“This is the last straw”: your next favorite song has been made by Artificial Intelligence

Your next favorite song could have been created with an AI. In recent months, new programs have been developed capable of learning and replicating the modulation and vocal intonation of any artist. Thus, users can make that ‘false’ voice generated by the application cover songs by other artists or sing a completely new letter. That possibility has led to a trend of experimentation that is becoming popular on platforms like TikTok, where Rihanna singing Beyonce’s “Cuff it” has gone viral; Ariana Grande covering Drake’s “Passionfruit”; and “Love Yourself” by Justin Bieber, reinterpreted by Kanye West. Creating songs with the voice of other artists has never been easier.

Frederic Font, a researcher in the Music Technology group at Pompeu Fabra University (UPF), warns that the potential is great and that, in the case of music, “it is still something very experimental.” The expert explains that “there is a part of AI research where the intention is not to completely replace the creation process, but to help or contribute to an artist thinking in a different way.” “From this point of view, it’s not so groundbreaking, it’s what has happened in the entire history of music as new elements have emerged,” he adds.

However, all this content is being generated without the consent of those whose vocal technique is plagiarized, something that worries both artists and record companies. Aware of this change, Universal Music Group has already asked the big streaming music platforms —including Spotify and Apple Music— not to allow AI companies to access their catalog without permission. “We will not hesitate to take action to defend our rights and those of our artists,” they threatened.

Font, however, points out that “in the field of music you cannot train with as much data as in other fields such as images or text, and then the quality is not as good.”

A week ago, a TikTok user posted “Heart on my sleeve,” a song created with AI to replicate the voices of Drake and The Weeknd. After accumulating more than seven million views, the song jumped to Spotify and Apple Music, where more than 600,000 people listened to it in less than 24 hours. Two days later the song could no longer be heard on any of those platforms or on YouTube, although it is finding other ways to stay in circulation.

It is unknown if the record companies would have asked to remove that song and what intentions are behind that mysterious account. Is it a ploy by Drake and Universal to get our attention or a copyright infringement by some opportunist? TikTok has not wanted to give details, but has recalled that its policies do not allow “content that violates the intellectual property rights of other people.” In a hectic week in the AI ​​/ music relationship, we have also witnessed the ‘resurrection’ of Oasis thanks to artificial intelligence. This false reunion of the Gallagher brothers has been baptized as AISIS [AI son las siglas en inglés de Artificial Intelligence]. Does AI pose a threat to artists and creators? «I don’t think so, you can assume that things are transformed. This will also depend on the attitude of the listener. The link between the listener and the artist goes beyond the music itself, the audio file itself, and what these systems do in the end is generate audio files”, answers Font. Universal’s request is part of an accentuated concern after the emergence of generative AI. The fear that the automation caused by tools like ChatGPT will affect more and more professions has led to a reaction from those who may be affected. Illustrators and graphic companies have already denounced image generators such as Stability AI, DevianArt or Midjourney for violating copyright. Record companies can follow that path. «The traditional concept of copyright has already been challenged in the digital age. These systems are one more step”, defends Font. “Perhaps what has to be questioned is the concept of copyright,” she adds.

The impact of AI on music goes beyond voice generators. Google has developed a language model, MusicLM, capable of generating music from text descriptions. Although the company has not published it due to the “risks associated with the possible misappropriation of creative content”, that possibility will present another challenge for the music industry. The first book that explored the idea that computers could help compose music was written in 1959. Evolution has been accelerating a possibility that many artists have preferred to see as an opportunity. This is the case with Arca, Holly Herndon, Toro y Moi or Ash Koosha, but David Bowie also did it —although in a more primitive phase—, who experimented with a random phrase generator. In the aforementioned cases, it is the artists who clone their voices with AI to innovate. The risk comes when it is a third party who appropriates it. More recently, DJ and producer David Guetta used this technology on one of his live shows to recreate the voice of rapper Eminem. The complex debate generated around it led the Frenchman to specify that he would not market this creation.

Faced with this enormous challenge, the music industry will have to adapt. Some platforms have been doing this for a long time. This is the case of Spotify, which is working on a series of AI tools that will allow users to compose synthetic music and remix music from various artists, but also Apple Music, which last year bought a startup expert in this technology. In the United Kingdom the authorities are even studying whether artificial music could have intellectual rights.