting exceptions of using copyrighted works without permission, and here, synthetic music is not rooted in human expression or creativity. Whether or not these AI companies can use the fair use defense, and to what extent, is yet to be determined by courts. These future decisions, and potential legislation, will undoubtedly have a huge impact on how AI models are and will continue to be developed.
______________________________________________________________________
Sources:
- https://www.wired.com/story/ai-music-generators-suno-and-udio-sued-for-copyright-infringement/
- https://www.riaa.com/wp-content/uploads/2024/06/Suno-complaint-file-stamped20.pdf
- https://news.justia.com/record-companies-file-lawsuits-against-ai-music-generators/
TAGS:
RECENT POSTS