By Boluwatife Daniel-Adebayo

Artificial Intelligence (“AI”) and Machine learning are the next phase of technological advancement. It is thus important that AI is used to push boundaries and create more opportunities. However, with this role, there is a tendency for AI technology implementation to abuse the rights of others. This possibility has made it important that the implementation of AI technology be done ethically and legally. As such, the laws that bind humans should also bind AI.

On the 4th of June 2024, the Recording Industry Association of America (“RIAA/PLAINTIFF”) on behalf of the major record labels filed two suits against AI music generators: Suno and Udio. The arguments of the plaintiff in both cases centred around the protection of sound recordings that are generated by AI tools. Suno and Udio, being AI music generators, train their AI models to generate output that imitates the qualities of actual sound recordings. These AI models are trained with copies of copyrighted music and upon prompt, they produce closely similar copyrighted materials.

A popular example of the use of AI in music is seen in 2024 during the rap battle between Drake and Kendrick Lamar. In one of the diss tracks titled ‘Taylor Made Freestyle’ released by Drake, the Canadian-born rapper included AI-generated vocals of both Tupac Shakur and Snoop Dogg on the track. Another popular example happened the year earlier in 2023 when an AI song called ‘Heart on My Sleeve’ was released featuring Drake and The Weeknd. The entire song was created by AI and it gathered over 8.5 million views on TikTok and over 250,000 listens on Spotify.

The reality is that AI tools depend heavily on copyrighted input. In the case of Suno, the company itself alleged that the training data it uses are music files of reasonable quality that are accessible on the open Internet—abiding by paywalls, password protections, and the like—combined with similarly available text descriptions. This use of copyrighted input to train the AI is not transformative in nature but a way of stealing from creatives without reference or compensation. It can be likened to sampling without clearance.

AI start-ups, however, claim that this process is in line with the doctrine of fair use which is provided for in the Copyright Act. This doctrine allows for copies of a protected work to be made without permission from the copyright holder under certain considerations. This doctrine is used in furtherance of public interests over individual interests. For example, in the case of Authors Guild v. Google, Inc., the court held that it was fair use to copy all of the books in numerous university libraries in order to create a commercial, full-text searchable index of the assembled corpus.

Although useful, the fair use doctrine defence used by AI start-ups does not adequately support their usage of copyrighted works in training their AI models. This is because AI models not only train on these copyrighted works but also produce outputs from prompts with what they have been trained on. These outputs can be uploaded on streaming platforms and can generate income for an external third-party user. As such, if the AI models only stopped at training with the copyrighted works, the doctrine of fair use might have been a suitable defence—but not anymore.

In 2023, UMG told streaming platforms, including Spotify and Apple, to block artificial intelligence services from scraping melodies and lyrics from their copyrighted songs. Furthermore, these AI start-ups charge a fee for the usage of their models.

The onus thus lies in the hands of labels to adequately protect the rights of their artists. If AI start-ups insist that newly generated output does not infringe the copyright of the works, labels could look at expanding the personality rights of artists to include sound. Moreover, AI start-ups that have trained their AI models on copyrighted music should seek to co-operate with the artists and their labels to agree on a licensing framework. Music is an emotional, cultural expression and not just a pattern of sounds—and should be treated as such.

Perhaps more so than with many other technologies, there is both promise and peril with AI. If developed with the permission and participation of copyright owners, AI models can assist creatives in producing new and innovative music. But if used without regard for fundamental copyright laws, they threaten the entire existence of the music ecosystem.