The rapid deployment of musical “deepfakes’’ on social media illustrates just how quickly artificial intelligence (AI) is already reshaping the music industry. Deepfakes are a type of AI that can create convincing replicas of real people, whether in the image, video, or audio format. In early 2023, social media users began using this technology to create and share new versions of popular songs, sung by a different artist (for example, SZA’s ‘Kill Bill’ sung by Ariana Grande). These uses continued to proliferate, coming to a head in April, when a user going by Ghostwriter shared a song called ‘Heart on My Sleeve’, written using AI and sung in the voices of The Weeknd and Drake. The track went viral, prompting the artists’ label, Universal Music Group, to ask Spotify to remove AI-generated songs and bar AI from training on its catalogue.
While this tech may be relatively new, it is simply accelerating an ongoing trend: the consumerisation of music-making. The barrier to entry for making music has been lowering for years, but previous advancements mostly made it easier for aspiring artists to pursue music. AI will put music-making in the hands of the masses. Viewed this way, deepfakes were an inevitable next step for creator culture, and it is inevitable that consumer-created music will flood streaming platforms. There are already various deepfake generators; some have been shut down (Drayk.it, specifically for generating Drake songs) while others, like covers.ai, remain online at time of publication. What the music industry does (and does not do) in response to this will have profound implications.
Music rightsholders will do their best to apply existing copyright law, but it was not designed for AI. It will be impossible (or at least, a fool’s errand) to play the whack-a-mole game of pursuing thousands of plagiarism and copyright infringement cases. The most scalable solution for music rightsholders will be to fix the problem at the top, by ensuring that generative-AI tools only learn from what they have permission to learn from. There is an entirely new licensing opportunity in pre-authorised datasets. In doing so, major labels can still benefit from this shift, but it will take a collaborative approach rather than one of enforcement.